Datasets:

Hendrik-a

/

arxiv

like 0

we consider a charged scalar field in a-dimensional de sitter spacetime and investigate pair creation by a schwinger mechanism in a constant electric field background . using a semiclassical approximation the current of the created pairs has been estimated . we find that the semiclassical current of the created pairs in the strong electric field limit responds as . going further but restricting to dimensional de sitter spacetime , the quantum expectation value of the spacelike component of the induced current has been computed in the in - vacuum state by applying an adiabatic subtraction scheme . we find that , in the strong electric field limit , the current responds as . in the weak electric field limit the current has a linear response in and an inverse dependence on the mass of the scalar field . in the case of a massless scalar field , the current varies with which leads to a phenomenon of infrared hyperconductivity . a new relation between infrared hyperconductivity , tachyons , and conformality is discussed , and a scheme to avoid an infrared hyperconductivity regime is proposed . in dimension , we eventually presented some first estimates of the backreaction of the schwinger pairs to the gravitational field , and we find a decrease of the hubble constant due to the pair creation .

1602.06556

the purpose of this paper is to look at schwinger pair creation in @xmath6 dimensional de sitter spacetime ( @xmath7 ) , with special emphasis on the case of @xmath2 dimension . the schwinger effect , i.e. , pair production by a strong electric field , is a nonperturbative effect of quantum field theory in flat spacetime which was initially discovered in the pioneers works @xcite ; for a recent review see , e.g. , @xcite . despite tremendous efforts on the experimental point of view , it has never been detected so far . the main reason is that the number of pairs created is exponentially damped before a critical value for the electric field @xmath8 @xcite . new laser facilities @xcite are planned to be operational in the next ten years and might approach this critical electric field . in the meantime , another proposal is to change the perspective : whereas all the experiments aiming at detecting the schwinger effect are conducted on earth , one could look for the schwinger effect in astrophysical and cosmological systems . the review @xcite describes some of these ideas , and the stress is on the backreaction of the created pairs and its application to astrophysics . during inflationary magnetogenesis strong electric fields are also produced @xcite , which provides motivation for considering schwinger pair production in this context . furthermore , pairs can also be created by gravitational fields , e.g. , in ds @xcite and quasi - ds @xcite , which is sometimes referred to as the cosmological schwinger effect @xcite . these effects can also be essential for the interaction and balance between matter - field and cosmological constant in the universe evolution @xcite . recently , the combination of electrical and gravitational pair creation was studied in depth for various types of particles and spacetime dimension @xcite . in @xcite and @xcite , the authors computed the schwinger effect for a charged scalar test field in @xmath9 and @xmath10 , respectively . in @xcite and @xcite , the generalization to @xmath9 and @xmath10 , respectively , for fermionic particles was performed aiming at checking if the known equivalence in flat spacetime between boson and fermion for a constant electric field still holds . the answer was that there was a difference between boson and fermion . to see that , it was necessary to compute the induced current which turns out , as also noted in @xcite , to be the right quantity to describe the schwinger effect in curved spacetime . indeed , it is not plagued by the need of the notion of particle in the adiabatic future which allows one to explore a broader parameter space . but to cure infinities arising from momentum integration , this current needs to be renormalized . the adiabatic subtraction is the most used method . the pauli - villars method was implemented in @xcite and can be shown to agree with the adiabatic subtraction . in @xcite , the point - splitting method was shown to agree with the adiabatic subtraction in @xmath10 for the boson . in @xcite , an adiabatic subtraction method was used to regularize the current . to further explore the validity of the use of the adiabatic subtraction , it is legitimate to look at @xmath11 and see which kinds of divergences arise and how they are cured . in @xcite , the schwinger effect in @xmath11 has been investigated as an example of odd dimensional ds , and it has been shown that no particle production occurs in odd dimensional global ds . however , in our study on the poincar patch together with an electric field , we report particle production . one more motivation to look in depth at an odd dimensional quantum field theory is that those theories sometimes exhibit strange behaviors . for instance , dimensional regularization shows no one - loop ultraviolet divergences because it only registers logarithmic divergences and all the divergences are power law in an odd dimension . for @xmath2 in curved spacetime with an electric field , we will , however , report a linear divergence arising from the electromagnetic side of the theory . regarding infrared phenomena , we will also report a different behavior than in an even dimension which is also the case in flat spacetime . our work aims at completing the picture of the schwinger effect in @xmath11 by computing the induced current . all the works so far describing pair creation in @xmath12 under the influence of a strong electric field @xcite assumed two backgrounds , i.e. , electromagnetic and gravitational . in @xcite and @xcite , the backreaction effects due to the created schwinger pairs on the electromagnetic field background have been investigated , while in this case , there is no report on the backreaction to the gravity sector . in order to investigate the backreaction to the gravitational background one needs the energy - momentum tensor of the created particles . in the absence of the electromagnetic background , i.e. , in the presence of a purely gravitational field , the energy - momentum tensor of the created scalar @xcite and fermion @xcite particles in a spatially flat friedmann - lemaitre - robertson - walker universe ( considering @xmath10 as a special case ) have been computed . in @xcite , the author examined backreaction to the gravitational field and found that the particle creation leads to a decrease of the cosmological constant , whereas in @xcite , the authors mainly developed the renormalization theory . in the last part of this article , we will present some preliminary results on backreaction to the gravitational field , namely , an estimate of the variation of the hubble constant due to the influence of the gravitational field . we hope to come back to this issue in the future by solving fully the einstein equation . the article is organized as follows : in sec . [ sec : ds ] , solutions of the klein - gordon equation are obtained . the pair creation is examined in sec . [ sec : creat ] . in sec . [ sec : curent ] the quantum vacuum expectation value of induced current is investigated . section [ sec : irhc ] is devoted to a discussion about the phenomenon of infrared hyperconductivity . in sec . [ sec : gravity ] , we present some first results on backreaction of the created pairs in @xmath0 dimension to the gravity sector of theory . eventually , in sec . [ sec : concl ] , we give some concluding remarks . for the sake of clarity , we relegated some of the technical calculations to the appendixes . in appendix [ app : sem ] , an alternative method to derive the main result of sec . [ sec : creat ] is given . in appendix [ app : math ] , some useful properties of mathematical functions are represented . in appendix [ app : int ] the computation of the current integral is reviewed . to study the vacuum expectation value of the current operator of a charged scalar field coupled to a constant electric field background in a @xmath7 , the field operator is needed . the field operator contains mode functions which are obtained by solving the klein - gordon equation . the @xmath7 metric can be read from the line element in the half of @xmath7 manifold @xmath13 it corresponds to the line element of the flat robertson - walker universe with its scale factor @xmath14 , @xmath15 is the proper time , and @xmath16 is the hubble constant . in terms of the conformal time @xmath17 the line element ( [ line ] ) takes the form @xmath18 where @xmath19 revealing that this portion of ds is conformal to a portion of minkowski spacetime . in order to obtain solutions of the klein - gordon equation in the presence of a constant electric field background in @xmath7 , we consider the action of scalar qed , @xmath20 where @xmath21 is a complex scalar field with mass @xmath22 and electrical charge @xmath23 . the @xmath7 metric @xmath24 reads from eq . ( [ metric ] ) , @xmath25 is the absolute value of its determinant , @xmath26 is the scalar curvature , and @xmath27 is the dimensionless conformal coupling . the introduction of the conformal coupling is done to make the theory more general and arises naturally in string inflation framework @xcite . the vector potential describing a constant electric field background is @xmath28 where @xmath4 is constant . our convention for indices is that we label the spatial dimension with arabic - persian numerals , e.g. , @xmath29 , and that letters , e.g. , @xmath30 , are reserved for fourier space . then , the only nonzero components of the electromagnetic field strength tensor are @xmath31 we derive the equation of motion for the scalar field @xmath32 by using the euler - lagrange equation for the lagrangian coming from the action ( [ action ] ) . we then obtain the klein - gordon equation @xmath33 where we defined @xmath34 after substituting explicit expressions of the @xmath7 metric and the vector potential given in eqs . ( [ metric ] ) and ( [ vector ] ) , respectively , eq . ( [ kgeq ] ) takes the form @xmath35\varphi(x)=0.\ ] ] if we define @xmath36 it can be shown that eq . ( [ phieq ] ) leads to @xmath37\tilde{\varphi}(x)=0.\ ] ] based on the invariance of eq . ( [ vphieq ] ) under translations along the spatial directions , let @xmath38 where the superscript @xmath39 denotes the positive and negative frequency solutions , respectively . substituting ( [ fpm ] ) into eq . ( [ vphieq ] ) leads to @xmath40 where the variables @xmath41 and @xmath42 are defined by @xmath43 with @xmath44 . in terms of dimensionless parameters @xmath45 the coefficients @xmath46 and @xmath47 read @xmath48 in secs . [ sec : ds ] and [ sec : creat ] of this article we consider that @xmath49 , and then the coefficient @xmath47 becomes purely imaginary ; in this case , we use the convention @xmath50 . equation ( [ feq ] ) is the whittaker differential equation , and its most general solution in terms of the conventions of @xcite can be written as @xmath51 where @xmath52 and @xmath53 are arbitrary constant coefficients . in view of eqs . ( [ varphi ] ) , ( [ fpm ] ) , and ( [ general ] ) the corresponding solutions of eq . ( [ phieq ] ) for positive and negative frequency solutions are @xmath54 where the choice of the sign for the @xmath47 parameter will follow without ambiguity from the discussion in sec . [ sec : mode ] . we need mode functions that determine the creation and annihilation operators and hence the vacuum state of the quantum field theory . this vacuum will be determined by specifying the asymptotic form of the mode functions @xcite . in order to determine the mode functions at early times , which is approached as @xmath55 , we impose that the functions @xmath56 , given by eq . ( [ general ] ) , asymptotically take the form @xmath57 as @xmath58 . a comparison with the minkowski spacetime mode functions shows that the functions @xmath59 and @xmath60 are positive and negative frequency mode functions , respectively . by the virtue of asymptotically expansions of the whittaker functions @xmath61 and @xmath62 as @xmath63 [ see eqs . ( [ win ] ) and ( [ min ] ) , respectively ] , the normalized positive and negative frequency mode functions are @xcite , respectively , @xmath64 a similar discussion is possible in the asymptotic future ( @xmath65 ) . the desired asymptotic form is @xmath66 , leading with eqs . ( [ mout ] ) and ( [ wout ] ) to the mode functions @xcite @xmath67 the subscripts @xmath68 denote that these mode functions have the desired asymptotic form at early / late times , and the corresponding vacuum state is referred to as the in vacuum and out vacuum , respectively . since the orthonormality of the mode functions should be independent of time , there exists a conserved scalar product . between two scalar functions @xmath69 and @xmath70 it is defined in @xmath6 dimension by @xmath71 where the integral is taken over a constant @xmath72 hypersurface @xcite . if @xmath69 and @xmath70 are solutions of the field equation ( [ kgeq ] ) which vanish at spacial infinity , then @xmath73 is conserved @xcite . the mode functions ( [ uin])-([vout ] ) will be orthonormal with respect to the scalar product ( [ produc ] ) integrated over a constant @xmath74 hypersurface . then , subsequent orthonormality relations are satisfied @xmath75 in sec . [ sec : ds ] , two complete sets of orthonormal mode functions were obtained , i.e. , @xmath76 given by eqs . ( [ uin ] ) and ( [ vin ] ) and @xmath77 given by eqs . ( [ uout ] ) and ( [ vout ] ) . in this section , we will derive a first result describing the schwinger pair creation rate : the decay rate that we will derive via a bogoliubov transformation ; see also @xcite . analogous methods were used to compute the pair creation rate in time - dependent field in minkowski spacetime @xcite and without an electric field for bosons in ds in @xcite . in @xcite the connection of this bogoliubov technique to kinetic theory was shown . we will then compute the semiclassical conduction current . the conduction current will be computed in the full generality in sec . [ sec : curent ] , and a comparison with the semiclassical expression will be performed . the scalar field operator @xmath78 may be expanded in terms of the @xmath76 set in the form @xmath79,\ ] ] where @xmath80 annihilates particles described by the mode function @xmath81 , and @xmath82 creates antiparticles described by the mode function @xmath83 . the quantization of the theory is implemented by adopting the commutation relations @xmath84=\big[b_{in\k},b_{in\k'}^{\dag}\big]=(2\pi)^{d}\delta^{(d)}(\k-\k').\ ] ] the vacuum state is defined as @xmath85 and then the construction of the fock space can be done similarly to the minkowski spacetime case . however , there is no @xmath86 killing vector to define positive frequency mode functions , and consequently a unique mode decomposition of the scalar field operator @xmath78 does not exist . therefore , @xmath78 may be expanded in terms of a second complete set of orthonormal mode functions in the form @xmath87,\ ] ] where @xmath88 annihilates particles described by the mode function @xmath89 , and @xmath90 creates antiparticles described by the mode function @xmath91 . in this case , the commutation relations are @xmath92=\big[b_{out\k},b_{out\k'}^{\dag}\big]=(2\pi)^{d}\delta^{(d)}(\k-\k').\ ] ] the decomposition of @xmath78 in eq . ( [ phiout ] ) defines a new vacuum state @xmath93 and a new fock space . since both sets are complete , the orthonormal mode functions @xmath89 can be expanded in terms of the first complete set of orthonormal mode functions . hence @xmath94.\ ] ] by virtue of the orthonormality relations ( [ orthonorm ] ) the bogoliubov coefficients @xmath95 and @xmath96 will be determined by @xmath97 where the bogoliubov coefficients satisfy the relations @xmath98 & = ( 2\pi)^{d}\delta^{(d)}\big(\k'-\k''\big ) , \nn\\ \int\frac{d^{d}k}{(2\pi)^{d}}\big[\alpha^{\ast}_{\k,\k'}\beta_{\k,\k''}-\beta_{\k,\k'}\alpha^{\ast}_{\k,\k''}\big]&=0.\end{aligned}\ ] ] as a consequence of eqs . ( [ phiin ] ) , ( [ phiout ] ) , and ( [ bogolu ] ) the late time annihilation operator @xmath88 is related to the early time annihilation operator @xmath80 by a bogoliubov transformation @xmath99.\ ] ] using @xmath88 and the vacuum state @xmath100 we can calculate the expectation value of the particle number operator . ] @xmath101 therefore , if @xmath102 , then particles are created . in order to obtain the density of the created pairs an explicit expression of the bogoliubov coefficients is needed . to identify the bogoliubov coefficients , the orthonormal mode functions , given by eqs . ( [ uin])-([vout ] ) , should be substituted into eq . ( [ bogolab ] ) . we then obtain @xmath103 and the normalization condition @xmath104 is satisfied . the expected number of the created pairs , with a given comoving momentum @xmath105 , in the in vacuum is given by eq . ( [ nout ] ) . after a short calculation , eq . ( [ beta ] ) results in @xmath106 volume of @xmath7 in a box with dimensions @xmath107 . then , the number of created pairs per comoving @xmath108 volume , with given comoving momentum @xmath105 is @xmath109 using the mathematical formulas ( [ formul ] ) , ( [ area ] ) , and ( [ areaint ] ) , the number of created pairs per unit @xmath108 volume is @xmath110 where @xmath111 is the modified bessel function ; see appendix [ app : bessel ] . this integral is not finite , since it takes into account the total number of created pairs from the infinite past to the infinite future . however , the number of created pairs per unit of time is finite . thus , we convert the @xmath112 integral in eq . ( [ intk ] ) into a @xmath74 integral . to have physically acceptable particles states , one needs to have particle states well defined in the asymptotic past and future ; that is , the background gravitational and electric fields vary slowly . this is called the adiabatic condition and is a semiclassical approximation . in the case of positive frequencies , the mode equation ( [ feq ] ) can be rewritten @xmath113 where the momentum dependent frequency is @xmath114 and we have used @xmath115 . the adiabatic condition requires that at all times , the frequency @xmath116 satisfies the relations @xmath117 where dots refer to the partial derivative with respect to the conformal time @xmath74 . in the infinite past @xmath58 the frequency approaches @xmath118 , and hence the adiabatic condition ( [ condition ] ) is satisfied . in the infinite future @xmath119 we have @xmath120 under the condition that @xmath121 the adiabatic condition is satisfied . in our investigation , the spacetime dimension is not too large , i.e. , @xmath122 . hence , the condition ( [ assumption ] ) implies @xmath123 the condition ( [ assumption ] ) justifies our assumption about the range of parameters @xmath124 and @xmath125 which leads to @xmath49 . observe that assuming ( [ assumption ] ) implies that , if one does not want to have trivial flat spacetime results , one needs to assume as well @xmath126 the proof of the previous statement is that if one assumes @xmath127 , together with the semiclassical condition ( [ assume ] ) , it will be equivalent to assume @xmath128 . in this limit the scalar field , the electromagnetic field , and the de sitter spacetime are conformally invariant , leading to flat spacetime results ; see the discussion in sec . [ sec : ur ] . two regimes can be discussed then under the semiclassical approximation : strong electric field , @xmath129 which will give the flat spacetime results , and heavy scalar field , @xmath130 . under ( [ assume ] ) , i.e. , @xmath131 , it can be verified @xcite that the extremum of @xmath132 occurs around the time @xmath133 more on converting momentum to the time integral , in the context of schwinger pair creation , can be found in @xcite . as a consequence of eq . ( [ extremum ] ) , the @xmath112 integral in the eq . ( [ intk ] ) can be converted into a @xmath74 integral @xmath134 the number of created pairs per unit of @xmath0 volume of @xmath7 or the decay rate is then given by @xmath135 where @xmath136 is the slice of @xmath0 volume in the conformal time interval @xmath137 . using eqs . ( [ cosh ] ) and ( [ sinh ] ) , it can be shown that in the cases @xmath138 and @xmath139 , eq . ( [ rate ] ) gives the same result as @xcite and @xcite , respectively . the decay rate ( [ rate ] ) is independent of time , and as a consequence the number density in the comoving frame at time @xmath74 reads @xmath140 it is constant with respect to time ; therefore , the number of pairs created by the background electric and gravitational fields is exactly balanced by the expansion of the ds . provided that eq . ( [ assume ] ) is satisfied , then the bogoliubov coefficient ( [ betasq ] ) is approximated as @xmath141 ) , @xmath142 implying @xmath143 , so the first term in the right - hand side of eq . ( [ betaapprox ] ) is smaller than the second one . then , to leading order , we find @xmath144.\ ] ] therefore , under the semiclassical condition ( [ assume ] ) , @xmath145 is nonzero for both @xmath146 and @xmath147 @xcite . in the language of nucleation of bubbles , considering @xmath148 and taking the particle with charge @xmath149 to the right of the particle with charge @xmath150 , the pairs can nucleate in both the screening and the antiscreening orientations ( corresponding to @xmath147 and @xmath146 , respectively ) because of the gravitational effects @xcite . hence , creating charges in the screening orientation tends to decrease the background electrical field while creating them in the antiscreening orientation tends to increase it . usually screening and antiscreening orientations are referred to as downward and upward tunneling @xcite . the minkowski spacetime limit is obtained in the limit @xmath151 . the decay rate ( [ rate ] ) in this limit approaches @xmath152 which is the same result with the schwinger pair production rate in @xmath0-dimensional minkowski spacetime @xcite . in @xmath7 , the pair production rate is higher than in flat spacetime , due to the gravitational pair production contribution . because of the electrical field , the newly created pairs start to move and hence create a conductive current . in this subsection we present a first semiclassical expression for it , following similar steps as in @xcite . in general , the relation between the current @xmath153 and the density @xmath154 of the semiclassical particles with charge @xmath23 and velocity @xmath155 is @xmath156 . the density of created pairs can be read from eqs . ( [ rate ] ) and ( [ density ] ) . hence , the semiclassical current is determined by @xmath157 for a semiclassical particle with a comoving momentum @xmath158 that interacts with the background vector potential ( [ vector ] ) , the components of the physical momentum vector @xmath159 can be written as @xmath160 and then the magnitude of the velocity reads @xmath161 . in the strong electric field regime , the relation @xmath129 is satisfied . using eqs . ( [ extremum ] ) and ( [ momentum ] ) one can show that @xmath162 where @xmath163 is given by eq . ( [ cordinat ] ) , and consequently @xmath164 . hence , in the strong electric field regime the created particles are ultrarelativistic . using an asymptotic expansion of the modified bessel function ( [ infty ] ) , it can be shown that in the strong electric field regime , @xmath129 , eq . ( [ semi ] ) is approximated as @xmath165 an explicit comparison to the flat spacetime is possible in the strong electric field regime . indeed , under the same assumptions , the flat spacetime limit in @xmath0 dimensions reads @xmath166 with @xmath15 being the minkowski time @xcite . in the expanding ds , accounting for the spacetime dilution in the comoving frame can be done by substituting @xmath167 . in the limit @xmath168 , the exponential factor in eq . ( [ semiur ] ) approaches unity ; then @xmath153 becomes independent of the scalar field mass and responds as @xmath1 . the result corresponds to the result of massless scalar fields in flat spacetime . in the heavy scalar field regime , the relation @xmath130 is satisfied . it was shown in @xcite that due to the background electric field , the charged particles have a terminal physical momentum at late times which is determined as @xmath169 in the heavy scalar field regime , we consider the terminal value for the physical momentum that leads to @xmath170 and consequently the terminal velocity becomes @xmath171 @xcite . hence , in the heavy scalar field regime , @xmath130 , the leading order term of the expansion of eq . ( [ semi ] ) is obtained as @xmath172 thus , for heavy , i.e. , nonrelativistic charged particles the semiclassical current is exponentially suppressed . we will give a more rigorous derivation of these results in @xmath2 dimension below : we will see that , in the strong electric field regime , the semiclassical current @xmath153 agrees with the expectation value of the current operator , whereas in the heavy scalar field regime , they are exponentially different from each other . in this section , we confine ourselves to the case of @xmath11 and compute the induced current and the conductivity without imposing the condition indicated in eq . ( [ assume ] ) . whereas the particle number has no meaning when the adiabatic future does not exist , the current is well defined and is indeed the right quantity to describe the schwinger effect in this context . it can be shown that the current operator of the charged scalar field @xmath173 is conserved , i.e. , @xmath174 @xcite . using eqs . ( [ phiin])-([vacout ] ) it can be shown that in the in vacuum and out vacuum , @xmath175 . however , in the in - vacuum state , the expectation of the spacelike component of the current operator is @xmath176 in order to compute the vacuum expectation value of the current operator ( [ operator ] ) we choose the in - vacuum state because this state is hadamard @xcite . hence , the expectation value has a uv behavior similar to the flat spacetime . substituting explicit expressions , the integral ( [ vev ] ) can be rewritten as @xmath177 where @xmath178 and @xmath179 is an upper cutoff on momentum @xmath112 introduced for convenience and that will be taken to infinity at the end of the calculation . we also have introduced @xmath180 the details of computation of the integral ( [ curent ] ) are reviewed in appendix [ app : int ] . the final result is @xmath181,\end{aligned}\ ] ] where @xmath182 denotes the digamma function and the coefficient @xmath183 is defined as @xmath184 in order to remove the uv divergence term from the expression ( [ unreg ] ) we need to apply a renormalization scheme . in the context of quantum field theory in curved spacetime various regularization and renormalization methods have been developed @xcite . the adiabatic subtraction or regularization method is achieved by subtracting terms computed in the limit of slowly varying backgrounds to obtain a finite expression . the idea of slow varying backgrounds is implemented by introducing adiabatic orders which in our problem will be nothing but counting time derivatives in a given quantity . more details about adiabatic subtraction in the context of schwinger pair creation in curved spacetime are given in ref . we will hence perform the adiabatic expansion of the mode functions up to the minimal order which makes the original expression ( [ unreg ] ) finite . to do so , we express the solution of the mode equation ( [ refeq ] ) as a wentzel - kramers - brillouin ( wkb ) type solution @xmath185,\ ] ] where in order to fulfill eq . ( [ refeq ] ) , the function @xmath186 satisfies the equation @xmath187 provided that the adiabatic condition ( [ condition ] ) holds , derivative terms in eq . ( [ weq ] ) will be negligible compared to @xmath188 terms . as we will see in this subsection , the zeroth order of the adiabatic expansion is enough to remove the uv divergent term from ( [ unreg ] ) . the zeroth order adiabatic expansion of @xmath186 is @xmath189 where the superscript denotes the adiabatic order . the last term in @xmath188 [ see eq . ( [ omega ] ) ] can be rewritten in the form @xmath190 revealing that this term is of adiabatic order 2 . therefore , @xmath191 in eq . ( [ wzero ] ) is given by @xmath192 by virtue of eqs . ( [ varphi ] ) , ( [ fpm ] ) , ( [ fa ] ) , ( [ wzero ] ) , and ( [ omegaz ] ) , the zeroth order adiabatic expansion of the positive frequency @xmath193 and of the negative frequency @xmath194 mode functions are @xmath195 , \nn\\ v_{\mathrm{a};-\k}(x)&=&\omega^{-\frac{1}{2}}(\tau)\big(2\omega_{0}\big)^{-\frac{1}{2 } } \exp\big[i\k\cdot\x+i\int^{\tau}\omega_{0}(\tau')d\tau'\big].\end{aligned}\ ] ] we use this complete set of orthonormal mode functions to expand the charged scalar field operator , then substituting into eq . ( [ operator ] ) leads to the zeroth order adiabatic expansion of the current operator @xmath196 we emphasize that in the expression ( [ ja ] ) there is no finite term or @xmath197-independent contribution . applying the adiabatic subtraction scheme , @xmath198 gives the regularized current as @xmath199.\end{aligned}\ ] ] by virtue of the modified bessel function property given in eq . ( [ parity ] ) , one can show that @xmath200 is an odd function under the transformation @xmath201 , illustrating that if one inverts the electrical field sense , the particles move in the opposite direction . after computing the renormalized current , we consider the conductivity defined as @xmath202 we present a plot of the current ( [ reg ] ) and of the conductivity ( [ sigma ] ) in figs . [ fig1 ] and [ fig2 ] , respectively . the general features of these figures are that in the strong electric field regime @xmath129 all the curves have the same asymptotic behavior , and in the weak electric field regime @xmath203 the current and conductivity are suppressed for increasing scalar field mass . for the case of a massless minimally coupled scalar field , i.e. , @xmath204 , for @xmath205 , the current and conductivity are increasing as the electric field is decreasing . this phenomenon was dubbed infrared - hyperconductivity ( ir - hc ) in @xcite . in the following subsections , we analytically investigate the limiting behaviors of the current and the conductivity . in this analysis , for simplicity , we use the sign conventions @xmath206 and @xmath207 . , the normalized quantum vacuum expectation value of the induced current @xmath208 and the semiclassical current @xmath209 in @xmath2 dimension are plotted as a function of @xmath125 with solid and dashed lines , respectively . ] taking @xmath210 in the current expression ( [ reg ] ) with @xmath124 fixed , the leading order term is @xmath211 the results ( [ strong ] ) analytically describe the behaviors of the current and conductivity shown by figs . [ fig1 ] and [ fig2 ] , respectively . as illustrated in the figures , in this limit , the current and conductivity become increasing functions of electric field @xmath4 and independent of @xmath212 . in the cases of @xmath9 @xcite and @xmath10 @xcite , the authors showed that the current responds as @xmath213 and @xmath214 , respectively , in this regime . indeed , in this limit the semiclassical computation is a good approximation , and as we found in sec . [ sec : semc ] , the current responds as @xmath1 in this regime . to compare the quantum vacuum expectation value of the induced current ( [ reg ] ) with the semiclassical current ( [ semi ] ) , in @xmath2 dimension , we plot @xmath153 as a function of the electric field in fig . this figure illustrates that in the strong electric regime @xmath129 , the semiclassical current approaches the induced current . however , as one decreases @xmath125 and goes away from the strong electric regime , there is a large discrepancy between the semiclassical current and the induced current due to the exponential mass suppression factor in @xmath153 ; see eq . ( [ seminr ] ) and the discussion in sec . [ sec : heavy ] . , the normalized conductivity @xmath215 is plotted as a function of @xmath125 . the phenomenon of infrared hyperconductivity appears for @xmath216 . ] the behavior of the current ( [ reg ] ) in the weak electric field regime @xmath203 is obtained by a series expansion around @xmath217 with @xmath124 fixed . in the case of heavy particles , i.e. , @xmath218 , the leading order terms are @xmath219 and in the case of light particles , i.e. , @xmath127 , leading order terms are given by @xmath220 the results given by eqs . ( [ wheavy ] ) and ( [ wlight ] ) are in agreement with the curves shown in figs . [ fig1 ] and [ fig2 ] . as illustrated in fig . [ fig1 ] the current monotonically increases for increasing electric field @xmath4 . figure [ fig2 ] shows that the conductivity is independent of the electric field @xmath4 . for both the current and the conductivity , the general feature is an inverse dependence on the scalar field mass parameter @xmath212 . in the case of @xmath9 the authors @xcite showed that the current responds as @xmath221 for heavy particles and behaves as @xmath222 for light particles , in this regime . in the case of @xmath10 , it has been shown that for the cases of heavy and light particles , the current behaves as @xmath222 @xcite . the behavior of the current ( [ reg ] ) in the heavy scalar field regime @xmath130 is obtained by taking the limit @xmath223 with @xmath125 held fixed . we then obtain the leading order terms as @xmath224 which are the same as the result ( [ wheavy ] ) . in fig . [ fig3 ] , the current ( [ reg ] ) and its limiting form ( [ heavy ] ) are plotted together . this figure illustrates a good agreement between the exact and the approximated results . the semiclassical current ( [ seminr ] ) , as it is exponentially suppressed , can not appear in the plot of fig . [ fig3 ] and exponentially disagrees with the current ( [ reg ] ) . the comparison of the results , in this regime , shows that in the @xmath138 dimension for heavy scalar @xcite and for heavy fermion @xcite fields , the current scales as @xmath225 , i.e. , is suppressed exponentially . in @xmath139 dimension , for heavy scalar @xcite and for heavy fermion @xcite fields , the current scales as @xmath222 . therefore , in the heavy scalar field regime , due to an exponentially mass suppression factor , the semiclassical current does not agree with the induced current . , the normalized current ( [ reg ] ) and its approximation ( [ heavy ] ) are plotted as a function of @xmath124 with dashed and dotted lines , respectively . ] in the case of a massless minimally coupled scalar field , i.e. , @xmath204 , we now examine the behavior of the current in two limiting regimes . in the limit @xmath210 , the current and conductivity are approximated by eq . ( [ strong ] ) , whereas in the limit @xmath226 we find the leading order terms @xmath227 the results ( [ hyper ] ) agree with the asymptotic behavior of the red curves corresponding to @xmath204 in figs . [ fig1 ] and [ fig2 ] . in the regime @xmath228 the current and the conductivity are not bounded from above and increase as @xmath125 decreases , as illustrated in figs . [ fig1 ] and [ fig2 ] , respectively . this divergence signals that the framework used to derive this result is not valid anymore and backreaction to the reservoir fields needs to be taken into account . more about that regime will be given in sec . [ sec : irhc ] . ) ( in the blue dashed line ) and its approximation ( [ light ] ) ( in the red dotted line ) are plotted as a function of @xmath125 with @xmath229 . ] we now study the behavior of the current ( [ reg ] ) for a light scalar field case , i.e. , @xmath127 and more specifically @xmath216 . in the regime @xmath128 , the current and conductivity scale as indicated by eq . ( [ strong ] ) . numerical analyses show that in the regime @xmath230 the current and conductivity behave as @xmath231 in sec . [ sec : irhc ] we will derive eq . ( [ light ] ) analytically . in fig . [ fig4 ] , we plot the current ( [ light ] ) together with the current ( [ reg ] ) , in the ir - hc regime . this figure illustrates the quite good agreement between the numerical and analytical results . the current has a local minimum in @xmath232 , where @xmath233 is a small positive parameter , and a local maximum in @xmath234 . in the interval @xmath235 the phenomenon of ir - hc occurs ; i.e. , the current increases for decreasing @xmath125 . beyond @xmath236 , in the interval @xmath237 , the current has a linear response for @xmath238 which agrees with eq . ( [ wlight ] ) . from this and sec . [ sec : hyper ] , we can conclude that for decreasing @xmath125 , if @xmath47 becomes real , then there would be a period of ir - hc ; if one keeps decreasing @xmath125 , it will be followed by a linear behavior for @xmath238 or continued unbounded for @xmath204 . in this section , we computed the current and described it in different limiting cases . especially a ir - hc regime has been reported , and we propose a discussion and a summary of what is known about ir - hc in sec . [ sec : irhc ] . ir - hc is a regime where for a given interval of the electric field , a decreasing electric field gives an increasing conductivity . ir - hc was first reported in @xcite . the authors showed there , in the case of @xmath9 that the current responds as @xmath239 for small electric fields and ir - hc was present for @xmath240 . in the case of @xmath10 the renormalization scheme introduces a term of the form @xmath241 @xcite in the regularized current expression which arises from the second order adiabatic expansion . therefore , it signals this renormalization method was not applicable for the case of exactly massless scalar field in @xmath10 @xcite ; see also discussions in @xcite . hence it was not possible to discuss ir - hc for the massless case but it was present for @xmath242 . in @xcite , the point - splitting method was shown to agree with the adiabatic subtraction in @xmath10 for boson . in @xmath11 , we report the same behavior @xmath239 in the regime of small electric field and massless minimally coupled charged particles and report ir - hc for @xmath243 . these results lead us to propose a procedure to avoid an ir - hc regime by setting the value of the conformal coupling to a specific range . in @xmath7 , the _ nonrenormalized _ in - vacuum state expectation of the spacelike component of the current operator is @xmath244 generalizing to @xmath0 dimensions the step performed between eq . ( [ vev ] ) and eq . ( [ curent ] ) , the integral ( [ vevd ] ) can be conveniently rewritten as @xmath245 where @xmath246 is given by eq . ( [ aelement ] ) . as pointed out in @xcite , in a ir - hc regime , the population of created pairs is dominated by ir contribution and no longer by the pairs created within a hubble time . hence the asymptotic behavior of the wave function , in the limit @xmath247 will give the dominant term in a ir - hc regime . the whittaker function @xmath61 as @xmath248 , has an asymptotic form @xcite given by @xmath249 and in this regime @xmath47 is real and by convention is positive . as a consequence , the integral ( [ integrald ] ) in the limit @xmath250 behaves as @xmath251 then power counting shows that in the regime @xmath252 the current integrand diverges in the limit @xmath247 . however , since @xmath253 , the total current integral remains finite . from eq . ( [ ircon ] ) and the definition of @xmath47 , given by eq . ( [ gamma ] ) , we find first that @xmath254 observe that setting @xmath204 , one recovers the behavior @xmath255 . figure [ fig4 ] shows a plot of the current in the ir - hc regime together with the analytical result of ( [ eq : approx ] ) , and the curves agree reasonably well . similar plots could be produced for @xmath138 or @xmath139 . second , again from eqs . ( [ gamma ] ) and ( [ ircon ] ) it is also possible to deduce that ir - hc occurred when @xmath256 therefore , a sufficient condition to avoid ir - hc is @xmath257 and we define thus @xmath258 . the previous condition implies also a minimal bound to avoid ir - hc for the conformal coupling @xmath27 which in the case of a massless scalar field reads @xmath259 we see that in nonconformally coupled theories , a conformal coupling with values larger than @xmath260 can be used to avoid the ir - hc regime . conversely , @xmath261 , ir - hc would appear for @xmath262 , with @xmath263 these results are summarized in table [ tableirhc ] . .minimum values of @xmath264 and @xmath27 to avoid ir - hc . the interval @xmath265 is the range of @xmath266 for which ir - hc would appear after turning on the conformal coupling . [ cols="^,^,^,^",options="header " , ] in this last section , we present our first results on the gravitational backreaction . more specifically , our main goal is to naively estimate the variation of the hubble constant in the heavy scalar field regime . the numerous works on the schwinger effect in @xmath12 @xcite always assumed that the created pairs do not backreact to the background metric . this assumption holds as far as the energy density of the pairs is much smaller than the background hubble energy @xcite . for this paper , we will focus on a semiclassical computation of the stress energy - momentum tensor . we assume that the effects of the pair creation to the einstein equation are small ; they give rise to an effective cosmological constant @xmath267 in the einstein equation . then , the einstein equation can be written as @xmath268 where @xmath269 is the @xmath7 ricci tensor and @xmath270 @xcite is the gravitational constant in @xmath0 dimensions , with @xmath271 being the planck mass . now , we wish to compute the semiclassical energy - momentum tensor on the right - hand side of the einstein equation ( [ einstein ] ) . similar to , e.g. , @xcite the semiclassical energy - momentum tensor of the schwinger pairs can be defined as @xmath272 ( [ betasq ] ) is the distribution function and @xmath159 ( [ momentum ] ) is the physical momentum vector of the created particle . to perform the integral on the right - hand side of eq . ( [ emtdef ] ) , we follow the same integration procedure used in sec . [ sec : den ] : impose the relation ( [ extremum ] ) to convert the @xmath112 integral into a @xmath74 integral . in the heavy scalar field regime , @xmath130 , we consider a terminal value of the physical momentum . hence , substituting expressions ( [ terminalm ] ) and ( [ terminale ] ) into eq . ( [ emtdef ] ) leads to @xmath273 where @xmath274 is given by @xmath275 the hubble parameter is defined as @xmath276 considering the metric ( [ metric ] ) , in terms of the hubble parameter @xmath277 , the components of the ricci tensor are obtained @xmath278 and the ricci scalar is @xmath279 the trace of the einstein equation ( [ einstein ] ) gives @xmath280 and in the heavy scalar field regime , we find that the leading order terms for the einstein equation ( [ einstein ] ) involve @xmath281 : using eqs . ( [ riccitensor])-([lambda ] ) it leads to @xmath282 the above equation determines the evolution of the hubble parameter with respect to the conformal time @xmath74 . in order to compare with the existing literature , we now work in cosmic time @xmath15 : using eqs . ( [ line ] ) and ( [ metric ] ) , it can be shown that the evolution of the hubble parameter with respect to the cosmic time @xmath15 is @xmath283 which agrees with @xcite . thus , the schwinger pair creation leads to a decay of the hubble constant and , as consequence of eq . ( [ lambda ] ) , a decay of the cosmological constant . this decay of the cosmological constant begins with the pair creation and continues until @xmath284 . in this picture , as a classical black hole being evaporated into hawking radiation or the coherent energy of an electric field being dissipated into @xmath285 @xmath286 pairs , the coherent vacuum energy is dissipated into a cloud of scalar pairs . the decay of the hubble constant affects @xmath287 for @xmath288 . for @xmath289 the gravitational constant decays until it reaches zero and for @xmath290 the gravitational constant increases . similar to @xcite , the time scale for evolution of the hubble constant can be estimated by @xmath291 a series expansion of the time scale expression ( [ timescale ] ) around @xmath217 , with @xmath124 fixed , leads to the leading order term @xmath292 which is independent of @xmath125 . in @xcite , the time scale has been computed in the global patch of @xmath10 , without electric field , and the author showed there , in the limit @xmath293 , the time scale behaves as @xmath294 . hence , in @xmath139 dimension , the result ( [ behavior ] ) agrees with the time scale obtained in the ref . @xcite up to a factor of 2 in the exponent . this factor could come from the different definitions for the energy - momentum tensor . observe that the calculation carried out in this section is not valid for @xmath295 as there is no einstein gravity in 1 + 1 dimension . observe beside that under our working assumption , i.e. , heavy scalar field regime , @xmath130 , we find @xmath296 which still allow for a long inflation . furthermore , we argue that this decay of the hubble constant presents similarities with generic models of slow roll inflation where a scalar field sees its potential energy slowly decaying into kinetic energy to ultimately exhibit coherent oscillations around the minimum of its potential which unleash a reheating phase . we want to explore further this issue in a future paper @xcite . the next step is to consider the expectation value of the energy - momentum operator , which as the current will present divergences . the computation of this tensor is much more involving and is beyond the scope of this paper . we have seen in sec . [ sec : result ] that the semiclassical estimates agreed in the strong field regime , but were exponentially different in the heavy scalar regime , so we argue that those results have to be checked by further study , mainly the exact computation of the energy - momentum tensor in order to see if those first estimates agree with the general case . for instance , a very recent work @xcite , without electric field , in @xmath139 dimension , with a slightly different method , discovered an enhancement of the hubble constant . the same exponential behavior as in eq . ( [ timescale ] ) was also found but with a different prefactor . we argue that those changes are due to the renormalization procedure they carried out which gives different results than the replacement of the @xmath112 integral into a @xmath74 integral we performed here . we have investigated pair creation by the schwinger mechanism in @xmath7 . specifically , we considered a charged massive scalar field coupled to a constant background electric field in @xmath7 . after the canonical quantization , bogoliubov coefficients were obtained , and then the decay rate and the density of created pairs were computed ; see these main results in eqs . ( [ rate ] ) and ( [ density ] ) . also , using a semiclassical method the decay rate and the density were computed ; see appendix [ app : sem ] . both methods agree to say that in the semiclassical approximation , the screening orientation stays and the antiscreening ordination is suppressed . the density of created pairs is constant with respect to time . it signals that the pair creation in @xmath7 from electric and gravitational fields exactly balances the dilution from the expansion of the universe . under the semiclassical condition we computed the conduction current of the created particles in any dimension . we find that in the strong electric field regime , @xmath129 , the semiclassical current becomes independent of the scalar field mass and responds as @xmath1 , and in the heavy scalar field regime , @xmath130 , due to the presence of a boltzmann mass suppression factor it exponentially damped . our main goal has been to study the induced quantum vacuum expectation value of the conduction current of the created pairs . thus , in the case of a @xmath2 dimensional ds , the expectation value of the spacelike component of the current operator has been computed in the in - vacuum state . as expected , a linear uv divergence appeared . applying an adiabatic subtraction regularization scheme the divergent term was removed and a finite expression was obtained for the current and the corresponding conductivity . they have been plotted in figs . [ fig1 ] and [ fig2 ] , respectively . the current and conductivity have been also analytically investigated . we find that in the strong electric field regime , @xmath129 , the current responds as @xmath3 and becomes independent of scalar field mass parameter @xmath212 ( [ mds ] ) . in the weak electric field regime , @xmath203 , the current has a linear response in @xmath4 and is inversely proportional to @xmath212 . for the case of a massless minimally coupled scalar field , i.e. , @xmath204 , for @xmath297 , the current varies as @xmath5 . consequently , in this regime , the current and conductivity are increasing unbounded for decreasing electric field , which leads to the phenomenon of ir - hc . the regime of ir - hc has been extensively discussed in sec . [ sec : irhc ] from both numerical and analytical points of view . it has been shown that ir - hc happens for @xmath298 with @xmath299 for @xmath300 and @xmath233 positive but unbounded in @xmath139 . this difference comes from the renormalization scheme used in @xmath139 . the behavior of the current has also been derived in the ir - hc regime for any dimension in eq . ( [ eq : approx ] ) up to the renormalization factors . a proposed relation of ir - hc with conformality and tachyonicity remains also to be further explored but is beyond the scope of this paper . until sec . [ sec : gravity ] , the gravitational and electric fields were treated as an external field , and one important next step is to take into account backreactions of the created pairs to those two fields . indeed , as soon as the energy of the population of the schwinger created pairs becomes of the order of the energy carried by the constant electric field or of the gravitational energy , backreaction effects become unavoidable . investigating these effects could be used to find specific forms of electric fields or specific classes of spacetimes which favor or disfavor pair creation . furthermore , it could also be a fruitful way to make cosmological statements about magnetogenesis , matter - antimatter asymmetry , primordial gravitational waves , or the way inflation is driven and ends . those issues are currently under investigation @xcite . our first results on gravitational backreaction effects were depicted in sec . [ sec : gravity ] . using a semiclassical approach the energy - momentum tensor of the schwinger pairs has been computed in the heavy scalar field regime ; see eq . ( [ emt ] ) . we showed that creation of particles leads to a decay of the hubble constant . in the limit of zero electric field , our result is consistent with a previous study @xcite up to a factor of 2 in the exponent but disagrees with a very recent work @xcite . a more consistent calculation of this effect must dynamically study the evolution of the hubble constant @xmath16 through einstein equations , and this will explicitly break de sitter invariance by introducing a preferred time slicing . we argue that it should be possible to compute it together with the corrections from the schwinger effect and the presence of an electric field to the vacuum fluctuation during an inflationary phase . this could in turn affect the power spectrum at the end of inflation , as it was already suggested in the conclusion of @xcite . after the evolution of the primordial power spectrum through the reheating and the radiation dominated era , in principle it could be measured by cosmic microwave background experiments . x. thank professor r. ruffini for discussions on general relativity , cosmology and physics of pair productions . c. s. is supported by the erasmus mundus joint doctorate program by grant no . 2013 - 1471 from the eacea of the european commission . e. b. is supported by the university of kashan grant no . in this appendix we compute the pair creation rate using the semiclassical scattering method . starting from eq . ( [ refeq ] ) , it is possible to write the equation of motion for the scalar particles as a harmonic oscillator one with a time dependent frequency given by eq . ( [ omega ] ) . from here it is possible to do a bogoliubov transformation and reformulate the equation of motion ( [ refeq ] ) in term of the bogoliubov coefficients @xmath301 and @xmath302 . this technique is inspired from well known flat spacetime techniques ( see , e.g. , @xcite ) and was already applied to the equivalent problem in @xmath139 dimension @xcite . the result will be similar up to dimensional factors . the semiclassical scattering method was usually referred to as the wkb method , but as detailed in @xcite , it is more precise to call it the scattering semiclassical method to differentiate with other wkb inspired methods @xcite . the starting point is to implement the bogoliubov transformation using an ansatz inspired by a wkb expansion , @xmath303,\end{aligned}\ ] ] where @xmath304 to preserve the commutation relation , it is necessary to impose the wronskian condition @xmath305 . in this basis , the momentum spectrum of the pair creation rate reads @xmath306 with the boundary conditions being plane waves in the positive frequency direction as the past asymptotic behavior of eq . ( [ refeq ] ) suggests @xmath307 it is possible to find a first order coupled differential equation for the bogoliubov coefficients @xmath308 note that at this point , the equations derived are still exact . aiming at finding the momentum spectrum ( [ spectrum ] ) , it is possible to integrate formally eqs . ( [ alphadot ] ) and ( [ betadot ] ) by using the boundary condition ( [ boundary ] ) . one finds @xcite @xmath309 each of these integrals can be calculated using a saddle point approximation . those integrals are dominated by the regions around the turning point , i.e. , @xmath310 , where the superscript @xmath39 denotes the two conjugate pairs in the complex plane of @xmath74 . more precisely , by deforming the contour of integration , we consider the singularities for the turning point for which @xmath311<0.\ ] ] from now on , the subscript @xmath39 will be dropped , and we will consider @xmath312 the turning point which corresponds to ( [ impart ] ) . following @xcite , it is possible to describe the behavior of @xmath313 near the turning point assuming first order singularity which is the case contemplating eq . ( [ omega ] ) , @xmath314 with @xmath315 being a constant which can be calculated . one can find then an expression for @xmath316 near the turning point @xmath317 changing variables to @xmath318 and @xmath319 one gets an approximate expression for the integrals @xmath320 where @xmath321 the final results read then @xmath322 for the semiclassical approximation to hold , one needs the notion of adiabatic vacuum in the asymptotic future . hence , the semiclassical approximation holds if the relation ( [ assume ] ) is satisfied . the remaining step is to compute the integral ( [ keq ] ) . the turning point is given by @xmath323,\ ] ] where the coefficients @xmath324 and @xmath46 have been defined in eqs . ( [ lambda ] ) and ( [ kappa ] ) , respectively . then one can find the imaginary part of @xmath316 @xmath325=-\pi\big(\rho+\lambda r\big)\theta\big(-\lambda r\big),\ ] ] where the heaviside step function @xmath326 is there to ensure that the condition ( [ impart ] ) holds . recall our convention @xmath327 , and hence eq . ( [ findim ] ) implies that , e.g. , a particle with charge @xmath328 is only created with a momentum @xmath146 . again , we see that in the semiclassical limit , the upward tunneling is suppressed and only the screening direction or downward tunneling stays . the number of pairs in the semiclassical limit is eventually given by @xmath329\theta(-\lambda r).\ ] ] the pair creation rate is defined as @xmath330 where @xmath331 is defined by eq . ( [ slice ] ) . as before , the procedure then is to transform the @xmath112 integral into a @xmath74 integral by using an estimate for the time when most of the particles are created ; see eq . ( [ extremum ] ) . using eqs . ( [ formul ] ) , ( [ formula ] ) , and ( [ area ] ) , it is then possible to present the final expression for the scalar pair creation rate in @xmath7 under the influence of a constant electric field , @xmath332 a common feature regardless of the number of spatial dimensions and of the bosonic or fermionic nature of the particle is that the physical number density defined in eq . ( [ density ] ) is constant with respect to conformal time . it signals that pair creation in @xmath7 from electric and gravitational fields exactly balances the dilution from the expansion of the universe . this implies that the population of scalars is always dominated by the particle created within a hubble time @xcite . this observation is important when it comes to study the backreaction to the electric field @xcite . it is interesting to note that when one changes the space dimension , what changes is the prefactor before the exponential . indeed , the semiclassical approximation is an expansion in @xmath333 to first order . the exponential factor is the classical trajectory which is not a function of the dimension @xmath108 . however , the one loop integration depends on @xmath108 , and hence the prefactor to the classical trajectory is a function @xmath108 . before concluding , we should remark that under the semiclassical condition ( [ assume ] ) the result ( [ rate ] ) obtained using standard methods reduced to the semiclassical result ( [ smifinal ] ) . therefore , the flat spacetime limit of the semiclassical pair creation rate ( [ smifinal ] ) is equal to the result presented in eq . ( [ limit ] ) . in this appendix , we have represented some useful relations and properties of mathematical functions needed in this article . more relations can be found in , e.g. , @xcite . the whittaker differential equation is @xmath334 it has the two linearly independent solutions , namely , @xmath61 and @xmath62 . the needed connection formulas are @xmath335 the asymptotical expansion of the whittaker functions as @xmath63 are given by @xmath336 where @xmath337 is an arbitrary small positive constant . in the limit @xmath338 , the asymptotically expansions are given by @xmath339 finally , some useful wronskians are @xmath340 the modified bessel function has integral representation along the real line @xmath341 if @xmath154 is an integer , then @xmath342 when @xmath343 is fixed and @xmath248 , @xmath344 when @xmath343 is fixed and @xmath345 , @xmath346 in the cases @xmath347 and @xmath348 , the relations @xmath349 are satisfied . the following mathematical formulas can be shown : @xmath350 in order to evaluate the integrals ( [ intk ] ) and ( [ semidif ] ) , we make use of the spherical coordinates to decompose the momentum vector @xmath105 in the flat @xmath108-dimensional euclidean space . hence , in this space the volume element is @xmath351 where @xmath246 is the area element of the unit sphere in the @xmath108-dimensional euclidean space . convenient coordinates on this sphere are specified by @xmath352 where @xmath353 for @xmath354 and @xmath355 . then , the metric on the sphere is @xmath356 and consequently , the area element is @xmath357 therefore the area of the sphere is @xmath358 using eqs . ( [ bessel ] ) , ( [ formul ] ) , ( [ aelement ] ) , and ( [ area ] ) the following formula can be shown : @xmath359 in this appendix , the computation of the current integral ( [ curent ] ) is reviewed . we follow the same integration procedure as performed in @xcite for a one - dimensional and @xcite for a three - dimensional momentum integral . we deal with the following integral : @xmath360 we will use the mellin - barnes representation of the whittaker function @xmath361 where the contour of integration separates the poles of @xmath362 from those of @xmath363 @xcite . based on the definition ( [ gamma ] ) , it depends on the range of the involved parameters whether @xmath47 is real or purely imaginary . however , by virtue of the relation ( [ conectionw ] ) , in the integral ( [ integral ] ) we have @xmath364 . then , the integral ( [ integral ] ) can be rewritten @xmath365 where @xmath366 is defined as @xmath367 if we choose both @xmath368 and @xmath15 integration contours to run in a similar way as ref . @xcite , then we obtain the final result @xmath369 where @xmath183 is given by eq . ( [ br ] ) . a. landete , j. navarro - salas and f. torrenti , phys . d * 89 * , 044030 ( 2014 ) [ arxiv:1311.4958 [ gr - qc ] ] ; a. landete , j. navarro - salas and f. torrenti , phys . rev . d * 88 * , 061501 ( 2013 ) [ arxiv:1305.7374 [ gr - qc ] ] ; s. ghosh , phys . d * 91 * , no . 12 , 124075 ( 2015 ) [ arxiv:1506.06909 [ gr - qc ] ] ; s. ghosh , phys . d * 93 * , no . 4 , 044032 ( 2016 ) [ arxiv:1601.05518 [ gr - qc ] ] .

we survey some recent results on long - standing conjectures regarding hamilton cycles in directed graphs , oriented graphs and tournaments . we also combine some of these to prove the following approximate result towards kelly s conjecture on hamilton decompositions of regular tournaments : the edges of every regular tournament can be covered by a set of hamilton cycles which are ` almost ' edge - disjoint . we also highlight the role that the notion of ` robust expansion ' plays in several of the proofs . new and old open problems are discussed . # 1 # 1

1006.0590

the decision problem of whether a graph has a hamilton cycle is np - complete and so a satisfactory characterization of hamiltonian graphs seems unlikely . thus it makes sense to ask for degree conditions which ensure that a graph has a hamilton cycle . one such result is dirac s theorem @xcite , which states that every graph on @xmath0 vertices with minimum degree at least @xmath1 contains a hamilton cycle . this is strengthened by ore s theorem @xcite : if @xmath2 is a graph with @xmath0 vertices such that every pair @xmath3 of non - adjacent vertices satisfies @xmath4 , then @xmath2 has a hamilton cycle . dirac s theorem can also be strengthened considerably by allowing many of the vertices to have small degree : psa s theorem @xcite states that a graph on @xmath5 vertices has a hamilton cycle if its degree sequence @xmath6 satisfies @xmath7 for all @xmath8 and if additionally @xmath9 when @xmath10 is odd . again , this is best possible none of the degree conditions can be relaxed . chvtal s theorem @xcite is a further generalization . it characterizes all those degree sequences which ensure the existence of a hamilton cycle in a graph : suppose that the degrees of the graph @xmath2 are @xmath11 . if @xmath12 and @xmath7 or @xmath13 for all @xmath14 then @xmath2 is hamiltonian . this condition on the degree sequence is best possible in the sense that for any degree sequence @xmath11 violating this condition there is a corresponding graph with no hamilton cycle whose degree sequence dominates @xmath15 . these four results are among the most general and well - known hamiltonicity conditions . there are many more often involving additional structural conditions like planarity . the survey @xcite gives an extensive overview ( which concentrates on undirected graphs ) . in this survey , we concentrate on recent progress for directed graphs . though the problems are equally natural for directed graphs , it is usually much more difficult to obtain satisfactory results . additional results beyond those discussed here can be found in the corresponding chapter of the monograph @xcite . in section [ digraph ] , we discuss digraph analogues and generalizations of the above four results . the next section is devoted to oriented graphs these are obtained from undirected graphs by orienting the edges ( and thus are digraphs without @xmath16-cycles ) . section [ tournament ] is concerned with tournaments . section [ general ] is devoted to several generalizations of the notion of a hamilton cycle , e.g. pancyclicity and @xmath17-ordered hamilton cycles . the final section is devoted to the concept of ` robust expansion ' . this has been useful in proving many of the recent results discussed in this survey . we will give a brief sketch of how it can be used . in this paper , we also use this notion ( and several results from this survey ) to obtain a new result ( theorem [ cover ] ) which gives further support to kelly s conjecture on hamilton decompositions of regular tournaments . in a similar vein , we use a result of @xcite to deduce that the edges of every sufficiently dense regular ( undirected ) graph can be covered by hamilton cycles which are almost edge - disjoint ( theorem [ regularii ] ) . for an analogue of dirac s theorem in directed graphs it is natural to consider the _ minimum semidegree @xmath18 _ of a digraph @xmath2 , which is the minimum of its minimum outdegree @xmath19 and its minimum indegree @xmath20 . ( here a directed graph may have two edges between a pair of vertices , but in this case their directions must be opposite . ) the corresponding result is a theorem of ghouila - houri @xcite . [ hamgh ] every strongly connected digraph on @xmath10 vertices with @xmath21 contains a hamilton cycle . in particular , every digraph with @xmath22 contains a hamilton cycle . ( when referring to paths and cycles in directed graphs we usually mean that these are directed , without mentioning this explicitly . ) for undirected regular graphs , jackson @xcite showed that one can reduce the degree condition in dirac s theorem considerably if we also impose a connectivity condition , i.e. every @xmath16-connected @xmath23-regular graph on @xmath10 vertices with @xmath24 contains a hamilton cycle . hilbig @xcite improved the degree condition to @xmath25 unless @xmath2 is the petersen graph or another exceptional graph . the example in figure 1 shows that the degree condition can not be reduced any further . clearly , the connectivity condition is necessary . [ digrapregular ] -regular @xmath16-connected graph @xmath2 on @xmath26 vertices with no hamilton cycle . to construct @xmath2 , start with @xmath27 disjoint cliques on @xmath28 vertices each . in the @xmath29th clique choose disjoint sets @xmath30 and @xmath31 with @xmath32 and @xmath33 and @xmath34 . remove a perfect matching between @xmath30 and @xmath31 for each @xmath29 . add 2 new vertices @xmath35 and @xmath36 , where @xmath35 is connected to all vertices in the sets @xmath30 and @xmath36 is connected to all vertices in all the sets @xmath31.,title="fig : " ] we believe that a similar result should hold for directed graphs too . [ diregconj ] every strongly @xmath16-connected @xmath23-regular digraph on @xmath10 vertices with @xmath24 contains a hamilton cycle . replacing each edge in figure 1 with two oppositely oriented edges shows that the degree condition can not be reduced . moreover , it is not hard to see that the strong @xmath16-connectivity can not be replaced by just strong connectivity . woodall proved the following digraph version of ore s theorem , which generalizes ghouila - houri s theorem . @xmath37 denotes the outdegree of a vertex @xmath38 , and @xmath39 its indegree . [ woodall ] let @xmath2 be a strongly connected digraph on @xmath40 vertices . if @xmath41 for every pair @xmath3 of vertices for which there is no edge from @xmath38 to @xmath42 , then @xmath2 has a hamilton cycle . woodall s theorem in turn is generalized by meyniel s theorem , where the degree condition is formulated in terms of the total degree of a vertex . here the _ total degree @xmath43 of @xmath38 _ is defined as @xmath44 . let @xmath2 be a strongly connected digraph on @xmath40 vertices . if @xmath45 for all pairs of non - adjacent vertices in @xmath2 , then @xmath2 has a hamilton cycle . the following conjecture of bang - jensen , gutin and li @xcite would strengthen meyniel s theorem by requiring the degree condition only for dominated pairs of vertices ( a pair of vertices is _ dominated _ if there is a vertex which sends an edge to both of them ) . [ bglconj ] let @xmath2 be a strongly connected digraph on @xmath40 vertices . if @xmath45 for all dominated pairs of non - adjacent vertices in @xmath2 , then @xmath2 has a hamilton cycle . [ digraphf ] : let @xmath46 be the digraph obtained from the complete digraph @xmath47 and a complete digraph on 3 other vertices @xmath48 as follows : remove the edge from @xmath38 to @xmath49 , add all edges in both directions between @xmath38 and @xmath50 and all edges from @xmath42 to @xmath50.,title="fig : " ] an extremal example @xmath46 can be constructed as in figure 2 . to see that @xmath46 has no hamilton cycle , note that every hamilton path in @xmath51 has to start at @xmath49 . also , note that the only non - adjacent ( dominated ) pairs of vertices are @xmath49 together with a vertex @xmath52 in @xmath50 and these satisfy @xmath53 . some support for the conjecture is given e.g. by the following result of bang - jensen , guo and yeo @xcite : if we also assume the degree condition for all pairs of non - adjacent vertices which have a common outneighbour , then @xmath2 has a @xmath54-factor , i.e. a union of vertex - disjoint cycles covering all the vertices of @xmath2 . there are also a number of degree conditions which involve triples or 4-sets of vertices , see e.g. the corresponding chapter in @xcite . nash - williams @xcite raised the question of a digraph analogue of chvtal s theorem quite soon after the latter was proved : for a digraph @xmath2 it is natural to consider both its outdegree sequence @xmath55 and its indegree sequence @xmath56 . throughout , we take the convention that @xmath57 and @xmath58 without mentioning this explicitly . note that the terms @xmath59 and @xmath60 do not necessarily correspond to the degree of the same vertex of @xmath2 . [ nw ] suppose that @xmath2 is a strongly connected digraph on @xmath12 vertices such that for all @xmath61 * @xmath62 or @xmath63 , * @xmath64 or @xmath65 then @xmath2 contains a hamilton cycle . it is even an open problem whether the conditions imply the existence of a cycle through any pair of given vertices ( see @xcite ) . the following example shows that the degree condition in conjecture [ nw ] would be best possible in the sense that for all @xmath5 and all @xmath66 there is a non - hamiltonian strongly connected digraph @xmath2 on @xmath10 vertices which satisfies the degree conditions except that @xmath67 are replaced by @xmath68 in the @xmath17th pair of conditions . to see this , take an independent set @xmath69 of size @xmath66 and a complete digraph @xmath50 of order @xmath70 . pick a set @xmath71 of @xmath17 vertices of @xmath50 and add all possible edges ( in both directions ) between @xmath69 and @xmath71 . the digraph @xmath2 thus obtained is strongly connected , not hamiltonian and @xmath72 is both the out- and indegree sequence of @xmath2 . in contrast to the undirected case there exist examples with a similar degree sequence to the above but whose structure is quite different ( see @xcite and @xcite ) . this is one of the reasons which makes the directed case much harder than the undirected one . in @xcite , the following approximate version of conjecture [ nw ] for large digraphs was proved . [ ckkochvatal ] for every @xmath73 there exists an integer @xmath74 such that the following holds . suppose that @xmath2 is a digraph on @xmath75 vertices such that for all @xmath61 * @xmath76 or @xmath77 ; * @xmath78 or @xmath79 . then @xmath2 contains a hamilton cycle . this improved a recent result in @xcite , where the degrees in the first parts of these conditions were not ` capped ' at @xmath1 . the earlier result in @xcite was derived from a result in @xcite on the existence of a hamilton cycle in an oriented graph satisfying a certain expansion property . capping the degrees at @xmath1 makes the proof far more difficult : the conditions of theorem [ ckkochvatal ] only imply a rather weak expansion property and there are many types of digraphs which almost satisfy the conditions but are not hamiltonian . the following weakening of conjecture [ nw ] was posed earlier by nash - williams @xcite . it would yield a digraph analogue of psa s theorem . [ nw2 ] let @xmath2 be a digraph on @xmath12 vertices such that @xmath80 for all @xmath81 and such that additionally @xmath82 when @xmath10 is odd . then @xmath2 contains a hamilton cycle . the previous example shows the degree condition would be best possible in the same sense as described there . the assumption of strong connectivity is not necessary in conjecture [ nw2 ] , as it follows from the degree conditions . theorem [ ckkochvatal ] immediately implies a corresponding approximate version of conjecture [ nw2 ] . in particular , for half of the vertex degrees ( namely those whose value is @xmath1 ) , the result matches the conjectured value . another sufficient condition for hamiltonicity in undirected graphs which is just as fundamental as those listed in the introduction is the chvtal - erds theorem @xcite : suppose that @xmath2 is an undirected graph with @xmath83 vertices , for which the vertex - connectivity number @xmath84 and the independence number @xmath85 satisfy @xmath86 , then @xmath2 has a hamilton cycle . currently , there is no digraph analogue of this . given a digraph @xmath2 , let @xmath87 denote the size of the largest set @xmath88 so that @xmath88 induces no edge and let @xmath89 be the size of the largest set @xmath88 so that @xmath88 induces no cycle of length 2 . so @xmath90 . @xmath87 is probably the more natural extension of the independence number to digraphs . however , even the following basic question ( already discussed e.g. in @xcite ) is still open . is there a function @xmath91 so that every digraph with @xmath92 and @xmath93 contains a hamilton cycle ? here the connectivity @xmath84 of a digraph is defined to be the size of the smallest set of vertices @xmath88 so that @xmath94 is either not strongly connected or consists of a single vertex . the following result shows that the analogous function for @xmath89 does exist . [ jacksonthm ] if @xmath2 is a digraph with @xmath95 then @xmath2 has a hamilton cycle . the proof involves a ` reduction ' of the problem to the undirected case . as observed by thomassen and chakroun ( see @xcite again ) , there are non - hamiltonian digraphs with @xmath96 and @xmath97 . but it could well be that every digraph satisfying @xmath98 has a hamilton cycle . even the following weaker conjecture is still wide open . [ hampath ] if @xmath2 is a digraph with @xmath99 , then @xmath2 contains a hamilton cycle . ( in fact , they even conjectured that @xmath2 as above is pancyclic . ) since the problem seems very difficult , even ( say ) a bound on @xmath100 which is polynomial in @xmath101 in theorem [ jacksonthm ] would be interesting . recall that an _ oriented graph _ is a directed graph with no @xmath16-cycles . results on oriented graphs seem even more difficult to obtain than results for the digraph case ( the caccetta - hggkvist conjecture on the girth of oriented graphs of large minimum outdegree is a notorious example of this kind ) . in particular , most problems regarding hamiltonicity of such graphs were open until recently and many open questions still remain . thomassen @xcite raised the natural question of determining the minimum semidegree that forces a hamilton cycle in an oriented graph . thomassen initially believed that the correct minimum semidegree bound should be @xmath102 ( this bound is obtained by considering a ` blow - up ' of an oriented triangle ) . however , hggkvist @xcite later gave a construction which gives a lower bound of @xmath103 : for @xmath10 of the form @xmath104 where @xmath105 is odd , we construct @xmath2 on @xmath10 vertices as in figure 3 . [ extremal2 ] : partition the vertices into @xmath106 parts @xmath107 , with @xmath108 , @xmath109 and @xmath110 . each of @xmath111 and @xmath112 spans a regular tournament , @xmath113 and @xmath114 are joined by a bipartite tournament ( i.e. an orientation of the complete bipartite graph ) which is as regular as possible . we also add all edges from @xmath111 to @xmath113 , from @xmath113 to @xmath112 , from @xmath112 to @xmath114 and from @xmath114 to @xmath111.,title="fig : " ] since every path which joins two vertices in @xmath114 has to pass through @xmath113 , it follows that every cycle contains at least as many vertices from @xmath113 as it contains from @xmath114 . as @xmath115 this means that one can not cover all the vertices of @xmath2 by disjoint cycles . this construction can be extended to arbitrary @xmath10 ( see @xcite ) . the following result exactly matches this bound and improves earlier ones of several authors , e.g. @xcite . in particular , the proof builds on an approximate version which was proved in @xcite . [ main ] there exists an integer @xmath116 so that any oriented graph @xmath2 on @xmath117 vertices with minimum semidegree @xmath118 contains a hamilton cycle . jackson conjectured that for regular oriented graphs one can significantly reduce the degree condition . [ jacksonconj ] for each @xmath119 , every @xmath23-regular oriented graph @xmath2 on @xmath120 vertices has a hamilton cycle . the disjoint union of two regular tournaments on @xmath1 vertices shows that this would be best possible . note that the degree condition is smaller than the one in conjecture [ diregconj ] . we believe that it may actually be possible to reduce the degree condition even further if we assume that @xmath2 is strongly @xmath16-connected : is it true that for each @xmath119 , every @xmath23-regular strongly @xmath16-connected oriented graph @xmath2 on @xmath121 vertices has a hamilton cycle ? a suitable orientation of the example in figure 1 shows that this would be best possible . hggkvist @xcite also made the following conjecture which is closely related to theorem [ main ] . given an oriented graph @xmath2 , let @xmath122 denote the minimum degree of @xmath2 ( i.e. the minimum number of edges incident to a vertex ) and set @xmath123 . [ haggconj ] every oriented graph @xmath2 on @xmath10 vertices with @xmath124 contains a hamilton cycle . ( note that this conjecture does not quite imply theorem [ main ] as it results in a marginally greater minimum semidegree condition . ) in @xcite , conjecture [ haggconj ] was verified approximately , i.e. if @xmath125 , then @xmath2 has a hamilton cycle ( note this implies an approximate version of theorem [ main ] ) . the same methods also yield an approximate version of ore s theorem for oriented graphs . [ thm : ore ] for every @xmath126 there exists an integer @xmath127 such that every oriented graph @xmath2 of order @xmath128 with @xmath129 whenever @xmath2 does not contain an edge from @xmath38 to @xmath42 contains a hamilton cycle . the construction in figure 3 shows that the bound is best possible up to the term @xmath130 . it would be interesting to obtain an exact version of this result . song @xcite proved that every oriented graph on @xmath131 vertices with @xmath132 and @xmath133 whenever @xmath2 does not contain an edge from @xmath38 to @xmath42 is pancyclic ( i.e. @xmath2 contains cycles of all possible lengths ) . in @xcite he also claims ( without proof ) that the condition is best possible for infinitely many @xmath10 as @xmath2 may fail to contain a hamilton cycle otherwise . note that theorem [ thm : ore ] implies that this claim is false . in @xcite a construction was described which showed that there is no satisfactory analogue of psa s theorem for oriented graphs : as soon as we allow a few vertices to have a degree somewhat below @xmath134 , then one can not guarantee a hamilton cycle . the question of exactly determining all those degree sequences which guarantee a hamilton cycle remains open though . it is also not clear whether there may be a version of the chvtal - erds theorem for oriented graphs . a _ tournament _ is an orientation of a complete graph . it has long been known that tournaments enjoy particularly strong hamiltonicity properties : camion @xcite showed that we only need to assume strong connectivity to ensure that a tournament has a hamilton cycle . moon @xcite strengthened this by proving that every strongly connected tournament is even pancyclic . it is easy to see that a minimum semidegree of @xmath135 forces a tournament on @xmath10 vertices to be strongly connected , leading to a better degree condition for hamiltonicity than that of @xmath136 for the class of all oriented graphs . a _ hamilton decomposition _ of a graph or digraph @xmath2 is a set of edge - disjoint hamilton cycles which together cover all the edges of @xmath2 . not many examples of graphs with such decompositions are known . one can construct a hamilton decomposition of a complete graph if and only if its order is odd ( this was first observed by walecki in the late 19th century ) . tillson @xcite proved that a complete digraph @xmath2 on @xmath10 vertices has a hamilton decomposition if and only if @xmath137 . the following conjecture of kelly from 1968 ( see moon @xcite ) would be a far - reaching generalization of walecki s result : [ kelly ] every regular tournament on @xmath10 vertices can be decomposed into @xmath138 edge - disjoint hamilton cycles . in @xcite we proved an approximate version of kelly s conjecture . moreover , the result holds even for oriented graphs @xmath2 which are not quite regular and whose ` underlying ' undirected graph is not quite complete . [ kellythm ] for every @xmath139 there exist @xmath140 and @xmath141 such that the following holds . suppose that @xmath2 is an oriented graph on @xmath128 vertices such that @xmath142 . then @xmath2 contains at least @xmath143 edge - disjoint hamilton cycles . we also proved that the condition on the minimum semidegree can be relaxed to @xmath144 . this is asymptotically best possible since the construction described in figure 3 is almost regular . some earlier support for kelly s conjecture was provided by thomassen @xcite , who showed that the edges of every regular tournament can be covered by at most @xmath145 hamilton cycles . in this paper , we improve this to an asymptotically best possible result . we will give a proof ( which relies on theorem [ kellythm ] ) in section [ coversec ] . [ cover ] for every @xmath146 there exists an integer @xmath147 so that every regular tournament @xmath2 on @xmath148 vertices contains a set of @xmath149 hamilton cycles which together cover all the edges of @xmath2 . kelly s conjecture has been generalized in several ways , e.g. bang - jensen and yeo @xcite conjectured that every @xmath17-edge - connected tournament has a decomposition into @xmath17 spanning strong digraphs . a bipartite version of kelly s conjecture was also formulated by jackson @xcite . thomassen made the following conjecture which replaces the assumption of regularity by high connectivity . for every @xmath150 there is an integer @xmath151 so that every strongly @xmath151-connected tournament has @xmath17 edge - disjoint hamilton cycles . a conjecture of erds ( see @xcite ) which is also related to kelly s conjecture states that almost all tournaments @xmath2 have at least @xmath18 edge - disjoint hamilton cycles . similar techniques as in the proof of the approximate version of kelly s conjecture were used at the same time in @xcite to prove approximate versions of two long - standing conjectures of nash - williams on edge - disjoint hamilton cycles in ( undirected ) graphs . one of these results states that one can almost decompose any dense regular graph into hamilton cycles . [ regular ] for every @xmath152 there is an integer @xmath153 so that every @xmath23-regular graph on @xmath148 vertices with @xmath154 contains at least @xmath155 edge - disjoint hamilton cycles . in section [ coversec ] we deduce the following analogue of theorem [ cover ] : [ regularii ] for every @xmath156 there is an integer @xmath147 so that every @xmath23-regular graph @xmath2 on @xmath148 vertices with @xmath157 contains at most @xmath158 hamilton cycles which together cover all the edges of @xmath2 . one of the earliest results on tournaments ( and the probablistic method ) , was obtained by szele @xcite , who showed that the maximum number @xmath159 of hamilton paths in a tournament on @xmath10 vertices satisfies @xmath160 and @xmath161 . the lower bound is obtained by considering a random tournament . the best upper bound is due to friedgut and kahn @xcite who showed that @xmath162 , where @xmath163 is slightly less than @xmath164 . the best current lower bound is due to wormald @xcite , who showed that @xmath165 . so in particular , @xmath159 is not attained for random tournaments . also , he conjectured that this bound is very close to the correct value . similarly , one can define the maximum number @xmath166 of hamilton cycles in a tournament on @xmath10 vertices . note that by considering a random tournament again , we obtain @xmath167 . unsurprisingly , @xmath166 and @xmath159 are very closely related , e.g. we have @xmath168 . in particular , the main result in @xcite states that @xmath169 , where @xmath163 is the same as above . this implies the above bound on @xmath159 , since alon @xcite observed that @xmath170 . also , wormald @xcite showed that @xmath171 . ( note this also follows by combining alon s observation with the lower bound on @xmath159 in @xcite . ) of course , in general it does not make sense to ask for the minimum number of hamilton paths or cycles in a tournament . however , the question does make sense for regular tournaments . friedgut and kahn @xcite asked whether the number of hamilton cycles in a regular tournament is always at least @xmath172 . the best result towards this was recently obtained by cuckler @xcite , who showed that every regular tournament on @xmath10 vertices contains at least @xmath173 hamilton cycles . this also answers an earlier question of thomassen . asking for the minimum number of hamilton paths in a tournament @xmath174 also makes sense if we assume that @xmath174 is strongly connected . busch @xcite determined this number exactly by showing that an earlier construction of moon is best possible . the related question on the minimum number of hamilton cycles in a strongly @xmath16-connected tournament is still open ( see @xcite ) . sumner s universal tournament conjecture states that every tournament on @xmath175 vertices contains every tree on @xmath10 vertices . in @xcite an approximate version of this conjecture was proved and subsequently in @xcite , the conjecture was proved for all large trees ( see e.g. @xcite for a discussion of numerous previous results ) . the proof in @xcite builds on several structural results proved in @xcite . [ sumnerapprox ] there is an integer @xmath116 such that for all @xmath148 every tournament @xmath2 on @xmath175 vertices contains any directed tree @xmath174 on @xmath10 vertices . while this result is not directly related to the main topic of the survey ( i.e. hamilton cycles ) , there are several connections . firstly , just as with many of the new results in the other sections , the concept of a robust expander is crucial in the proof of theorem [ sumnerapprox ] . secondly , the proof of theorem [ sumnerapprox ] also makes direct use of the fact that a robust expander contains a hamilton cycle ( theorem [ expanderthm ] ) . suitable parts of the tree @xmath174 are embedded by considering a random walk on ( the blow - up of ) such a hamilton cycle . in @xcite , we also proved that if @xmath174 has bounded maximum degree , then it suffices if the tournament @xmath2 has @xmath176 vertices . this is best possible in the sense that the ` error term ' @xmath130 can not be completely omitted in general . but it seems possible that it can be reduced to a constant which depends only on the maximum degree of @xmath174 . if @xmath174 is an orientation of a path , then the error term can be omitted completely : havet and thomass @xcite proved that every tournament on at least 8 vertices contains every possible orientation of a hamilton path ( for arbitrary orientations of hamilton cycles , see section [ arbitraryorient ] ) . in this section , we discuss several natural ways of strengthening the notion of a hamilton cycle . recall that a graph ( or digraph ) is _ pancyclic _ if it contains a cycle of every possible length . dirac s theorem implies that a graph on @xmath0 vertices is pancyclic if it has minimum degree greater than @xmath1 . ( to see this , remove a vertex @xmath38 and apply dirac s theorem to the remaining subgraph to obtain a cycle of length @xmath177 . then consider the neighbourhood of @xmath38 on this cycle to obtain cycles of all possible lengths through @xmath38 . ) similarly , one can use ghouila - houri s theorem to deduce that every digraph on @xmath10 vertices with minimum semidegree greater than @xmath1 is pancyclic . in both cases , the complete bipartite ( di-)graph whose vertex class sizes are as equal as possible shows that the bound is best possible . more generally , the same trick also works for meyniel s theorem : let @xmath2 be a strongly connected digraph on @xmath40 vertices . if @xmath178 for all pairs of non - adjacent vertices @xmath179 in @xmath2 , then @xmath2 is pancyclic . ( indeed , the conditions imply that either @xmath2 contains a strongly connected tournament or contains a vertex @xmath38 with @xmath180 , in which case we can proceed as above . ) if @xmath10 is even , the bound @xmath181 is best possible . for @xmath10 is odd , it follows from a result of thomassen @xcite that one can improve it to @xmath182 . for oriented graphs the minimum semidegree threshold which guarantees pancyclicity turns out to be @xmath136 , i.e. the same threshold as for hamiltonicity ( see @xcite ) . the above trick of removing a vertex does not work here . instead , to obtain ` long ' cycles one can modify the proof of theorem [ main ] . a triangle is guaranteed by results on the caccetta - hggkvist conjecture e.g. a very recent result of hladk , krl and norine @xcite states that every oriented graph on @xmath10 vertices with minimum semidegree at least @xmath183 contains a 3-cycle . short cycles of length @xmath184 can be guaranteed by a result in @xcite which states that for all @xmath185 every oriented graph @xmath2 on @xmath10 vertices with @xmath186 contains an @xmath187-cycle . this is best possible for all those @xmath188 which are not divisible by 3 . surprisingly , for some other values of @xmath187 , an @xmath187-cycle is forced by a much weaker minimum degree condition . in particular , the following conjecture was made in @xcite . [ con : short ] let @xmath188 be a positive integer and let @xmath17 be the smallest integer that is greater than @xmath16 and does not divide @xmath187 . then there exists an integer @xmath189 such that every oriented graph @xmath2 on @xmath128 vertices with minimum semidegree @xmath190 contains an @xmath187-cycle . the extremal examples for this conjecture are always ` blow - ups ' of cycles of length @xmath17 . possibly one can even weaken the condition by requiring only the outdegree of @xmath2 to be large . it is easy to see that the only values of @xmath17 that can appear in conjecture [ con : short ] are of the form @xmath191 with @xmath192 , where @xmath193 is a prime and @xmath194 a positive integer . as mentioned earlier , the most natural notion of a cycle in a digraph is to have all edges directed consistently . but it also makes sense to ask for hamilton cycles where the edges are oriented in some prescribed way , e.g. to ask for an ` antidirected ' hamilton cycle where consecutive edges have opposite directions . surprisingly , it turns out that both for digraphs and oriented graphs the minimum degree threshold which guarantees a ` consistent ' hamilton cycle is approximately the same as that which guarantees an arbitrary orientation of a hamilton cycle . there exists an @xmath116 so that every digraph @xmath2 on @xmath195 vertices with minimum semidegree @xmath196 contains every orientation of a hamilton cycle . in @xcite , they conjectured an analogue of this for oriented graphs , which was recently proved by kelly . for every @xmath126 there exists an integer @xmath127 such that every oriented graph @xmath2 on @xmath128 vertices with minimum semidegree @xmath197 contains every orientation of a hamilton cycle . the proof of this result uses theorem [ main ] as the well as the notion of expanding digraphs . interestingly , kelly observed that the thresholds for various orientations do not coincide exactly : for instance , if we modify the example in figure 3 so that all classes have the same odd size , then the resulting oriented graph has minimum semidegree @xmath136 but no antidirected hamilton cycle . thomason @xcite showed that for large tournaments strong connectivity ensures every possible orientation of a hamilton cycle . more precisely , he showed that for @xmath198 , every tournament on @xmath10 vertices contains all possible orientations of a hamilton cycle , except possibly the ` consistently oriented ' one . ( note that this also implies that every large tournament contains every orientation of a hamilton path , i.e. a weaker version of the result in @xcite mentioned earlier . ) the bound on @xmath10 was later reduced to 68 by havet @xcite . thomason conjectured that the correct bound is @xmath199 . suppose that we require our ( hamilton ) cycle to visit several vertices in a specific order . more formally , we say that a graph @xmath2 is _ @xmath17-ordered _ if for every sequence @xmath200 of distinct vertices of @xmath2 there is a cycle which encounters @xmath200 in this order . @xmath2 is _ @xmath17-ordered hamiltonian _ if it contains a hamilton cycle with this property . kierstead , srkzy and selkow @xcite determined the minimum degree which forces an ( undirected ) graph to be @xmath17-ordered hamiltonian . [ kordered ] for all @xmath150 , every graph on @xmath201 vertices of minimum degree at least @xmath202 is @xmath17-ordered hamiltonian . the extremal example consists of two cliques intersecting in @xmath203 vertices if @xmath17 is even and two cliques intersecting in @xmath204 vertices if @xmath17 is odd . the case when @xmath10 is not too large compared to @xmath17 is still open . the corresponding ore - type problem was solved in @xcite . here the ore - type result does not imply the dirac - type result above . many variations and stronger notions have been investigated ( see e.g. @xcite again ) . directed graphs form a particularly natural setting for this kind of question . the following result gives a directed analogue of theorem [ kordered ] . [ thm : ordham ] for every @xmath192 there is an integer @xmath205 such that every digraph @xmath2 on @xmath195 vertices with minimum semidegree @xmath206 is @xmath17-ordered hamiltonian . note that if @xmath10 is even and @xmath17 is odd the bound on the minimum semidegree is slightly larger than in the undirected case . however , it is best possible in all cases . in fact , if the minimum semidegree is smaller , it turns out that @xmath2 need not even be @xmath17-ordered . again , the family of extremal examples turns out to be much richer than in the undirected case . note that every hamiltonian digraph is @xmath16-ordered hamiltonian , so the case when @xmath207 in theorem [ thm : ordham ] is covered by ghouila - houri s theorem . it would be interesting to obtain an ore - type or an oriented version of theorem [ thm : ordham ] . another natural way of generalizing dirac s theorem is to ask for a certain set of vertex - disjoint cycles in @xmath2 which together cover all the vertices of @xmath2 ( note this also generalizes the notion of pancyclicity ) . for large undirected graphs , abassi @xcite determined the minimum degree which guarantees @xmath17 vertex - disjoint cycles in a graph @xmath2 whose ( given ) lengths are @xmath208 , where the @xmath209 sum up to @xmath10 and where the order @xmath10 of @xmath2 is sufficiently large . as in the case of hamilton cycles , the corresponding questions for directed and oriented graphs appear much harder than in the undirected case and again much less is known . keevash and sudakov @xcite recently obtained the following result . there exist positive constants @xmath210 and an integer @xmath116 so that whenever @xmath2 is an oriented graph on @xmath195 vertices with minimum semidegree at least @xmath211 and whenever @xmath212 are so that @xmath213 , then @xmath2 contains vertex - disjoint cycles of length @xmath212 . in general , one can not take @xmath214 . in the case of triangles ( i.e. when all the @xmath215 ) , they show that one can choose @xmath216 . this comes very close to proving a recent conjecture formulated independently by cuckler and yuster @xcite , which states that every regular tournament on @xmath217 vertices contains vertex - disjoint triangles covering all the vertices of the tournament . similar questions were also raised earlier by song @xcite . for instance , given @xmath218 , he asked for the smallest integer @xmath219 so that all but a finite number of strongly @xmath219-connected tournaments @xmath174 satisfy the following : let @xmath10 be the number of vertices of @xmath174 and let @xmath220 . then @xmath174 contains vertex - disjoint cycles of length @xmath212 . chen , gould and li @xcite proved the weaker result that every sufficiently large @xmath218-connected tournament @xmath2 contains @xmath218 vertex - disjoint cycles which together cover all the vertices of @xmath2 . this proved a conjecture of bollobs . sarkzy , komls and szemerdi @xcite showed that every sufficiently large graph @xmath2 on @xmath10 vertices with minimum degree at least @xmath221 contains the @xmath17th power of a hamilton cycle . extremal examples are complete @xmath222-partite graphs with classes of almost equal size . it appears likely that the situation for digraphs is similar . however , just as for ordinary hamilton cycles , it seems that for oriented graphs the picture is rather different . ( both for digraphs and oriented graphs , the most natural definition of the @xmath17th power of a cycle is a cyclically ordered set of vertices so that every vertex sends an edge to the next @xmath17 vertices in the ordering . ) [ squareconj ] for every @xmath223 there is an integer @xmath224 so that every oriented graph @xmath2 on @xmath117 vertices with minimum semidegree at least @xmath225 contains the square of a hamilton cycle . a construction which shows that the constant @xmath226 can not be improved is given in figure 4 . [ square ] : the set sizes are @xmath227 , @xmath228 , @xmath229 , @xmath230 and @xmath231 , where @xmath105 is even . @xmath113 , @xmath112 and @xmath114 induce regular tournaments , while @xmath111 and @xmath232 induce independent sets . a single arrow ( e.g. from @xmath113 to @xmath112 ) indicates an orientation of the complete bipartite graph from the initial set towards the final set . a double edge ( e.g. between @xmath113 and @xmath232 ) indicates an orientation of the complete bipartite graph so that within each set , the in- and outdegrees of the vertices differ by at most one.,title="fig : " ] we claim that the square of any hamilton cycle would have to visit a vertex of @xmath113 in between two visits of @xmath232 . since @xmath233 , this shows that the graph does not contain the square of a hamilton cycle . to prove the claim , suppose that @xmath46 is a squared hamilton cycle and consider a vertex @xmath234 of @xmath46 which lies in @xmath232 . then the predecessor of @xmath234 lies in @xmath112 or @xmath113 , so without loss of generality we may assume that it is a vertex @xmath235 . again , the predecessor of @xmath236 lies in @xmath112 or @xmath113 ( since it must lie in the common inneighbourhood of @xmath112 and @xmath232 ) , so without loss of generality we may assume that it is a vertex @xmath237 . the predecessor of @xmath238 can now lie in @xmath111 , @xmath113 or @xmath112 . if it lies in @xmath113 we are done again , if it is a vertex @xmath239 , we consider its predecessor , which can again only lie in @xmath111 , @xmath113 or @xmath112 . since @xmath46 must visit all vertices , it follows that we eventually arrive at a predecessor @xmath240 whose successor on @xmath46 is some vertex @xmath241 . the predecessor of @xmath35 on @xmath46 must lie in the common inneighbourhood of @xmath35 and @xmath163 , so it must lie in @xmath113 , as required . for the case of tournaments , the problem was solved asymptotically by bollobs and hggkvist @xcite . given a tournament @xmath174 of large order @xmath10 with minimum semidegree at least @xmath242 , they proved that ( for fixed @xmath17 ) @xmath174 contains the @xmath17th power of a hamilton cycle . so asymptotically , the semidegree threshold for an ordinary hamilton cycle in a tournament is the same as that for the @xmath17th power of a hamilton cycle . roughly speaking , a graph is an expander if for every set @xmath88 of vertices the neighbourhood @xmath243 of @xmath88 is significantly larger than @xmath88 itself . a number of papers have recently demonstrated that there is a remarkably close connection between hamiltonicity and expansion ( see e.g. @xcite ) . the following notion of robustly expanding ( dense ) digraphs was introduced in @xcite . let @xmath244 . given any digraph @xmath2 on @xmath10 vertices and @xmath245 , the _ @xmath246-robust outneighbourhood @xmath247 of @xmath88 _ is the set of all those vertices @xmath38 of @xmath2 which have at least @xmath248 inneighbours in @xmath88 . @xmath2 is called a _ robust @xmath249-outexpander _ if @xmath250 for all @xmath245 with @xmath251 . as the name suggests , this notion has the advantage that it is preserved even if we delete some vertices and edges from @xmath2 . we will also use the more traditional ( and weaker ) notion of a _ @xmath249-outexpander _ , which means @xmath252 for all @xmath245 with @xmath253 . [ expanderthm ] let @xmath116 be a positive integer and @xmath254 be positive constants such that @xmath255 . let @xmath2 be a digraph on @xmath195 vertices with @xmath256 which is a robust @xmath249-outexpander . then @xmath2 contains a hamilton cycle . theorem [ expanderthm ] is used in @xcite to give a weaker version of theorem [ ckkochvatal ] ( i.e. without the degrees capped at @xmath1 ) . in the same paper it is also applied to prove a conjecture of thomassen regarding a weak version of conjecture [ kelly ] ( kelly s conjecture ) . one can also use it to prove e.g. theorem [ thm : ore ] and thus an approximate version of theorem [ main ] . ( indeed , as proved in @xcite , the degree conditions of theorem [ thm : ore ] imply expansion , the proof for robust expansion is similar . ) as mentioned earlier , it is also used as a tool in the proof of theorem [ sumnerapprox ] . finally , we will also use it in the next subsection to prove theorem [ cover ] . in @xcite , theorem [ expanderthm ] was deduced from a result in @xcite . the proof of the result in @xcite ( and a similar approach in @xcite ) in turn relied on szemerdi s regularity lemma and a ( rather technical ) version of the blow - up lemma due to csaba @xcite . a ( parallel ) algorithmic version of theorem [ expanderthm ] was also proved in @xcite . below , we give a brief sketch of a proof of theorem [ expanderthm ] which avoids any use of the blow - up lemma and is based on an approach in @xcite . the _ density _ of a bipartite graph @xmath2 with vertex classes @xmath111 and @xmath113 is defined to be @xmath257 , where @xmath258 denotes the number of edges between @xmath111 and @xmath113 . given @xmath259 , we say that @xmath2 is @xmath260-_regular _ if for all subsets @xmath261 and @xmath262 with @xmath263 and @xmath264 we have that @xmath265 . we also say that @xmath2 is @xmath266-_super - regular _ if it is @xmath260-regular and furthermore every vertex @xmath240 has degree at least @xmath267 and similarly for every @xmath268 . these definitions generalize naturally to non - bipartite ( di-)graphs . we also need the result that every super - regular digraph contains a hamilton cycle . [ fk ] suppose that @xmath269 and @xmath2 is an @xmath266-super - regular digraph on @xmath195 vertices . then @xmath2 contains a hamilton cycle . lemma [ fk ] is a special case e.g. of a result of frieze and krivelevich @xcite , who proved that an @xmath266-super - regular digraph on @xmath10 vertices has almost @xmath270 edge - disjoint hamilton cycles if @xmath10 is large . here we also give a sketch of a direct proof of lemma [ fk ] . we first prove that @xmath2 contains a 1-factor . consider the auxiliary bipartite graph whose vertex classes @xmath111 and @xmath113 are copies of @xmath271 with an edge between @xmath272 and @xmath268 if there is an edge from @xmath35 to @xmath36 in @xmath2 . one can show that this bipartite graph has a perfect matching ( by hall s marriage theorem ) , which in turn corresponds to a @xmath54-factor in @xmath2 . it is now not hard to prove the lemma using the ` rotation - extension ' technique : choose a 1-factor of @xmath2 . now remove an edge of a cycle in this 1-factor and let @xmath273 be the resulting path . if the final vertex of @xmath273 has any outneighbours on another cycle @xmath112 of the 1-factor , we can extend @xmath273 into a longer path which includes the vertices of @xmath112 ( and similarly for the initial vertex of @xmath273 ) . we repeat this as long as possible ( and one can always ensure that the extension step can be carried out at least once ) . so we may assume that all outneighbours of the final vertex of @xmath273 lie on @xmath273 and similarly for the initial vertex of @xmath273 . together with the @xmath260-regularity this can be used to find a cycle with the same vertex set as @xmath273 . eventually , we arrive at a hamilton cycle . * sketch proof of theorem [ expanderthm]*. choose @xmath274 to satisfy @xmath275 . the first step is to apply a directed version of szemerdi s regularity lemma to @xmath2 . this gives us a partition of the vertices of @xmath2 into clusters @xmath276 and an exceptional set @xmath277 so that @xmath278 and all the clusters have size @xmath105 . now define a ` reduced ' digraph @xmath279 whose vertices are the clusters @xmath276 and with an edge from @xmath280 to @xmath281 if the bipartite graph spanned by the edges from @xmath280 to @xmath281 is @xmath260-regular and has density at least @xmath23 . then one can show ( see lemma 14 in @xcite ) that @xmath279 is still a @xmath282-outexpander ( this is the point where we need the robustness of the expansion in @xmath2 ) with minimum semidegree at least @xmath283 . this in turn can be used to show that @xmath279 has a @xmath54-factor @xmath46 ( using the same auxiliary bipartite graph as in the proof of lemma [ fk ] ) . by removing a small number of vertices from the clusters , we can also assume that the bipartite subgraphs spanned by successive clusters on each cycle of @xmath46 are super - regular , i.e. have high minimum degree . for simplicity , assume that the cluster size is still @xmath105 . moreover , since @xmath2 is an expander , we can find a short path in @xmath2 between clusters of different cycles of @xmath46 and also between any pair of exceptional vertices . however , we need to choose such paths without affecting any of the useful structures that we have found so far . for this , we will consider paths which ` wind around ' cycles in @xmath46 before moving to another cycle . more precisely , a _ shifted walk _ from a cluster @xmath111 to a cluster @xmath113 is a walk @xmath284 of the form @xmath285 where @xmath286 , @xmath287 , @xmath288 is the cycle of @xmath46 containing @xmath289 , and for each @xmath290 , @xmath291 is the predecessor of @xmath289 on @xmath288 and the edge @xmath292 belongs to @xmath279 . we say that @xmath293 as above _ traverses @xmath218 cycles _ ( even if some @xmath288 appears several times in @xmath293 ) . we also say that the clusters @xmath294 are the _ entry clusters _ ( as this is where @xmath293 ` enters ' a cycle @xmath288 ) and the clusters @xmath295 are the _ exit clusters _ of @xmath293 . note that * for any cycle of @xmath46 , its clusters are visited the same number of times by @xmath296 . using the expansion of @xmath279 , it is not hard to see that * for any clusters @xmath111 and @xmath113 there is a shifted walk from @xmath111 to @xmath113 which does not traverse too many cycles . indeed , the expansion property implies that the number of clusters one can reach by traversing @xmath218 cycles is at least @xmath297 as long as this is significantly less than the total number @xmath17 of clusters . now we will ` join up ' the exceptional vertices using shifted walks . for this , write @xmath298 . for each exceptional vertex @xmath299 choose a cluster @xmath300 so that @xmath299 has many outneighbours in @xmath300 . similarly choose a cluster @xmath301 so that @xmath299 has many inneighbours in @xmath301 and so that * no cluster appears too often as a @xmath300 or a @xmath301 . given a cluster @xmath71 , let @xmath302 be the predecessor of @xmath71 on the cycle of @xmath46 which contains @xmath71 and let @xmath303 be its successor . form a ` walk ' @xmath293 on @xmath304 which starts at @xmath305 , then moves to @xmath306 , then follows a shifted walk from @xmath306 to @xmath307 , then it winds around the entire cycle of @xmath46 containing @xmath307 until it reaches @xmath308 . then @xmath293 moves to @xmath309 , then to @xmath310 using a shifted walk as above until it has visited all the exceptional vertices . proceeding similarly , we can ensure that @xmath293 has the following properties : * @xmath293 is a closed walk which visits all of @xmath277 and all of @xmath311 . * for any cycle of @xmath46 , its clusters are visited the same number of times by @xmath293 . * every cluster appears at most @xmath312 times as an entry or exit cluster . \(b ) follows from ( i ) and ( c ) follows from ( ii ) and ( iii ) . the next step towards a hamilton cycle would be to find a cycle @xmath112 in @xmath2 which corresponds to @xmath293 ( i.e. each occurrence of a cluster in @xmath293 is replaced by a distinct vertex of @xmath2 lying in this cluster ) . unfortunately , the fact that @xmath277 may be much larger than the cluster size @xmath105 implies that there may be clusters which are visited more than @xmath105 times by @xmath293 , which makes it impossible to find such a @xmath112 . so we will apply a ` short - cutting ' technique to @xmath293 which avoids ` winding around ' the cycles of @xmath46 too often . for this , we now fix edges in @xmath2 corresponding to all those edges of @xmath293 that do not lie within a cycle of @xmath46 . these edges of @xmath293 are precisely the edges in @xmath293 at the exceptional vertices as well as all the edges of the form @xmath313 where @xmath111 is used as an exit cluster by @xmath293 and @xmath113 is used as an entrance cluster by @xmath293 . for each edge @xmath314 at an exceptional vertex we choose an edge @xmath315 , where @xmath38 is an outneighbour of @xmath299 in @xmath300 . we similarly choose an edge @xmath316 from @xmath301 to @xmath299 for each @xmath317 . we do this in such a way that all these edges are disjoint outside @xmath277 . for each occurrence of @xmath313 in @xmath293 , where @xmath111 is used as an exit cluster by @xmath293 and @xmath113 is used as an entrance cluster , we choose an edge @xmath318 from @xmath111 to @xmath113 in @xmath2 so that all these edges are disjoint from each other and from the edges chosen for the exceptional vertices ( we use ( c ) here ) . given a cluster @xmath111 , let @xmath319 be the set of all those vertices in @xmath111 which are the final vertex of an edge of @xmath2 fixed so far and let @xmath320 be the set of all those vertices in @xmath111 which are the initial vertex of an edge of @xmath2 fixed so far . so @xmath321 . let @xmath322 be the bipartite graph whose vertex classes are @xmath323 and @xmath324 and whose edges are all the edges from @xmath323 to @xmath324 in @xmath2 . since @xmath293 consists of shifted walks , it is easy to see that the vertex classes of @xmath322 have equal size . moreover , it is possible to carry out the previous steps in such a way that @xmath322 is super - regular ( here we use ( c ) again ) . this in turn means that @xmath322 has a perfect matching @xmath325 . these perfect matchings ( for all clusters @xmath111 ) together with all the edges of @xmath2 fixed so far form a @xmath54-factor @xmath326 of @xmath2 . it remains to transform @xmath326 into a hamilton cycle . we claim that for any cluster @xmath111 , we can find a perfect matching @xmath327 in @xmath322 so that if we replace @xmath325 in @xmath326 with @xmath327 , then all vertices of @xmath322 will lie on a common cycle in the new @xmath54-factor @xmath326 . to prove this claim we proceed as follows . for every @xmath328 , we move along the cycle @xmath329 of @xmath326 containing @xmath35 ( starting at @xmath35 ) and let @xmath330 be the first vertex on @xmath329 in @xmath331 . define an auxiliary digraph @xmath332 on @xmath333 such that @xmath334 . so @xmath332 is obtained by identifying each pair @xmath335 into one vertex with an edge from @xmath335 to @xmath336 if @xmath322 has an edge from @xmath330 to @xmath36 . since @xmath322 is super - regular , it follows that @xmath332 is also super - regular . by lemma [ fk ] , @xmath332 has a hamilton cycle , which clearly corresponds to a perfect matching @xmath327 in @xmath322 with the desired property . we now repeatedly apply the above claim to every cluster . since @xmath321 for each cluster @xmath111 , this ensures that all vertices which lie in clusters on the same cycle of @xmath46 will lie on the same cycle of the new @xmath54-factor @xmath326 . since by ( a ) @xmath293 visits all clusters , this in turn implies that all the non - exceptional vertices will lie in the same cycle of @xmath326 . since the exceptional vertices form an independent set in @xmath337 , it follows that @xmath337 is actually a hamilton cycle . here we give proofs of theorems [ cover ] and [ regularii ] . the proof of theorem [ cover ] uses theorems [ kellythm ] and [ expanderthm ] . * proof of theorem [ cover ] . * choose new constants @xmath338 such that @xmath339 . consider any regular tournament @xmath2 on @xmath195 vertices . apply theorem [ kellythm ] to @xmath2 in order to obtain a collection @xmath340 of at least @xmath143 edge - disjoint hamilton cycles . let @xmath46 be the undirected graph consisting of all those edges of @xmath2 which are not covered by the hamilton cycles in @xmath340 . note that @xmath46 is @xmath17-regular for some @xmath341 . by vizing s theorem the edges of @xmath46 can be coloured with at most @xmath342 colours and thus @xmath46 can be decomposed into at most @xmath343 matchings . split each of these matchings into at most @xmath344 edge - disjoint matchings , each containing at most @xmath345 edges . so altogether this yields a collection @xmath346 of at most @xmath347 matchings covering all edges of @xmath46 . it is enough to show that for each @xmath348 there exists a hamilton cycle of @xmath2 which contains all the edges in @xmath349 . so consider any @xmath348 . as observed in @xcite ( see the proof of corollary 16 there ) , any regular tournament is a robust @xmath249-outexpander . let @xmath114 be the digraph obtained from @xmath2 by ` contracting ' all the edges in @xmath349 , i.e. by successively replacing each edge @xmath350 with a vertex @xmath351 whose inneighbourhood is the inneighbourhood of @xmath38 and whose outneighbourhood is the outneighbourhood of @xmath42 . using that @xmath349 consists of at most @xmath345 edges and that @xmath352 , it is not hard to check that @xmath114 is still a robust @xmath282-outexpander and @xmath353 . so theorem [ expanderthm ] implies that @xmath114 contains a hamilton cycle , which corresponds to a hamilton cycle in @xmath2 containing all edges of @xmath349 , as required . note that we can not simply apply theorem [ main ] instead of theorem [ expanderthm ] at the end of the proof , because @xmath114 may not been an oriented graph . however , instead of using theorem [ expanderthm ] , one can also use the following result of thomassen @xcite : for every set @xmath232 of @xmath354 independent edges in a regular tournament on @xmath10 vertices , there is a hamilton cycle which contains all edges in @xmath232 . theorem [ regularii ] can be proved in a similar way , using ghouila - houri s theorem instead of theorem [ expanderthm ] . * proof of theorem [ regularii ] . * choose a new constant @xmath355 such that @xmath356 and apply theorem [ regular ] to find a collection of at least @xmath357 edge - disjoint hamilton cycles . let @xmath46 denote the subgraph of @xmath2 consisting of all edges not lying in these hamilton cycles . then @xmath46 is @xmath17-regular for some @xmath358 . choose a collection @xmath346 of matchings covering all edges of @xmath46 as in the the proof of theorem [ cover ] . so each matching consists of at most @xmath359 edges . as before , for each @xmath360 it suffices to find a hamilton cycle of @xmath2 containing all edges of @xmath349 . let @xmath361 be the digraph obtained from @xmath2 by orienting each edge in @xmath349 and replacing each edge in @xmath362 with two edges , one in each direction . let @xmath114 be the digraph obtained from @xmath361 by ` contracting ' the edges in @xmath349 as in the the proof of theorem [ cover ] . then @xmath361 has minimum semidegree at least @xmath1 and thus contains a hamilton cycle by ghouila - houri s theorem ( theorem [ hamgh ] ) . this hamilton cycle corresponds to a hamilton cycle in @xmath2 containing all edges of @xmath349 , as required . we are grateful to demetres christofides , peter keevash and andrew treglown for their comments on a draft of this paper .

we present a structural and photometric analysis of 42 elliptical / lenticular galaxies in ( e / s0 + s ) pairs observed in the color bands . the aim of the analysis is to empirically determine the effects of interactions on their morphology , structure and stellar populations as seen from the light concentration ( ) , asymmetry ( ) , and clumpiness ( ) parameters . we further compare these values to a control sample of 67 mostly isolated non - interacting e / s0 galaxies . we find that the paired e / s0 galaxies occupy a more scattered loci in space than non - interacting e / s0 s , and that the effects of interactions on e / s0 s are minor , in contrast to disk galaxies involved in interactions . this suggests that observational methods for recognizing interactions at high redshift , such the methodology of conselice ( 2003 ) , would hardly detect e / s0 s involved in interactions ( related to early phases of the so called ` dry - mergers ' ) , and that the majority of interacting galaxies identified at high redshift must be gas dominated systems . we however find statistical differences in the asymmetry index when comparing isolated and interacting e / s0s . in the mean , paired e / s0 galaxies have values times larger than the ones of non - interacting e / s0 s . for the subset of presumably strongly interacting e / s0 s , and can be several times larger than the typical values of the isolated e / s0 s . we show that the asymmetries are consistent with several internal and external morphological distortions . we conclude that the subset of interacting e / s0s should be dense , gas poor galaxies in systems spaning a wide range of interaction stages , with typical merging timescales gyr . we use the observed phenomenology of this subsample to predict the approximate loci of ` dry pre - mergers ' in the parameter space .

astro-ph0603032

understanding how elliptical / lenticular galaxies and the bulges of spiral galaxies form and evolve is still an open question in astronomy . in comparison with galactic disks , the properties of spheroids suggest violent or strong perturbative processes in their formation . there are two ways to define the formation epoch of a spheroid : when most of its stars formed or when the stellar spheroid acquired its dynamical properties in violent or secular processes ( avila - reese & firmani 1999 ) . for the monolithic collapse mechanism ( eggen , lynden - bell & sandage 1962 ; chiosi & carraro 2002 ) , both epochs coincide , as spheroids form as a result of an early violent collapse and a burst of star formation ( sf ) . however , in the context of the popular hierarchical cold dark matter ( cdm ) cosmogony , disks galaxies are generic structures and spheroids may form by the mergers of these disks ( e.g. , kauffmann , white & guiderdoni 1993 ; baugh et al . 1996 ) , or by secular dynamical evolution ( van den bosch 1998 ; avila - reese & firmani 1999,2000 ) . if major mergers occur at high redshifts when the disks are mostly gaseous , as appears to be the case ( e.g. , conselice et al . 2003 ; conselice 2006 ) , then the situation is conceptually similar to the monolithic collapse . at low redshifts most galactic disks have already transformed a large fraction of their gas into stars . in this case , spheroids may assemble by dissipationless major / minor mergers and/or by disk secular evolution . but the picture is even more complex in the hierarchical cosmogony as galaxy morphology may be continuously changing ( e.g. , steinmetz & navarro 2002 ) , depending on the mass aggregation history ( smooth accretion and violent mergers ) and environment . thus , a relevant question to answer is how the mass of early - type galaxies has been assembled . the morphological properties of nearby elliptical / lenticular ( hereafter e / s0 ) galaxies in different environments reveal important clues for understanding the nature of spheroids and their formation . the detection of fine structures , dust lanes , blue cores , nuclear disks , and peculiar kinematics in early - type galaxies , considered evidence of recent merging / accretion events ( e.g. , malin & carter 1983 ; lauer 1985 ; abraham et al . 1999 ; menanteau , abraham & ellis 2001 ; papovich et al . 2003 ; lauer et al . 2005 ) is most frequently found among galaxies in the field than in cluster members ( schweizer 1992 ; reduzzi et al . 1996 ; kuntschner et al . 2002 ) . on the other hand , while the bulk of the stars in luminous cluster ellipticals are old and coeval ( e.g. , bender et al . 1996 ) , in low density environments the colors and gradient colors of ellipticals suggest that recent bursts of star formation occurred ( menanteau et al . 2001 ; stanford et al . 2004 ; menanteau et al . blue clumps in early - type galaxies have also suggested to be evidence of recent accretion episodes ( elmegreen , elmegreen & ferguson 2005 ; pasquali et al . the last authors find that the fraction of early - type galaxies with blue clumps increases at high redshift . it is also possible that elliptical galaxies , at least the massive ones , evolve through the so - called ` dry - mergers ' -major mergers between galaxies without the presence of gas , such as ellipticals . late interactions and dissipationless mergers may have a profound effect upon the population and internal morphology of early - type galaxies ( e.g. , van dokkum 2005 ; tran et al . 2005 ; bell et al . 2005,2006 ) , yet we know very little about how these effects occur , or their timescales . the results of numerical simulations show indeed that the outcome of the fusion of early type galaxies ( dry merger ) are anisotropic , slowly rotating , boxy spheroids ( naab , khochfar & burkert 2006 ) , while low mass highly rotating , disky spheroids are produced typically in inequal mass spiral mergers ( naab et al . 1999 ; naab & burkert 2003 ) . in this paper , we will study several photometric properties of 42 e / s0 galaxies in mixed pairs , where the companion is a spiral galaxy close in luminosity to the e / s0 galaxy . these early - type galaxies ( at least the most interacting ) are affected gravitationally by the perturber companion , allowing us to explore the effects of interactions in early phases of mergers involving early - type galaxies ( dry pre - mergers ) . results have shown thus far that nearby ellipticals in pairs and groups reveal a large variety of morphological and kinematic peculiarities , such as off - centering of inner versus outer isophotes , small gaseous disks , shells , counter - rotating cores , and the occurrence of sf in the recent past ( schweizer & seitzer 1992 ; longhetti et al . 2000 ; tantalo & chiosi 2004 ) . there is also evidence that weak interactions ( e.g. , thomson 1991 ) have played a significant role in determining the final structure of spheroids . in addition to the traditional method of studying systems through their gross photometric properties , namely color , we study the structures of these galaxies through the concentration of stellar light ( @xmath1 ) , the asymmetry in the light distribution ( @xmath2 ) , and a measure of the clumpiness of light distributions ( @xmath3 ) . these three structural and morphological indices constitute the so - called @xmath4 system , which has been proposed to distinguish galaxies at different stages of evolution ( conselice 2003 , and references therein ; see also lotz , primack & madau 2004 , who present two more morphological parameters ) . several authors are using now the @xmath4 system to study galaxy evolution in photometric redshift surveys ( e.g. , moustakas et al . 2004 ; conselice , blackburne & papovich 2005 ; cassata et al . 2005 ; menanteau et al . 2006 ) . we use here the @xmath4 system to address how the structures of early - type galaxies change during an interaction . hernndez - toledo et al . ( 2005 , hereafter h2005 ) confirmed that the @xmath4 parameters are tracers of structural , morphological and sf properties of disk galaxies during an interaction . to investigate how e / s0 galaxies respond to , and evolve , during an interaction , we use a sample of e / s0 galaxies in mixed pairs ( e / s0 + s ) , which we compare to non - interacting e / s0 s and galaxies involved in ongoing gas rich mergers ( ulirgs ) . we also discuss dynamical evolutionary mechanisms that might produce the slight structural changes we find in early - type systems involved in interactions . this paper is organized as follows : section 2 summarizes the main characteristics of our paired e / s0 sample , and the non - interacting e / s0 galaxy comparison sample . a brief description of the @xmath4 parameters and the methodology for measuring them in our samples is also described . section 3 presents the measured @xmath4 parameters in the @xmath7 band for paired e / s0 galaxies , and compares their loci in @xmath4 space with three other samples , namely a reference sample of e / s0 non - interacting galaxies , an interacting / merging ulirg sample , and a sample of dwarf e / s0 galaxies . we also carry out a comparison of the @xmath4 parameters of the paired sample at different wavelenghts and explore the internal morphological properties of a subsample of the most interacting e / s0s . our main results are discussed and interpreted in 4 . section 5 gives the main conclusions of the paper . for our study we analyze a collection of 42 _ mixed pairs _ ( an elliptical / lenticular , e / s0 , galaxy + a spiral , s , galaxy ) selected from one of the most complete and homogeneous sample of interacting galaxies in the literature , the catalog of isolated pairs of galaxies in the northern hemisphere ( karachentsev 1972 ) . karachentsev ( 1972 ) used a strong pair isolation criterion in terms of the apparent angular separation and angular diameter of galaxies to create a list of @xmath8 non - merging , but mostly _ dynamically bounded _ pairs . binaries selected in this way show a wide range of separations : for @xmath9 of them , the projected separation does not exceed the sum of the diameters of the member galaxies , and only @xmath10 of the systems have a separation greater than @xmath11 @xmath12kpc ( but less than @xmath13 @xmath12kpc ) . for our sample of 42 mixed pairs , the mean projected separation is @xmath14 @xmath15kpc , showing a wide range of morphological types , and interaction - induced features ( franco - balderas et al . 2003 , 2004 , 2005 ) . since our sample is restricted to isolated environments only the intrinsic properties of individual galaxies , and the effects of mutual interactions between pair components , should affect the structures of these galaxies . the sample was observed in the johnson - cousins @xmath0 photometric system with the only criterion of availability of time and good weather conditions . the images were acquired using two telescopes : the observatorio astronmico nacional ( oan ) 1.5 meter and 0.84 meter telescopes at san pedro mrtir , baja california , in mxico , and under reasonable seeing conditions ( @xmath16 ) . the data are sensitive enough to detect faint stellar tidal structures and at a typical size of @xmath17 arcsec / galaxy , an average of 30 resolution elements / galaxy are reasonable enough to avoid underestimations of the @xmath4 parameters . a more detailed description of the selection criteria , completeness and global optical emission properties can be found in hernndez - toledo et al . we show a mosaic of four representative e / s0+s pairs in the @xmath18 band to illustrate the range of apparent separations and galaxy morphologies in the sample ( figure 1 ) . as a reference sample of _ non - interacting _ e / s0 galaxies , we use images from three different sources : frei et al . ( 1996 ) , colbert et al . ( 2001 ) , and hernndez - toledo & salazar - contreras ( 2006 ) . the frei et al . ( 1996 ) sample consists mainly of bright , high surface brightness galaxies of all morphological types in the field and in nearby clusters of galaxies . the images were acquired in two telescopes , the lowell 1.1 meter and the palomar 1.5 meter , with a typical resolution of @xmath19 . from the 25 e / s0 galaxies in the frei et al . sample , we have selected 18 galaxies ( mostly isolated ) in the @xmath7 band with no companions within a circle of 30 arcmin radius around ( projected on the sky ) . the colbert et al . ( 2001 ) sample is a compilation of 19 isolated and poor - group ( non - interacting ) early - type galaxies from the third reference catalogue of bright galaxies ( rc3 , de vaucouleurs et al . e / s0 galaxies were selected with no catalogued galaxies with known redshifts within a projected radius of @xmath20 mpc and a velocity of @xmath21 . the @xmath22band images were obtained at the las campanas 1 m telescope under photometric conditions and a typical 1.5 arcsec seeing . the hernndez - toledo & salazar - contreras ( 2006 ) sample is a set of 40 isolated e / s0 galaxies extracted from the hyperleda database , recently observed in the johnson - cousins @xmath0 photometric system at the 0.84 m telescope of the oan - spm observatory with a typical resolution of @xmath23 . the basic selection criteria were : not belonging to any catalogued rich group and cluster of galaxies , not having companions in a neighborhood of 0.5 mpc ( projected on the sky ) and relative redshift lower than 600 km / s . after inspecting each galaxy in sky survey images for significant companions in a 30 arcsec radius ( projected on the sky ) a final sample of @xmath24 isolated e / s0 galaxies was considered for the analysis . the final sample of non - interacting ( mostly isolated ) e / s0 galaxies amounts to 67 galaxies : 18 from frei et al . ( 1996 ) , 19 from colbert et al . ( 2001 ) , and 30 from hernndez - toledo & salazar - contreras ( 2006 ) . a typical size of @xmath25 arcsec / galaxy for all the e / s0 s in this final reference sample renders an average number of 20 - 30 resolution elements / galaxy , enough to avoid underestimations of the @xmath4 parameters ( see conselice et al . homogeneity of the @xmath4 estimates for all the e / s0 galaxies in this study is guaranteed from the procedure described below . as a comparison sample of galaxies involved in gas rich major - mergers we use images of ultra luminous infrared galaxies ( ulirgs ) . ulirgs are thought to be galaxies involved in extreme interactions , either during or after the merging process ( borne et al . 2000 ; canalizo & stockton 2001 ) . several groups obtained hubble space telescope ( @xmath26 ) images of ulirgs in the f814w ( hereafter @xmath27 ) and f555w ( hereafter @xmath28 ) bands ; 66 of which have their quantitative @xmath4 structural parameters presented in conselice ( 2003 ; hereafter c2003 ) . alternative descriptions of these galaxies can be found in ferrah et al . ( 2001 ) . to properly compare the @xmath4 values with the @xmath22band images used in the comparison sample , a quantitative morphological @xmath29correction to estimate the @xmath7-band value for each morphological index was applied ( c2003 ) . considering a typical fwhm of @xmath30 and a size of @xmath31 arcsec / ulirg , an average number of 90 resolution elements / ulirg avoids underestimations of their @xmath4 parameters . finally , we include in the comparative analysis the @xmath4 parameters for a sample of 19 dwarf elliptical galaxies imaged in the wiyn telescope with the s2 kb detector in the @xmath7 band . the observational details for these galaxies are described in conselice et al . ( 2002 ) . the main comparison presented in this paper is between the e / s0 s in pairs and the non - interacting e / s0 galaxy sample . to avoid inconsistencies that may compromise the direct comparison of these samples , we test whether their blue absolute magnitude and morphological type distributions are similar ( fig . 2 ) . a kolmogorov - smirnov ( k - s ) test indicates that the hypothesis that the morphological type and magnitude distributions of both samples come from the same parental population can not be significantly discarded ( at a significant level of 0.96 and 0.98 respectively ) . notice however that the sudden cut at the low luminosity end in mixed pairs is real , and has been interpreted as a morphology - density relation for sub - dwarf / dwarf ellipticals ( @xmath32 ) in the field ( hernndez - toledo et al . early - type morphologies are equally represented in the e / s0 galaxies in pairs and in the reference e / s0 sample . we conclude that the distributions of absolute magnitude and morphological - type are similar enough to allow a fair comparison between both samples in the @xmath4 space . we briefly review each one of the @xmath4 parameters and discuss the reliability of measuring them on our samples of isolated and interacting e / so galaxies . * concentration of light @xmath1 : * the concentration index @xmath1 is defined as the ratio of the 80% to 20% curve of growth radii ( @xmath33 , @xmath34 ) , within 1.5 times the petrosian inverted radius at r(@xmath35 ) ( @xmath36 ) normalized by a logarithm : @xmath37 . for a detailed description of how this parameter is computed see bershady et al . ( 2000 ) , conselice et al . ( 2002 ) , and c2003 . the concentration of light is related to the galaxy light ( or stellar mass ) distributions . low ( high ) concentrations are expected for extended ( compact ) galaxies ( bershady et al . 2000 ; graham et al . 2001 ; c2003 ) . * asymmetry @xmath2 : * the asymmetry index is the number computed when a galaxy is rotated @xmath38 from its center and then subtracted from its pre - rotated image , and the summation of the intensities of the absolute value residuals of this subtraction are compared with the original galaxy flux . this parameter is also measured within @xmath36 . for a full description see conselice et al . ( 2000a , b ) . the @xmath2 index is sensitive to any feature that produces asymmetric light distributions . this includes galaxy interactions / mergers , large star - forming regions , and projection effects such as dust lanes ( conselice 1997 ; conselice et al . 2000a ) . * clumpiness @xmath3 : * galaxies undergoing sf are very patchy and contain large amounts of light at high spatial frequency . to quantify this , the clumpiness index @xmath3 is defined as the ratio of the amount of light contained in high frequency structures to the total amount of light in the galaxy within @xmath36 ( c2003 ) . the @xmath3 parameter , because of its morphological nature , is sensitive to dust lanes and inclination ( c2003 ) . * sep parameter : * we express the apparent projected separation in paired galaxies ( @xmath39 ) in units of the primary component diameter ( @xmath40 ) where @xmath40 is the 25-mag / arcsec@xmath41 isophote diameter in the @xmath42band . thus , a quantity @xmath43 is defined . we sort our ( e / s0+s ) sample into wide ( @xmath44 ) and close ( @xmath45 ) pairs . light contamination effects are expected in paired galaxies of similar diameters , with @xmath46 or in paired galaxies with different diameters and @xmath47 . if the light of the companion enters within the petrosian radius @xmath48 of a given galaxy , the observed @xmath48 could be biased to a larger value , depending on the type of deformation induced to a `` pre - contaminated '' light profile . this affects the measured values of the @xmath4 parameters as shown in h2005 . to avoid biases from this effect , the e / s0 galaxies in the closest mixed pairs were `` decontaminated '' following a correcting procedure described below . the @xmath4 parameters of the ulirg and dwarf elliptical samples ( in the @xmath7 band ) were taken from conselice ( 2003 ; 2003a ) . * measurement of @xmath4 parameters . * the measurement of the @xmath4 parameters for the e / s0 galaxies was carried out in several steps : ( i ) close field and overlapping stars were removed from each image ; ( ii ) sky background was removed from the images by fitting a polynomial function that yielded the lowest residual after subtraction ; ( iii ) the center of each galaxy was considered as the barycenter of the light distribution and the starting point for measurements ( conselice 2003 ) ; ( iv ) the @xmath4 parameters for paired e / s0 s with @xmath44 , as well as with @xmath45 but where the companion galaxy is beyond 2 times the @xmath40 of the primary , were estimated directly , i.e. individual components were not considered to be influenced by light contamination from the companion ; ( vi ) closest pairs ( those with @xmath45 and where the companion galaxy is within 1.5 times the @xmath40 of the primary ) are considered as light - contaminated by the companion . in these cases , a model of the pair component @xmath49 , called _ model a _ is first built by using the task bmodel in iraf . then we subtract this model from the original image , creating an _ image 2_. from _ image 2 _ , we estimated @xmath4 parameters for component @xmath50 . this same procedure is applied to component @xmath50 of each pair to estimate @xmath4 for component @xmath49 . bmodel creates a model of the light distribution in a galaxy by taking into account all the information from an isophotal analysis . the resultant image , after subtracting a model galaxy to the original galaxy , yielded in most of the cases , traces of underlying structures . in such cases , these remaining features were masked or interpolated before measuring the @xmath4 parameters . in a few cases , the overlapping degree makes it difficult to apply this procedure and thus we caution the reader by marking these pairs in our analysis . * intrinsic ellipticity and/or inclination effects . * galaxies with high inclinations or axis ratios could introduce systematic biased trends in the values of the @xmath4 parameters ( c2003 ) , but usually only for disk galaxies . in the case of e / s0 galaxies the diversity of apparent axial ratios is therefore not expected to cause a strong effect in the @xmath4 estimates , in particular in the earliest types , where these ratios are intrinsic rather than geometrical . in any case , we have evaluated the influence of the apparent axial ratios on the estimated parameters by adopting a measure of the `` inclination to the line - of - sight '' for e / s0 s as outlined in paturel et al . no significant trend of @xmath4 parameters with axis ratio for both the non - interacting and paired samples were found . notice that paired galaxies whose apparent axial ratios yield `` inclinations '' larger than @xmath51 will be marked with a skeletal circle on the following plots . the @xmath4 parameters in the @xmath18 , @xmath28 , @xmath7 and @xmath27 bands for our sample of 42 e / s0 galaxies in mixed pairs were estimated and then analyzed for any relative trend . the median and average of the @xmath4 parameters for the e / s0 components show marginal wavelength differences among the different bands . both the kolmogorov - smirnov ( k - s ) test and a conventional statistics for measuring the significance of a difference of means ( student s @xmath52test ) between the @xmath18 and @xmath27 band @xmath4 parameters show no significant differences . after confirming that the corresponding variances are not significantly different , a paired student s @xmath52test , which takes into account point - by - point effects in the compared samples , indicates that the @xmath1 and @xmath2 parameters could be marginally different at the 93% level in the sense that the e / s0 components tend to be more concentrated and less asymmetric in the @xmath27 band than in the @xmath18 band . such a trend is more marked in the non - interacting e / s0 s , at least for the 30 e / s0 galaxies compiled from the hyperleda database and observed by us ( see 2 ) . conselice ( 1997 ) and conselice et al . ( 2000a ) have also reported a similar difference in bands for the frei+ sample . no differences with band were found in the @xmath3 parameter in any of our tests . in the following , the @xmath7 photometric band is adopted to perform our comparative @xmath4 analysis . table 1 shows our estimate of the @xmath7 band @xmath4 parameters for the e / s0 galaxies in mixed pairs . the @xmath4 parameters for the isolated e / s0 galaxies will be presented along with their @xmath0 photometric properties in a forthcoming paper ( hernndez - toledo & salazar - contreras 2006 ) . in this section we perform a comparative analysis of the @xmath4 parameters for the paired and non - interacting ( mostly isolated ) e / s0 galaxies . although both samples are in no ways complete , this comparison allows us to identify possible systematic differences between interacting and non - interacting e / s0 galaxies . we further interpret the properties of the more interacting e / s0s as close to the ones of early - type galaxies in a phase of dry pre - merger . table 2 shows the median and the lower ( 25% ) and upper ( 75% ) quartiles of the @xmath4 parameters for e / s0 galaxies in pairs and for the sample of non - interacting e / s0 galaxies . given the shape of the distributions of @xmath4 parameters , we adopt the medians and quartiles as more representative of the distributions . table 2 also includes the results for the subsample of paired e / s0 galaxies in close ( @xmath45 ) pairs . in fig . 3 we show cumulative fractions of the @xmath4 paramaters for the paired and non - interacting e / s0 s samples in the @xmath7 band . the major differences between both samples are for the asymmetry parameter . for @xmath53 of the galaxies in each sample , @xmath2 is on average @xmath54 times larger in paired galaxies than in the non - interacting , while for the remaining @xmath55 galaxies with the highest @xmath2 values , the difference increases even more . according to table 2 , the median @xmath2 value of paired e / s0 s is @xmath56 times larger than the corresponding median of the non - interacting e / s0 s . for the averages and their standard deviations , the diference is @xmath57 . a student s @xmath52test confirms the statistical significance of the difference between the averages at the 99.5% level , while a k - s test also shows different distributions at a significance level of @xmath58 . furthermore , non - parametric two - sample tests , namely two versions of the wilcoxon test ( gehans generalized and peto & peto ) , and the logrank test , reinforce the significance of this result with a probability of 0.0001 . the statistical significant differences in the @xmath2 values of paired and non - intearcting e / s0 galaxies found here imply that interactions do induce systematically asymmetries in the structure of these galaxies . however , the level of these asymmetries is typically moderate . the median and average concentration values of the e / s0 galaxies in pairs are similar to those of the e / s0 non - interacting galaxies . a student s @xmath52test shows that the difference between both samples is not statistically significant and a k - s test also shows no different distributions at a significance level of @xmath59 . the non - parametric wilcoxon and logrank two - sample tests also confirm this result . notice however a slight difference for the subset with the highest @xmath1 values ( @xmath60 ) in both samples : @xmath1 is slightly larger in paired e / s0 s than in non - interacting e / s0 s ( fig . it seems that interactions do not seriously affect the concentration of spheroids . however , a possibility is that the concentration could be affected , depending on the dynamical stage of the interaction ( see 4.1 ) , such that in some cases @xmath1 increases and in another ones @xmath1 decreases , with negligible neat differences . the clumpiness values of e / s0 galaxies in pairs are small ( @xmath61 ) , in spite of the interaction . the cumulative distributions of @xmath3 for both , the paired and non - interacting e / s0 samples , show that @xmath3 for the former is in most of cases similar or larger than for the latter ( fig . 3 ) ; the differences in the medians is of a factor of 1.92 . however , statistical tests do not show evidence for any significant difference between both samples . the non - parametric two - sample tests also confirm this result . thus , interactions , on average , seem not to systematically increase sf activity in e / s0 galaxies , likely because there is little gas available . nevertheless , for the subsets of the @xmath62 fraction of galaxies with the largest @xmath3 parameter in both samples , @xmath3 becomes significantly higher in the paired e / s0 s than in the non - interacting ones , suggesting possible cross - fueling of gas and induced sf activity in these cases . recent studies have shown that a high fraction of e / s0 s in mixed morphological pairs show signatures of starbursts , agn activity and emission excess in the 20 cm continuum , which also points towards cross - fueling of gas from the spiral components ( domingue et al . 2003 ; 2005 ) . figure 4 shows the @xmath4 parameters of the paired e / s0 sample versus the apparent separation @xmath63 ( left panel ) and versus the ratio of the @xmath27 band absolute magnitudes of the e / s0 galaxy and its @xmath3 companion , @xmath64 ( right panel ) . filled circles are for systems with apparent separation @xmath45 . according to fig . 4 , there is no significant correlation of the @xmath4 parameters with @xmath63 or the @xmath64 ratio . in more detail , one sees that most galaxies with the highest values of @xmath2 have @xmath45 and@xmath65or @xmath66 . in fig . 5 the loci of paired e / s0 galaxies in the @xmath22band @xmath67 , @xmath68 , and @xmath69 planes of the @xmath4 space are shown . the data are sorted into two groups : close ( @xmath45 , circles ) and wide ( @xmath44 , squares ) paired galaxies . for reference , the mean projected separations of the former and latter subsets are @xmath70kpc and @xmath71kpc , respectively . the solid circles represent decontaminated @xmath4 values for the eight closest pairs after applying the procedure described in 2.5 and h2005 . the lenticulars are marked with a small horizontal line crossing the corresponding symbol . galaxies with `` inclinations '' greater than @xmath72 are marked with a cross . for comparison , the @xmath4 values of the non - interacting sample are plotted as crosses . additional error bars correspond to the average and @xmath73 dispersion of the @xmath4 parameters of ulirgs and dwarf e / s0 s , from conselice ( 2003 ; 2003a ) . the loci in the @xmath4 space of most e / s0 s in pairs coincide roughly with that of the non - interacting e / s0 galaxies , but the former are much more scattered in this space than the latter ( fig . 5 ) . for a subset of the paired e / s0 s with the highest asymmetries , a weak trend is seen such that as @xmath2 is larger , @xmath1 is larger . notice that most of these galaxies had to be decontaminated by the presence of a close companion ( see 2.5 and h2005 ) . in the @xmath69 plane , a trend of increasing @xmath3 as @xmath1 decreases is observed for the non - interacting e / s0 s ( fig . the paired e / s0 s also show this tendency although with more scatter and with a subset shifted to high @xmath3 values . most of the galaxies of this subset are s0s . furthermore , in the @xmath68 plane , the weak trend of increasing @xmath2 as @xmath3 increases is seen for both non - interacting and paired galaxies . for the paired e / s0 s with the highest @xmath2 values ( mainly e s ) , the values of @xmath3 are not as high as one would expect by extrapolating the trend of the rest of the galaxy sample . a view of the @xmath4 planes reveals a subsample of e / s0s in pairs whose behavior clearly departs from the rest of the paired e / s0s ( and of the non - interacting ones ) . these galaxies have typically @xmath2 and @xmath3 values larger than the average or median ( see also 3.2 and fig . 4 ) . the regions enclosed in solid lines in fig . 5 are an attempt to represent the approximate loci of these galaxies in the @xmath4 planes . in 4.2 we explore more closely the morphological appearance of these galaxies and show that they have features consistent with high values of @xmath2 and @xmath3 parameters . finally , we can see in fig . 5 that the @xmath4 parameters of interacting e / s0 s are far from those of the ulirg and dwarf e samples . ulirgs imply mergers with a significant fraction of gas , more easily detected through their larger asymmetry and clumpiness ( e.g. , c2003 ; lotz et al . 2004 ) . figure 6 shows the @xmath4 parameters vs. corrected absolute @xmath42band magnitude m@xmath74 , @xmath75 color index , central velocity dispersion @xmath76 in logarithmic scale , and the continuum 20 cm of the luminosity l@xmath77 also in logarithmic scale , for paired e / s0 ( open symbols ) and non - interacting e / s0 sample ( crosses ) galaxies . color data for the paired galaxies were taken from our photometric studies ( franco - balderas et al . 2003 , 2004 , 2005 ) , while absolute magnitudes and velocity dispersions were taken from the hyperleda database . magnitudes , colors , and velocity dispersions for the reference sample were also taken from hyperleda database , while @xmath78 luminosity was estimated from data compiled from the nrao / vla sky survey at 20 cm ( nvss ) for both samples . for both the paired and non - interacting e / s0 s samples there are no significant correlations between @xmath4 parameters and m@xmath74 , @xmath75 , @xmath76 , and l@xmath77 . the loci of paired e / s0 s galaxies in all the diagrams of fig . 6 shows again more scatter than that of non - interacting galaxies but clear and systematic differences are not appreciated , except for the @xmath4 vs @xmath75 color plots . while no non - interacting e / s0 galaxies have @xmath79 , a high fraction of the most asymmetric and clumped paired e / s0 galaxies are bluer than @xmath80 . in general , paired e / s0 s contain a larger distribution of colors ( @xmath81 ) than the non - interacting e / s0 s ( @xmath82 ) , and on average the former are slightly bluer than the latter ( @xmath83 , and @xmath84 , respectively ) , showing that sf in the recent past could likely be induced to some level by gas accreted from the spiral companion . this conclusion is somehow reinforced by the surprising nvss radio detection fraction found in the e / s0 components in our sample of mixed pairs ( see last panel of figure 6 and domingue et al . 2005 ) suggesting that the gas giving rise to this effect may be acquired from the gas rich companion . if e / s0 galaxies in mixed pairs originate in a similar way as the isolated e / s0 ones and therefore have similar initial @xmath4 parameters , then it is important to understand how , as a consequence of the interaction , @xmath2 becomes larger at least for a subset of the paired e / s0 s . these galaxies may well represent the properties of early - type galaxies in the phase of `` dry '' pre - merging . numerical simulations of e+e interpenetrating encounters have shown that @xmath85 deformations are common when the pericenter is comparable or less than the characteristic size of the system . at large interacting distances the @xmath85 term disappears and the bi - symmetric ( @xmath86 ) term dominates ( combes et al . the signature of @xmath85 deformations include off - centering of inner isophotes compared to the outer ones , which on its own will cause an asymmetry signature . the numerical simulations also show that the intermediate and outer envelopes remain off - centered or asymmetric for only a few @xmath87 years after the pericenter passage ( combes et al . 1995 ; see also aguilar & white 1985 ) . although very noisy , a loose trend of decreasing @xmath2 as @xmath63 is larger ( fig . 4 ) is observed , possibly reflecting the short time scales of @xmath85 deformations . it could be expected that collisionless spheroids expand during interactions as orbital energy of the perturber is transferred to internal energy of the spheroid ( e.g. , aguilar & white 1985 ; namboodiri 1995 ; vergne & muzzio 1995 ; evstigneeva , reshetnikov & sotnikova 2002 ) . however , as all these authors show , the process is much more complex and depends on several initial and physical conditions of the pair , for example the mass ratio of the intervening galaxies , orbital parameters , strength of the interaction , and initial structure of the components . furthermore , the shape of the surface density profile is expected to change during the interaction , reaching its equilibrium state after several crossing times . in this scenario , a unique global parameter like @xmath1 will not be sensitive enough to describe all ( global and local ) structural changes and transient processes associated with the interaction . we have found that the concentration index tends to be the same in paired e / s0 galaxies as in the non - interacting e / s0 s , but with a larger dispersion for the e / s0 sample , showing this the diversity of dynamical evolutionary stages through which the pairs are observed . this is consistent with results from several numerical simulations . in these simulations there are cases when the galaxy s surface density profile becomes more compact . simulations of spheroids with density profiles given by the de vaucouleurs models that are tidally perturbed by encounters show that strong collisions produce a final shrinkage in the effective radius and a brightening in the effective surface brightness , whereas weak collisions have the opposite effect ( aguilar & white 1985 ) . however , during early stages of the interaction the profiles flatten externally and deform . aguilar & white ( 1985 ) conclude that the effects of tidal encounters in the luminosity profiles can only be recognized immediately after a close passage . on the other hand , we caution that the concentration index does not describe in detail the diversity of luminosity distributions . it could be that in some cases the expansion is efficient only in the outer , less bounded regions , in such a way that @xmath88 increases but @xmath89 remains almost constant or even decreases ( e.g. , see namboodiri 1995 for head - on e+e simulations ) . therefore , the concentration index will `` formally '' increase in spite of the profile shows an outer expansion . the @xmath4 parameters of e / s0 s do not change much during an interaction , in comparison to interacting disk galaxies . the structures of spheroid dominated galaxies also show less distortion during a merger than a pure disk system in n - body models ( conselice 2006 ) . the small observed change of e / s0 galaxy properties with interaction agrees with the fact that early - type galaxies are typically dense , concentrated , dynamically hot structures with low gas fractions which are slowly rotating . furthermore , the interacting s0 galaxies have @xmath4 values closer to the non - interacting ellipticals rather than to the interacting spirals , implying thick , hot and gas - poor disks , which do not strongly react to the interaction . finally , we consider the possibility that the differences seen between e / s0s in mixed pairs and non - interacting e / s0 s are due to differences in their formation processes rather than to the interaction as disscused above . the e / s0 s in pairs could originate from disks ( sa - sb galaxies ) by secular mechanisms amplified by interactions . however , the interacting spirals have @xmath4 parameters ( see h2005 ) that are very different than those in the paired e / s0 galaxies ; unless the transition is very quick , which is unlikely , this result does not support the secular mechanism of e / s0 formation in pairs . in the secular scenario it is also expected that the formed pseudo - spheroid will have a significant level of rotation . numerical simulations show that the level of asymmetry , including off - centering , in spheroids is much more pronounced if they are rotating ( combes et al . thus , the relative low level of asymmetry measured in the paired e / s0 galaxies suggests that the rotation of these galaxies is probably slow , and thus do not originate from latter hubble types . conselice ( 2003 ) , and more recently h2005 , have inferred statistical criteria for identifying galaxy major pre - mergers based on their @xmath4 parameters . thus , interacting disk galaxies can be identified by their @xmath4 values in an automated way in high - redshift samples . it would be useful to have the analogous criteria for finding pre - merging e / s0 s , and whether they meet the criteria for interaction established in c2003 , and conselice et al . ( 2000a , b ) . for our sample of e / s0 galaxies in mixed pairs , we have found some evidence of systematically higher asymmetries than in the non - interacting e / s0 galaxies . in particular , for the apparently most interacting e / s0 galaxies , the @xmath2 parameter can be several times larger than the typical @xmath2 value of non - interacting early - type galaxies . however , even in these cases , the value of @xmath2 is not larger than 0.35 , which was suggested as a rough criterion of ongoing merger or interaction for disk galaxies ( c2003 ; h2005 ) . this indicates that the structures of e / s0 s are very robust to changes due to interactions because of their high stellar mass density and likely dark matter concentrations , and/or a lack of gas content and slow rotation . this is an indication that mergers between evolved galaxies without significant gas , the so - called ` dry mergers ' , would be underestimated in merger studies that rely on morphology . studies that use pair techniques to find mergers ( e.g. , bundy et al . 2004 ; lin et al . 2004 ) might be necessary to fully account for these systems . the introduction of more morphological indicators , as the gini coefficient and @xmath90 , seem to be also of utility ( lotz et al . 2004,2006 ) . as a result , it is very likely that the progenitors of mergers found at high redshift are gas - rich disk galaxies rather than ellipticals ( conselice et al . 2003a , b ) . based on our results , we can develop a loose structural criteria for finding early - type galaxies perturbed by a companion of nearly similar luminosity ( mass ) . in our non - interacting sample of e / s0 s , approximately @xmath91 have asymmetries smaller than 0.05 , while for our paired e / s0 sample , @xmath92 of the galaxies have @xmath2 values smaller than 0.05 . all of these galaxies are in pairs with morphological indices that reflect weak evidence of tidal distortions . one may therefore roughly define @xmath93 as the transition value from non - interacting to interacting e / s0 galaxies , although reaching this level of accuracy in the asymmetry index at high redshift is difficult ( conselice et al . 2000a ) . since there is some trend of @xmath2 with @xmath3 for both the paired and non - interacting e / s0 galaxies ( fig . 5 ) , it is important to explore the deviations of paired from non - interacting galaxies in the @xmath68 plane . we define the normalized deviation @xmath94 as @xmath95 , where the numerator is the residual of the observed @xmath2 value of a paired e / s0 galaxy with respect to a linear bisector fitting to the sample of non - interacting e / s0 galaxies in the @xmath68 plane , @xmath96 , and @xmath97 is the variance of the fitting . the deviations of paired e / s0 s from the non - interacting ones in the @xmath68 plane can be as large as @xmath98 . in fig . 7 we plot @xmath94 vs @xmath2 for our e / s0 pair sample . most galaxies have @xmath99 , and @xmath94 increases as @xmath2 is higher , i.e. the @xmath3 parameter of paired e / s0 s increases with @xmath2 at a slower rate than in the case of the non - interacting e / s0 s . this is evidence that the sf does not increase significantly even for galaxies with relatively large asymmetries , i.e. galaxies with significant interaction . in fig . 7 galaxies are coded according to the interaction indices introduced by karachentsev ( 1972 ) . the index assignment is based on a visual inspection of morphological characteristics of the system components . the @xmath100 index is designated for pairs with components in a common luminous halo with an amorphous , shredded , or asymmetric structure . @xmath101 pairs show evidence of tidal bridges , tails or both in discernible components . @xmath102 pairs show evidence of structural distortion in one or both of the separated components . finally , a @xmath103 index is introduced for wide pairs with no obvious morphological distortions . the order @xmath104 has been suggested ( hernndez - toledo et al . 2001 ) as a sequence from strongest to weakest evidence for tidal distortion in the pair or , alternatively , most to least dynamically evolved . from fig . 7 one infers that a statistical criterion for finding presumably interacting e / s0 galaxies might be @xmath105 and @xmath106 , the former being the stronger criterion . the galaxies below these criteria belong to pairs with weak evidence of interaction ( interaction indexes @xmath102 and @xmath103 , and @xmath44 ) . from an inspection of fig . 5 , one might complement the above criteria of presumably interacting early - type galaxies with the extra conditions @xmath107 ( valid also for the non - interacting e / s0 s ) , and @xmath108 ( see regions enclosed by solid lines in fig . 5 ) . we further analyze the subsample of e / s0 galaxies with @xmath109 to look for the presence of any internal / external morphological signatures . notice that some of them are identified as e / s0s in close pairs ( @xmath110 ) and their light had to be decontaminated by the presence of the spiral companion ( see 2.5 ) . observations of close encounters involving an early - type galaxy reveal that morphological signatures of interactions are often weak ; tidal tails are broad and of low surface brightness . our ability to detect low surface brightness features is limited by the extent of the features , confusion with neighboring objects , contrast of the image and other effects . we proceed to look for the presence of tidal external signatures by analyzing our imges via unsharp masking to the complete image frame ( including the spiral companion ) . next , the light distribution in each e / s0 companion is modelled by using the galfit package ( peng et al . this elliptical model is then subtracted from the original image to look for internal structure in a residual image . additionally , a @xmath111 color - index map is obtained to complement our search of tidal / localized features . figure 10 illustrate the results of our image analysis for 3 representative galaxies from the subsample of the most interacting e / s0 galaxies with @xmath109 . for the subsample of the most interacting e / s0 galaxies with @xmath109 , 5 e / s0s show clear central elongated or disky structures , 6 e / s0s show 1 or 2 localized clumps along the face of the galaxy , and 3 e / s0s show evidence of shell - like structures . these results are consistent with recent studies of local isolated ellipticals surveyed for internal structures ( colbert et al . 2001 ; reda et al . 2004 ) and elliptical galaxies in the ultra deep field ( elmegreen et al . 2005 ; pasquali et al . 2006 ) . whether these structures are produced by the interaction or not , the presence of central disks ( in a few cases ) that might react to the interaction as well as the presence of localized clumps out of the central regions could lead to a formal increase of the asymmetry parameter . this is further confirmed by viewing the residual map from the asymmetry computation . the cases with localized blue clumps are probably related to enhanced sf in a particular region of the galaxy . however , this can not be associated with a global increase of the sf activity , as typically happens in interacting spiral galaxies , where the @xmath3 parameter is systematically enhanced overall the galaxy image . thus , from our preliminary image analysis of the subset of presumably interacting e / s0 galaxies , we conclude that the observed internal / external structures deviced in these galaxies are consistent with their observed levels of asymmetry and clumpiness . if the above morphological features are interpreted as signatures of past and ongoing interactions in these pairs , it is important then to have an estimate of their timescale of merging . following patton et al . ( 2000 ) , we assume that the timescale for merging of inequal mass galaxies can be approximated by the dynamical friction timescale . we use the mean values of the @xmath42band magnitude , projected separation ( it gives a lower limit to the physical separation @xmath112 ) , and relative velocity of the subsample of 17 e / s0 s with @xmath113 : @xmath114 , 22 kpc , and 125 km / s , respectively . thus , assuming @xmath115 for these e / s0 galaxies , the average dynamical friction ( merging ) timescale of the subsample is @xmath116 gyr , which is relatively short . for nearly - equal mass pairs , an alternative estimate for the merger timescale is provided by just the circular orbit timescale , @xmath117 , where @xmath118 is the galaxy circular velocity , related to the effective velocity dispersion by @xmath119 . this effective velocity dispersion is smaller than the central one , @xmath76 . assuming an effective velocity disspersion @xmath120km / s for the subsample of highly assymetric e / s0 s , we estimate an average merger timescale of @xmath121 gyr . on the ground of semi - analytic models and simulations , it was proposed that the local population of ellipticals assembled a significant fraction of their masses via late major ( `` dry '' ) mergers ( kauffmann & haehnelt 2000 ; khochfar & burkert 2003 ; domnguez - tenreiro et al . non - dissipative mergers seem to be the main mechanism to produce the present - day massive anisotropic , slowly rotating and boxy ellipticals ( naab et al . recently , from an analysis of tidal features associated with bright red galaxies , van dokkum ( 2005 ; see also tran et al . 2005 ) has found that @xmath122 of early - type galaxies experienced a `` dry '' major merger involving more than 20% of its final mass . the e / s0 s in mixed pairs analyzed here , at least those with clear signatures of interactions , may serve as local examples of the population of early - type galaxies in a previous stage to a major merger at higher redshifts . thus , according to our results , the @xmath4 criteria of pre - merger precursors of bright ellipticals should be @xmath113 , and @xmath108 ( see above and fig . 5 ) . we note that several of the interacting e / s0 s have colors slightly bluer than the non - interacting e / s0 s ( 3.4 and fig . 6 ) . finally , in fig . 9 we ressume how the classification of various galaxies is envisaged from the point of view of @xmath4 planes . the average and @xmath73 dispersion @xmath4 values of non - interacting isolated ( crossed bars ) and interacting ( circled bars ) galaxies in different morphological - type ranges are shown . for the interacting galaxies , we use the subsets of paired e / s0 s ( this paper ) , and paired sasb s and scsm s ( from h2005 ) that obey the corresponding interaction criteria . for completness , we also plot the data for the ulir , starburst , and dwarf e galaxy samples ( dotted error bars , see c2003 ) . different galactic environments ( from mostly isolated to mergers ) can be differentiated in this structural - morphological space . if the @xmath4 parameters can be measured reliably at various s / n ratios , resolutions and redshifts , this diagram can be a useful tool for exploring different galaxy populations at larger redshifts . we have analyzed 42 images in the optical @xmath0 bands of e / s0 galaxies in mixed morphology pairs selected from the karachentsev ( 1972 ) catalogue . we measured concentration ( @xmath1 ) , asymmetry ( @xmath2 ) and clumpiness ( @xmath3 ) ( @xmath4 ) parameters on these images using the c2003 methodology . our goal was to explore the effects of interactions on paired e / s0 galaxies by comparing their structural parameters to those of mostly isolated non - interacting e / s0 s ( our isolated sample comprissed 67 galaxies , half of them observed by us ) . our main results are : \1 . the difference in the means of the asymmetry parameter @xmath2 between paired and non - interacting e / s0 s is statistically significant , their ratio being @xmath5 . for the @xmath123 subset of galaxies with the highest asymmetries in each sample , @xmath2 can be several times larger for the paired e / s0 s with respect to the non - interacting ones . the concentration parameter @xmath1 of both samples is statistically similar . for the subsets with the highest @xmath1 values in both samples , the @xmath1 values are slightly larger for the paired e / s0s than for the non - interacting ones . for the @xmath3 parameter , the the median @xmath3 value of the former is @xmath19 times larger than the median of the latter , but the difference is not statistically significant . the paired e / s0 galaxies occupy a more scattered loci in @xmath4 space than non - interacting e / s0 s , but still quite narrow with respect to paired disk galaxies . there is a subset of paired e / s0 s ( presumably the most interacting ones ) that deviate in the @xmath4 planes from the rest of paired e / s0 and non - interacting galaxies in the direction of higher values of @xmath2 and @xmath3 . there are no significant correlations between the @xmath4 parameters and magnitude , total @xmath75 color , central velocity dispersion , and 20 cm continuum luminosity for both paired and non - interacting e / s0 s . the scatter in all of these diagrams are larger for the former than for the latter , but there are no systematic differences , with the exception of the @xmath4 vs @xmath75 color diagrams , where a subset of blue paired e / s0 s clearly appears . our results indicate that early - type galaxies in a pre - merging phase would not be easily recognized through standard interaction / merger criteria that are suitable for gas rich galaxies . we find that a rough statistical criterion for finding presumably pre - merging e / s0 galaxies might be : @xmath113 ( alternativley , @xmath124 ) and @xmath108 . for the e / s0 s galaxies in our sample of mixed pairs that obey these criteria , an image analysis show indeed morphological evidence of interaction . a very rough estimate of the average dynamical friction ( major merging ) timescale of these pairs is @xmath6 gyr . @xmath125 interactions do not produce significant changes in the morphological / structural / sf properties of present - day e / s0 galaxies , suggesting that most e / s0 galaxies in pairs are dense , dynamically hot spheroids with low gas fractions and probably slow rotation . @xmath125 the main effect of interactions on e / s0 galaxies are to make them moderatively asymmetric , probably through @xmath85 tidal deformations , and to increase the clumpiness for the most interacting cases . @xmath125 in the case the differences in the @xmath4 parameters between pair and isolated e / s0 s were dominated by differences in the formation processes rather than by the interaction , our results show some evidence against a formation scenario of e / s0s based only on the secular ( enhanced by interactions ) mechanism . @xmath125 the loci of our strongly interacting e / s0 s in the @xmath4 space might correspond to the loci of ( `` dry '' ) pre - merger early - type precursors of present - day bright ellipticals . thus , our results can serve to characterize these pre - merger early - type population at higher redshifts . we thank j. colbert for sharing his @xmath7-band observations of 20 truly isolated e / s0 galaxies . we acknowledge the anonymous referee for a careful reading of the manuscript and suggestions that improved it . support for this work was provided by conacyt grant 42810/a-1 to h.h.t . support from an nsf astronomy and astrophysics fellowship is acknowledged by c.j.c . jerjen , h. , tammann , g. a. , binggeli , b. 1992 in `` morphological classification of galaxies '' , longo g. , capaccioli m. , busarello g. ( eds ) astrophys . space sci . vol . 178 . kluwer , dordrecht , p. 17 kpg101b & -21.44 & 4.14@xmath1260.15 & 0.02@xmath1260.03 & 0.03@xmath1260.09 & + kpg129a & -20.78 & 4.93@xmath1260.23 & 0.02@xmath1260.01 & 0.08@xmath1260.09 & + kpg162a & -21.40 & 4.22@xmath1260.19 & 0.00@xmath1260.08 & 0.02@xmath1260.17 & + kpg191a & -20.43 & 3.94@xmath1260.58 & 0.06@xmath1260.01 & 0.07@xmath1260.18 & + kpg202a & -20.31 & 4.46@xmath1260.16 & 0.00@xmath1260.00 & -0.02@xmath1260.08 & + kpg229a & -20.11 & 4.84@xmath1260.42 & 0.08@xmath1260.02 & 0.07@xmath1260.23 & + kpg234a & -21.04 & 4.35@xmath1260.10 & 0.05@xmath1260.01 & -0.02@xmath1260.13 & + kpg239a & -20.02 & 4.43@xmath1260.43 & 0.07@xmath1260.01 & 0.11@xmath1260.07 & + kpg254b & -20.74 & 4.56@xmath1260.37 & 0.08@xmath1260.01 & 0.03@xmath1260.15 & + kpg260b & -20.34 & 4.54@xmath1260.20 & 0.10@xmath1260.00 & 0.21@xmath1260.11 & + kpg29b & -20.66 & 4.16@xmath1260.12 & 0.01@xmath1260.02 & -0.03@xmath1260.18 & + kpg303a & -21.28 & 4.40@xmath1260.21 & 0.04@xmath1260.01 & -0.004@xmath1260.01 & + kpg339b & -20.66 & 4.60@xmath1260.26 & 0.03@xmath1260.02 & -0.02@xmath1260.05 & + kpg353b & -21.13 & 4.57@xmath1260.08 & 0.05@xmath1260.00 & -0.01@xmath1260.09 & + kpg363a & -20.28 & 5.06@xmath1260.20 & 0.22@xmath1260.01 & 0.05@xmath1260.07 & + kpg386b & -19.77 & 4.33@xmath1260.39 & 0.03@xmath1260.02 & -0.01@xmath1260.07 & + kpg38a & -19.80 & 4.40@xmath1260.16 & 0.01@xmath1260.03 & 0.01@xmath1260.07 & + kpg38b & -19.80 & 4.77@xmath1260.13 & 0.01@xmath1260.04 & -0.02@xmath1260.20 & + kpg392a & -19.84 & 4.44@xmath1260.25 & 0.06@xmath1260.01 & 0.12@xmath1260.06 & + kpg393a & -21.02 & 4.13@xmath1260.37 & 0.07@xmath1260.01 & 0.06@xmath1260.02 & + kpg394a & -21.54 & 4.10@xmath1260.30 & -0.04@xmath1260.04 & -0.06@xmath1260.03 & + kpg407b & -21.58 & 5.08@xmath1260.12 & 0.49@xmath1260.00 & 0.16@xmath1260.21 & + kpg408a & -21.08 & 4.26@xmath1260.28 & 0.01@xmath1260.00 & -0.01@xmath1260.09 & + kpg416b & -20.46 & 5.18@xmath1260.25 & 0.01@xmath1260.02 & 0.05@xmath1260.10 & + kpg419b & -20.15 & 4.65@xmath1260.35 & 0.04@xmath1260.01 & -0.06@xmath1260.07 & + kpg429b & -20.42 & 4.95@xmath1260.37 & 0.02@xmath1260.01 & 0.14@xmath1260.03 & + kpg432a & -21.25 & 4.47@xmath1260.37 & 0.00@xmath1260.01 & -0.01@xmath1260.11 & + kpg445a & -21.03 & 4.65@xmath1260.24 & 0.01@xmath1260.01 & 0.00@xmath1260.06 & + kpg460b & -20.79 & 4.48@xmath1260.33 & 0.16@xmath1260.01 & 0.26@xmath1260.30 & + kpg487b & -21.13 & 5.43@xmath1260.17 & 0.24@xmath1260.03 & 0.02@xmath1260.22 & + kpg526b & -19.68 & 5.14@xmath1260.18 & 0.03@xmath1260.01 & -0.07@xmath1260.05 & + kpg542b & -19.70 & 4.26@xmath1260.30 & 0.18@xmath1260.00 & 0.07@xmath1260.09 & + kpg548b & -20.51 & 4.48@xmath1260.16 & 0.02@xmath1260.01 & 0.00@xmath1260.12 & + kpg552b & -20.12 & 3.87@xmath1260.32 & 0.04@xmath1260.00 & 0.10@xmath1260.05 & + kpg553a & -19.28 & 4.44@xmath1260.27 & 0.04@xmath1260.02 & 0.00@xmath1260.14 & + kpg572a & -20.17 & 3.80@xmath1260.53 & 0.06@xmath1260.02 & 0.06@xmath1260.24 & + kpg591a & -17.85 & 3.43@xmath1260.11 & 0.20@xmath1260.01 & 0.27@xmath1260.15 & + kpg61b & -19.61 & 4.55@xmath1260.31 & 0.02@xmath1260.01 & -0.01@xmath1260.02 & + kpg62a & -19.26 & 4.33@xmath1260.21 & -0.00@xmath1260.02 & 0.03@xmath1260.05 & + kpg81b & -20.86 & 5.60@xmath1260.19 & 0.13@xmath1260.05 & 0.26@xmath1260.36 & + kpg83a & & 4.25@xmath1260.13 & 0.19@xmath1260.01 & 0.05@xmath1260.12 & + kpg86b & -18.34 & 4.57@xmath1260.08 & 0.07@xmath1260.02 & 0.04@xmath1260.15 & + @xmath1 & 4.656/4.597 & 4.439/4.282 & 4.190/4.082 & 4.621 & 4.476 & 4.171 + @xmath2 & 0.067/0.150 & 0.035/0.047 & 0.009/0.015 & 0.039 & 0.019 & 0.007 + @xmath3 & 0.065/0.068 & 0.025/0.020 & -0.022/-0.018 & 0.057 & 0.014 & -0.029 +

we present the first results from a high sampling rate , multi - month reverberation mapping campaign undertaken primarily at mdm observatory with supporting observations from telescopes around the world . the primary goal of this campaign was to obtain either new or improved h reverberation lag measurements for several relatively low luminosity agns . we feature results for ngc 4051 here because , until now , this object has been a significant outlier from agn scaling relationships , e.g. , it was previously a outlier on the relationship between the broad - line region ( blr ) radius and the optical continuum luminosity the relationship . our new measurements of the lag time between variations in the continuum and h emission line made from spectroscopic monitoring of ngc 4051 lead to a measured blr radius of light days and black hole mass of . this radius is consistent with that expected from the relationship , based on the present luminosity of ngc 4051 and the most current calibration of the relation by . we also present a preliminary look at velocity - resolved h light curves and time delay measurements , although we are unable to reconstruct an unambiguous velocity - resolved reverberation signal .

0904.0251

recent theoretical and observational studies have provided strong evidence suggesting a connection between supermassive black hole ( smbh ) growth and galaxy evolution ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . to better understand this connection , we need more direct measurements of smbh masses across cosmological distances . unfortunately , measuring smbh masses directly with dynamical methods requires high angular resolution , so use of these methods is limited to relatively nearby sources . this resolution problem can be obviated by studying smbhs in type 1 active galactic nuclei ( agns ) . in this case , the technique of reverberation mapping @xcite may be applied to measure the smbh mass , as has been done for more than 3 dozen type 1 agns to date ( e.g. , see recent compilation by * ? ? ? reverberation mapping relies on time resolution rather than angular resolution , since it takes advantage of the presence of a time delay , @xmath7 , between continuum and emission line flux variations observed through spectroscopic monitoring . this time delay corresponds to the light travel time across the broad - line region ( blr ) , and thus measurements of @xmath7 provide an estimate of the size of the region , @xmath8 . because the blr gas is well within the sphere of influence of the black hole and studies have provided evidence for virialized motions within this region ( e.g. , * ? ? ? * and references therein ) , @xmath3 can be related to the mass of the smbh through the velocity dispersion of the blr gas . although stellar and gas dynamical modeling and reverberation mapping are invaluable for measuring smbh masses directly , these methods are observationally taxing , due to resolution requirements for dynamical methods and time requirements for reverberation mapping campaigns . it is currently impossible to measure smbh masses directly for large statistical samples of galaxies . however , these direct mass measurements have led to the discovery of scaling relationships that relate smbh mass to other galaxy or agn observables that can be used to investigate the connection between smbh mass and galaxy evolution . in particular , some relations show connections between properties of the smbh ( i.e. , its mass ) and global properties of the host galaxy . examples include the correlation between smbh mass and bulge / spheroid stellar velocity dispersion , i.e. the @xmath9@xmath10 relation for agns @xcite and quiescent galaxies @xcite , and the correlation between smbh mass and galaxy bulge luminosity @xcite . other relations connect various agn properties . one such relation is the correlation between black hole mass and optical luminosity @xcite , which relates directly to the accretion rates of agns . there is also a correlation between blr radius and agn luminosity , i.e. , the @xmath3@xmath4 relation @xcite , which has proven to be very powerful for making indirect smbh mass estimations from single - epoch spectra ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? these indirect mass estimates can then be related to other properties of the host galaxy through direct measurements or separate scaling relations . although scaling relations have become widely used for statistical studies , it is important to understand that the indirect mass estimates determined by these relations are only as reliable as the direct mass measurements used to calibrate them . therefore , establishing a secure calibration across a wide dynamic range in parameter space and better understanding any intrinsic scatter in these relations is essential . to accomplish this , we must continue to make new direct measurements as well as to check previous results that are , for one reason or another , suspect . ngc 4051 , an sab(rs)bc galaxy with a narrow - line seyfert 1 ( nls1 ) nucleus at redshift @xmath11 , is a case in point . measurements of the blr radius and optical luminosity @xcite place it above the @xmath3@xmath4 relation , i.e. , the blr radius is too large for its luminosity ( cf . figure 2 of * ? ? ? . it also appears to be accreting mass at a lower eddington rate than other nls1s ( cf . figure 16 of * ? ? ? these two anomalies together suggest that perhaps the blr radius has been overestimated by @xcite ; indeed an independent reverberation measurement of the blr radius in ngc 4051 by ( * ? ? ? * hereafter s03 ) is about half the value measured by ( * ? ? ? * hereafter p00 ) . furthermore , neither the p00 nor s03 data sets are particularly well sampled on short time scales , so neither set is suitable for detection of smaller time lags ( e.g. , @xmath1223 days ) . in addition , @xcite reports a tully - fisher distance to ngc 4051 that is @xmath150% larger than that inferred from its redshift ( i.e. , 15.2 mpc versus 10.0 mpc , respectively ) . this suggests that the luminosity derived in past studies from the redshift could be an underestimate and might also be a contributing factor to the placement of ngc 4051 above the @xmath3@xmath4 relation . in this work , we present an analysis of new , optical spectroscopic and photometric observations of ngc 4051 , which represent the first results from a densely sampled reverberation mapping campaign that began in early 2007 . the campaign spanned more than 4 months , during which time we consistently obtained multiple photometric observations per night and spectroscopic observations nearly every night from a combination of five different observatories around the globe . the immediate goal of this campaign is to remeasure the h@xmath0 reverberation lag measurements for several objects on the low - luminosity end of @xmath3@xmath4 scaling relationship with poorly determined reverberation lags and , consequently , poorly determined black hole masses . we will also add to the overall sample of reverberation mapped agns by measuring lags for new objects . another goal for this extensive collection of homogeneous data is to reconstruct the observed response of the h@xmath0 emission line to an outburst from the variable continuum source by modeling the response as a function of both line - of - sight velocity and light travel time , i.e. , a `` velocity delay map '' ( for a tutorial , see * ? ? ? creation of a velocity delay map will provide novel insight into the structure and kinematics of the blr . though we have not yet attempted to reconstruct a full velocity delay map , we present preliminary velocity - resolved lag measurements for ngc 4051 . complete results for ngc 4051 and other campaign targets will be presented in future work . most data acquisition and analysis practices employed here follow closely those described by @xcite and laid out by @xcite . the reader is referred to these works for additional details and discussions . throughout this work , we assume the following cosmological parameters : @xmath13 , @xmath14 , and @xmath15 km sec@xmath16 mpc@xmath16 . spectra of the nuclear region of ngc 4051 were obtained from both the 1.3-meter telescope at mdm observatory and the 2.6-meter shajn telescope of the crimean astrophysical observatory ( crao ) . the mdm observations utilized the boller and chivens ccd spectrograph , where 86 observations were taken over the course of 89 nights between jd2454184 and jd2454269 , targeting the h@xmath17 and [ oiii]@xmath18 emission line region of the optical spectrum . the position angle was set to @xmath19 , with a slit width of 50 projected on the sky , resulting in a spectral resolution of 7.6 across this emission - line region . figure [ fig : meanrms ] shows the mean and rms spectra of ngc 4051 based on the mdm observations . we acquired 22 crao spectra over 34 nights between jd2454266 and jd2454300 with the nasmith spectrograph and spec-10 @xmath20 pixel ccd . for these observations a 30 slit was utilized , with a @xmath21 position angle . spectral wavelength coverage for this data set was from @xmath138006000 , with a dispersion of 1.8 / pix and a spectral resolution of 7.5 . note that ( 1 ) the dispersion varies with wavelength : the value 1.8 / pix is given for 5100 , and ( 2 ) the real wavelength coverage is slightly greater than given but the red and blue edges of the ccd frame are unusable ( too low s / n ratio ) because of vignetting . a relative flux calibration of each set of spectra was performed based on the constancy of the narrow [ oiii]@xmath22 line flux . because this line emission originates in the extended , low - density narrow line region , it can be assumed constant over the timescale of this campaign and therefore serves as the basis for a reliable relative flux calibration . however , the data quality is not identical from night to night due to , e.g. , seeing , weather conditions , atmospheric transparency , etc . this affects not only the integrated line flux in each observation , but also properties of the spectrum such as s / n and resolution . consequently , we employ the spectral scaling algorithm of @xcite for the [ oiii]@xmath22 flux calibration . this algorithm determines the best scaling through @xmath23 minimization of residuals rather than simply calculating a simple multiplicative scale factor to scale the spectral fluxes . following this method , we created a reference spectrum by averaging all spectra from a given data set . we then formed a difference spectrum by subtracting the reference spectrum from each individual spectrum . the algorithm uses a least squares method to minimize the residuals of the [ oiii]@xmath22 line flux in each difference spectrum by making small zero - point wavelength calibration adjustments , correcting for resolution differences , and applying a multiplicative scale factor to the [ oiii]@xmath22 line flux of the individual spectrum . because this method is based on minimizing residuals between each individual spectrum and the reference spectrum , there is a small residual dispersion in the line fluxes after calibration . this dispersion is related to the data quality and the ability of the scaling algorithm to mitigate night to night differences between individual spectra , related largely to seeing and s / n . tests of the original algorithm by @xcite estimated errors in the scaled fluxes of better than 5% , however , past studies employing somewhat improved versions of this same scaling algorithm typically achieved dispersions across a data set of @xmath12% ( e.g. , * ? ? ? * ) . we measure a dispersion of @xmath11.5% , demonstrating the high level of homogeneity that we have been able to achieve in the current data set . in addition to spectral observations , we obtained supplemental @xmath24-band photometry from the 2.0-m multicolor active galactic nuclei monitoring ( magnum ) telescope at the haleakala observatories in hawaii , the 70-cm telescope of the crao , and the 0.4-m telescope of the university of nebraska . the magnum observations were imaged with the multicolor imaging photometer ( mip ) as described by @xcite , @xcite , and @xcite . photometric fluxes measured from 23 observations between jd2454182 and jd2454311 within an aperture of 83 . photometric reduction of ngc 4051 was similar to that described for other sources by @xcite and @xcite , except the host - galaxy contribution to the flux within the aperture was not subtracted and the filter color term was not corrected because these photometric data were later scaled to the mdm continuum light curve ( as described below ) . also , minor corrections ( of order 0.01 mag or less ) due to the seeing dependence of the host - galaxy flux were ignored . the 76 crao photometric observations were collected between j245d4180 and jd2454299 with the ap7p ccd mounted at the prime focus of the 70-cm telescope ( @xmath25 cm ) . in this setup , the @xmath26 pixels of the ccd field covers a 15 @xmath27 15 field of view . photometric fluxes were measured within an aperture of 150 . for further details of the crao @xmath24-band observations and reduction , see the similar analysis described by @xcite . the university of nebraska observations were conducted by taking and separately measuring a large number of one - minute images ( @xmath120 ) each of 28 nights between jd2454195 and jd2454290 . details of the observing and reduction procedure are as described by @xcite . comparison star magnitudes were calibrated following @xcite and @xcite . to minimize the effects of variations in the image quality , fluxes were measured through an aperture of radius 8 arcseconds . the errors given for each night are the errors in the means . light curves of the h@xmath0 flux were made based on integrated fluxes measured in the mdm and crao spectra between 48154920 and over a linearly interpolated continuum defined between the average flux density in each of the following regions blueward and redward of h@xmath0 , respectively : 47704780 and 50905130 . the crao h@xmath0light curve was then scaled to the mdm light curve with a multiplicative constant based on the average flux ratio between the four pairs of closely spaced points in the mdm and crao h@xmath0 light curves separated by no more than 0.5 day . this scaling is necessary to account for differences in the amount of [ oiii]@xmath22 emission line flux that enters the slit in the different data sets due to seeing and aperture affects . the lower panel of figure [ fig : lcsep ] shows the h@xmath0 light curve derived from both data sets after scaling the crao fluxes to those measured from mdm spectra . a continuum light curve was created with observations from each @xmath24-band photometric data set and the average continuum flux density measured over 50905130 ( i.e. , rest frame @xmath15100 ) in each spectrum of the spectroscopic data sets . first , the multiplicative scale factor determined for above to scale the crao h@xmath0 fluxes to the mdm light curve was also applied to the crao continuum fluxes , since the calibration of these fluxes is also susceptible to the same seeing and aperture affects as the h@xmath0 flux calibration . next , this light curve as well as the individual photometric light curves ( see fig . [ fig : lcsep ] , upper panel ) were scaled , one by one , to the same flux scale as the mdm light curve by making an additive , relative flux adjustment to each . this additive correction is necessary for both spectroscopic and photometric data because of differences in host galaxy starlight that enters the different aperture sizes of the various data sets . for the photometric observations , there is an additional component ( also additive ) due to the larger width of the filter bandpass . each light curve was merged with the parent light curve to which it was scaled before the next light curve was scaled , thus building up a larger , more well - sampled light curve in the following order : mdm , magnum , crao photometry , university of nebraska , and crao spectroscopy . this was done so that the smaller , and in some cases shorter , light curves could be scaled to a longer and more densely sampled parent light curve . the scale factor applied to each secondary light curve to scale it to its parent light curve was calculated based on the difference between a linear least squares fit to this light curve and to the parent light curve before it , starting with the mdm light curve as the initial parent light curve . the fits to each light curve , both parent and secondary were limited to using only observations within the same overall temporal range , so that , when necessary , the beginning and/or ends of the light curves were truncated during the fitting process . assignment of uncertainties to the photometric fluxes is described above in the text or in references describing the photometric data sets . however , we calculated the uncertainties in the mdm and crao spectroscopic fluxes after creating the light curves for these data sets but before intercalibration . typically , we determine uncertainties in our light curve flux measurements by applying a mean fractional error to all points . this fractional error is determined by comparing the average flux difference between closely spaced pairs of observations , assuming that flux differences across these short times scales are due to noise rather than genuine variability . because real variability has been confirmed by @xcite to occur in ngc 4051 on time scales shorter than 2 days , and our sampling rate is @xmath11 day , on average , we could not use this method to determine the relative errors on our spectroscopic flux measurements for this object . instead , we took advantage of the observations of other higher - luminosity agns that we monitored as part of this larger campaign ( i.e. , same telescopes , instrumental setup , and observing conditions ; results in preparation ) . unlike ngc 4051 , these objects neither exhibit variability on such short time scales nor have such short measured lags . therefore , the uncertainties assigned to the ngc 4051 spectroscopic observations seen in figure [ fig : lcsep ] are an average of the uncertainties in the flux measurements calculated as described above from closely spaced observations ( separations of @xmath122.0 days ) of these other objects ( e.g. , typical fractional errors in the range @xmath28 and @xmath29 for the continuum and line fluxes , respectively ) . the merged continuum and h@xmath0 light curves shown in figure [ fig : lc4cc ] are used for the subsequent time series analysis . these differ from simply combining the individual light curves , shown in figure [ fig : lcsep ] , in the following ways : * first , we applied an absolute flux calibration to both light curves by applying a single multiplicative scale factor determined from the ratio of the [ oiii]@xmath22 emission line flux determined by p00 to that measured in the reference spectrum used above for the relative flux calibration . unlike the emission line flux in our reference spectrum , the p00 flux measurement of @xmath30[oiii]@xmath31 erg s@xmath16 @xmath32 was taken from observations obtained under photometric conditions and , consequently , @xmath18% larger than our measured value . additionally , the p00 measurement was made employing observing strategies and measurement practices similar to what we present here , thus validating this direct comparison . this additional flux calibration does not affect the reverberation results but is necessary for accurately measuring the 5100 continuum luminosity . * second , we subtracted the host galaxy starlight contribution to the continuum flux , determined using the methods of @xcite to be @xmath33(5100)@xmath34 erg s@xmath16 @xmath32 @xmath16 . * third , we binned closely spaced observations as a weighted average and applied this to continuum flux measurements separated by @xmath350.25 days and h@xmath0 flux measurements separated by @xmath350.5 days . fluxes for individual observations ( i.e. , before time binning ) from all sources are listed in table [ tab : fluxes ] . values listed represent the flux of each observation after completing all flux calibrations described above ( i.e. , relative calibration to intercalibrate all data sets onto the mdm flux scale , followed by absolute calibration based on the p00 [ oiii]@xmath22 line flux and removal of host starlight contamination ) . column 1 gives the julian date of each observation . the 5100 continuum or @xmath24-band flux and integrated h@xmath0 flux are given in columns ( 2 ) and ( 3 ) , respectively , and column ( 4 ) lists the source of each measurement . photometric and spectroscopic observations from crao can be differentiated by noting that no h@xmath0 flux values are present for photometric observations . table [ tab : lcstats ] displays statistical parameters describing the final light curves shown in figure [ fig : lc4cc ] . column ( 1 ) gives the spectral feature represented by each light curve , and the number of data points in each light curve is shown in column ( 2 ) . columns ( 3 ) and ( 4 ) are mean and median sampling intervals , respectively , between data points . the mean flux with standard deviation is given in column ( 5 ) , while column ( 6 ) shows the mean fractional error in these fluxes . column ( 7 ) gives the excess variance , calculated as @xmath36 where @xmath37 is the variance of the observed fluxes , @xmath38 is their mean square uncertainty , and @xmath39 is the mean of the observed fluxes @xcite . finally , column ( 8) is the ratio of the maximum to minimum flux in the light curves . a times series analysis of the continuum and h@xmath0 light curves was performed to determine the mean light travel time lag between continuum and emission line variations . we used two cross correlation schemes designed for data sets with uneven time sampling : 1 . an interpolation scheme @xcite with an interval of 0.2 day . a cross - correlation function ( ccf ) is constructed from the mean value of the correlation coefficient , @xmath40 , computed from cross correlating both the interpolated line light curve with the original continuum light curve and then the interpolated continuum light curve with the original line light curve , a process during which a range of possible lags , @xmath7 , are imposed on the h@xmath0 light curve . 2 . a time binning scheme @xcite with a bin size of 1.0 day . here , a discrete correlation function ( dcf ) is produced , which determines @xmath40 as a function of lag , similar to the ccf . in this scheme , however , only the actual data are cross correlated , and the resulting values of @xmath40 for all discretely correlated pairs are binned as a function of lag . the dcf that results gives the mean value of @xmath40 in each bin , where the corresponding uncertainty is assigned in a statistical manner ( see * ? ? ? this method prevents possible spurious lag determinations that could potentially arise in the interpolation method due to under - sampling or large gaps in the data . the resulting ccf and dcf are shown in figure [ fig : ccf ] , along with the auto - correlation function ( acf ) computed by cross correlating the continuum with itself . we characterize the time delay between the continuum and emission line variations using two parameters derived from the ccf ; @xmath41 is the lag that corresponds to the largest correlation coefficient , @xmath42 , and @xmath43 is the centroid of the ccf based on all points with @xmath44 . time dilation corrected values of @xmath41 and @xmath45 determined from the ccf in figure [ fig : ccf ] are given in table [ tab : results ] . uncertainties in both lag parameters are computed via model - independent monte - carlo simulations that employ the bootstrap method of @xcite , with the additional modifications of @xcite . our measured h@xmath0 lag of @xmath46 days from this work is consistent , within the errors , to the most recent results for this object by s03 , who measured a lag of @xmath47 days . it is not clear how meaningful a direct comparison might be , however , because s03 measured the time delay between variations in the @xmath16800 continuum flux density and the integrated h@xmath48 flux rather than between the @xmath15100 continuum and h@xmath0 . we also note that the median sampling rate of s03 was larger than our measured lag , suggesting to us that the s03 light curves are undersampled . furthermore , s03 only perform a cross correlation analysis based on the dcf method , which sacrifices time resolution . our new time delay measurements are inconsistent , however , with the previous measurement of @xmath49 days by @xcite using data from p00 . these differences are unlikely a luminosity effect , since the average luminosity states of ngc 4051 were similar during this and the p00 campaigns ( log@xmath50 = 41.82 and log@xmath50 = 41.87 , respectively ) that of the peterson et al . campaign as well . ] . therefore , we carefully re - examined the light curves used by p00 to better understand possible causes for the observed inconsistency . @xcite suggested that the cause for a similar inconsistency between lag measurements from two reverberation mapping campaigns of ngc 5548 @xcite was due to different continuum variability timescales observed in the separate campaigns : longer continuum variability timescales lead to larger lag measurements . however , this explanation is unlikely to be the cause for the current inconsistency between our measured lag and that of p00 because the prominent variability timescales observed in both the p00 and current continuum light curves are similar ( @xmath14050 days ) . instead , the simplest explanation for the inconsistency between our measured lag and that of p00 is random error . we investigated this possibility by performing monte carlo simulations using the `` subset 1 '' h@xmath0 and 5100 continuum flux light curves from p00 with the goal of estimating the likelihood that a lag of @xmath51 = 5.8 days would be measured , even if the actual blr radius of the h@xmath0 emission was 2.7 light days , as expected from the @xcite @xmath52@xmath4 relation for the average luminosity of ngc 4051 during this time period . in each simulation we created a simulated h@xmath0 light curve by convolving a modified continuum light curve with a transfer function that assumed a blr with a thin spherical shell geometry of radius 2.7 light days . the sampling was increased in the modified continuum light curve over that of the original subset 1 continuum light curve by interpolating between the actual points on a 0.5 day scale . noise was added to the flux of each interpolated point using a random gaussian deviate . the size of each deviate was based on the average uncertainty in flux of the closest ` real ' continuum point on each side of the interpolated point . the simulated emission - line light curve was then sampled identically to the original subset 1 h@xmath0 light curve . we cross correlated this new emission - line light curve with the original subset 1 continuum light curve to determine a reverberation lag . the simulation was repeated 10,000 times , and lags were measured similarly for each iteration . we found that the average lag recovered was 2.7 days ( reassuring , since this was our input radius ) . however , we were unable to reproduce even a single lag of 5.8 days . in fact , the largest lag our simulations recovered was 4.0 days . a couple of possibilites suggest themselves : * the blr has physically changed in the 11-year interval between the time of p00 s subset 1 and the time of our recent campaign . this is a physical possibility since the dynamical timescale of the blr in ngc 4051 is @xmath53 years . * the p00 data are undersampled and there are really unresolved variations occurring on timescales shorter than the typical sampling interval of 2.2 days in subset 1 , the best - sampled part of the p00 light curve . we have no particular reason to believe the former possibility . however , the latter is suggested by how p00 established the relative uncertainties of their fluxes , namely by assuming that there are no true variations on time scales shorter that the typical sampling time scales and that any differences between closely spaced measurements reflect random errors only , not true variability . the estimates of the relative flux errors in the p00 subset 1 based on comparing measurements separated by 2 days or less are about 3.2% for both the contiuum and the line . in our new data set , obtained with the same instrument , we find relative errors of about 1.4% and 2.1% in the continuum and the line , respectively , using the method described in section [ s : lightcurves ] . we conclude that the flux uncertainties of the p00 data were overestimated due to short timescale variability . proceeding with the assumption that the p00 light curves are undersampled , we isolated the portion of the light curves that has the highest sampling across the sharpest features . we made this selection in an attempt to avoid occurrences of undersampling more complex variability . from the initial light curve , reproduced in figure [ fig : p00lc ] , we removed the first 10 observations that exhibit a broad inflection in the flux with a poorly defined peak . we perform a cross correlation analysis on these shortened light curves and determine a shorter lag , @xmath54 days . this lag determination is consistent with both the expected radius of 2.7 light days from the @xmath52@xmath4 relation , the current results , and the results of s03 . if we continue with the assumption that the light curves of p00 are undersampled , and the p00 flux uncertainties are overestimated , their assigned uncertainties would act to decrease the significance of short timescale variability , likely attributing it to noise instead . if we impose the average continuum flux uncertainty measured from our current data set ( given above ) on the shortened p00 continuum light curve , we can further improve the precision of this revised lag measurement to @xmath55 days . we then conducted another simulation in which we applied the sampling rate from the p00 light curves to the light curves from this work . by undersampling our current light curves , we can determine the probability that undersampling could lead to an overestimated lag similar to that measured by p00 . this type of simulation can provide further evidence that the lag measured by p00 was an overestimate and a consequence of undersampling . at the same time , it could diminish the possibility that the discrepancy in lag measurements is due to a difference in the physical conditions or structure of the blr during the p00 campaign compared to the present . using the continuum and h@xmath0 light curves shown in figure [ fig : lc4cc ] as the starting point , we modified them similarly to the continuum light curve described for our first set of simulations ( i.e. , increasing the sampling by interpolating between data points and adding noise to these points with a gaussian deviate ) , but this time we interpolated both the continuum and emission line light curves from this work on a 0.1 day interval . we then drew sample light curves from this parent light curve with the same length and sampling pattern as the full p00 light curve shown in figure [ fig : p00lc ] . we applied the same cross - correlation analysis ( as described in section [ s : lagresults ] ) to measure lags from these sample light curves . the parent light curves cover a longer time span than the sample light curves and therefore allow for multiple iterations of sample light curves to be chosen from different subsets of the parent light curves . the first iteration of sample light curves are created from the subset of the parent light curves where the beginning points match up , but the ends of the parent light curves are discarded . we then build up multiple iterations by shifting the starting point of the sample light curve in time by one time step , i.e. , 0.1 day , across the parent light curves . in this way , we were able to build up 330 sample light curves , where in the last iteration , the sample light curves begin in the middle of the parent light curves , but both sets of light curves end at the same time . based on the cross correlation analysis from these 330 sample light curves , the probability of measuring @xmath565.0 days is 0.6% ( 2 out of 330 ) , and the probability of measuring @xmath573.5 days ( i.e , the lag we calculated above from only a portion of the p00 light curve ) is @xmath18% ( 25 out of 330 ) . we conclude that undersampling is at least a plausible explanation for the difference between the p00 results and those reported here . applying virial assumptions to the reverberating gas in the blr , the mass of the black hole can be defined by @xmath58 where @xmath7 is the measured emission - line time delay , so that c@xmath7 represents the blr radius , and @xmath59 is the blr velocity dispersion @xcite . the dimensionless factor @xmath60 depends on the structure , kinematics , and inclination of the blr and is of order unity . we estimate the blr velocity dispersion from the line width of h@xmath0emission line . the line width can be characterized by either the fwhm or the line dispersion , i.e. , the second moment of the line profile . the fwhm and the line dispersion , @xmath61 , were measured from both the mean and the rms spectra of ngc 4051 shown in figure [ fig : meanrms ] . here , we have measured both quantities and their uncertainties employing methods described in detail by @xcite . all measured values of the h@xmath0 line width are listed in table [ tab : results ] . typically , the narrow - line emission component of the line should be removed before measuring the line width ( see * ? ? ? * ) ; however , this component could not be reliably isolated from the rest of the line profile in this object . as a result the line widths measured in the mean spectrum , particularly the fwhm , are less reliable for these purposes than the widths measured from the rms spectrum . we calculate the black hole mass for ngc 4051 using @xmath43 , for the time delay , @xmath7 , and the line dispersion , @xmath62 , measured from the h@xmath0 emission line in the rms spectrum , for the emission - line width , @xmath59 . we utilize the calibration of the reverberation mass scale of @xcite for this choice of lag and line width parameters and therefore adopt a scale factor value of @xmath60=5.5 . we then use equation 2 to estimate the black hole mass of ngc 4051 to be @xmath63 . here , statistical and observational uncertainties have been included , but intrinsic uncertainties from sources such as unknown blr inclination can not be accurately ascertained . @xcite have considered the effect of radiation pressure on smbh mass estimates and provide a new prescription for calculating the smbh mass that includes a correction factor to account for radiation pressure . radiation pressure acts to partially counteract the force of gravity on the blr gas , since the outward radiation force has the same radial dependence ( @xmath64 ) as the inward gravitational force . as a result the blr gas motions are effectively under the influence of an apparently smaller smbh mass , which leads to an underestimate of the true smbh mass . @xcite determine that although taking radiation pressure into account is most important for black holes radiating near the eddington limit , it should be considered even for systems with @xmath65 . on the other hand , in a comparison study of type 1 versus type 2 black hole masses determined with independent methods , @xcite sees better agreement between the mass and @xmath66 distributions of these two populations if radiation pressure forces are neglected . netzer concludes that either the effects of radiation pressure on the blr gas in these objects is negligible or that blr column densities must be significantly larger , i.e. , @xmath67 , than assumed by marconi et al . ( 2008 ; @xmath68 ) . in a more recent paper , @xcite reinvestigate the results of @xcite and support their findings on the dependence of the effect of radiation pressure on column density but conclude that , until it is possible to determine the nature of the apparent dependence of @xmath69 on source properties ( e.g. , @xmath66 ) , one should always consider the possibility that radiation forces are important , and black hole masses should consequently be determined using the correction of @xcite . the importance of radiation pressure forces on black hole mass determinations is still under debate . therefore , in addition to our virial mass estimate for ngc 4051 given above , we also estimate the smbh mass in ngc 4051 taking radiation pressure into consideration ( cf . equation 6 of * ? ? ? * ) , with @xmath70 and log@xmath71 , which are derived by marconi et al . from fits to smbh masses from reverberation mapping studies . with these scale factors and the same line width and blr radius measurements used above , we calculate a mass of @xmath72 . contrary to the expectations from the physical arguments , we calculate a mass smaller than that determined in the case where we did not consider the effect of radiation pressure . because ngc 4051 is a low - luminosity agn radiating well below the eddington limit ( @xmath73 ) , the correction to the mass due to radiation pressure is small ( smaller even than the formal observational uncertainties on the mass ) , adding only @xmath74 to the mass . however , this radiation - corrected mass estimate is smaller than our original estimate because the value of @xmath60 derived by marconi et al . is a factor of 1.8 smaller than the @xcite value we adopted above . because this scale factor was derived in a statistical sense by both onken et al . and marconi et al . , the difference in @xmath60 values between these two studies has the potential to affect the mass of an individual object more than would be expected for a statistical sample , particularly for the low accretion - rate objects that need only a small radiation pressure correction . the lag measurements between the continuum and h@xmath0 emission in table [ tab : results ] represent the average time delay across the blr . because the blr is an extended region and the velocity of the gas is most likely a function of position , gas in different locations of the blr should respond to variations in the ionizing continuum flux on slightly different time scales . the observable result should be a difference in the reverberation lag measurement in different parts of the line profile ( i.e. , separated in velocity space ) . velocity - resolved reverberation mapping thus gives us information about the kinematics of gas in the blr . previous studies of time delay differences between multiple emission lines and the velocity dependence of the lag within a single emission line have shown that the blr is virialized and commonly contains an additional inflow component ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) ; however , the creation of full velocity delay maps ( see * ? ? ? * ) is an aspect of the reverberation mapping technique that has not yet been fully realized ( though see @xcite , @xcite , @xcite , and @xcite for previous attempts ) . by resolving the velocity - dependent reverberation response of the blr better than has been done in the past , we can reconstruct and analyze the velocity delay map to gain further insights into the geometry and kinematics of the blr . we searched for a velocity - dependent reverberation signal by dividing the h@xmath0 emission line flux into 8 velocity - space bins . the blue and red sides of the line were separately divided into 4 bins of equal velocity width , covering only the most variable portions of the line profile , i.e. , the outer - most wavelength boundaries were reduced from those considered for the analysis of the full profile to only include flux within the range 48404900 ( roughly @xmath752,000 km s@xmath16 ) . light curves were created from measurements of the integrated h@xmath0flux in each bin and then cross correlated with the continuum light curve following the same procedures described above . the top panel of figure [ fig : reslags ] shows the division of the h@xmath0 line profile from the rms spectrum into the eight velocity bins , and the bottom panel shows the lag measurements for each of these bins . error bars in the velocity direction represent the bin width . the evidence for a velocity - stratified blr response to continuum variations is present , but marginal . in particular , the lags measured for bins 1 - 2 and 7 - 8 are consistent with zero and might simply reflect correlated errors due to continuum contamination . although the shape of the velocity - resolved signal in figure [ fig : reslags ] supports our virial assumptions , since the higher velocity gas varies on shorter time scales than the low velocity gas , there are no strong indications for either outflow or inflow . outflow could be suggested by the larger lag measured in bin 5 ( red side of the line ) compared to bin 4 ( blue side of the line ) , but the difference is very marginal . even with the high sampling rate and spectral resolution we achieved during this campaign , observing a velocity - resolved signal for ngc 4051 as clearly as that detected for arp 151 by @xcite would have been rather surprising for the following reasons . first , the precision with which we can measure a lag is somewhat dependent on the median observational sampling interval , which , for ngc 4051 , was still not much shorter than the measured lag . this indicates that in order to better resolve a velocity - dependent signal , we need even higher time resolution for this object . second , because the h@xmath0 line is particularly narrow in this object , there are only a few velocity resolution elements across the line . based on our simulations , reanalysis of the light curves from p00 , and the additional arguments we presented above , we conclude that the inconsistency between the past measurements of the h@xmath0 reverberation lag in ngc 4051 and our present measurements most likely results from an overestimation of the lag by p00 . therefore , we adopt the results from the current reverberation campaign over previous campaign results measuring this lag in ngc 4051 . using the lag measurement of @xmath46 days presented here and the tully - fisher distance of @xcite , ngc 4051 is no longer an outlier on the @xmath3@xmath4 relationship . figure [ fig : b08rlrelation ] replicates the most recent version of this relationship by @xcite with both the previous and current lag values of ngc 4051 marked . secure placement of low - luminosity objects such as ngc 4051 on the @xmath3@xmath4 relationship is important for supporting the extrapolation of this relationship to the even lower - luminosity regime potentially populated by intermediate - mass black holes . additional results from the present campaign ( denney et al . , in preparation ) , as well as results from a recent monitoring campaign at the lick observatory ( e.g. , * ? ? ? * ) , aim to further populate this low - luminosity end of the @xmath3@xmath4 relationship , thus solidifying the calibration in a relatively under - sampled region of the relation . a reliable calibration of this relationship is imperative for large studies of black hole masses and galaxy evolution , since it allows for the calculation of black hole masses from single - epoch spectra and provides luminosity and radius estimates that help constrain parameter space in the search for intermediate - mass black holes . our new results provide a measure of the blr radius of ngc 4051 that is closer to the value naively expected from its luminosity . however , these new results do not resolve the unexpected location of this object on the @xmath9@xmath4 relation : nls1s tend to lie on a locus of this relation with relatively high eddington ratios ( @xmath76 ; see figure 16 of * ? ? ? * ) . however , based on the black hole mass of @xmath77 that we have calculated , the eddington ratio of ngc 4051 is still only @xmath73 . it seems unlikely that we have overestimated the mass of the black hole ( and thus underestimated @xmath66 ) , since our mass measurement is already a factor of a few lower than predicted by the @xmath78@xmath10 relationship @xcite . in this case , the narrowness of the balmer lines in the spectrum of ngc 4051 might be due at least in part to the inclination of the blr relative to our line of sight indeed , inclination has been invoked as one possible way to explain the nls1 phenomenon since the early days of research on these sources ( e.g. , * ? ? ? * ; * ? ? ? * ) . from observations of the narrow - line region in ngc 4051 , @xcite estimate that the inclination of this source is @xmath79 , near the maximum expected for a type 1 active nucleus in unified models . it is entirely reasonable to suppose that the blr and accretion disk are at the same inclination as the narrow - line region . even so , accounting for this high inclination would increase the line width by only about a factor of 2 , still within the nls1 regime . clearly further investigation is required to understand the low value of @xmath66 in ngc 4051 compared to other nls1s . we would like to thank the anonymous referee for useful suggestions that improved the clarity of this manuscript . we acknowledge support for this work by the national science foundation though grant ast-0604066 to the ohio state university . cmg is grateful for support by the national science foundation through grants ast 03 - 07912 and ast 08 - 03883 . mv acknowledges financial support from hst grants hst - go-10417 , hst - ar-10691 , and hst - go-10833 awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa , under contract nas5 - 26555 . vtd acknoledges the support of the russian foundation for basic research ( project no . 06 - 02 - 16843 ) to the crimean laboratory of the sternberg astronomical institute . sgs acknowledges support through grant no . 5 - 20 of the `` cosmomicrophysics '' program of the national academy of sciences of ukraine to crao . the crao ccd cameras have been purchased through the us civilian research and development foundation for the independent states of the former soviet union ( crdf ) awards up1 - 2116 and up1 - 2549-cr-03 . 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 . , w. f. , martino , d. l. , kawaguchi , g. , & kollatschny , w. 2007 , in astronomical society of the pacific conference series , vol . 373 , the central engine of active galactic nuclei , ed . l. c. ho & j .- w . wang , 29+ 4180.30 & 5.03 @xmath75 0.18 & & crao + 4181.37 & 5.66 @xmath75 0.16 & & crao + 4182.02 & 5.47 @xmath75 0.09 & & magnum + 4182.41 & 5.07 @xmath75 0.22 & & crao + 4184.79 & 5.41 @xmath75 0.20 & 5.22 @xmath75 0.11 & mdm + 4185.71 & 4.51 @xmath75 0.19 & 4.96 @xmath75 0.10 & mdm + 4186.49 & 4.51 @xmath75 0.22 & & crao + 4186.71 & 4.89 @xmath75 0.20 & 4.15 @xmath75 0.09 & mdm + 4187.38 & 4.82 @xmath75 0.22 & & crao + 4187.85 & 5.27 @xmath75 0.20 & 4.95 @xmath75 0.10 & mdm + 4188.37 & 4.32 @xmath75 0.25 & & crao + 4188.71 & 4.81 @xmath75 0.20 & 4.90 @xmath75 0.10 & mdm + 4189.39 & 4.75 @xmath75 0.26 & & crao + 4189.61 & 4.78 @xmath75 0.19 & 4.62 @xmath75 0.10 & mdm + 4189.96 & 5.53 @xmath75 0.21 & 4.88 @xmath75 0.10 & mdm + 4189.96 & 4.94 @xmath75 0.05 & & magnum + 4190.41 & 4.82 @xmath75 0.23 & & crao + 4190.72 & 4.82 @xmath75 0.20 & 4.73 @xmath75 0.10 & mdm + 4191.38 & 4.59 @xmath75 0.32 & & crao + 4191.62 & 4.91 @xmath75 0.20 & 4.51 @xmath75 0.09 & mdm + 4191.91 & 5.27 @xmath75 0.20 & 4.77 @xmath75 0.10 & mdm + 4192.43 & 4.84 @xmath75 0.32 & & crao + 4192.75 & 4.02 @xmath75 0.19 & 4.44 @xmath75 0.09 & mdm + 4193.61 & 4.86 @xmath75 0.20 & 4.34 @xmath75 0.09 & mdm + 4193.92 & 4.41 @xmath75 0.19 & 4.57 @xmath75 0.10 & mdm + 4194.73 & 4.62 @xmath75 0.19 & 4.60 @xmath75 0.10 & mdm + 4195.45 & 4.49 @xmath75 0.16 & & unebr . + 4195.63 & 4.30 @xmath75 0.19 & 4.06 @xmath75 0.09 & mdm + 4196.62 & 4.59 @xmath75 0.19 & 4.22 @xmath75 0.09 & mdm + 4197.78 & 4.22 @xmath75 0.19 & 4.43 @xmath75 0.09 & mdm + 4198.44 & 4.34 @xmath75 0.14 & & unebr . + 4198.78 & 4.51 @xmath75 0.19 & 4.80 @xmath75 0.10 & mdm + 4198.93 & 4.46 @xmath75 0.06 & & magnum + 4199.36 & 4.68 @xmath75 0.17 & & crao + 4199.40 & 4.78 @xmath75 0.15 & & unebr . + 4199.86 & 4.55 @xmath75 0.19 & 4.37 @xmath75 0.09 & mdm + 4200.40 & 4.52 @xmath75 0.15 & & crao + 4200.72 & 4.19 @xmath75 0.19 & 4.29 @xmath75 0.09 & mdm + 4201.31 & 4.84 @xmath75 0.23 & & crao + 4201.73 & 4.59 @xmath75 0.19 & 4.49 @xmath75 0.09 & mdm + 4202.37 & 4.56 @xmath75 0.18 & & crao + 4202.95 & 4.93 @xmath75 0.08 & & magnum + 4204.40 & 4.53 @xmath75 0.17 & & crao + 4204.73 & 4.35 @xmath75 0.19 & 4.58 @xmath75 0.10 & mdm + 4205.34 & 4.34 @xmath75 0.16 & & crao + 4205.62 & 3.97 @xmath75 0.18 & 4.37 @xmath75 0.09 & mdm + 4205.83 & 4.32 @xmath75 0.03 & & magnum + 4205.91 & 4.28 @xmath75 0.19 & 4.54 @xmath75 0.10 & mdm + 4206.36 & 4.19 @xmath75 0.18 & & crao + 4206.40 & 4.23 @xmath75 0.23 & & unebr . + 4206.62 & 4.31 @xmath75 0.19 & 4.46 @xmath75 0.09 & mdm + 4207.40 & 4.30 @xmath75 0.15 & & unebr . + 4207.82 & 4.40 @xmath75 0.19 & 4.60 @xmath75 0.10 & mdm + 4208.34 & 4.42 @xmath75 0.15 & & crao + 4208.40 & 4.26 @xmath75 0.17 & & unebr . + 4208.62 & 4.24 @xmath75 0.19 & 4.58 @xmath75 0.10 & mdm + 4208.88 & 4.25 @xmath75 0.03 & & magnum + 4209.40 & 4.46 @xmath75 0.15 & & crao + 4209.78 & 4.81 @xmath75 0.20 & 4.81 @xmath75 0.10 & mdm + 4210.62 & 5.01 @xmath75 0.20 & 4.99 @xmath75 0.10 & mdm + 4211.41 & 4.70 @xmath75 0.31 & & crao + 4212.34 & 4.64 @xmath75 0.15 & & crao + 4212.72 & 5.11 @xmath75 0.20 & 5.01 @xmath75 0.11 & mdm + 4212.75 & 4.78 @xmath75 0.03 & & magnum + 4213.34 & 5.11 @xmath75 0.19 & & crao + 4213.74 & 4.90 @xmath75 0.20 & 4.50 @xmath75 0.09 & mdm + 4214.34 & 4.65 @xmath75 0.20 & & crao + 4214.73 & 4.92 @xmath75 0.20 & 5.15 @xmath75 0.11 & mdm + 4215.40 & 4.72 @xmath75 0.21 & & crao + 4215.74 & 4.84 @xmath75 0.20 & 4.84 @xmath75 0.10 & mdm + 4216.32 & 5.15 @xmath75 0.21 & & crao + 4216.73 & 4.73 @xmath75 0.19 & 5.09 @xmath75 0.11 & mdm + 4217.35 & 4.49 @xmath75 0.23 & & crao + 4217.73 & 4.73 @xmath75 0.19 & 5.22 @xmath75 0.11 & mdm + 4218.31 & 4.58 @xmath75 0.19 & & crao + 4218.80 & 4.31 @xmath75 0.19 & 4.74 @xmath75 0.10 & mdm + 4218.91 & 4.70 @xmath75 0.04 & & magnum + 4219.32 & 4.73 @xmath75 0.21 & & crao + 4219.40 & 4.71 @xmath75 0.16 & & unebr . + 4219.83 & 5.06 @xmath75 0.20 & 4.70 @xmath75 0.10 & mdm + 4220.31 & 4.78 @xmath75 0.25 & & crao + 4220.40 & 5.05 @xmath75 0.16 & & unebr . + 4220.74 & 5.18 @xmath75 0.20 & 4.89 @xmath75 0.10 & mdm + 4221.35 & 5.23 @xmath75 0.38 & & crao + 4221.74 & 4.91 @xmath75 0.20 & 4.75 @xmath75 0.10 & mdm + 4221.99 & 4.78 @xmath75 0.08 & & magnum + 4222.40 & 5.08 @xmath75 0.26 & & crao + 4222.74 & 4.91 @xmath75 0.20 & 4.83 @xmath75 0.10 & mdm + 4223.38 & 4.47 @xmath75 0.25 & & crao + 4223.74 & 4.73 @xmath75 0.19 & 5.22 @xmath75 0.11 & mdm + 4224.37 & 5.12 @xmath75 0.19 & & crao + 4224.73 & 4.81 @xmath75 0.20 & 5.41 @xmath75 0.11 & mdm + 4225.35 & 4.46 @xmath75 0.16 & & crao + 4225.80 & 4.30 @xmath75 0.19 & 4.56 @xmath75 0.10 & mdm + 4225.88 & 4.43 @xmath75 0.03 & & magnum + 4226.29 & 4.46 @xmath75 0.20 & & crao + 4226.75 & 3.88 @xmath75 0.18 & 4.55 @xmath75 0.10 & mdm + 4227.43 & 4.28 @xmath75 0.23 & & crao + 4227.74 & 4.12 @xmath75 0.19 & 4.58 @xmath75 0.10 & mdm + 4228.80 & 3.81 @xmath75 0.18 & 4.54 @xmath75 0.10 & mdm + 4229.37 & 3.91 @xmath75 0.15 & & crao + 4229.78 & 3.91 @xmath75 0.18 & 4.31 @xmath75 0.09 & mdm + 4230.73 & 4.38 @xmath75 0.19 & 4.36 @xmath75 0.09 & mdm + 4231.36 & 4.07 @xmath75 0.16 & & crao + 4231.59 & 4.58 @xmath75 0.28 & & unebr . + 4231.75 & 4.81 @xmath75 0.20 & 4.61 @xmath75 0.10 & mdm + 4232.28 & 4.47 @xmath75 0.16 & & crao + 4232.38 & 4.59 @xmath75 0.22 & & unebr . + 4232.73 & 4.22 @xmath75 0.19 & 4.53 @xmath75 0.09 & mdm + 4233.32 & 4.51 @xmath75 0.15 & & crao + 4233.44 & 4.88 @xmath75 0.18 & & unebr . + 4233.73 & 4.64 @xmath75 0.19 & 4.72 @xmath75 0.10 & mdm + 4234.32 & 4.17 @xmath75 0.16 & & crao + 4234.73 & 4.51 @xmath75 0.19 & 4.69 @xmath75 0.10 & mdm + 4234.85 & 4.29 @xmath75 0.03 & & magnum + 4235.31 & 4.35 @xmath75 0.14 & & crao + 4235.46 & 4.80 @xmath75 0.36 & & unebr . + 4235.73 & 4.43 @xmath75 0.19 & 4.56 @xmath75 0.10 & mdm + 4236.31 & 4.53 @xmath75 0.15 & & crao + 4236.73 & 3.96 @xmath75 0.18 & 4.54 @xmath75 0.09 & mdm + 4237.31 & 4.17 @xmath75 0.14 & & crao + 4237.52 & 4.13 @xmath75 0.28 & & unebr . + 4237.73 & 3.94 @xmath75 0.18 & 4.28 @xmath75 0.09 & mdm + 4238.49 & 4.38 @xmath75 0.17 & & unebr . + 4238.73 & 4.16 @xmath75 0.19 & 4.42 @xmath75 0.09 & mdm + 4238.93 & 4.05 @xmath75 0.06 & & magnum + 4239.35 & 4.19 @xmath75 0.17 & & crao + 4239.43 & 4.34 @xmath75 0.18 & & unebr . + 4239.75 & 3.81 @xmath75 0.18 & 4.40 @xmath75 0.09 & mdm + 4240.29 & 3.69 @xmath75 0.14 & & crao + 4240.47 & 3.69 @xmath75 0.18 & & unebr . + 4240.72 & 3.97 @xmath75 0.18 & 4.37 @xmath75 0.09 & mdm + 4241.31 & 3.74 @xmath75 0.16 & & crao + 4241.38 & 3.92 @xmath75 0.24 & & unebr . + 4241.73 & 3.75 @xmath75 0.18 & 4.30 @xmath75 0.09 & mdm + 4242.29 & 3.75 @xmath75 0.14 & & crao + 4242.75 & 3.56 @xmath75 0.18 & 3.91 @xmath75 0.08 & mdm + 4243.29 & 3.35 @xmath75 0.19 & & crao + 4243.74 & 3.39 @xmath75 0.18 & 4.05 @xmath75 0.09 & mdm + 4244.78 & 3.68 @xmath75 0.18 & 3.90 @xmath75 0.08 & mdm + 4245.35 & 4.38 @xmath75 0.25 & & crao + 4245.75 & 4.36 @xmath75 0.19 & 4.03 @xmath75 0.09 & mdm + 4245.77 & 4.21 @xmath75 0.06 & & magnum + 4246.34 & 4.95 @xmath75 0.18 & & crao + 4246.40 & 4.49 @xmath75 0.17 & & unebr . + 4246.74 & 4.06 @xmath75 0.19 & 4.12 @xmath75 0.09 & mdm + 4247.74 & 4.57 @xmath75 0.19 & 4.20 @xmath75 0.09 & mdm + 4248.33 & 4.50 @xmath75 0.21 & & crao + 4248.73 & 4.50 @xmath75 0.19 & 4.48 @xmath75 0.09 & mdm + 4249.45 & 4.36 @xmath75 0.22 & & crao + 4249.74 & 4.40 @xmath75 0.19 & 4.52 @xmath75 0.09 & mdm + 4250.41 & 3.56 @xmath75 0.33 & & crao + 4250.74 & 4.07 @xmath75 0.19 & 4.44 @xmath75 0.09 & mdm + 4251.32 & 3.74 @xmath75 0.30 & & crao + 4251.74 & 4.02 @xmath75 0.19 & 4.39 @xmath75 0.09 & mdm + 4252.39 & 4.47 @xmath75 0.27 & & crao + 4252.73 & 4.15 @xmath75 0.19 & 4.22 @xmath75 0.09 & mdm + 4252.88 & 4.17 @xmath75 0.05 & & magnum + 4253.73 & 4.44 @xmath75 0.19 & 4.19 @xmath75 0.09 & mdm + 4254.35 & 4.62 @xmath75 0.16 & & crao + 4254.39 & 4.52 @xmath75 0.17 & & unebr . + 4255.38 & 4.73 @xmath75 0.24 & & crao + 4255.76 & 4.40 @xmath75 0.19 & 4.50 @xmath75 0.09 & mdm + 4256.35 & 5.02 @xmath75 0.14 & & crao + 4256.71 & 4.75 @xmath75 0.19 & 4.56 @xmath75 0.10 & mdm + 4257.40 & 4.80 @xmath75 0.17 & & crao + 4257.74 & 4.64 @xmath75 0.19 & 4.47 @xmath75 0.09 & mdm + 4258.35 & 4.52 @xmath75 0.16 & & crao + 4258.50 & 5.06 @xmath75 0.32 & & unebr . + 4258.76 & 4.83 @xmath75 0.20 & 4.91 @xmath75 0.10 & mdm + 4259.34 & 5.15 @xmath75 0.17 & & crao + 4259.75 & 4.92 @xmath75 0.20 & 4.97 @xmath75 0.10 & mdm + 4259.84 & 4.81 @xmath75 0.13 & & magnum + 4260.30 & 4.96 @xmath75 0.17 & & crao + 4260.75 & 4.36 @xmath75 0.19 & 4.70 @xmath75 0.10 & mdm + 4261.31 & 5.08 @xmath75 0.16 & & crao + 4261.42 & 5.15 @xmath75 0.14 & & unebr . + 4261.74 & 4.82 @xmath75 0.20 & 5.16 @xmath75 0.11 & mdm + 4262.30 & 4.78 @xmath75 0.16 & & crao + 4262.74 & 4.39 @xmath75 0.19 & 4.72 @xmath75 0.10 & mdm + 4263.35 & 4.68 @xmath75 0.17 & & crao + 4263.39 & 4.47 @xmath75 0.22 & & unebr . + 4263.72 & 4.52 @xmath75 0.19 & 5.01 @xmath75 0.11 & mdm + 4263.88 & 4.61 @xmath75 0.11 & & magnum + 4264.76 & 4.73 @xmath75 0.19 & 5.12 @xmath75 0.11 & mdm + 4265.76 & 5.43 @xmath75 0.20 & 5.15 @xmath75 0.11 & mdm + 4266.34 & 4.84 @xmath75 0.27 & 5.18 @xmath75 0.23 & crao + 4266.76 & 5.55 @xmath75 0.21 & 5.29 @xmath75 0.11 & mdm + 4267.30 & 4.43 @xmath75 0.26 & 5.43 @xmath75 0.24 & crao + 4267.75 & 4.55 @xmath75 0.19 & 5.24 @xmath75 0.11 & mdm + 4268.29 & 4.33 @xmath75 0.26 & 5.34 @xmath75 0.23 & crao + 4268.75 & 4.38 @xmath75 0.19 & 5.17 @xmath75 0.11 & mdm + 4269.32 & 4.57 @xmath75 0.26 & 5.16 @xmath75 0.23 & crao + 4269.75 & 4.56 @xmath75 0.19 & 5.39 @xmath75 0.11 & mdm + 4269.84 & 4.37 @xmath75 0.06 & & magnum + 4270.36 & 5.09 @xmath75 0.27 & 5.12 @xmath75 0.23 & crao + 4270.37 & 4.37 @xmath75 0.20 & & unebr . + 4271.31 & 4.52 @xmath75 0.26 & 5.38 @xmath75 0.24 & crao + 4272.85 & 4.54 @xmath75 0.04 & & magnum + 4274.31 & 4.78 @xmath75 0.26 & 5.16 @xmath75 0.23 & crao + 4275.81 & 4.31 @xmath75 0.03 & & magnum + 4276.41 & 4.80 @xmath75 0.20 & & unebr . + 4277.29 & 4.42 @xmath75 0.26 & 4.79 @xmath75 0.21 & crao + 4278.29 & 4.01 @xmath75 0.25 & 4.99 @xmath75 0.22 & crao + 4278.48 & 3.66 @xmath75 0.37 & & unebr . + 4278.84 & 4.14 @xmath75 0.06 & & magnum + 4279.31 & 4.33 @xmath75 0.26 & 4.74 @xmath75 0.21 & crao + 4279.32 & 4.22 @xmath75 0.21 & & crao + 4280.32 & 4.61 @xmath75 0.26 & 4.87 @xmath75 0.21 & crao + 4280.34 & 4.38 @xmath75 0.43 & & crao + 4281.32 & 4.34 @xmath75 0.24 & & crao + 4281.32 & 4.38 @xmath75 0.26 & 4.88 @xmath75 0.22 & crao + 4282.33 & 3.96 @xmath75 0.24 & & crao + 4282.35 & 4.22 @xmath75 0.25 & 4.47 @xmath75 0.20 & crao + 4283.31 & 3.78 @xmath75 0.25 & 4.63 @xmath75 0.20 & crao + 4283.33 & 3.75 @xmath75 0.24 & & crao + 4283.42 & 3.56 @xmath75 0.20 & & unebr . + 4284.30 & 3.64 @xmath75 0.24 & 4.50 @xmath75 0.20 & crao + 4284.31 & 3.87 @xmath75 0.18 & & crao + 4285.77 & 3.70 @xmath75 0.09 & & magnum + 4289.29 & 4.20 @xmath75 0.25 & 4.22 @xmath75 0.19 & crao + 4289.45 & 3.65 @xmath75 0.25 & & unebr . + 4290.30 & 3.85 @xmath75 0.25 & 4.41 @xmath75 0.19 & crao + 4290.40 & 4.27 @xmath75 0.22 & & unebr . + 4291.30 & 3.53 @xmath75 0.24 & 4.34 @xmath75 0.19 & crao + 4294.29 & 4.63 @xmath75 0.18 & & crao + 4295.32 & 4.51 @xmath75 0.34 & & crao + 4296.29 & 4.35 @xmath75 0.26 & 4.67 @xmath75 0.21 & crao + 4296.30 & 4.57 @xmath75 0.17 & & crao + 4298.32 & 4.78 @xmath75 0.26 & 4.38 @xmath75 0.19 & crao + 4299.28 & 4.29 @xmath75 0.26 & 4.65 @xmath75 0.20 & crao + 4299.31 & 4.87 @xmath75 0.16 & & crao + 4300.28 & 4.14 @xmath75 0.25 & 4.68 @xmath75 0.21 & crao + 4304.79 & 4.66 @xmath75 0.11 & & magnum + 4311.76 & 4.79 @xmath75 0.09 & & magnum cc @xmath43 & @xmath95 days + @xmath41 & @xmath96 days + @xmath97 & @xmath98 km s@xmath16 + fwhm ( mean ) & @xmath99 km s@xmath16 + @xmath100 & @xmath101 km s@xmath16 + fwhm ( rms ) & @xmath102 km s@xmath16 + @xmath9 & @xmath103 + @xmath104 & @xmath105 +

we present results from a monitoring campaign performed with the _ chandra x - ray observatory _ of the gravitationally lensed quasars rx j1131 and he 1104 . we detect significant x - ray variability in all images of both quasars . the flux variability detected in image a of rx j1131 is of particular interest because of its high amplitude ( a factor of 20 ) . we interpret it as arising from microlensing since the variability is uncorrelated with that of the other images and the x - ray flux ratios show larger changes than the optical as we would expect for microlensing of the more compact x - ray emission regions . the differences between the x - ray and optical flux ratios of he 1104 are less dramatic , but there is no significant soft x - ray or dust absorption , implying the presence of x - ray microlensing in this system as well . combining the x - ray data with the optical light curves we find that the x - ray emitting region of he 1104 is compact with a half - light radius , where the gravitational radius is r = 3.6 10 cm , thus placing significant constraints on agn corona models . we also find that the microlensing in he 1104 favors mass models for the lens galaxy that are dominated by dark matter . finally , we better characterize the massive foreground cluster near rx j1131 , set limits on other sources of extended x - ray emission , and limit the fluxes of any central odd images to be 3050 ( 3 ) times fainter than the observed images . # 1 # 1

0805.4492

resolving the emission regions of distant quasars is beyond the current capabilities of present - day telescopes , as direct imaging of accretion disks requires angular resolutions on the order of tens of nanoarcseconds at @xmath7 @xmath1 1 . until the spatial resolution of telescopes reaches this limit , we will have to rely on indirect methods of mapping the emission regions of quasars . such methods include light - travel time arguments , reverberation mapping of the broad line region ( blandford & mckee 1982 ; peterson 1993 , netzer & peterson 1997 ) , reverberation mapping of the fe k@xmath8 emission region ( young & reynolds 2000 ) , occultation measurements of the central x - ray source by orbiting compton thick clouds ( risaliti et al , 2007 ) , and microlensing of the continuum and line emission regions ( e.g. , grieger et al . 1988 and 1991 ; schneider , ehlers & falco 1992 ; gould & gaudi 1997 ; agol & krolik 1999 ; yonehara et al . 1999 ; mineshige & yonehara 1999 ; chartas et al . 2002a ; popovic et al . 2003 ; blackburne et al . 2006 ; pooley et al . 2006 , 2007 ; kochanek et al . 2007 ; jovanovic et al . 2008 ; morgan et al . 2008a , 2008b ) . we observe microlensing as uncorrelated variability in the brightnesses of the images of a lensed quasar , where the amplitude of the variability is determined by the size of the emission regions ( e.g. , lewis et al . 1998 ; popovic & chartas 2005 ) . the largest components , the radio , dust and optical line emission regions , should show little or no microlensing variability . the thermal continuum emission from the disk should show greater variability at shorter wavelengths corresponding to smaller disk radii and higher temperatures , and this has been observed by poindexter et al . ( 2008 ) and anguita et al . the non - thermal x - ray emission is thought to be dominated by inverse compton scattering of uv photons from the disk ( e.g. , reynolds & nowak 2003 ) , some of which is reprocessed into fe k@xmath8 line emission ( e.g. , george & fabian 1991 ; mushotzky et al . 1993 ) but the geometry and scale of these emission regions is not well understood . the effects of microlensing on the x - ray fluxes of lensed images have been reported for many lenses at this point , including q0957 + 561 ( e.g. , chartas et al . 1995 ) , rx j0911 + 0511 ( e.g. , morgan et al . 2001 ; chartas et al . 2001 ) , h1413 + 117 ( e.g. , chartas et al . 2004 , 2007 ) , pg 1115 + 080 ( pooley et al . 2006 ; morgan et al . 2008b ) , and rx j1131@xmath01231 ( e.g. , blackburne et al . 2006 ; kochanek et al . 2007 ; dai et al . 2008 ) . the differences between the x - ray flux ratios and optical flux ratios can be used to estimate the source size after accounting for absorption by the interstellar medium of the lens ( e.g. , dai & kochanek 2008 ) and the effects of substructures ( satellites ) in the lens halo ( mao & schneider 1998 ) . pooley et al . ( 2007 ) used observed x - ray and optical flux ratios for ten four - image lensed quasars to show that the x - ray source is much more strongly microlensed than the optical source , indicating that the x - ray emission region is more compact than the optical . they also found that their estimates of the optical source sizes were significantly larger than would be expected based on the observed , magnification - corrected optical fluxes . morgan et al . ( 2008a ) measured the sizes of the optical emission regions of 11 quasars using microlensing to find that the sizes scaled with black hole mass as expected from thin disk theory ( shakura & sunyaev 1973 ) and had approximately the sizes expected from thin disk theory , but confirmed the pooley et al . ( 2007 ) result that these sizes , both microlensing and theoretical , are larger than expected given the observed optical flux and the same disk model . accurate microlensing estimates of the sizes of the x - ray emission regions requires x - ray light curves rather than isolated epochs . first , the absolute magnifications of lensed images are not well understood because substructure in the lens galaxy can modify the fluxes from the predictions of lens models ( e.g. , mao & schneider 1998 ; dalal & kochanek 2002 ) . while comparing optical and x - ray flux ratios provides estimates for the difference in size between the emission regions , the lack of a secure estimate of the intrinsic flux ratios makes it difficult to set an absolute scale . second , quasars have intrinsic variability that appears in the images with relative time delays , which means that instantaneous flux ratios are contaminated by the effects of time variability modulated by the delays . third , while the source size ultimately determines the amplitude of microlensing variability , the instantaneous values are a combination of the source size and the location of the source in the complex microlensing magnification patterns . monitoring the variability minimizes these problems and should lead to far more accurate estimates of the sizes . we have been monitoring roughly 25 lenses in the optical to measure time delays ( e.g. , kochanek et al . 2006 ) and to study quasar structure ( e.g. , morgan et al . 2008a ; poindexter et al . 2008 ) . from this sample we selected a small subset with reasonable x - ray fluxes for monitoring with the _ chandra x - ray observatory_. given a well - sampled optical light curve that can be used to determine the size of the optical / uv emission region and act as a microlensing variability `` reference , '' we can afford to sample the x - ray microlensing variability relatively sparsely . in this paper we present the x - ray observations of the lensed quasars rx j1131@xmath01231 ( @xmath9 ) and he 1104@xmath01805 ( @xmath10 . the optical data for rx j1131@xmath01231 and he 11041805 are presented in morgan et al . ( 2006 ) and poindexter et al . ( 2007 ) respectively , and the optical microlensing results are presented in morgan et al . ( 2008a ) and poindexter et al . ( 2008 ) . here we carry out a microlensing analysis of he 1104@xmath01805 , while a companion paper , dai et al . ( 2009 ) , does so for rx j1131@xmath01231 . we discuss the observations and data analysis in 2 , study the microlensing of he 11041805 in 3 and summarize our results in 4 . throughout this paper we adopt a @xmath11-dominated cosmology with @xmath12 = 70 km s@xmath13 mpc@xmath13 , @xmath14 = 0.7 , and @xmath15 = 0.3 . we monitored rx j1131@xmath01231 and he 1104@xmath01805 with the advanced ccd imaging spectrometer ( acis ; garmire et al . 2003 ) on board the _ chandra x - ray observatory _ ( hereafter _ chandra _ ) using short ( @xmath1 5 ksec ) snapshot observations . we obtained 5 epochs for each lens in 2006 placing the sources on the back - illuminated s3 chip of acis . we combined these with archival observations of rx j1131@xmath01231 in april 2004 and he 1104@xmath01805 in february 2000 . a log of the observations that includes observation dates , observed count rates , total exposure times , and observation identification numbers is presented in table 1 . we analyzed the data using the standard software ciao 4.0 provided by the cxc . we used standard cxc threads to screen the data for status , grade , and time intervals of acceptable aspect solution and background levels . we removed the @xmath16 025 spatial randomization applied to the event positions by the standard pipeline and instead used the sub - pixel resolution techniques developed by tsunemi et al . ( 2001 ) and mori et al . ( 2001 ) in order to improve the image resolution . the _ chandra _ spectra of rx j1131@xmath01231 and he 1104@xmath01805 were fit with a variety of models employing ` xspec ` version 12 ( arnaud 1996 ) . for all spectral models of rx j1131@xmath01231 and he 1104@xmath01805 we included galactic absorption due to neutral gas ( dicky & lockman 1990 ) with column densities of @xmath17= 3.6 @xmath4 10@xmath18 @xmath19 and 4.6 @xmath4 10@xmath18 @xmath19 , respectively . we use 90% confidence level uncertainties unless otherwise stated . for estimating the x - ray counts of images b , c , and d of rx j1131@xmath01231 we extracted events from circular regions with radii of 1.5 arcsec slightly off - center from the images to reduce contamination from nearby images ( see panel ( c ) of fig . 1 ) . to estimate the x - ray counts of image a of rx j1131@xmath01231 , which lies between images b and c , we extracted events within a 0.75 arcsec circular region centered on a. aperture corrections were applied to all images to account for counts not included in the extraction regions . for estimating the x - ray counts of images a and b of he 1104@xmath01805 we extracted events from circular regions centered on the images with radii of 1.5 arcsec . the backgrounds for rx j1131@xmath01231 and he 1104@xmath01805 were determined by extracting events within an annulus centered on the mean location of the images with inner and outer radii of 7.5 arcsec and 50 arcsec , respectively . contamination between images was minimized and made insignificant by the application of the sub - pixel resolution technique and the use of appropriately sized and off - center extraction apertures . we also corrected for pile - up , " an instrumental effect that occurs when two or more x - ray photons strike individual or neighboring ccd pixels within one frame time . neglecting these corrections can cause spectral distortion , grade migration and distortion of the image psf . we used the forward spectral - fitting tool lynx ( chartas et al . 2000 ) to estimate the fraction of events lost due to the pile - up effect . table 1 summarizes the observed 0.210 kev band source counts . the pile - up corrections needed to correctly estimate these counts can be significant . for example the counts for images b and c in the 2004 april 12 observation of rx j1131@xmath01231 have pile - up corrections of 37% and 16% respectively . in their analysis of the 2004 observation of rx j1131@xmath01231 , blackburne et al . ( 2006 ) did not include pile - up corrections , so our present results should be more reliable . in particular , pile - up usually leads to energy spectra that appear harder , and this likely explains the differences in the x - ray spectral hardness ratios reported by blackburne et al . ( 2006 ) and our present results . in figure 1 we show the lucy - richardson deconvolved images in the 0.28 kev bandpass of the _ chandra _ observations of rx j1131@xmath01231 and he 1104@xmath01805 . to estimate the relative x - ray image positions of rx j1131@xmath01231 and he 1104@xmath01805 we modeled the _ chandra _ images using point - spread functions ( psfs ) generated by the simulation tool ` marx ` ( wise et al 1997 ) . the x - ray event locations were binned with a bin - size of 00246 to sample the psf sufficiently ( an acis pixel subtends 0491 ) . the simulated psfs were fitted to the _ chandra _ data by minimizing the @xmath20-statistic formed between the observed and simulated images . in table 2 we compare the x - ray image separations to the observed nicmos values reported by morgan et al . ( 2006 ) for rx j1131@xmath01231and by lhar et al . ( 2000 ) for he 1104@xmath01805 . we conclude that the x - ray and optical image positions are consistent given the estimated uncertainties . the deconvolved images of rx j1131@xmath01231 and he 1104@xmath01805 do not show any additional lensed images . to obtain quantitative limits on possible additional images located at the centers of rx j1131@xmath01231 and he 1104@xmath01805we extracted the 0.210 kev counts within 0.4 arcsec circles centered on the images at the mean lens position of each quasar . the backgrounds in the central source extraction regions are dominated by the contamination from the images the instrumental and cosmic backgrounds in the 0.4 arcsec extraction circles are significantly lower than the contamination . to estimate the fractional contamination per image we used the 0.210 kev counts in 0.4 arcsec circular apertures placed 1 and 2 arcsec north of image b in rx j1131@xmath01231 and 1 and 2 arcsec northwest of image a in he 1104@xmath01805 . the detected 0.210 kev counts in the central 0.4 arcsec apertures are consistent with the estimated background and contamination from the bright images . by combining all the data for each lens we set 3@xmath6 upper limits on the 0.210 kev flux of any central , odd image of @xmath21 and @xmath22 ergs s@xmath13 @xmath19 for rx j1131@xmath01231 and he 1104@xmath01805 , respectively . these limits corresponded to a 3@xmath6 limit on the flux ratio of a factor of 50 relative to image c in rx j1131@xmath01231 , and a factor of 30 relative to image a in he 1104@xmath01805 . while relatively tight , these limits are not strong enough to constrain the central surface density of the lens ( keeton 2003 ) . for example , in the models we use for he 1104@xmath01805 below , the expected flux ratio between image a and any central image is @xmath23 . we performed fits to the individual spectra of the images of rx j1131@xmath01231 and he 1104@xmath01805 using events in the 0.48 kev energy range with a model that consisted of a simple power law modified by galactic absorption . due to the moderate s / n of the spectra , we performed these fits using the cash statistic which does not require binning of the data , although if we instead use @xmath24 statistics we find similar results . the best - fit parameters of these fits and the unabsorbed 0.22 kev and 2 - 10 kev fluxes are presented in tables 2 and 3 . we found no significant variability ( within the 90% errors ) of the photon indices ( @xmath25 ) of the spectra of the images of rx j1131@xmath01231 and he 1104@xmath01805 with the exception of the 2004 observation of rx j1131@xmath01231 that showed significant differences of @xmath25 between images a / b / c with @xmath26 and image d with @xmath27 . we also considered models that included neutral absorption at the redshifts of the sources or lenses of rx j1131@xmath01231 and he 1104@xmath01805 . we do not detect any additional neutral absorption at these redshifts . this is consistent with the marginal detection of absorption in he 1104@xmath01805 in our earlier analysis ( dai et al . 2006 ) . in figure 2 we show the _ chandra _ image of the 2004 observation of the lensed system rx j1131@xmath01231 and the surrounding field . to reduce background contamination and to enhance possible soft extended x - ray emission we filtered the image to include only photons with energies ranging between 0.4 and 3.0 kev . the image was binned with a bin size of 0.5 arcsec and adaptively smoothed with the tool csmooth developed by ebeling et al . csmooth smooths a two - dimensional image with a circular gaussian kernal of varying radius . we confirm the extended soft x - ray emission reported by morgan et al . ( 2006 ) centered at 11 32 1.6 , @xmath012 31 6.5 ( j2000 ) and about 158 arcsec northeast of image d of rx j1131@xmath01231 . this emission comes from a foreground cluster at @xmath28 . morgan et al . ( 2006 ) also reported a possible detection at a low significance level ( 3 @xmath6 ) of extended emission centered 33 arcsec southwest of rx j1131@xmath01231 at 11 31 50.1 , @xmath012 32 23 ( j2000 ) . we combined all six observations of rx j1131@xmath01231 listed in table 1 , for a total exposure time of 33.8 ksec , to better limit this possible source . all the observations from 2006 were performed in sub - array mode to mitigate pile - up and with the reduced field of view they do not include the foreground cluster to the northeast of the lens . figure 3 shows the adaptively smoothed image including only photons with energies from 0.4 to 3.0 kev . we find no significant sources of extended emission near the lens beyond the foreground cluster . the 0.22 . kev @xmath29 upper flux and luminosity limits are @xmath30 ergs @xmath19 s@xmath13 and @xmath31 ergs s@xmath13 , respectively , assuming a cluster with a temperature of @xmath32 kev at the lens redshift with an abundance of @xmath33 solar and an extraction radius of 30 arcsec . we fit the emission from the foreground cluster using a @xmath34 model for the cluster brightness profile combined with a uniform background of 0.005 events per pixel . prior to performing the fit we binned the image in 1 pixels and smoothed this with a gaussian ( @xmath6 = 3 ) . the fits were performed with the cxc software package ` sherpa ` . we find that the cluster center is @xmath35 = 1502 @xmath16 05 east and @xmath36 = 523 @xmath16 05 north of image d. the smoothed intensity distribution is nearly round with an ellipticity of @xmath37 0.10 @xmath16 0.3 . the best - fit values for @xmath34 and the core radius of the cluster are @xmath34 = 0.4 @xmath16 0.2 and r@xmath38 = 42@xmath39 ( 8 kpc ) , respectively . we extracted the spectrum of this @xmath40 cluster from a 50 arcsec radius circle centered on the x - ray cluster center . this spectrum was fit with a simple model consisting of an emission spectrum from hot diffuse gas based on the xspec model mekal modified by galactic absorption . we obtain best - fit values for the temperature and metal abundances of 1.2@xmath41 kev and a = 0.5@xmath42 , respectively ( both errors are at the 90% confidence level ) . the 210 kev luminosity of this cluster of galaxies is 1.7 @xmath4 10@xmath43 ergs sec@xmath13 . these are consistent with the morgan et al . ( 2006 ) estimates and the values for the temperature , @xmath34 and the core radius are consistent with the observed correlations of clusters ( e.g. jones & forman 1999 ) . we also stacked the images of the six observations of he 1104@xmath01805listed in table 1 for a total exposure time of 71.8 ksec . figure 4 shows the adaptively smoothed image of he 1104@xmath01805 including only photons with energies from 0.4 to 3.0 kev . we find no significant sources of extended emission . the 0.22 kev @xmath29 upper flux and luminosity limits are @xmath44 ergs @xmath19 s@xmath13 and @xmath45 ergs s@xmath13 , respectively , assuming a cluster with a temperature of @xmath32 kev at the lens redshift with an abundance of @xmath33 solar and an extraction radius of 30 arcsec . the 0.210 kev time - delay corrected light curves of the images of rx j1131@xmath01231 and he 1104@xmath01805 are shown in figures 5a and 5b respectively . significant x - ray flux variability is detected in all images . rx j1131@xmath01231 shows a mixture of intrinsic and microlensing variability . if we normalize the first epoch for image d to match the optical light curves , we find that images b , c and d roughly track the optical light curves and show similarities in their time variability . in figure 6a we show the flux ratios f@xmath46/f@xmath47 , f@xmath48/f@xmath47 , and f@xmath49/f@xmath47 in the 0.210 kev and @xmath50-bands of rx j1131@xmath01231 . the flux ratio f@xmath48/f@xmath47 is almost constant in both the x - ray and optical bands indicating that the observed variability in images b and c is correlated and therefore mostly intrinsic in origin . image a shows a completely different behavior . it is a factor of @xmath1 4 fainter in the x - rays than the optical relative to c / d in 2004 , comparable to the optical in the spring of 2006 , and a factor of @xmath1 5 brighter by the fall of 2006 . since the 2006 observations come in clusters with temporal separations of order the a / b / c time delays , we can be confident that these differences are not due to intrinsic variability modulated by the time delay . the significant rise of the f@xmath46/f@xmath47 ratio in the x - ray band compared to the one measured in the @xmath50-band supports our microlensing interpretation of the flux enhancement in image a of rx j1131@xmath01231 . we note that the flux ratios reported in our analysis for the 2004 observation of rx j1131@xmath01231 are different from those reported by blackburne et al . ( 2006 ) because of the correction for the loss of counts due to pile - up and the correction for the spectral distortion due to pile - up . we see no strong evidence for variations in the spectral hardness , but this is a weak statement given the limited count rates . he 1104@xmath01805is also affected by intrinsic x - ray variability . in particular , the peak in the optical flux near 4100 days seems likely to be associated with a peak in the x - ray flux . in figure 6b we show the flux ratios f@xmath46/f@xmath48 in the 0.210 kev and @xmath50-bands of he 1104@xmath01805 . the february / march and october / november epochs are roughly separated by the time delay , so we can reasonably conclude that the x - ray flux ratio near 4050 days ( observed a ) is approximately f@xmath46/f@xmath48@xmath51 and very different from the optical flux ratio of f@xmath46/f@xmath48@xmath52 . since there is no evidence for significant soft x - ray or dust absorption ( see 2.3 , dai et al . 2006 , poindexter et al . 2008 ) this difference in flux ratios must be due to microlensing . in 3 we proceed with fitting the light - curves of he 1104 - 1805 with a sophisticated microlensing model . the microlensing analysis of the light - curves of rx j 1131 - 1231 will be presented in a companion paper by dai et al . ( 2009 ) . as with rx j1131@xmath01231 , we see no evidence for variations in the spectral hardness , largely due to the large uncertainties . we modeled the microlensing in he 1104@xmath01805 as in morgan et al . ( 2008a ) and poindexter et al . ( 2008 ) using a fixed 162 day time delay , the @xmath50-band light curve constructed from our smarts data ( poindexter et al . 2007 ) , ogle ( wyrzykowski et al . 2003 ) , and ofek & maoz ( 2003 ) , and the delay - corrected x - ray flux measurement . the main free parameters of our lensing model for he 1104@xmath01805are the x - ray and uv ( rest - frame ) scale lengths , a microlens mass scale , a mass fraction of the local surface density comprised of stars , and a velocity vector describing the motion of the agn emission region across the microlensing caustics . we used the lens model sequence from poindexter et al . ( 2007 ) , where we start from a constant mass - to - light ratio model for the mass distribution , defined by @xmath53 , and then reduce its mass in 10% increments while adding an nfw halo , where @xmath54 would correspond to a pure halo model . the measured time delay requires @xmath55 . we made 8 random realizations of the star fields near each image and then generated @xmath56 pixel magnification maps using the methods of kochanek ( 2004 ) . the maps had an outer scale of @xmath57 cm and an inner scale of @xmath58 cm . based on its emission line widths , peng et al . ( 2006 ) estimated that the black hole mass in he 1104@xmath01805is @xmath59 , corresponding to a gravitational radius of @xmath60 cm that is well - resolved in the magnification maps . we modeled the surface brightness of the emission regions as a face - on , thin disk ( shakura & sunyaev 1973 ) without the central temperature depression , @xmath61^{-1 } \label{eqn : fnu}\ ] ] where the scale length @xmath62^{1/3}\ ] ] @xmath63 is the radius at which the disk temperature matches the rest wavelength of the observations , @xmath64 , @xmath65 is the accretion rate , @xmath66 is the luminosity in units of the eddington luminosity and @xmath67 is the accretion efficiency . we scale the results assuming a radiative efficiency of @xmath68 . while the available energy of accretion is larger , the division of that energy between radiative , advective and kinetic losses is not directly calculable at present . studies of the growth of black holes tend to find that the average radiative efficiency of accretion is less than 10% ( e.g. shankar et al . 2007 ) , and low radiative efficiencies are also needed to reconcile thin disk theory with microlensing accretion disk size estimates ( e.g. , morgan et al . 2007 ) . for the uv ( rest - frame ) emission we can neglect the central temperature depression provided @xmath69 where @xmath70 is the inner edge of the disk . while this thermal accretion disk model is not appropriate for the non - thermal x - ray emission , mortonson et al . ( 2005 ) have shown that microlensing essentially measures the half - light radius of the emission region , so we used the same profile for the x - rays to simplify the computations but regard it as a measurement of the half - light radius @xmath71 . the microlenses were given a power - law mass distribution @xmath72 with a dynamic range in mass of a factor of 50 , roughly matching the galactic disk mass function of gould ( 2000 ) , and we use a uniform prior on the mean microlens mass of @xmath73 . we modeled the probability distribution for the physical velocities as in poindexter et al . ( 2008 ) where our projected velocity onto the lens plane is @xmath74 km / s , the velocity dispersion of the lens is @xmath75 km / s based on the critical radius of the mass model , and the lens and source have rms peculiar velocities of @xmath76 and @xmath77 km / s respectively . the data were then modeled using the bayesian monte carlo method of kochanek ( 2004 ) . the microlensing data constrains the source velocity and size in `` einstein '' length units of @xmath78 cm . we can determine the true sizes using the prior on the microlens masses , the prior on the physical velocities or both . figure 7 shows the probability distribution found for @xmath79 assuming the prior on the physical velocities . while the distribution peaks at somewhat lower mass ( @xmath80 ) than the range of our prior ( @xmath81 to @xmath82 ) , the distribution is so broad ( @xmath83 to @xmath84 at 1-@xmath6 ) that we can view them as statistically consistent . for example , none of the results depend significantly on the inclusion of the mass prior . since the mass scales as @xmath85 where @xmath86 is the physical velocity from the prior and @xmath87 is the einstein - unit velocity estimated from the microlensing models , it is difficult to tightly constrain the mean mass with a single lens given the unknown peculiar velocities of the lens and source even with perfect knowledge of the einstein - unit velocity . figure 8 shows the raw estimates of the source sizes @xmath88 in einstein units , where @xmath89 . both of these are scaled source sizes @xmath90 defined by the emission profile presented earlier in this section . the consequence of adding the prior on the microlens masses is to drive the source sizes @xmath88 to be modestly smaller . in essence , a given level of microlensing variability can be created by a large source moving rapidly ( big @xmath88 and @xmath87 ) or a small source moving more slowly ( small @xmath88 and @xmath87 ) . but high effective velocities correspond to small mass scales given a fixed physical velocity , so imposing the mass prior given the distribution seen in figure 7 will favor lower velocities @xmath87 and smaller sources @xmath88 . it will have less effect on the physical source size @xmath89 because the smaller size @xmath88 is balanced by the larger mass scale @xmath91 ( see kochanek 2004 ) . while figure 8 truncates the probability distribution for the x - ray source at the pixel scale of the magnification patterns , it would not converge to zero if we used magnification patterns with smaller pixel scales to extend the distribution to smaller source sizes . this is a consequence of not having a well - sampled x - ray light curve . the source must be small enough to allow the x - ray flux ratio to differ from the optical , but still smaller sources are disfavored because even more discrepant flux ratios become more likely . however , reducing the size still further has no effect on the probabilities because it converges to the likelihood of the data points avoiding precise caustic crossings . unfortunately , this does mean that we will obtain robust upper limits on the x - ray source size but not lower limits . figure 9 shows the results for the rest - frame uv and x - ray source sizes @xmath90 assuming either uniform or logarithmic priors for the size . we used @xmath90 for both so that the sizes are easily compared . if we regard the source size at 2000 as measuring a projected area , then @xmath90 scales as @xmath92 for inclination angle @xmath93 of a thin inclined accretion disk . similarly , the x - ray size in figure 9 should be increased by a factor of @xmath94 to convert from @xmath90 to the half - light radius @xmath95 . the disk size at 2000 ( @xmath50-band in the rest frame at @xmath96 ) is @xmath97 for a logarithmic prior on the size and @xmath98 for a linear prior on the size . this agrees well with our earlier estimates in poindexter et al . ( 2008 ) and morgan et al . ( 2008a ) . the x - ray size is less certain and more prior - dependent , with an estimated half - light radius @xmath99 for a logarithmic prior on the size and @xmath100 for the linear prior on the size . we note that the difference in the sizes is entirely driven by the significant differences between the x - ray and optical flux ratios at the time of the x - ray observation . the microlensing data also prefer lens models with low stellar mass fractions , as shown in figure 10.the probability distribution peaks at @xmath101 , in relatively good agreement with the value of @xmath102 that agrees best with the measured time delay and a hubble constant of @xmath103 km s@xmath13 mpc@xmath13 . while this preliminary result does not lead to a microlensing constraint on @xmath104 that is tight enough to break the degeneracies between the radial mass distribution and the hubble constant ( see kochanek 2002 ) , it is encouraging that there is agreement with the time delays given the expected value of @xmath105 and that the microlensing requires a dark matter - dominated mass distribution . a detailed microlensing analysis of rx j1131@xmath01231 that provides similar constraints on the structure of the lens and the structure of the quasar is presented in dai et al . ( 2009 ) . figures 11 and 12 show an example of a trial that fits the data well . in this trial the einstein unit source sizes are @xmath106 and @xmath107 for the x - ray and uv rest - frame sources respectively and the effective velocity corresponds to a sensible microlens mass scale . we chose this solution mainly because of the curious origin of the initial @xmath50-band peak , where what might be interpreted as a single heavily smoothed caustic crossing is actually a pair of competing caustic crossings in the a and b images , where the a image dominates because of its shallower magnification gradients . the x - ray epoch then corresponds to time when the source is in a valley between caustics in both images . we note that this particular solution probably would not reproduce the wavelength dependence of the optical and ir flux ratios in he 1104@xmath01805 discussed and modeled poindexter et al . while there is a consensus that the origin of the kev x - ray continuum emission from quasars is inverse compton scattering ( e.g. , see the review of reynolds & nowak 2003 ) , the extent and geometric configuration of the emission region is less clear . in the standard disk - corona model ( e.g. , haardt & maraschi 1991 ; merloni 2003 ) , it is produced in a hot , extended corona surrounding the disk . in the general relativistic mhd simulations of hirose et al . ( 2004 , also machida & matsumoto 2003 ) the region of the corona with the highest current densities lies close to the inner edge of the disk . under the assumption that these are also the regions with the highest dissipation rates for heating the corona ( see hirose et al . 2004 ) , they will dominate the x - ray emission and we would expect the x - ray source size to track the inner edge of the disk . microlensing of the x - ray source allows us to observationally measure the size of the x - ray emission region . we start with well - sampled optical light curves that allow us to determine the size of the accretion disk . morgan et al . ( 2008a ) found a clear scaling of the disk size with black hole mass that is consistent with the @xmath108 scaling of simple thin disk theory for a fixed eddington accretion rate ( shakura & sunyaev 1973 ) . we can estimate the disk size by two routes other than microlensing . first we can estimate it from the observed flux at some wavelength . size estimates from the observed flux do not depend on the black hole mass but depend only on the assumed temperature profile and the assumption that the disk radiates as a black body at that temperature . second , thin disk theory predicts the scale length through the expression we presented in the beginning of 3 . size estimates from thin disk theory depend on the black - hole mass , the eddington factor and the wavelength . the disk size from our microlensing analysis is roughly consistent with thin disk theory if the accretion efficiency is relatively low , but both the microlensing and theoretical sizes are significantly larger than that estimated from the observed optical flux of the quasar and the same disk model . this mismatch between the sizes estimated from microlensing and from the flux was first pointed out by pooley et al . one solution to this problem may be to use a shallower temperature profile than the @xmath109 of simple thin disk theory , and poindexter et al . ( 2008 ) found some evidence for a shallower profile in he 1104@xmath01805 by examining the optical / near - ir wavelength dependence of microlensing in this system . our x - ray monitoring observations of rx j1131@xmath01231 and he 1104@xmath01805 show significant flux variability in all images of both quasars . the factor of @xmath110 x - ray flux variability in image a of rx j1131@xmath01231 is interesting not only because of the high amplitude but also because it is a highly magnified saddle point image . schechter & wambsganss ( 2002 ) pointed out that in a dark matter dominated galaxy , microlensing is in a regime where images at minima and saddle points of the time - delay surface behave differently . in particular , the saddle point images should show significantly higher variability amplitudes , just as we observe in rx j1131@xmath01231 . the relatively large sizes of accretion disks in the optical relative to the typical einstein radius of the stars seems to have masked much of this effect in the optical signatures of microlensing for example , where the x - ray flux ratio @xmath111 changed by a factor of 4 during our monitoring period , the optical flux ratio changed by only a factor of 1.4 . the differences between the x - ray and optical flux ratios of he 1104@xmath01805 also imply the presence of x - ray microlensing . neither system shows evidence for soft x - ray or dust absorption that could explain the differences . while we measure the source sizes by modeling the data in detail , the essence of the measurement is comparing the amplitude of the x - ray microlensing variability to that in the optical . if the amplitude for the x - ray microlensing is larger than in the optical , then the x - ray source must be more compact , and this is what we observed in both rx j1131@xmath01231 and he 1104@xmath01805 . given the estimated black hole mass of @xmath112 ( peng et al . 2006 ) , our fits to the x - ray and optical light curves using the microlensing model described in 2.5 indicate that the x - ray emission in he 1104@xmath01805 comes from a region smaller than @xmath113 and the uv emission region has a size of @xmath114 . we note that the source sizes of he 1104@xmath01805 were derived using a bayesian analysis which requires a prior . as shown in figures 8 and 9 our upper limit on the x - ray source size is robust to prior selection but not the lower limit . obtaining a more precise x - ray size estimate requires a more extensive x - ray light curve for the system . this is somewhat challenging for he 1104@xmath01805 because the long time delay of he 1104@xmath01805 ( 162 days , morgan et al . 2008a ) and the intrinsic variability we observe here means that this lens requires observations separated by the time delay in order to correct for intrinsic variability . this is less of a problem for almost all other candidates for x - ray monitoring because they have far shorter time delays . for example , in rx j1131@xmath01231 we observe some intrinsic variability but it is significant only on time scales much longer than the delays between the a / b / c images . given the morgan et al . ( 2008b ) estimates for pg 1115 + 080 , the dai et al . ( 2009 ) results for rx j1131@xmath01231 , and our present results for he 1104@xmath01805 , the x - ray microlensing results suggest that the size of the x - ray emission arises from close to the inner edge of the disk . thus , our estimated sizes for the x - ray and uv emission regions place significant constraints on agn models that assume extended coronal envelopes around accretion disks . reducing the uncertainties in these measurements and measuring the scaling of the size with the estimated masses of the black holes requires longer x - ray time series and more systems , but there is clearly no technical barrier to making the measurements and extending them to the dependence of the size on x - ray energy . combining the data from the time series also allows us to search for extended emission from groups or clusters near the lens , which can be important for lens models ( e.g. williams et al . 2006 , fassnacht et al . 2007 ) . here we analyzed the x - ray data for the foreground cluster found near rx j1131@xmath01231 by morgan et al . ( 2006 ) in more detail , and set limits on the presence of any other clusters near both systems ( see 2.4 ) . we also used the combined images to search for central or odd images , although this is challenging given the resolution of _ chandra _ and the expected faintness of the images . while we reached 1@xmath6 limits on the flux ratios of @xmath115 , significant constraints on the central images require limits better by another order of magnitude ( keeton 2003 ) . agol , e. , & krolik , j. 1999 , apj , 524 , 49 + anguita , t. , schmidt , r.w . , turner , e.l . , wambsganss , j. , webster , r.l . , loomis , k.a . , long , d. , & mcmillan , r. , 2008 , a&a , 480 , 327 + arnaud , k. a. 1996 , asp conf . ser . 101 : astronomical data analysis software and systems v , 5 , 17 + blackburne , j. a. , pooley , d. , & rappaport , s. 2006 , apj , 640 , 569 + blandford , r. d. , & mckee , c. f. 1982 , apj , 255 , 419 + chartas , g. , et al . 2000 , , 542 , 655 + chartas , g. , agol , e. , eracleous , m. , garmire , g. , bautz , m. w. , & morgan , n. d. 2002a , apj , 68 , 509 + chartas , g. , gupta , v. , garmire , g. , jones , c. , falco , e. e. , shapiro , i. i. , & tavecchio , f. 2002b , , 565 , 96 + chartas , g. , dai , x. , gallagher , s. c. , garmire , g. p. , bautz , m. w. , schechter , p. l. , & morgan , n. d. 2001 , apj , 558 , 119 + chartas , g. , eracleous , m. , agol , e. , & gallagher , s. c. 2004 , apj , 606 , 78 + chartas , g. , eracleous , m. , dai , x. , agol , e. , & gallagher , s. 2007 , apj , 661 , 678 + chartas , g. , falco , e. , forman , w. , jones , c. , schild , r. , & shapiro , i. 1995 , apj , 445 , 140 + dai , x. , kochanek , c.s . , chartas , g. , & mathur , s. , 2006 , apj , 637 , 53 + dai , x. , & kochanek , c.s . , 2008 , apj submitted [ arxiv:0803.1679 ] + dalal , n. , & kochanek , c. s. 2002 , , 572 , 25 + dickey , j. m. , & lockman , f. j. 1990 , , 28 , 215 + fassnacht , c.d . , kocevski , d.d . , auger , m.w . , lubin , l.m . , neureuther , j.l . , jeltema , t.e . , mulchaey , j.s . , & mckean , j.p . , 2007 , apj submitted [ arxiv:0711.2066 ] + garmire , g. p. , bautz , m. w. , ford , p. g. , nousek , j. a. , & ricker , g. r. 2003 , , 4851 , 28 + george , i. m. , & fabian , a. c. 1991 , , 249 , 352 + gould , a. , & gaudi , b. s. 1997 , apj , 486 , 687 + gould , a. , 2000 , apj , 535 , 928 + grieger , b. , kayser , r. , & refsdal , s. 1988 , a&a , 194 , 54 + grieger , b. , kayser , r. , & schramm , t. 1991 , a&a , 252 , 508 + haardt , f. , & maraschi , l. , 1991 , apj , 380 , 51 + hirose , s. , krolik , j. , de villiers , j .- p . , & hawley , j. , 2004 , apj , 606 , 1083 + jones , c. , & forman , w. 1999 , , 511 , 65 + jovanovi , p. , zakharov , a. f. , popovi , l. . , & petrovi , t. 2008 , , 386 , 397 + keeton , c.r . , 2003 apj , 582 17 + kochanek , c. s. , dai , x. , morgan , c. , morgan , n. , poindexter , s. c. g. & chartas , g. , 2007 , statistical challenges in modern astronomy iv , 371 , 43 + kochanek , c.s . , morgan , n.d . , falco , e.e . , et al . , 2006 , apj , 640 , 47 + kochanek , c. s. 2002 , , 578 , 25 + kochanek , c. s. 2004 , , 605 , 58 + lehr , j. , et al . 2000 , , 536 , 584 + lewis , g. f. , irwin , m. j. , hewett , p. c. , & foltz , c. b. 1998 , , 295 , 573 + machida , m. , & matsumoto , r. , 2003 , apj , 585 , 429 + mao , s. , & schneider , p. 1998 , , 295 , 587 + merloni , a. , 2003 , mnras , 341 , 1051 + mineshige , s. , & yonehara , a. 1999 , pasj , 51 , 497 + morgan , n. d. , chartas , g. , malm , m. , bautz , m. w. , burud , i. , hjorth , j. , jones , s. e. , & schechter , p. l. 2001 , apj , 555 , 1 + morgan , n. d. , kochanek , c. s. , falco , e. e. , & dai , x. 2006 , arxiv astrophysics e - prints , arxiv : astro - ph/0605321 + morgan , c. w. , eyler , m. e. , kochanek , c. s. , morgan , n. d. , falco , e. e. , vuissoz , c. , courbin , f. , & meylan , g. , 2008 , apj , 676 , 80 + morgan , c. w. , eyler , m. e. , kochanek , c. s. , morgan , n. d. , falco , e. e. , vuissoz , c. , courbin , f. , & meylan , g. 2008a , , 676 , 80 + morgan , c. w. , kochanek , c. s. , morgan , n. d. , & falco , e. e. 2007 , arxiv e - prints , 707 , arxiv:0707.0305 + morgan , c. w. , kochanek , c. s. , dai , x. , morgan , n. d. , & falco , e. e. 2008b , arxiv e - prints , 802 , arxiv:0802.1210 + mori , k. , tsunemi , h. , miyata , e. , baluta , c. , burrows , d. n. , garmire , g. p. , & chartas , g. 2001 , in asp conf . 251 , new century of x - ray astronomy , ed . inoue & h. kunieda ( san francisco : asp ) , 576 + mortonson , m.j . , schechter , p.l . , & wambsganss , j. , 2005 , apj , 628 , 594 + mushotzky , r. f. , done , c. , & pounds , k. a. 1993 , , 31 , 717 + netzer , h. , & peterson , b. m. 1997 , astronomical time series , 218 , 85 + ofek , e.o . & maoz , d. , 2003 , apj , 594 , 101 + peng , c.y . , impey , c.d . , rix , h .- w . , kochanek , c.s . , keeton , c.r . , falco , e.e . , lhar , j. , & mcleod , b.a . , 2006 , apj 649 , 616 + peterson , b. m. 1993 , pasp , 105 , 247 + poindexter , s. , morgan , n. , kochanek , c.s . , & falco , e.e . , 2007 , apj , 660 , 146 + pointdexter , s. mogan , n. , & kochanek , c.s . , 2008 , apj , 673 , 34 + pooley , d. , blackburne , j. a. , rappaport , s. , schechter , p. l. , & fong , w .- f . 2006 , apj , 648 , 67 + pooley , d. , blackburne , j. a. , rappaport , s. , & schechter , p. l. 2007 , apj , 661 , 19 + popovi , l. . , & chartas , g. 2005 , mnras , 357 , 135 + popovi , l. . , mediavilla , e. g. , jovanovi , p. , & muoz , j. a. 2003 , a&a , 398 , 975 + reynolds , c.s . & nowak , m.a . , 2003 , phr , 377 , 389 + risaliti , g. , elvis , m. , fabbiano , g. , baldi , a. , zezas , a. , & salvati , m. 2007 , apjl , 659 , l111 + schechter , p. l. , & wambsganss , j. 2002 , , 580 , 685 + schneider , p. , ehlers , j. , & falco , e. e. 1992 , gravitational lenses , xiv , springer - verlag berlin heidelberg new york . also astronomy and astrophysics library . + shankar , f. , weinberg , d. h. , & miralda - escude , j. 2007 , arxiv e - prints , 710 , arxiv:0710.4488 + tsunemi , h. , mori , k. , miyata , e. , baluta , c. , burrows , d. n. , garmire , g. p. , & chartas , g. 2001 , , 554 , 496 + williams , k.a . , momchemva , i. , keeton , c.r . , zabludoff , ai . , & lhar , j. , 2006 , apj , 646 , 85 wise , m. w. , davis , j. e. , huenemoerder , houck , j. c. , dewey , d. flanagan , k. a. , and baluta , c. 1997 , _ the marx 3.0 user guide , cxc internal document _ available at http://space.mit.edu/asc/marx/ + wryzkwoski , l. , et al . , 2003 acta astron . 53 229 + yonehara , a. , mineshige , s. , fukue , j. , umemura , m. , & turner , e. l. 1999 , a&a , 343 , 41 + young , a. j. , & reynolds , c. s. 2000 , apj , 529 , 101 + clcccccc + + & & & & & & & + & & _ chandra _ & exposure & & & & + epoch & observation & observation & time & @xmath116@xmath117 & @xmath118@xmath117 & @xmath119@xmath117 & @xmath120 @xmath117 + & date & i d & ( ks ) & counts & counts & counts & counts + & & & & & & & + + 1&2004 april 12 & 4814 & 10 . & 425 @xmath16 22 & 2950 @xmath16 54 & 839 @xmath16 29 & 211 @xmath16 15 + 2&2006 march 10 & 6913 & 4.9 & 393 @xmath16 20 & 624 @xmath16 25 & 204 @xmath16 14 & 103 @xmath16 10 + 3&2006 march 15 & 6912 & 4.4 & 381 @xmath16 20 & 616 @xmath16 25 & 233 @xmath16 15 & 93 @xmath16 10 + 4&2006 april 12 & 6914 & 4.9 & 413 @xmath16 20 & 507 @xmath16 23 & 146 @xmath16 12 & 131 @xmath16 12 + 5&2006 november 10 & 6915 & 4.8 & 3708 @xmath16 61 & 1411 @xmath16 38 & 367 @xmath16 19 & 155 @xmath16 13 + 6&2006 november 13 & 6916 & 4.8 & 3833 @xmath16 62 & 1618 @xmath16 40 & 415 @xmath16 20 & 115 @xmath16 11 + & & & & & & & + + + 1&2000 june 10 & 375 & 47.4 & 1024 @xmath16 32 & 547 @xmath16 23 + 2&2006 february 16 & 6918 & 5.0 & 38 @xmath16 6 & 86 @xmath16 9 + 3&2006 march 15 & 6917 & 4.6 & 40 @xmath16 6 & 59 @xmath16 8 + 4&2006 april 09 & 6919 & 4.9 & 33 @xmath16 6 & 61 @xmath16 8 + 5&2006 october 31 & 6920 & 5.0 & 60 @xmath16 8 & 40 @xmath16 6 + 6&2006 november 08 & 6921 & 4.9 & 52 @xmath16 7 & 57 @xmath16 8 + @xmath121background - subtracted source counts for events with energies in the 0.210 kev band . the counts for images b and c for the 2004 april 12 observation of rx j1131@xmath01231 are corrected for pile - up by factors of 37% and 16% , respectively . the counts for images a and b for the 2006 november 10 observation of rx j1131@xmath01231 are corrected for pile - up by factor of 32% and 13% , respectively . the counts for images a and b for the 2006 november 13 observation of rx j1131@xmath01231 are corrected for pile - up by factors of 33% and 17% , respectively . + ccccc + + & & & & + & & & & + & & & & + & & & & + + hst & 0,0 & 0.588@xmath160.003,1.120@xmath160.003 & 0.618@xmath160.003,2.307@xmath160.003 & @xmath02.517@xmath160.003 , 1.998@xmath160.003 + _ chandra _ & 0,0 & 0.59@xmath160.02 , 1.12@xmath160.02 & 0.62@xmath160.02 , 2.31@xmath160.02 & @xmath02.52@xmath160.02 , 2.00@xmath160.02 + & & & & + + hst & & 0,0 & 2.901@xmath160.003,@xmath01.332@xmath160.003 & + _ chandra _ & & 0,0 & 2.92@xmath160.02 , @xmath01.33@xmath160.02 & + cccccc + + & & & & & + & & & & & + & & image a@xmath122 & image b@xmath122 & image c@xmath122 & image d@xmath122 + & & & & & + 1 & @xmath123 & 1.44@xmath124 & 1.41@xmath125 & 1.47@xmath126 & 1.95@xmath127 + & 0.22 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 1.4@xmath129 & 9.1@xmath130 & 3.1@xmath131 & 0.9@xmath129 + & 210 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 2.7@xmath132 & 15.7@xmath133 & 3.8@xmath134 & 0.7@xmath129 + & @xmath135 & 556/787 & 661/787 & 583/787 & 368/787 + & @xmath136 & 25/31 & 105/97 & 28/41 & 15.4/15 + & @xmath137@xmath138 & 0.76 & 0.27 & 0.95 & 0.42 + & & & & & + 2 & @xmath123 & 1.58@xmath139 & 1.63@xmath140 & 1.62@xmath141 & 1.66@xmath142 + & 0.22 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 2.9@xmath143 & 5.0@xmath130 & 1.3@xmath144 & 0.8@xmath132 + & 210 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 4.5@xmath145 & 7.0@xmath146 & 2.3@xmath147 & 1.1@xmath148 + & @xmath135 & 531/838 & 653/838 & 435/838 & 300/838 + & @xmath136 & 33/28 & 61/52 & 14.4/16 & 2.4/6 + & @xmath137@xmath138 & 0.24 & 0.18 & 0.57 & 0.9 + & & & & & + 3 & @xmath123 & 1.61@xmath139 & 1.73@xmath140 & 1.57@xmath149 & 1.60@xmath150 + & 0.22 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 3.1@xmath151 & 5.6@xmath152 & 1.8@xmath131 & 0.8@xmath39 + & 210 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 4.6 @xmath145 & 6.7@xmath153 & 3.0@xmath154 & 1.2@xmath134 + & @xmath135 & 521/838 & 555/838 & 425/838 & 292/838 + & @xmath136 & 26/27 & 42/48 & 18.5/17 & 3.8/6 + & @xmath137@xmath138 & 0.52 & 0.73 & 0.36 & 0.7 + & & & & & + 4 & @xmath123 & 1.53@xmath155 & 1.64@xmath126 & 1.70@xmath156 & 1.62@xmath157 + & 0.22 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 3.0@xmath143 & 3.9@xmath158 & 1.2@xmath132 & 1.0@xmath132 + & 210 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 5.1@xmath159 & 5.4@xmath160 & 1.5@xmath134 & 1.5@xmath134 + & @xmath135 & 541/838 & 548/838 & 333/838 & 350/838 + & @xmath136 & 27/29 & 40/41 & 7.7/10 & 13/9 + & @xmath137@xmath138 & 0.58 & 0.52 & 0.66 & 0.15 + & & & & & + 5 & @xmath123 & 1.59@xmath125 & 1.82@xmath161 & 1.90@xmath139 & 1.54@xmath162 + & 0.22 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 29@xmath163 & 13@xmath164 & 3.4@xmath131 & 1.1@xmath144 + & 210 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 35@xmath165 & 11@xmath166 & 2.8@xmath147 & 1.9@xmath134 + & @xmath135 & 771/838 & 626/838 & 502/838 & 402/838 + & @xmath136 & 72/59 & 66/63 & 34/28 & 7/11 + & @xmath137@xmath138 & 0.1 & 0.37 & 0.19 & 0.7 + & & & & & + 6 & @xmath123 & 1.51@xmath125 & 1.82@xmath124 & 1.76@xmath155 & 2.28@xmath167 + & 0.22 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 30@xmath163 & 15@xmath163 & 3.2@xmath131 & 1.4@xmath131 + & 210 kev flux ( 10@xmath128 erg s@xmath13 @xmath19 ) & 36@xmath168 & 13@xmath163 & 3.6@xmath154 & 0.6@xmath169 + & @xmath135 & 768/838 & 658/838 & 489/838 & 276/838 + & @xmath136 & 65/59 & 62/70 & 29/33 & 9/8 + & @xmath137@xmath138 & 0.29 & 0.76 & 0.66 & 0.36 + & & & & & + @xmath121 model 1 consists of a power - law modified by galactic absorption . + @xmath170all errors are for 90% confidence unless mentioned otherwise with all parameters taken to be of interest except absolute normalization . + @xmath171@xmath137 is the probability of exceeding @xmath172 for @xmath173 degrees of freedom if the model is correct . + cccc + + & & & + & & & + & & image a@xmath122 & image b@xmath122 + & & & + 1 & @xmath123 & 1.59@xmath124 & 1.86@xmath126 + & 0.22 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 6.9@xmath143 & 4.28@xmath143 + & 210 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 10.6@xmath175 & 3.97@xmath176 + & @xmath135 & 623/790 & + & & & + 2 & @xmath123 & 1.4@xmath132 & 1.6@xmath144 + & 0.22 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 2.6@xmath177 & 6.5@xmath178 + & 210 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 5.8@xmath179 & 10.2@xmath180 + & @xmath135 & 192/838 & 284/1060 + & & & + 3 & @xmath123 & 1.53@xmath181 & 2.4@xmath132 + & 0.22 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 3.4@xmath182&8.4@xmath183 + & 210 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 5.9@xmath184 & 3.0@xmath185 + & @xmath135 & 186/838 & 195/1060 + & & & + 4 & @xmath123 & 1.76@xmath186 & 2.03@xmath187 + & 0.22 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 2.9@xmath188 & 6.5@xmath189 + & 210 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 3.2@xmath190 & 4.3@xmath191 + & @xmath135 & 160/838 & 224/1060 + & & & + 5 & @xmath123 & 1.59@xmath192 & 2.33@xmath193 + & 0.22 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 4.5@xmath194 & 5.0@xmath195 + & 210 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 6.9@xmath196 & 1.8@xmath197 + & @xmath135 & 236/838 & 165/1060 + & & & + 6 & @xmath123 & 1.61@xmath198 & 1.81@xmath199 + & 0.22 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 4.4@xmath200 & 4.6@xmath201 + & 210 kev flux ( 10@xmath174 erg s@xmath13 @xmath19 ) & 6.4@xmath202 & 4.7@xmath203 + & @xmath135 & 230/838 & 222/1060 + & & & + @xmath121 model 1 consists of a power - law modified by galactic absorption . the x - ray fluxes represent unabsorbed values . + @xmath170all errors are for 90% confidence unless mentioned otherwise with all parameters taken to be of interest except absolute normalization . +

we study the pointwise perturbations of countable markov maps with infinitely many inverse branches and establish the following continuity theorem : let and be expanding countable markov maps such that the inverse branches of converge pointwise to the inverse branches of as . then under suitable regularity assumptions on the maps and the following limit exists : where is the topological conjugacy between and and stands for the hausdorff dimension . this is in contrast with the fact that other natural quantities measuring the singularity of fail to be continuous in this manner under pointwise convergence such as the hlder exponent of or the hausdorff dimension for the preimage of the absolutely continuous invariant measure for . as an application we obtain a perturbation theorem in non - uniformly hyperbolic dynamics for conjugacies between intermittent manneville - pomeau maps when varying the parameter . 306494 and msca - if grant ) ]

1601.06591

countable markov maps , that is , interval maps with countably many expanding branches , have received much attention over the past several years . they appear in particular in diophantine approximation in the study of approximation rates of irrationals by rational numbers . the key examples here are the _ gauss map _ @xmath11 , which generates the continued fraction expansion @xcite , and the various _ lroth maps _ , which generate lroth expansions @xcite . moreover , countable markov maps appear naturally in the study of non - uniformly hyperbolic dynamical systems such as the intermittent _ manneville - pomeau maps _ @xcite , where often one considers induced countable markov maps of such systems . various examples are pictured in figure [ fig : countablemarkovmaps ] below . , the lroth map @xmath12 and the alternating lroth map @xmath13 , see @xcite for definitions . ] in this paper , we are interested in the changes to the dynamics of countable markov maps when small pointwise perturbations are applied . a possible way to evaluate the effect of such perturbations on the dynamics of these maps is to investigate the topological conjugacies between the original map and the perturbed map , where we recall that a homeomorphism @xmath14 between two topological dynamical systems is said to be a _ topological conjugacy _ if @xmath15 . in other words , every orbit under @xmath1 corresponds to an orbit under @xmath16 and vice versa . in the case of countable markov maps @xmath1 and @xmath16 the conjugacies will usually be strictly increasing , singular maps , otherwise known as slippery devil s staircases ( a term coined by mandelbrot @xcite ) . singular here means that the derivative is lebesgue - almost everywhere equal to zero : @xmath17 now the _ degree _ of the singularity of the conjugacy @xmath18 gives us a certain sense of how `` close '' the maps @xmath1 and @xmath16 are . natural ways to measure the degree of singularity are for example the hausdorff dimension @xmath19 or the hlder exponent of the conjugacy @xmath18 . perhaps the first well - studied example of a singular function is minkowski s question - mark function @xmath20\to [ 0,1]$ ] , which was constructed by h. minkowski in 1908 ( see @xcite ) . it is illustrated in figure [ fig : minkowskiluroth ] . this function was originally designed precisely to map all rational numbers in @xmath21 $ ] onto the dyadic rationals , and all algebraic numbers of degree two onto the non - dyadic rationals , in an order preserving way . the main idea was to illustrate the lagrange property of the algebraic numbers of degree two ( see theorem 28 in @xcite ) . the function @xmath22 was proved to be singular by denjoy @xcite , and was also studied by salem @xcite . more recently , kessebhmer and stratmann @xcite showed that the minkowski question - mark function can be thought of as the topological conjugacy between the gauss map and the alternating lroth map ( or , equivalently , between the classical farey map from elementary number theory and the tent map ) . moreover , they showed that the derivative can either take the value zero , be infinite , or else it does nt exist . they then applied previous thermodynamical results to compute the hausdorff dimension of the sets where the derivative is infinite and where it does nt exist , and these dimensions turn out to be equal @xcite . and the @xmath23-farey - minkowski function @xmath24 for the classical alternating lroth map @xmath12 , as shown above . ] the hausdorff dimension of the set of non - zero derivative for a variant of the minkowski question - mark function has been studied by li , xiao and dekking in @xcite , and for the case of expanding maps of the interval with finitely many increasing branches by kessebhmer _ _ in @xcite . a similar problem has also been studied in the case of singular functions which are increasing but not strictly increasing , such as for several variants of the cantor ternary function , see @xcite for example . moreover , similar results have been considered for topological conjugacies ( called @xmath9-_farey - minkowski functions _ ) between @xmath9-lroth maps by munday @xcite ( an example is shown in figure [ fig : minkowskiluroth ] ) and later by arroyo @xcite , where he considers the conjugacy maps between the gauss map and any @xmath9-lroth map . there is extensive literature on the perturbations of dynamical systems and their effect on entropy , dimension , and other statistical quantities under both random and deterministic perturbations . in our case we will study the following problem : how do the notions of singularity of the topological conjugacy @xmath18 between countable markov maps @xmath1 and @xmath16 behave when @xmath1 and @xmath16 are sufficiently close ? here by `` closeness '' we mean the relatively weak notion that the inverse branches of @xmath1 and @xmath16 are pointwise close . heuristically here one would expect that the conjugacies @xmath18 would share the properties of the identity mapping as @xmath18 is pointwise close to the identity . we will find out that for the hausdorff dimension of the set of @xmath25 with @xmath26 , we do have some continuity under pointwise perturbations ( see theorem [ mainthm ] below ) , but under other notions of singularity of @xmath18 , such as hlder exponents or hausdorff dimension of the @xmath18 image of the absolutely continuous invariant measure , the continuity fails to occur ( see propositions [ prop : holder ] , [ prop : hausdorff ] , [ prop : lyapunov ] below ) due to the non - compact nature of countable markov maps . to state our main result , let us first fix a little notation ( we refer to section [ sec : prelim ] for a more thorough exposition ) . let @xmath27 \to [ 0,1]$ ] be @xmath28 contractions for each @xmath29 and where either @xmath30 , @xmath31 for all @xmath29 and @xmath32 is a decreasing sequence with @xmath33 or we have that @xmath34 , @xmath35 for all @xmath29 and @xmath36 is a decreasing sequence . these maps are the inverse branches of a piecewise differentiable countable markov map @xmath1 . we assume some regularity on @xmath1 and a standard assumption in this setting is that the geometric potential @xmath37 has _ summable variations _ ( see section [ sec : thermo ] for a definition ) , that is , @xmath38 which is satisfied , for example , for the gauss map , jump transformations of the manneville - pomeau map , and for all @xmath9-lroth maps . we will fix such a system \{@xmath1 , @xmath39 } and consider perturbations of the system , in the following sense . for each @xmath40 we will consider a system with maps @xmath41 and @xmath42 satisfying the variation assumption above and where for each @xmath43 $ ] we have @xmath44 we need that @xmath41 have the same orientation as the maps @xmath45 . this means the dynamical systems @xmath42 and @xmath1 are topologically conjugate and we will denote the conjugacy by @xmath46 , that is the homeomorphism @xmath47\to [ 0,1]$ ] satisfies that @xmath48 . now the pointwise convergence of the inverse branches guarantee that when @xmath2 , we have that the conjugacy @xmath4 will flatten and converge pointwise to the identity mapping , see figure [ fig : limitperturbation ] for example . between two countable markov maps @xmath0 and @xmath1 . the map @xmath1 is the @xmath49-lroth map for the dyadic partition @xmath49 and @xmath0 is the @xmath9-lroth map for a @xmath50-adic partition for @xmath50 attaining the values @xmath51 , @xmath52 , @xmath53 and @xmath54 . the maps @xmath4 approach the identity pointwise when @xmath55 pointwise . ] thus one would expect that @xmath4 should share the properties of the identity in the limit . our main result shows that this happens for the hausdorff dimension of the set @xmath56 under suitable assumptions on the converging family of countable markov maps . [ mainthm ] suppose @xmath1 is a countable markov map with inverse branches @xmath45 such that the potential @xmath57 has summable variations . let @xmath58 be a sequence of countable markov maps with inverse branches @xmath41 . assume the following two assumptions on the tail and variations : * there exists @xmath59 with @xmath60|^t<\infty.\ ] ] * the potentials @xmath61 have summable variations with a uniform bound over @xmath62 : @xmath63 under these assumptions , if for any @xmath64 the inverse branches @xmath55 pointwise as @xmath2 , we have @xmath65 let us make a few remarks on the conditions ( 1 ) and ( 2 ) required in theorem [ mainthm ] . the condition ( 1 ) holds if the countable markov map @xmath1 has at most a polynomially fat tail , in the sense that the lengths @xmath66| = o(i^{-p})$ ] as @xmath67 for some @xmath68 . thus ( 1 ) yields in particular that the absolutely continuous invariant measure for @xmath1 has finite entropy , but it is not an equivalent condition . the condition ( 2 ) on variation in theorem [ mainthm ] is satisfied if the inverse branches of @xmath0 are linear , i.e. , when the maps @xmath0 are @xmath9-lroth maps for certain partitions @xmath9 in the notation of @xcite . thus our result gives rather general conditions to have such a perturbation theorem for @xmath9-lroth maps , provided that the map being perturbed has a thin enough tail . in the non - linear case , the gauss map will satisfy the tail assumption ( 1 ) we impose , so the perturbation theorem is valid provided we have a uniform bound ( 2 ) over the sums of variations on the family of maps converging to the gauss map . furthermore , the conditions in theorem [ mainthm ] are weak enough for us to apply theorem [ mainthm ] to the study of a certain family of intermittent maps in non - uniformly hyperbolic dynamics known as the _ manneville - pomeau maps _ @xmath69 \to [ 0,1]$ ] , @xmath70,\ ] ] for a parameter @xmath71 . the jump transformations ( in other words , `` accelerated dynamics '' or induced maps ) for @xmath72 give us countable markov maps that have polynomial tail and satisfy the assumptions of theorem [ mainthm ] when varying the parameter @xmath9 for the maps @xmath72 , since this means pointwise convergence of the inverse branches . thus we obtain the following corollary to theorem [ mainthm ] : [ cor : mp ] let @xmath73 . then as @xmath74 we have @xmath75 where @xmath76 is the topological conjugacy between the manneville - pomeau maps @xmath77 and @xmath72 . corollary [ cor : mp ] concerns the topological stability for @xmath72 when varying @xmath9 . a related area of study for manneville - pomeau maps is the measure theoretical _ statistical stability _ , where the behaviour of the absolutely continuous invariant measure for @xmath72 is studied when varying @xmath9 , see for example the recent works by freitas and todd @xcite and baladi and todd @xcite . there are also other natural ways to measure the singularity of the conjugacies @xmath4 and the effect of perturbations to them . however , we will see that the continuity as presented in theorem [ mainthm ] fails for these quantities . we will consider three possible examples below . firstly , observe that the topological conjugacies @xmath4 are all hlder continuous . thus one might expect that the hlder exponent @xmath78 of @xmath4 ( see section [ sec : prelim ] for definitions ) would converge to @xmath79 , which is the hlder exponent of the identity . however , this can be made to fail : [ prop : holder ] there exist examples of @xmath0 and @xmath1 satisfying the assumptions of theorem [ mainthm ] such that the hlder exponents @xmath78 of @xmath4 satisfy @xmath80 a similar behaviour can be observed also in the following setting . if @xmath7 is the absolutely continuous @xmath1-invariant measure , then one might also expect that the hausdorff dimensions @xmath6 of the @xmath4-preimages of the measure @xmath7 would converge to @xmath79 . on the other hand , the maps @xmath0 can be chosen such that the the dimensions do not converge to the expected value : [ prop : hausdorff ] there exist examples of @xmath0 and @xmath1 satisfying the assumptions of theorem [ mainthm ] such that the hausdorff dimensions of @xmath81 satisfy @xmath82 moreover , denoting by @xmath83 the absolutely continuous @xmath0-invariant measure , we also consider the entropy ( that is , the lyapunov exponent ) of the absolutely continuous invariant measures for the maps @xmath0 and @xmath1 respectively . if we would have that @xmath84 , instead of pointwise convergence of the inverse branches of @xmath0 , it would be considerably easier to prove the statement of the main result theorem [ mainthm ] . however , @xmath84 is too strong a property to be deduced from pointwise convergence , as the following result shows . [ prop : lyapunov ] there exist examples of @xmath0 and @xmath1 satisfying the assumptions of theorem [ mainthm ] such that the entropy @xmath85 but the limit @xmath86 we remark that in the uniformly hyperbolic compact case , i.e. , in the situation of finitely many branches with uniform expansion rate , all these notions can be shown to be continuous under pointwise perturbations . the heuristic reason for propositions [ prop : holder ] , [ prop : hausdorff ] and [ prop : lyapunov ] is that they represent notions that are very sensitive to the tail behaviour of the countable markov maps @xmath0 . on the other hand , the idea of the proof of theorem [ mainthm ] is that we approximate the infinite systems considered by a finite branch system and in this approximation the precise nature of the tails is not so important , except in terms of the tail of the limiting map @xmath1 ( the tail condition ( 1 ) of theorem [ mainthm ] ) . thus the hausdorff dimension of non - differentiability points will not be as sensitive to the tails as the hlder exponent @xmath78 , hausdorff dimension of @xmath81 or the entropy @xmath87 . the limit obtained in theorem [ mainthm ] does not tell us about the possible rate of the numbers @xmath88 converging to @xmath79 as @xmath89 approaches infinity . if we restrict the class of countable markov maps we consider , then this can be addressed and the hausdorff dimension can be explicitly computed . for this , we will consider a class of countable markov maps similar to those arising from the _ salem family _ considered in @xcite . fix @xmath90 and define the map @xmath91 to be the countable markov map with decreasing linear branches on each interval @xmath92 $ ] , @xmath62 . in the language of @xmath9-lroth maps @xcite , the map @xmath91 is the @xmath9-lroth map for the partition @xmath93 : k \in \n\}$ ] . we obtain the following theorem . [ mainthm2 ] fix @xmath94 and let @xmath95 be the topological conjugacy between @xmath91 and @xmath96 . then @xmath97 where @xmath98 due to the choice of the specific countable markov maps @xmath91 , the proof of theorem [ mainthm2 ] is reduced to the study of conjugacies between tent - like expanding maps with two full branches , one increasing and one decreasing . a similar result was obtained in ( * ? ? ? * theorem 1.1 ) , where the authors consider a family of expanding maps with finitely many increasing full branches . however , as we have one increasing and one decreasing branch , the proof in our situation is rather simpler than in @xcite . the paper is organised as follows . in sections [ sec : prelim ] and [ sec : thermo ] we will give all the necessary background results from dimension theory and thermodynamic formalism . in section [ sec : mainthm ] we will give the proof of theorem [ mainthm ] . in section [ sec : holderentropy ] we present how to achieve propositions [ prop : holder ] , [ prop : hausdorff ] and [ prop : lyapunov ] . in section [ sec : manpom ] we discuss the manneville - pomeau example further and prove corollary [ cor : mp ] , and finally , in section [ sec : geometric ] we prove theorem [ mainthm2 ] . a _ countable markov map _ @xmath100 \to [ 0,1]$ ] is defined with the help of its inverse branches . we consider the situation where for each @xmath29 , there exist maps @xmath101\to [ 0,1]$ ] which are continuous and strictly decreasing on @xmath21 $ ] and differentiable on @xmath102 . we further assume that there exists @xmath103 and @xmath104 such that for all @xmath105 we have that @xmath106 for all @xmath107 . we will also suppose that @xmath108 , @xmath109 for all @xmath29 and @xmath33 or alternatively that @xmath110 , @xmath111 for all @xmath29 and @xmath33 . thus @xmath112)=(0,1]$ ] and if @xmath113 then @xmath114 . we define an expanding map @xmath115\to [ 0,1]$ ] by setting @xmath116 given a countable markov map @xmath1 with inverse branches @xmath45 , @xmath29 , it is convenient to model our systems using symbolic dynamics . let @xmath117 and let @xmath118 be the usual left - shift transformation . we can relate this to our systems @xmath119 via projections @xmath120 $ ] . we define @xmath121 the factor map @xmath122 allow us to import the thermodynamical formalism from the shift space to measures invariant under @xmath1 . for a shift invariant measure @xmath7 , the push - forward measure @xmath123 will be @xmath1-invariant . moreover if @xmath7 is ergodic for the shift map then @xmath124 will be ergodic for @xmath1 . thus we can use the symbolic model @xmath125 and the geometric model @xmath126,t)$ ] interchangably . now if we have a sequence of countable markov maps @xmath0 with inverse branches @xmath127 satisfying the assumptions of theorem [ mainthm ] , we will shorten the notation by letting @xmath128 and @xmath129 . then the topological conjugacy @xmath4 between @xmath0 and @xmath1 will satisfy @xmath130.\ ] ] in other words , the conjugacy map between the systems @xmath1 and @xmath0 takes the point @xmath25 with coding given by @xmath1 and sends it to the point with the same coding , but now understood in terms of @xmath0 . let @xmath131 be the hausdorff dimension of a set @xmath132 and the @xmath133-dimensional hausdorff measures @xmath134 and the @xmath135-hausdorff content @xmath136 , see @xcite for a definition . for a radon measure @xmath137 on @xmath138 , the hausdorff dimension of @xmath137 is defined to be @xmath139 where @xmath140 is the _ lower _ local dimension of @xmath137 at @xmath25 , which is defined by @xmath141 if @xmath142 \to [ 0,1]$ ] is a function , then the _ hlder exponent _ @xmath143 of @xmath18 is defined to be the infimal @xmath144 such that for some @xmath145 the following inequality holds : @xmath146.\ ] ] now we will consider a fixed measure @xmath7 on @xmath21 $ ] and countable markov map @xmath1 and we will define the notions of lyapunov exponents and entropy for this measure . note that the lyapunov exponent depends upon the mapping @xmath1 as well as the measure @xmath7 . the _ lyapunov exponent _ of the measure @xmath7 is defined to be @xmath147 similarly , if @xmath148 $ ] , for @xmath149 , are the construction intervals generated by the countable markov map @xmath1 , the entropy of @xmath7 is defined as follows : the _ kolmogorov - sinai entropy _ ( with respect to @xmath1 ) of the measure @xmath7 is defined to be @xmath150 note that sometimes we also write @xmath151 or @xmath152 for a measure @xmath7 living on @xmath153 and then we just mean the values @xmath154 and @xmath155 respectively for the projected measure @xmath124 . if we just take the entropy of such @xmath7 with respect to the shift map @xmath156 on @xmath153 , we define @xmath157 like @xmath151 but we replace the intervals @xmath158 by the cylinders @xmath159 $ ] . now , given a countable markov map @xmath1 , the hausdorff dimensions of each of the @xmath122-projections of an ergodic shift - invariant measure can be computed using the following result : [ prop : measdim ] if @xmath7 is an ergodic @xmath1 invariant probability measure on @xmath21 $ ] and @xmath85 , then the hausdorff dimension of @xmath7 is given by @xmath160 the above result can be found as theorem 4.4.2 in the book @xcite by mauldin and urbaski . in this section we present the tools we will need from thermodynamical formalism . we mostly concentrate on the countable markov shift @xmath153 as this is where we will reformulate the problem , using the theory developed in a much more general setting in d. mauldin and m. urbaski @xcite and the series of works by o. sarig , see for example @xcite . first , recall that a potential @xmath161 is said to be _ locally hlder _ if there exist constants @xmath162 and @xmath163 such that for all @xmath164 the _ variations _ @xmath165 decay exponentially : @xmath166\ } \leq c\delta^n.\ ] ] note that since nothing is assumed in the case that @xmath167 , this does _ not _ imply that @xmath161 is bounded . the _ birkhoff sum _ @xmath168 of a potential @xmath169 is the potential defined by @xmath170 the _ pressure _ of a locally hlder potential @xmath161 is then the limit @xmath171 where @xmath172 is the periodic word repeating the word @xmath173 . define @xmath174 to be the collection of all @xmath156-invariant measures on @xmath153 . a deep and useful result which we will now state is the _ variational principle _ , which gives a representation of @xmath175 using the kolmogorov - sinai entropy : [ lma : varprinciple ] for any locally hlder potential @xmath161 we have that @xmath176 for a proof , see theorem 2.1.8 in @xcite . if there exists a measure @xmath177 which attains the supremum in lemma [ lma : varprinciple ] , then we call @xmath7 an _ equilibrium state _ for a potential @xmath161 . in the case of finite pressure more can be said about equilibrium states . let @xmath178 be a locally hlder potential . if @xmath175 is finite , then we call @xmath179 a _ gibbs measure _ for @xmath161 if there exists a constant @xmath145 such that @xmath180 \leq c\exp(s_n { \varphi}({\mathbf{j } } ) - np({\varphi}))\ ] ] for any @xmath173 , @xmath181 $ ] and @xmath182 . an example of such a measure is the _ bernoulli measure _ @xmath7 associated to weights @xmath183 $ ] , @xmath64 , with @xmath184 , which is the equilibrium state for the potential @xmath185 . then @xmath186 and @xmath187 = p_{i_1}\dots p_{i_n } = \exp(s_n { \varphi}({\mathbf{j } } ) ) , \quad \text{for } { \mathbf{j}}\in [ { \mathbf{i}}].\ ] ] the following proposition relates gibbs measures to equilibrium states . [ prop : uniqequilibrium ] let @xmath178 be a locally hlder potential . if @xmath188 then there exists a unique invariant probability measure , @xmath179 which is a gibbs measure for @xmath161 . moreover , if @xmath161 is integrable with respect to @xmath179 then @xmath179 is the unique equilibrium state for @xmath161 . for a proof of this result , see proposition 2.1.9 , theorem 2.2.9 and corollary 2.7.5 in @xcite . the case when @xmath161 is not integrable with respect to @xmath189 is the subject of the next lemma . [ lem : noequilibrium ] let @xmath178 be a locally hlder potential with @xmath190 . if @xmath161 is not @xmath179 integrable , then there exist no equilibrium states for @xmath161 . it is a result of sarig ( * ? ? ? * theorem 7 ) that the only possible equilibrium state is a fixed point for the ruelle operator ( see @xcite for a definition ) . it is then shown in the proof of ( * ? ? ? * theorem 1 ) that in the situation where the system satisfies the big image property ( see sarig s paper for the definition ; note that it includes the full shift ) such measures are gibbs measures . thus there can not exist equilibrium states for @xmath161 . all the above thermodynamic definitions can be formulated also for the finite alphabet @xmath191 , @xmath192 and it makes things considerably simpler . for instance , in the finite alphabet case it is known that unique equilibrium states always exist for hlder potentials and they are gibbs measures . this makes it convenient to restrict to the finite case and consider approximations for the pressure . given a locally hlder potential @xmath169 , we write @xmath193 to denote the pressure of @xmath161 restricted to the finite shift @xmath194 . then we have the following approximation result , which can be found as theorem 2.1.5 in @xcite . [ finiteapprox ] for any locally hlder potential @xmath161 , @xmath195 this theorem will allow us to use results which hold on the full shift with a finite alphabet ( or , more generally , on topologically mixing subshifts of finite type ) . these results can sometimes be extended to the infinite case , but due to the hypotheses needed it is more convenient to use theorem [ finiteapprox ] and the results in the finite alphabet case . the first of these results that we will need is the following lemma on the derivative of pressure , which is proposition 4.10 in @xcite . [ lem : thermoderivative ] let @xmath196 be hlder continuous functions and define the analytic function @xmath197 let @xmath198 be the gibbs measure on @xmath199 for the potential @xmath200 . then the derivative of @xmath201 is given by @xmath202 gibbs measures satisfy many statistical theorems similar to ones in probability theory . we will use one of these , namely , the _ law of the iterated logarithm_. before stating this theorem , we recall that a function @xmath203 is said to be _ cohomologous to a constant _ if there exists a constant @xmath204 and a continuous function @xmath205 such that @xmath206 moreover , @xmath207 is called a _ coboundary _ if the constant @xmath208 is equal to @xmath209 . [ lem : lawofiterated ] let @xmath196 be hlder potentials where @xmath207 is not cohomologous to a constant . then there exists @xmath210 such that for @xmath189-almost every @xmath25 , we have @xmath211 this is corollary 2 in @xcite . note that @xmath212 and it is shown in proposition 4.12 of @xcite that @xmath213 , with equality if and only if @xmath207 is cohomologous to a constant . the number @xmath214 is the variance of @xmath207 with respect to @xmath179 and is also the second derivative of the pressure function @xmath215 at @xmath216 . finally in this section we need the following result in the countable case regarding the behaviour of equilibrium states . [ cusp ] let @xmath217 $ ] be locally hlder such that @xmath218 , and let @xmath219 we have that 1 . there exists a sequence @xmath220 of compactly supported @xmath156-invariant ergodic measures such that @xmath221 2 . for any @xmath222 there exists @xmath223 such that if @xmath7 is ergodic , @xmath161 is integrable with respect to @xmath7 and @xmath224 , then @xmath225 let @xmath226 . we can always find @xmath227 such that @xmath228 . therefore we can find @xmath229 such that @xmath230 let @xmath231 be defined by @xmath232 , and observe that @xmath233 . also , by the mean value theorem and the convexity of pressure , @xmath234 . by lemma [ lem : thermoderivative ] the equilibrium state @xmath7 on @xmath199 for @xmath235 will satisfy that @xmath236 and @xmath237 . to complete the proof of the first part for each @xmath164 simply take @xmath238 to find the sequence of measures @xmath220 . now let @xmath239 . thus @xmath240 and so , by the variational principle , for any ergodic measure @xmath7 for which @xmath161 is integrable we have @xmath241 and since , by assumption , @xmath218 we have that @xmath242 . thus if @xmath243 then @xmath244 thus @xmath245 in other words , taking the contrapositive , we have that if @xmath246 then @xmath247 , and the proof is complete . in this section we will present the proof of theorem [ mainthm ] . to this end , fix the countable markov maps @xmath0 and @xmath1 and define the potentials @xmath248 for @xmath249 . recall that by the assumption theorem [ mainthm](2 ) these potentials have uniformly bounded sums of variations . our first step is to slightly simplify the problem by ` iterating ' these potentials to a suitable generation @xmath250 such that the distortion of @xmath251 and @xmath161 from analogous potentials coming from systems with linear branches is small . this is possible due to the bounded variations . for this purpose , let us fix a generation @xmath250 and denote by @xmath252 for @xmath253 the inverse branch corresponding to @xmath254 of the @xmath255-fold composition map @xmath256 . we define the branches @xmath257 similarly for the map @xmath258 . now these maps determine intervals @xmath259 ) \quad \text{and } \quad i_{\mathbf{i}}:= f_{\mathbf{i}}([0,1]).\ ] ] we denote the lengths of these intervals by @xmath260 and @xmath261 respectively . to bound the hausdorff dimension of the set @xmath262 of non - zero derivative for some @xmath62 , we must find a _ compactly supported _ ergodic measure @xmath7 on the shift space @xmath99 for which the @xmath263 projection of typical points will not have a derivative . moreover , we will aim to choose the measure @xmath7 such that its hausdorff dimension is close to @xmath79 when @xmath89 is large . this will be done in the following steps : * in lemma [ lma : step1 ] we will first iterate the potentials @xmath251 and @xmath161 to the @xmath255-th generation ( for some large @xmath250 ) by studying the potentials @xmath264 and @xmath265 and then use the absolutely continuous and invariant measure for @xmath1 to construct a @xmath266 bernoulli measure @xmath267 on @xmath99 which satisfies both that @xmath268 and that the @xmath263 projection of @xmath267 has dimension close to @xmath79 . the construction is possible due to the pointwise convergence of the inverse branches and the tail / variation assumptions in theorem [ mainthm ] . * the measure @xmath267 induces canonically a @xmath156-invariant measure @xmath269 of the same dimension as @xmath267 for which @xmath270 . the measure @xmath271 allows us to apply thermodynamic formalism ( lemmas [ lem : pressure ] and [ lem : pressurefinite ] ) and invoke finite approximation properties ( lemma [ step2 ] ) to find a _ compactly supported _ gibbs measure @xmath7 where @xmath272 but @xmath273 is not a coboundary , and @xmath7 still has dimension close to @xmath79 . * we will then essentially apply the law of iterated logarithms ( lemma [ lil ] ) and the coboundary condition to show that for typical points under the projection of the measure @xmath7 the derivative of @xmath4 does not exist and the dimension of the projection of this measure will be a lower bound for the dimension of the set of points with non - zero derivative . we then show that this dimension tends to @xmath79 as @xmath89 tends to infinity , which completes the proof . let us begin by constructing the bernoulli measure @xmath267 . [ lma : step1 ] for each @xmath274 there exists @xmath275 such that for any @xmath276 there exists @xmath277 such that for any @xmath278 there exists a @xmath266 ergodic measure @xmath267 on @xmath153 which satisfies @xmath279 for the proof of lemma [ lma : step1 ] , we will need the following two preliminary lemmas . we will let @xmath189 be the equilibrium state for @xmath178 ( and also recall that @xmath280 ) . since @xmath218 we have that @xmath281 . let us define the following quantities related to the entropy and lyapunov exponents . for @xmath250 , @xmath149 and a potential @xmath282 , let us write @xmath283\}\ ] ] and @xmath284\}.\ ] ] for the potential @xmath285 , define the numbers @xmath286 [ lma : approximation ] under the assumptions of theorem [ mainthm ] , we have the following approximations * the entropy of the measure @xmath189 is given by @xmath287 * there exists @xmath288 such that for any @xmath250 and @xmath253 we have @xmath289 \(1 ) by the definition of @xmath290 we have that @xmath291 the result then follows since @xmath292 \(2 ) fix @xmath250 and @xmath253 . let us first verify that @xmath293 for any @xmath294 $ ] . we will proceed by induction . for @xmath295 , this is the pointwise convergence assumption for the inverse branches of @xmath0 and @xmath1 . now suppose the claim holds for @xmath296 with @xmath297 . fix @xmath253 . by the mean value theorem , there exists a point @xmath298 $ ] on the interval where the derivative @xmath299 . since , according to assumption ( 2 ) for theorem [ mainthm ] , we have @xmath300 , this yields that @xmath301 for all @xmath253 and @xmath62 . the mean value theorem gives @xmath302 which decays to @xmath303 as @xmath2 by the induction assumption for @xmath296 . this completes the proof as latexmath:[\ ] ] putting these two estimates together , we obtain @xmath389 thus the proof is complete . now let us proceed with the proof of theorem [ mainthm ] . let @xmath390 and fix @xmath276 ( recall the choice of @xmath314 from lemma [ lma : step1 ] ) and write @xmath391 and define the auxiliary @xmath156 invariant measure @xmath392 where @xmath267 is the @xmath266 bernoulli measure determined in lemma [ lma : step1 ] . the measure @xmath271 satisfies the following properties : @xmath393 and the dimension @xmath394 lemma [ lma : step1 ] will allow us to deduce the following lower bound on the pressure function @xmath395 with a suitable choice of @xmath396 . [ lem : pressure ] if @xmath397 , then @xmath398 by lemma [ lma : step1 ] , we have @xmath399 thus we have that for all @xmath400 the following property @xmath401{\,{\ , d}}\eta+h(\eta,\sigma)>t\int{\varphi}_k{\,{\ , d}}\eta+h(\eta,\sigma)>0.\ ] ] on the other hand , if @xmath402 we first suppose that the potential @xmath251 has an equilibrium state @xmath403 . in this case as @xmath404 and @xmath405 we have @xmath406 thus by the variational principle , @xmath407 if @xmath251 does not have an equilibrium state then we must have that @xmath408 and by assumption @xmath409 therefore , if we let @xmath410 and apply the first part of lemma [ cusp ] to @xmath251 and the second part to @xmath161 , we can find a compactly supported @xmath156 invariant ergodic measure @xmath7 such that @xmath411 and @xmath412 therefore @xmath413 and so for all @xmath400 @xmath414 we can now use the approximation property of pressure to allow us to find suitable measures which are compactly supported . recall that the finite approximation property was given in lemma [ finiteapprox ] , and it states that @xmath415 , where @xmath193 is the pressure of @xmath161 restricted to the finite shift @xmath416 . [ lem : pressurefinite ] if @xmath397 , then then there exists @xmath417 with @xmath418 and @xmath419 first of all by taking @xmath403 as in the proof of previous lemma [ lem : pressure ] we have @xmath420 let us use these measures @xmath271 and @xmath403 to construct measures @xmath421 and @xmath422 satisfying similar properties but supported on a compact set @xmath199 for a large enough @xmath423 as follows . by birkhoff s ergodic theorem there exist words @xmath424 and indices @xmath425 such that @xmath426 thus if we let @xmath421 and @xmath422 be the measures supported on these @xmath427 and @xmath428 periodic orbits of @xmath254 and @xmath429 respectively , then there exists an index @xmath430 such that both @xmath431 are invariant measures on @xmath199 for all @xmath432 and we will have that @xmath433 thus if @xmath432 and we put @xmath434 then by the variational principle there exists @xmath162 such that @xmath435 for all @xmath436 $ ] and @xmath419 on the other hand , by the finite approximation property ( lemma [ finiteapprox ] ) and lemma [ lem : pressure ] we have that @xmath437 for all @xmath438 $ ] . now if for each @xmath164 we define the set @xmath439:p_n(q({\varphi}_k-{\varphi})-t{\varphi}_k)\leq 0\},\ ] ] then @xmath440 for for all @xmath164 . however , if we can find @xmath441 , then @xmath442 , which is a contradiction . thus @xmath443 and since each @xmath444 is compact there must exists @xmath432 such that @xmath445 . for this value of @xmath229 we must have that @xmath418 as claimed . now for the @xmath417 constructed in lemma [ lem : pressurefinite ] , we can formulate a key lemma : [ step2 ] if @xmath446 , then there exists @xmath229 such that 1 . @xmath447 is not a coboundary on @xmath199 . 2 . there exists a gibbs measure @xmath7 on @xmath199 such that @xmath448 by lemma [ lem : pressurefinite ] we know that there exists @xmath229 such that @xmath418 and @xmath419 the restrictions of @xmath251 and @xmath161 to @xmath199 are hlder continuous and so the function @xmath449 defined by @xmath450 is analytic with @xmath451 by lemma [ lem : thermoderivative ] , where @xmath452 is the gibbs measure on @xmath199 for @xmath453 . since @xmath454 we know by the definition of pressure that @xmath447 can not be a coboundary on @xmath199 therefore , as @xmath455 , there must exist @xmath456 such that @xmath457 . thus the gibbs measure @xmath458 on @xmath199 satisfies @xmath459 and by the variational principle ( since @xmath460 ) we have @xmath461 therefore , we have by the negativity of @xmath251 that @xmath462 as claimed . the key to the proof of the main theorem will be to combine the above result with the following simple application of the law of the iterated logarithm for function differences @xmath463 , which are not coboundaries . [ lil ] let @xmath464 be hlder continuous potentials such that @xmath463 is not a coboundary and let @xmath7 be a gibbs measure on @xmath199 where @xmath465 . we then have that @xmath466 for @xmath7 almost all @xmath467 . since @xmath463 is not cohomologous to a constant we can apply the law of the iterated logarithm , lemma [ lem : lawofiterated ] , to the functions @xmath463 and @xmath468 to conclude that for some positive constants @xmath469 the following asymptotic bounds hold : @xmath470 at @xmath7 almost every @xmath467 . in particular at these @xmath25 also @xmath471 let us now complete the proof of the main theorem . for any @xmath274 and @xmath276 by lemma [ lma : step1 ] , we can find @xmath472 such that for all @xmath40 with @xmath473 there exists a @xmath266-invariant ergodic measure @xmath267 on @xmath153 such that @xmath474 thus by lemma [ step2 ] applied to @xmath475 and for the @xmath417 given by that result , @xmath447 is not a coboundary on @xmath199 and we can find a gibbs measure @xmath7 supported on a compact set of @xmath153 ( i.e. @xmath199 embedded into @xmath153 ) such that @xmath476 therefore , by lemma [ lil ] , we may also assume that at @xmath7 almost all @xmath477 we have @xmath478 fix one such @xmath479 . recall that the projections @xmath480 $ ] map cylinder sets from @xmath153 onto @xmath0 and @xmath1 construction intervals respectively and the conjugacy @xmath4 between @xmath0 and @xmath1 satisfies @xmath481 now for each @xmath182 , let us define a word @xmath482 by @xmath483 then @xmath484 and so @xmath485 , where we emphasise the interval map @xmath0 or @xmath1 used . therefore , for all @xmath182 the distances @xmath486 and @xmath487 moreover , we have the lower bound @xmath488 so in all cases there is @xmath489 independent of @xmath374 satisfying @xmath490 similarly , for a suitable @xmath491 independent of @xmath374 the images satisfy @xmath492 thus as the numbers @xmath493 and @xmath208 are independent of @xmath374 we obtain by our choice of @xmath25 that @xmath494 and @xmath495 thus the derivative of @xmath4 at @xmath496 can not exist . since @xmath25 was @xmath7 typical , this means that @xmath497 gives full mass to the set of @xmath498 where @xmath499 does not exist . therefore , for all @xmath500 we have @xmath501:\theta_k'(y)\text { does not exist}\ } \geq \dim\pi_{k } \mu > \frac{1 - 3\delta}{1 + 3\delta}.\ ] ] the proof of theorem [ mainthm ] is therefore complete , since @xmath502 was chosen arbitrarily . let us now prove corollary [ cor : mp ] to theorem [ mainthm ] . fix @xmath503 with @xmath504 and let @xmath505 and @xmath506 be the jump transformations of @xmath72 and @xmath77 . that is , if @xmath507 is the first hitting time to the interval between @xmath508 $ ] , where @xmath509 is the solution to the equation @xmath510 on @xmath102 , then @xmath511 and similarly for @xmath506 . now the topological conjugacy @xmath512 between @xmath72 and @xmath77 agrees with the topological conjugacy between @xmath505 and @xmath506 . therefore , in order to prove corollary [ cor : mp ] , we need to establish the assumptions on theorem [ mainthm ] when @xmath74 . \(a ) pointwise convergence of the inverse branches of the induced maps can be established since when @xmath74 , we have that @xmath513 and the hitting times @xmath514 for a fixed @xmath515 $ ] . \(b ) now for the tail behaviour , that is , condition ( 1 ) in theorem [ mainthm ] , we will cite sarig @xcite and in particular the proof of proposition 1 there , where it is proved that if @xmath45 are the inverse branches of @xmath505 , then for any @xmath71 there exists @xmath516 with @xmath517|^{t(\alpha ) } < \infty.\ ] ] \(c ) finally , the variations will be uniformly bounded . fix any @xmath518 such that @xmath519 . for @xmath520 , write @xmath521 where we recall that @xmath522 maps cylinders @xmath159 $ ] onto intervals @xmath523 . then to check the uniform bound ( 2 ) in theorem [ mainthm ] on variations , we will need to establish @xmath524 where @xmath525 \subset ( 0,\infty)$ ] as this yields the assumption ( 2 ) in theorem [ mainthm ] for all sequences @xmath526 , where @xmath527 as @xmath2 . to do this , we just need to check that the mapping @xmath528 is bounded by a continuous function since the supremum is over a compact interval @xmath529 . this follows from nakaishi s work ( * ? ? ? * lemmas 2.1 and 2.2 ) where the following estimate can be established : @xmath530 for @xmath173 and @xmath311 $ ] and so @xmath531 . here the constants @xmath532 and @xmath533 depend continously on the parameter @xmath534 . hence @xmath535 , where @xmath536 is the riemann zeta function . thus the sum is bounded by a continuous function of @xmath534 , which is what we wanted . in this section we will prove propositions [ prop : holder ] , [ prop : hausdorff ] and [ prop : lyapunov ] by giving examples of countable markov maps @xmath0 and @xmath1 satisfying the conditions of theorem [ mainthm ] but with , respectively , the hlder exponents , hausdorff dimensions of the push - forward of the invariant measure for @xmath1 and lyapunov exponents failing to converge . all of the examples we give below come from the class of @xmath9-lroth maps , which were introduced in @xcite , so let us briefly recall the definition . we start with a sequence of real numbers @xmath537 with the property that @xmath538 and let @xmath539:n\in\n\}$ ] . we also denote the length of @xmath540 by @xmath541 . then the map @xmath9-lroth map @xmath542 is defined to be the countable markov map with inverse branches that map the unit interval affinely onto each partition element @xmath540 . two particular examples we will use below come from the partitions @xmath23 , defined by @xmath543 , and @xmath49 , which is given by @xmath544 . we start with the map @xmath545 as described above . then we modify the partition @xmath49 to obtain a sequence of @xmath9-lroth maps that converge pointwise to @xmath1 , in the following way . let @xmath546 be the partition where @xmath547 for all @xmath548 , and we modify the point @xmath549 in order to obtain the lengths @xmath550 and @xmath551 . then the conjugacy map @xmath4 between @xmath0 and @xmath1 is exactly the map studied in @xcite , where in particular it was shown in ( * ? ? ? * lemma 2.3 ) that the hlder exponent of @xmath4 is given by @xmath552 therefore , for our example , we see that the hlder exponent of @xmath4 is given by @xmath553 . this proves proposition [ prop : holder ] . in this case we choose @xmath1 to be the @xmath23-lroth map , so @xmath554 for all @xmath555 . therefore we have that the lyapunov exponent and the entropy @xmath556 now for each @xmath40 we make a modification to the partition @xmath23 to obtain a sequence of partitions @xmath546 as follows . fix the first @xmath89 elements of the partition , and then for @xmath557 let the partition elements have size @xmath558 letting @xmath559 , and the conjugacy between @xmath0 and @xmath1 again be denoted by @xmath4 , the conditions of theorem [ mainthm ] are readily seen to hold as @xmath560 is a piecewise constant function and the tail @xmath561 decays exponentially . however , for each @xmath89 we have that @xmath562 but the maps @xmath0 are constructed such that @xmath563 an application of proposition [ prop : measdim ] now finishes the proof of proposition [ prop : hausdorff ] . an example of maps where the lyapunov exponents to fail to converge is made by adapting the tails of the partition @xmath49 again , similarly to the trick for hlder exponents in section [ sec : holder ] . so , let @xmath545 , and recall that this means @xmath564 for all @xmath565 . thus for the entropy we have @xmath566 now let us define a partition @xmath546 by fixing the first @xmath567 elements to be equal to the first @xmath567 elements from the dyadic partition , letting @xmath568,\ ] ] and , for @xmath557 , letting @xmath569.\ ] ] thus , for @xmath557 , we have that @xmath570 and these decay for any fixed fixed @xmath62 with the rate @xmath571 as @xmath67 , which is far too slow to have finite entropy for @xmath83 . this yields that the entropy @xmath572 for all @xmath62 . in this section , we first aim to prove theorem [ mainthm2 ] . before we begin , we must introduce some preliminaries and notation ( for more details , we refer to @xcite and the references therein ) . to begin , suppose that we have two maps @xmath573\to [ 0,1]$ ] which have exactly two full branches , the first ( thinking left to right from the origin ) increasing and the second decreasing , and both branches are strictly contracting @xmath574 diffeomorphisms . we are interested , of course , in the topological conjugacy map @xmath18 between @xmath16 and @xmath1 , and the set @xmath575 . below , to mirror @xcite more closely , and to make it clearer where changing from increasing branches to the tent - like case we have here makes differences to the proof , we also use the sets @xmath576 and @xmath577 , which are defined to be the set of points where the derivative of said conjugacy map does not exist or is infinite , respectively . note that these are the only options for the derivative to be non - zero . we define two hlder continuous potentials @xmath578 by setting @xmath579 also , to simplify the notation later , we define @xmath580 . then , where we recall that @xmath581 denotes the topological pressure , we can define a function @xmath582 implicitly through the pressure equation @xmath583 we let @xmath584 denote the equilibrium measure for the the potential @xmath585 , which always exists and is unique . recall from the preliminaries given in section 2 that this means @xmath584 achieves the supremum in the variational principle : @xmath586 further , @xmath587 if we suppose that @xmath161 and @xmath207 are cohomologically independent , there also exists a unique @xmath588 such that @xmath589 . . we will now give a sequence of lemmas which give the necessary geometric information about the derivative , and how the differential quotient can be transferred to a sort of `` symbolic derivative '' . we also need some notation : we write @xmath593:=\{y=(y_1 , y_2 , \ldots):y_i = x_i\text { for all } 1\leq i \leq n\}$ ] for the symbolic cylinder sets and write @xmath594)$ ] for the projection of the cylinder set @xmath593 $ ] to a subinterval of @xmath595 $ ] . we also recall the definition of the variations of @xmath161 , @xmath596}|{\varphi}(x ) - { \varphi}(y)|,\ ] ] and note that here , since we are in a compact metric space , @xmath597 is finite . since the potentials @xmath161 and @xmath207 are hlder continuous , the variations of both are exponentially decaying and thus summable . let @xmath602 . observe that for our maps with one increasing and one decreasing full branch , the projection of the cylinder sets from @xmath603 works as follows : if @xmath604 is such that @xmath605 is odd , then @xmath606 splits into @xmath607 , written in order , left to right , whereas if the sum of the digits @xmath608 is even , the @xmath609-level cylinders project the other way around , namely , to @xmath610 . this implies that in order for @xmath598 , we must have that @xmath611 . that is , the words @xmath612 and @xmath613 can only be different at the @xmath614-th letter . thus we have that @xmath615 and @xmath616 ( here , the @xmath617 term occurs precisely for the one difference at the @xmath614-th letter . ) thus we can take @xmath618 let us denote the inverse branches of @xmath16 by @xmath624 and @xmath625 . then if we apply the mean value theorem to the map @xmath626 , it follows that there exists a point @xmath627 such that @xmath628).\ ] ] then , if we set @xmath629 , we have that @xmath630 $ ] and @xmath631)\asymp e^{s_n{\varphi}(y)}.\ ] ] the result then follows since @xmath632 the second part of the result follows by exactly the same method . first note that as the points @xmath25 and @xmath498 are different in the unit interval ( not just having different symbolic codes ) , this @xmath374 always exists in other words , at level @xmath636 the points @xmath25 and @xmath498 are in neighbouring subintervals of @xmath21 $ ] , and at level @xmath374 there is at least one interval between them . by the previous two results it then follows that @xmath637 and @xmath638 the result immediately follows . let @xmath25 be such that @xmath641 and @xmath642 . for each @xmath164 we can find @xmath643 such that @xmath644\cap [ y(n)_1,\ldots , y(n)_k]={\varnothing}.$ ] by lemma [ derivative ] we have that @xmath645 part ( a ) of the lemma follows immediately . to prove ( b ) , suppose that @xmath646 . let @xmath643 be a sequence such that @xmath647 where each @xmath648 . let @xmath649\cap [ y(n)_1,\ldots , y(n)_j]={\varnothing}\}.$ ] we have that @xmath650 and @xmath651 thus since @xmath652 we have that @xmath653 and the proof is finished . now , recalling the definition of @xmath588 given above it can be shown by adapting the methods in @xcite ( the start of section 2.2 ) that @xmath655 and by the arguments in section 2.3 of @xcite that @xmath656 and so @xmath657 the second part is essentially identical to the work in @xcite so we only show here how to adapt the proof of @xmath658 if @xmath659 , setting @xmath660 and letting @xmath661 be the periodic element in the cylinder set @xmath662 $ ] , we then have for infinitely many @xmath164 that @xmath663 it follows that @xmath664 for infinitely many @xmath164 . so for all @xmath229 the union for @xmath665 of @xmath374th level cylinders @xmath662 $ ] where @xmath666 forms a cover of the set @xmath667 . we will denote this cover by @xmath668 . to finish the proof of ( [ starp8 ] ) in light of remark [ remp10 ] it suffices to show that @xmath677 however , this again can be done precisely as in @xcite , so we omit the details . thus the proof of proposition [ prop : salemthermo ] is complete . let us now show how to prove theorem [ mainthm2 ] using proposition [ prop : salemthermo ] . first , fix @xmath678 and denote by @xmath679 the conjugacy map between @xmath91 and @xmath96 . in order to apply proposition [ prop : salemthermo ] , we observe that the conjugacy @xmath18 coincides with the conjugacy map between the farey maps @xmath680 and @xmath681 , where @xmath680 is defined for @xmath43 $ ] by @xmath682 for more details on these maps , we refer to @xcite . that the conjugacies coincide is a direct consequence of the fact that @xmath91 is the jump transformation of @xmath683 . we apply proposition [ prop : salemthermo ] with @xmath684 and @xmath685 the tent map . in this case the potentials @xmath686 and @xmath687 are given by @xmath688 and similarly for @xmath689 . also , since these maps are linear , we have that the measure @xmath690 is given by the @xmath691-bernoulli measure @xmath692 such that @xmath693 let us find the precise value for this @xmath694 as follows . the fraction is @xmath695 and so @xmath696 furthermore , as @xmath697 is an equilibrium measure , we have that @xmath698 as claimed . the proof of proposition [ prop : salemthermo ] is similar to how in @xcite the _ salem family _ @xmath699 is analysed , where the salem maps are interval maps with two increasing branches with slopes @xmath700 and @xmath701 . 31 a. arroyo . preprint , available on arxiv : http://arxiv.org/pdf/1407.0354.pdf j. barrionuevo , r. m. burton , k. dajani and c. kraaikamp . _ acta arith . _ , * lxxiv * ( 4 ) , 311 - 327 , 1996 . v. baladi , m. todd : linear response for intermittent maps , _ comm . ( to appear ) , 2015 , preprint at http://arxiv.org/abs/1508.02700 r. cawley and r.d . 92 , 196236 , 1992 . k. dajani and c. kraaikamp . . carus math . assoc . of america , washington , dc , 2002 . 119 , 105108 , 1993 . a. denjoy . sur une fonction relle de minkowski . , 17:105151 , 1938 . m. denker and w. philipp . 4 , 541552 , 1984 . k. falconer . john wiley , new york , 1990 . k. falconer . 136 , 167174 , 2004 . j. m. freitas , m. todd : statistical stability for equilibrium states . _ dynamics , games and science . ii _ , 317321 , springer proc . math . , 2 , springer , heidelberg , 2011 . t. jordan , m. kessebhmer , m. pollicott and b.o . stratmann . 206:161183 , 2009 . s. kalpazidou , a. knopfmacher and j. knopfmacher . metric properties of alternating lroth series . , 319325 , 1991 . m. kessebhmer and b.o . a multifractal analysis for stern - brocot intervals , continued fractions and diophantine growth rates . , 133 - 163 , 2007 . m. kessebhmer and b.o . 2 , 489503 , 2009 . m. kessebhmer and b.o . 9 , 26632686 , 2008 . m. kessebhmer , s. munday and b.o . , 32 no . 3:9891017 , 2012 . i. ya . chicago press , 1964 . w. li , d. xiao and f. dekking . 2 , 345355 , 2002 . j. lroth . . _ _ 21 , 411423 , 1883 . b. mandelbrot . . freeman , san francisco , 1977 . manneville , y. pomeau , intermittent transition to turbulence in dissipative dynamical systems , _ comm . _ 74(2 ) , 189 - 197 , 1980 . r.d . mauldin and m. _ graph directed markov systems : geometry and dynamics of limit sets _ , cambridge university press , 2003 . h. minkowski . _ geometrie der zahlen_. gesammelte abhandlungen , vol . 2 , 1911 ; reprinted by chelsea , new york , 4352 , 1967 . s. munday . _ discrete cont . syst . _ 34(2 ) , 709732 , 2014 . k. nakaishi . . _ ergod . _ 20 , 843857 , 2000 w. parry and m. pollicott . _ zeta functions and the periodic orbit structure of hyperbolic dynamics . _ astrisque no . 187 - 188 ( 1990 ) . r. salem . on some singular monotonic functions which are strictly increasing . 3 , 427439 , 1943 . o. sarig . thermodynamic formalism for countable markov shifts , _ ergodic theory and dynamical systems _ 19(6 ) , 1565 1593 , 1999 o. sarig . phase transitions for countable markov shifts , _ commun . _ 217 , 555 577 , 2001 o. sarig . existence of gibbs measures for countable markov shifts . soc . _ 131(6 ) , 17511758 , 2003 s. troscheit . 2 , 295311 , 2014 .

in gou et al . ( 2011 ) , we reported that the black hole primary in the x - ray binary cygnus x-1 is a near - extreme kerr black hole with a spin parameter ( ) . we confirm this result while setting a new and more stringent limit : at the ( 99.7% ) level of confidence . the earlier work , which was based on an analysis of all three useful spectra that were then available , was possibly biased by the presence in these spectra of a relatively strong compton power - law component : the fraction of the thermal seed photons scattered into the power law was% , while the upper limit for reliable application of the continuum - fitting method is% . we have subsequently obtained six additional spectra of cygnus x-1 suitable for the measurement of spin . five of these spectra are of high quality with in the range 10% to 19% , a regime where the continuum - fitting method has been shown to deliver reliable results . individually , the six spectra give lower limits on the spin parameter that range from to , allowing us to conservatively conclude that the spin of the black hole is ( ) .

1308.4760

the x - ray binary cygnus x-1 was discovered in the early days of x - ray astronomy @xcite , and its compact primary was the first black hole candidate to be established via dynamical observations @xcite . recently , in three sequential papers on cygnus x-1 , we reported accurate values of the source distance @xmath7 @xcite , black hole mass @xmath8 and orbital inclination angle @xmath9 @xcite , and an extreme value for the black hole s spin parameter , @xmath0 ( @xmath10 ; * ? ? ? * ) with @xmath11 , where @xmath12 is the angular momentum of the black hole . ] . we measured the spin of the black hole by fitting the thermal x - ray continuum spectrum of the accretion disk to the thin - disk model of @xcite . the key fit parameter is the radius of the inner edge of the accretion disk , which is equivalent to the radius of the innermost stable circular orbit @xmath13 @xcite . in turn , @xmath14 is directly related to the dimensionless spin parameter @xmath15 @xcite . the continuum - fitting method of measuring spin is simple : it is strictly analogous to measuring the radius of a star whose flux , temperature and distance are known . by this analogy , it is clear that one must have accurate estimates of @xmath7 , @xmath8 and @xmath9 in order to obtain an accurate estimate of @xmath15 by fitting the x - ray spectrum . the robustness of the continuum - fitting method is demonstrated by the dozens or hundreds of independent and consistent measurement of spin that have been obtained for several black holes ( e.g. , * ? ? ? * ) , and through careful consideration of many sources of systematic errors ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . herein , using the continuum - fitting method @xcite and precisely the same methodologies that are described in gou et al . ( 2011 ; hereafter gou11 ) but using data of much higher quality we confirm our conclusion that cygnus x-1 s black hole is a near - extreme kerr hole , a result that has received support via the independent fe - line method of measuring spin ( see section 7.1 ) . importantly , these new data allow us to obtain a more stringent limit on the spin parameter , namely @xmath2 ( @xmath10 ) . for reliable application of the continuum - fitting method , it is essential that the thermal disk component dominate over the compton power - law component @xcite , which is always present in the spectra of x - ray binaries . it is by this criterion that the present data are of much higher quality than those analyzed in gou11 , as we now explain . the strength of the complicating compton component is parameterized by the scattering fraction @xmath5 , which is the fraction of the thermal seed photons that are scattered into the power - law component @xcite . ideally , @xmath5 is a few percent , while the limit for reliable application of the continuum - fitting method , based on a thorough investigation of two black hole binaries , has been shown to be @xmath16% @xcite . the extreme spin reported in gou11 is based on an analysis of the only three spectra of cygnus x-1 that were then available and suitable for measurement of spin via the continuum - fitting method . one spectrum was marginally within the limit ( @xmath17% ) and the other two were above the limit ( both with @xmath18% ) . herein , we report on spin results for six new spectra , five of which have much more favorable scattering fractions in the range @xmath1910%-19% . each of the six spectra individually confirms the spin limit set by gou11 ( @xmath0 at @xmath1 ) . it is challenging to measure the spin of cygnus x-1 not only because the compton component is always relatively strong for this source ( e.g. , see section 4.3 in * ? * ) , but also for two additional reasons : ( 1 ) it is essential to have spectral data that span a broad energy range , @xmath20 kev , in order to simultaneously constrain the unusually soft thermal component ( @xmath21 kev ) and the compton power - law and reflected components ( see section 2 and figure [ fig : third ] in gou11 ) , and such broadband data are rare ; and ( 2 ) the source dwells in its soft state only a small fraction of the time . in mid-2010 , cygnus x-1 again entered the soft state . seizing this opportunity , we observed the source with _ chandra _ , _ swift _ , _ suzaku _ , and _ rxte _ and obtained the spectra with moderate values of @xmath5 that are mentioned above . the times of these various observations are indicated by arrows in the x - ray light curve shown in figure [ fig : first ] . the paper is organized as follows : in section 2 we describe the observations and data reduction , and in section 3 the data analysis and our spectral model . presented in sections 4 , 5 and 6 respectively are our results , a discussion of their robustness , and a comprehensive analysis of the errors . in section 7 we first discuss spin results obtained using the fe - line method and then compare cygnus x-1 to two other well - studied persistent black hole systems . we offer our conclusions in the final section . in late 2010 and during 2011 , we made the five soft - state observations listed in table 1 using _ chandra _ , _ swift _ , _ suzaku _ and _ rxte_. for the four _ chandra _ and _ swift _ observations , the essential high - energy coverage was provided via simultaneous observations made using the proportional counter array ( pca ) aboard _ rxte_. because _ rxte _ observations are segmented by earth occultations and because we require that the _ rxte _ observations be strictly simultaneous ( with those of _ chandra _ or _ swift _ ) , we chopped the five observations into ten observation intervals , each providing one of the spectra s1s10 that are listed in table 1 . here and throughout , `` spectrum '' refers to a segment of an observation , as schematically defined in table 1 . while two spectra may be part of a single contiguous observation , any two observations were obtained at disjoint time intervals and correspond to distinct pointings . we adhere to this language throughout . observation no . 1 , which corresponds to spectra s1s5 ( table 1 ) , is by far the most important observation because the compton component is relatively faint , much fainter than during observations 25 , and also much fainter than during the three observations reported on in gou11 . for this crucial observation , we show in figure [ fig : second ] the count rates measured by _ rxte _ in 16 s bins and _ chandra _ in 100 s bins . we now discuss in turn the observations and data reduction procedures for _ chandra _ and _ swift _ , and then for _ rxte _ , which provides the complementary high - energy coverage . in the final subsection , we discuss observation no . 5 , which was performed solely by _ suzaku _ , with the high - energy coverage provided by _ suzaku _ s hard x - ray detector ( hxd ) . table 1 gives for each observation basic information including the energy range used in analyzing the data for a given detector , the gross observation times , the effective exposure times , the intensity of the source in crab units , the spectral hardness ( figure [ fig : first ] ) and the orbital phase of the binary system . the orbital phase is useful for assessing the likelihood that an observation is affected by absorption dips , which are observed in both the hard and soft states of cygnus x-1 near phase zero ( e.g. , * ? ? ? * ; * ? ? ? this key observation ( obsid=12472 ) was obtained in the acis cc mode . as indicated in figure [ fig : second ] and discussed above , the observation , which has a total duration of 24 ks , was parceled up into five data segments . the start and stop times for each data segment , which are given in table 1 , are the same as those for the corresponding _ rxte _ pca observation ( figure [ fig : second ] ) . the individual pca observation times range from 1.5 ks to 3.4 ks , while the corresponding _ net exposure times are @xmath22 times shorter ( table 1 ) due to the telemetry saturation . for this _ chandra _ observation , as well as for observation no . 2 ( see section 2.2 ) , we ( i ) used the high - energy transmission grating ( hetg ) and the advanced camera for imaging and spectroscopy ( acis - s ; * ? ? ? * ; * ? ? ? * ) ; ( ii ) binned the data to achieve a minimum number of counts per channel of 200 ; and ( iii ) made no allowance for systematic error because the statistical error is completely dominant for all our _ chandra _ data ; e.g. , adding in each channel a systematic error of 1% in quadrature with the statistical error leaves our fit results unchanged . observing a bright source such as cygnus x-1 is challenging because of the effects of `` pileup , '' i.e. , the arrival of two or more photons in the same or adjacent pixel within a single frame time . the cc mode has the virtue of a short 2.85 ms frame time that is achieved by continuously transferring the data from the image array to the frame - store array . while this largely solves the problem of pileup ( see section 5.2.1 ) , it results in the collapse of the 2d image into a 1d image and a consequent loss of information on the spatial distribution of photons ( also see section 2 in gou11 ) . telemetry limitations are also a consideration in observing a bright source . accordingly , we only transmitted the data for the high energy grating ( heg ; -1 order ) and medium energy grating ( meg ; + 1 order ) components of the hetg . we used the standard procedures for extracting the data , which ( apart from the 1.32.0 kev chip gap in the meg spectrum ) were fitted over the energy range 0.88.0 kev . in reducing these te - mode data ( obsid=13219 ) , we followed the method described by @xcite while again using the orders of the heg and meg spectra mentioned above . for the te - mode data we also used the readout `` streak '' spectra located alongside the heg and meg spectra . we followed the recommended procedures in extracting the streak and background spectra . although the net exposure times for the two te - mode spectra s6 and s7 are respectively 3.6 ks and 0.9 ks , the effective exposure times for the streak spectra are only about 19.2 s and 5.0 s , respectively . as in gou11 , we used the full 0.510 kev bandwidth for the streak spectra , which has a pileup fraction that is less than 3% over the whole range . for the heg and meg spectra , we used the energy ranges 0.70.9 kev and 7.010.0 kev and confirmed that the pileup fraction in these energy intervals is less than 5% . ( see section 5.2 for a discussion of pileup effects . ) three _ swift / rxte _ observations were performed on ut october 8 , 24 and 26 . we disregard the observation of october 24 because the _ rxte _ data were not simultaneous and the source was highly variable during this period . the wt mode was used to minimize the effects of pileup . the data were extracted using the procedures outlined in @xcite . we used for the background region an extraction aperture of @xmath23 pixels and for the source region @xmath24 pixels ( i.e. , 40 pixels along the image strip and 20 pixels transverse to it ; 1 pixel = 2.36 arcsec ) . despite our use of the wt mode , the data are strongly affected by pileup . pileup is negligible below 100 counts s@xmath25 and moderate below 300 counts s@xmath25 @xcite . however , the count rate exceeded 800 counts s@xmath25 for all of our observations . we therefore excluded a @xmath26 pixel region in the center of the source extraction region to ensure that pileup effects are small ( see section 5.2.3 for details ) . we netted three simultaneous _ swift_+_rxte _ observations , each @xmath27 ks in duration ( table 1 ) , that we used to measure spin . although the gap between the two observations is only @xmath28 min , we chose not to combine them because our model fits show strong source variability , with the source intensity increasing from 0.59 crab to 0.90 crab ( table 1 ) and the scattering fraction increasing from 31% to 50% ( section 3 ) . we binned all the _ swift _ data to achieve a minimum of 200 counts per channel , and we included a systematic error of 0.5% in the count rates in each pha channel . as in gou11 , for _ rxte _ we used only the data for pcu2 , which is widely regarded as the best - calibrated detector . meanwhile , it is unimportant whether one uses pcu2 alone or all the pcus ( gou11 ) . all the _ rxte _ spectra have been reprocessed using the latest pca calibrations available in nasa software release heasoft 6.13 . in particular , we generated new response files and used the latest assignments for converting pulse - height channel to energy . in addition , we used a revised pca background model , `` pca_bkgd_cmvle_emv20111129.mdl '' , which we obtained from the pca instrument team . furthermore , we corrected the effective area of the pca using the @xcite spectrum of the crab nebula precisely as described in section 2 in gou11 , thereby obtaining for observation nos . 1/2/3/4 normalization correction factors @xmath29 of 1.128/1.133/1.123/1.123 and power - law slope correction factors @xmath30 0.022/0.024/0.023/0.023 ; the respective dead time correction factors are 1.029/1.039/1.048/1.048 . finally , as customary for pca observations of bright sources , we included an allowance of 0.5% for systematic error . we fitted the _ spectra over the energy range 2.950 kev ( pulse - height channels 4 to 83 ) . both the x - ray imaging spectrometer ( xis ) and the hard x - ray detector ( hxd ) were used for this observation with a gross observing time of @xmath31 ks ( table 1 ) . we reduced the data using the standard procedures described in @xcite . there is no fast readout mode for the xis detector , and the effects of pileup are large , even though we excluded the counts in the central source region within a radius of 60 pixels . to achieve an acceptable fit ( @xmath322.0 ) , for the xis we ignored the energy ranges : 1.71.9 kev and 2.12.3 kev , and for the hxd we ignored the energy range below 20 kev . we furthermore added the 2% customary systematic error for the xis . ( no systematic error was included for the hxd . ) given ( 1 ) that the fit we were able to achieve is relatively poor with @xmath33 , ( 2 ) the lack of any constraint on the reflection component in the 1020 kev band , and ( 3 ) the significant effects of pileup we do not use the _ suzaku _ spectrum to estimate spin , although for completeness we list the observation in table 1 . a soft - state spectrum of cygnus x-1 consists of thermal , power - law and reflected components , which are illustrated in figure [ fig : third ] . the latter component includes the fe k@xmath34 emission line . a schematic sketch of the physical structures that generate the three spectral components , namely the accretion disk and the corona , are shown in figure 2 in gou11 . the spectral fitting package xspec ] version 12.6.0 @xcite was used for all data analysis and model fitting . unless otherwise indicated , the error on a single parameter is reported at the @xmath35 ( 68.3% ) level of confidence . in this section and the one that follows , the input parameters @xmath7 , @xmath9 and @xmath8 are fixed at their fiducial values ( see section 4 ) . in gou 11 , we analyzed three spectra of cygnus x-1 using a progression of seven preliminary models . the first three models , nr1nr3 , were nonrelativistic , with the accretion disk component modeled using diskbb . the results for the physically most realistic of these models , nr3 , were adopted . we obtained consistent values of inner - disk temperature and radius for the three spectra , which are reported in table 7 in gou11 : @xmath36 , and @xmath37 ( std . dev . ; n=3 ) . next , we analyzed the spectra using four preliminary relativistic models , r1r4 . the principal component of these models is kerrbb2 @xcite , which is a fully relativistic model of a thin accretion disk . like diskbb , kerrbb2 returns two fit parameters , @xmath15 and the mass accretion rate @xmath38 ( instead of @xmath39 and @xmath40 ) . the models r1r4 progress toward our adopted model , where r1 is the most primitive model . the four models and our adopted model all gave very similar results for the key parameter @xmath15 . in gou11 , we presented a full set of results for models r1r4 to show clearly that our results for the spin parameter are insensitive to the analysis details , as expected given the dominance of the thermal component . in this paper , we employ a single model , namely the one adopted in gou11 , which is the most physically realistic of the eight models considered by gou11 . the structure of the model , showing all the components of which it is comprised , is expressed as follows : @xmath41}\end{aligned}\ ] ] in overview , the power - law component is generated by simplr operating on the thermal seed photons supplied by kerrbb2 , while the reflection component is generated in turn by ireflect operating on the power - law component . the fit returns a single value of @xmath15 , a key parameter that appears in kerrbb2 , kerrdisk and kerrconv . we now discuss the principal components of the model ( thermal , power - law and reflected ) and their relationships . ( for futher details and a complete description of each component , see gou11 . ) _ thermal component : _ the core component is the fully - relativistic thin - disk model kerrbb2 @xcite . the effects of spectral hardening are incorporated into the basic model kerrbb via a pair of look - up tables for the hardening factor @xmath42 corresponding to two representative values of the viscosity parameter : @xmath43 and 0.1 ( for details , see @xcite ) . throughout this work we use @xmath44 unless stated otherwise @xcite . as noted earlier , the two fit parameters of kerrbb2 are @xmath15 and @xmath38 , which along with @xmath8 determine the eddington - scaled bolometric luminosity of the disk , @xmath45 . as in gou11 , we turn on the effects of returning radiation and limb darkening , set the torque at the inner disk radius to zero , fix the normalization to unity , allow @xmath38 to vary freely , and fit directly for @xmath15 . _ power - law component : _ the term simplr@xmath46kerrbb2 models the power - law plus the observed thermal component . simplr @xcite has the same two parameters as its parent model simpl @xcite : the power - law photon index @xmath47 and the scattered fraction @xmath5 . however , simplr has one additional parameter , namely , the fraction of the power - law photons that strike the disk . in this application , simplr models a corona that scatters half the thermal seed photons outward and the remainder downward toward the disk , thereby generating the reflected component . _ reflected component _ : the remaining two additive terms inside the square brackets model the reflected component . simplc , which is the isolated compton component that illuminates the disk , is equivalent to simplr@xmath46kerrbb2 minus the unscattered portion of the thermal component @xcite . the reflected spectrum generated by ireflect acting on simplc contains numerous sharp absorption features but no emission lines . we supplement this partial reflection model by employing the line model kerrdisk and the convolution smearing model kerrconv @xcite . we model the emissivity profile as a single power law with index @xmath48 , and tie together all the common parameters of these two models , including the two principal parameters @xmath15 and @xmath48 . ( for further details concerning assumed values of elemental abundances , disk temperature , etc . , see gou11 ) . the three multiplicative model components are ( 1 ) crabcor , which corrects for calibration deviations relative to toor & seward ( see section 2 in gou11 and * ? ? ? * ) ; ( 2 ) const , which allows for discrepancies in the calibrations of the various detectors ( the normalization of the _ rxte_/pcu2 detector is fixed to unity and the normalizations of the _ chandra _ and _ swift _ detectors are allowed to float ) ; and ( 3 ) tbabs a standard low - energy absorption model @xcite . comparing figure [ fig : third ] with the corresponding figure 3 in gou11 , one sees at a glance that spectra s1s5 ( with @xmath49% ) are much more strongly disk - dominated than spectra sp1sp3 in gou11 ( with @xmath3% ) . for spectra s1s5 , the peak flux in the thermal component is 510 times the peak flux in the power - law component , and it is @xmath50 times the peak flux in the reflected component . in this section , we present results with the key input parameters fixed at their fiducial values : @xmath51 kpc , @xmath52 , and @xmath53 deg @xcite . the fit results for all ten spectra , s1s10 , are summarized in tables 2 and 3 . before broadly discussing the results , we focus on the value of the scattering fraction , @xmath5 ( tables 2 and 3 ) , and we strictly follow the data selection criterion @xmath54% @xcite . therefore , we henceforth consider only the six spectra s1s6 for which @xmath55% , and we disregard the remaining spectra ( s7s10 ) . before focusing solely on spectra s1s6 , however , we note that the results for the rejected spectra are , in detail , consistent with those of the selected spectra . the most notable difference is the depressed value of @xmath15 for s7 ( 0.972 vs. 0.999 for the other nine spectra ) ; but note the poor fit achieved for s7 ( @xmath56/dof@xmath57 vs. a mean value of 1.11 for s1s6 ) . meanwhile , a comparison of the mean values of the parameters for spectra s1s6 with their corresponding mean values for spectra s7s10 shows that in most cases the mean values differ by @xmath58% ( tables 2 and 3 ) . the two notable exceptions ( apart from of course the scattering fraction ) are the steeper power - law slope ( @xmath59 and significantly weaker fe line for the four rejected spectra . finally , we note that the values of @xmath5 for three of the rejected spectra ( s7 , s8 and s10 ) are very nearly the same as for the two inferior spectra used in gou11 ( sp2 and sp3 ) , namely @xmath60% . we now direct our attention hereafter solely to spectra s1s6 with values of @xmath61% . the fits are all good , with @xmath62/dof ranging for s1 from 0.95/628 to 1.40/491 for s6 . the spin parameter is very high and is pegged at the @xmath63 limit of the kerrbb2 model @xcite , which is the principal result of this section . the luminosity of the disk component is low and uniform , @xmath64% , and it easily meets a key data selection criterion for successful application of the continuum - fitting method , namely @xmath65% @xcite . correspondingly , the disk is expected to be geometrically thin at all radii ( @xmath66 ; see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . meanwhile , the spectral hardening factor @xmath42 is well determined ( @xmath67 ) because the disk luminosity is sufficiently high . the column density is statistically well determined with uncertainties of only 12% , , @xmath68 , agrees very well with the values derived from the 21-cm line in the direction of cygnus x-1 , which is @xmath69 , a weighted average from both lab and dl maps @xcite . ] while it varies by 3.3% ( std.deviation ; n=6 ) . the variability is unsurprising since @xmath70 varies by several percent for all three well - studied supergiant black - hole binaries : m33 x-7 @xcite , cygnus x-1 @xcite , and lmc x-1 @xcite . the power - law slope is well determined and quite stable , @xmath71 ( std . deviation ; n=6 ) , and its value is the expected one for the steep power - law state ( @xmath72 ; * ? ? ? the ionization parameter is modest and in the range @xmath73 . in gou11 , we discuss many factors that might affect our principal result , namely the extreme spin of cygnus x-1 ; we find that none of them are significant . here , we review these matters briefly . for further details , see sections 5 and 7 in gou11 , and also see section 5 in @xcite . sections 5.5 and 5.6 below are wholly new and discuss our adopted reflection model in relation to the recently - released reflection model xillver @xcite . section 5.2 on pileup and section 5.8 on the effect of dust scattering are likewise new . the accuracy of continuum - fitting results ultimately depends on the reliability of the novikov - thorne model . the key assumption of this model is that the torque , and hence the flux , vanishes at the isco @xcite . the effects of this approximation on spin measurements have been quantitatively investigated via general relativistic magnetohydrodynamic ( grmhd ) simulations of thin disks by several authors @xcite . the general consensus is that the zero - torque approximation introduces uncertainties in spin estimates of around @xmath74 for low spin values ( @xmath75 ) and much smaller errors as @xmath76 . these error estimates , which are for geometrically thin disks ( @xmath77 , or @xmath78 ) are in practice less than the observational errors in the parameters @xmath7 , @xmath8 and @xmath9 . for more details concerning the novikov - thorne model and a discussion of other sources of model errors , see section 5 in @xcite . we estimated the pileup fraction for each spectrum from eqn . 2 in the _ chandra _ abc guide to pileup using the observed photon flux as the input . there can be no clear - cut prescription for what level of pileup is acceptable because its effects depend in complex ways on detector performance and science goals . the guideline on pileup for _ chandra _ data stated in the chandra proposer s observatory guide ( section 6.15.12 ) is : `` if one s scientific objective demands precise flux calibration , then the pileup fraction should probably be kept well below 10% . '' a specific concern for this paper is that pileup effects , which tend to harden a continuum spectrum , might significantly boost the value of the spin parameter . we find this not to be the case , as we now discuss . for the five cc - mode spectra ( s1s5 ) , the effects of pileup are small , @xmath79% over the full fitting range of 0.88.0 kev ( figure [ fig : forth ] ) , because of the gratings and the nominal 2.85 ms frame time . we nevertheless made two tests to assess the effects of pileup , both of which show that they are negligible . first , we refitted the five spectra ignoring the _ chandra _ data above 2.0 kev ( i.e. , the data that determine @xmath47 ) , and we found that the values of the key parameter , @xmath15 , remained unchanged and pegged at the physical limit , while @xmath47 and the scattering fraction in all cases changed by less than 1.1% and 4.0% , respectively . the small change in @xmath47 is as expected , since the number of _ chandra _ counts in the spectrum above 2 kev is only @xmath80% of the _ rxte _ counts . secondly , we performed a marx simulation to quantitatively estimate how the pileup fraction affects the power - law component for a single _ chandra _ spectrum ( i.e. , excluding its companion _ rxte _ spectrum ) . because marx does not support the cc mode we relied on simulations of te - mode data . we simulated a parent te - mode spectrum using parameters that describe a typical cygnus x-1 spectrum . we then used the simulated spectrum to generate four fake spectra with pileup fractions ( at peak flux ) of 1.5% , 3% , 5% and 10% . we fitted these spectra using our nonrelativistic model nr3 ( gou11 ; the power - law component is poorly constrained for the relativisitc model ) , excluding in this case the reflection component pexriv , and we compared the results to the results obtained by fitting the parent spectrum ( @xmath81 , @xmath5=0.187 , @xmath82 , @xmath83 ) . the photon index @xmath47 increased respectively by 0.8% , 2.2% , 2.1% and 6.4% , and the fractional change in the scattering fraction was 6.7% , 15% , 22% , 59% . meanwhile , concerning the thermal component , the disk temperature decreased respectively by 0.1% , 0.5% , 0.3% and 2.3% , while the corresponding normalization constant decreased by 2.2% , 3.6% , 8.1% and 9.9% . because @xmath40 is proportional to the square root of the disk normalization , the fractional change in @xmath40 is even smaller , decreasing respectively by 1.1% , 1.8% , 4% , and 5% . we conclude therefore that the spin is likely to be only very moderately overestimated . these results give reasonable assurance that our fit results ( tables 2 and 3 ) are negligibly affected by pileup , given that the peak pileup fraction for the five cc - mode spectra is @xmath79% and for the te - mode streak spectrum is @xmath84% ( see below ) . our two te - mode spectra ( s6 and s7 ) suffer more than the cc - mode spectra from the effects of pileup , especially the heg and meg components of the spectrum . we estimated the pileup fraction using the kernel pileup in isis , whose mathematical formulation is also eqn . 2 in the _ chandra _ abc guide to pielup . the pileup fraction for the streak spectrum is @xmath84% over our entire fitting range of 0.510 kev . in order to ensure a pileup fraction of @xmath85% for the meg and heg spectra , we only used data in two restricted energy ranges : 0.70.9 kev and 7.010.0 kev . as a test for the effects of pileup , we refitted spectrum s6 excluding the meg and heg data . the largest effect on the fit parameters was an @xmath86 change in the column density , which decreased from @xmath87 to @xmath88 . the changes in the best - fit values of the other parameters are much less than 1% . the three xrt wt - mode spectra ( s8s10 ) have the same format as the _ chandra _ cc - mode spectra ; i.e. , they are collapsed one - dimensional strips rather than images . we reduced the effects of pileup by excluding the central @xmath26-pixel region ( i.e. , a 20-pixel - wide swath extending 15 pixels along the image strip ) , a choice validated by @xcite . these authors performed pileup tests with the excluded region ranging from 0 pixels to 15 pixels ( i.e. , from 0@xmath8920 pixels to 15@xmath8920 pixels ) and for five levels of source intensity . they found no pileup effects ( i.e. , spectral distortion ) for count rates in the range 0100 counts s@xmath25 , and only moderate effects in the range 100300 counts s@xmath25 . in our case , we therefore expect minimal pileup effects because our count rate ( with the central region excluded ) is only @xmath90 counts s@xmath25 after the exclusion . nevertheless , we performed one additional test : we refitted spectrum s8 ignoring the xrt data above 3 kev ( while fitting jointly with the _ rxte _ data , as before ) and found our fit results to be the same as those reported in table 3 . in gou11 , we showed that the fe line and other reflection features in soft - state spectra of cygnus x-1 are cosmetic and have a negligible effect on the continuum - fitting measurement of spin . specifically , we refitted the three spectra considered in gou11 excluding the 5.0 - 10.0 kev band and the fe - line component kerrdisk . this removed the energy range covering the fe k@xmath34 line and edge as well as a feature in the residuals near 9 kev .. ] we found that our spin results were essentially unchanged , as expected given the modest equivalent widths of these features and the relative faintness of the reflected component ( see section 5.6 and figure 3 ) . in section 5.2 of gou11 , we showed that the energy coverage of the pca , which extends to 45 kev , is sufficient to adequately constrain both the power - law and reflection components . we did this by refitting one spectrum including _ rxte _ hexte data , which cover the range 20 kev to 150 kev . this result is not surprising since coverage to 45 kev is more than adequate to determine the power - law component and the reflection component is decreasing rapidly at 45 kev ( figure 3 ) . as in section 5.3 in gou11 , we replaced our reflection component kerrconv@xmath46ireflect@xmath46simplc+kerrdisk with kerrconv@xmath46reflionx @xcite , which is widely used in measuring spin via the fe k@xmath34 line . as in gou11 , we again find that the effects on the spin parameter are essentially nil . more recently , a new and improved reflection model xillver has become available @xcite . this version of xillver ( like reflionx ) is intended for use when the thermal disk flux is faint compared to the incident power - law flux , and it is therefore not well - suited to our case . nevertheless , as with reflionx , we performed a test by replacing our reflection component with kerrconv@xmath46xillver . the fits are poorer with reduced chi - square ranging from 1.9 to 2.2 for s1-s5 , but the effects on the spin parameter were again found to be negligible ( less than 0.2% ) . in computing the reflected component , we rely on ireflect , which generates a spectrum containing sharp absorption features and no emission lines . figure 20 in @xcite shows that ( ignoring line emission ) ireflect / pexriv is a good approximation to the sophisticated model xillver at low ionization , @xmath91 ( left panel ) , while it is a very poor approximation at high ionization , @xmath92 ( right panel ) . in figure [ fig : fifth ] , we show that for an intermediate case , @xmath93 , which corresponds to the moderately ionized disk of cygnus x-1 ( see tables 2 and 3 ) , ireflect / pexriv is in reasonable agreement with xillver . considering further that the peak reflected flux is @xmath50 times fainter than the peak thermal flux ( figure [ fig : third ] ) , it is not surprising that our estimate of spin is insensitive to the choice of reflection model . in all the fits we have fixed the disk temperature in the reflection model at @xmath94 k , which corresponds to 0.52 kev . the disk temperature is quite constant at this value for spectra s1-s6 and the three spectra in gou11 ( see section 3 ) . meanwhile , increasing the disk temperature by 50% to @xmath95 k or halving it has a negligible effect on the spin and other key parameters ( apart from the ionization parameter ) . reanalyzing the data using @xmath43 instead of our adopted value of @xmath44 has a very slight effect on our results , and doing so only increases the already extreme value of spin . the effects of varying metallicity are likewise very small , whether one grossly decreases its value to a tenth solar or considers the suprasolar values implied by the ireflect fits ( tables 1 and 2 ) . in the former / latter case , the spin is depressed / increased , but only very slightly ( see section 5.4 in gou11 ) . an analysis of high resolution optical spectra of the donor star indicates that fe is somewhat overabundant relative to solar @xcite . in determining the spins of supermassive black holes via the fe k@xmath34 method , careful modeling of absorption by intervening warm gas is usually crucial ( e.g. , * ? ? ? however , we have shown , via a continuum - fitting analysis of _ chandra _ hetg spectra , that the effect of warm absorbers is unimportant in estimating the spin of cygnus x-1 ( see section 7.6 in gou11 ) . the dust scattering halo of cygnus x-1 ( e.g. , * ? ? ? * ) has an effect on the source spectrum that is equivalent to direct absorption . in order to assess the effects of dust scattering on our results , we used the only relevant model that is presently available in xspec , namely dust . the model assumes that the source flux is scattered into a uniform disk whose size and total flux vary respectively as @xmath96 and @xmath97 . the simple model dust is a good approximation to more accurate models ( e.g. , weingartner & draine 2001 ) at energies in the bandpass of interest , namely @xmath98 kev ( table 1 ) . we reanalyzed spectra s1s5 as before , but this time we included the multiplicative model component dust . the model has two parameters that specify at 1 kev ( 1 ) the fraction of photons scattered by dust grains , and ( 2 ) the size of the halo in units of the detector beam size . if both parameters are allowed to vary , neither can be constrained . we therefore initially fixed the scattering parameter to 0.17 , which was obtained by extrapolating the value 0.12 at 1.2 kev given by ( * ? ? ? * see their figure 10 ) . the results obtained for the key parameters @xmath15 and @xmath5 for each of the five spectra are essentially identical to those that appear in table 2 , although the column density @xmath70 is reduced by @xmath99% . even if one increases the dust scattering parameter from 0.17 to 0.3 , the values of @xmath15 and @xmath5 are essentially unchanged , while in this case @xmath70 is reduced by @xmath50% . we conclude that the effects of dust scattering are unimportant . in section 7.4 in gou11 , we considered a principal source of uncertainty in the continuum - fitting method , namely , whether the black hole s spin axis and the inner disk will align with the orbital plane . if , as some evidence suggests , the persistent supergiant systems are formed by direct , kickless collapse @xcite , then spin - orbit alignment would be expected for these systems . ( for full discussions on the topic of spin - orbit alignment , see section 1 in @xcite , and section 5.4 in @xcite ) . in any case , as we demonstrate for cygnus x-1 in figure 5 in gou11 , even for a misalignment angle as large as , e.g. , 16 deg the spin parameter is still @xmath1000.95 ( ignoring the uncertainties in @xmath7 , @xmath8 and @xmath9 ) . the dominant error in all continuum - fitting measurements of spin is attributable to the observational uncertainties in the source distance , black hole mass and disk inclination . for cygnus x-1 , we have determined accurate values for these quantities : @xmath101 kpc @xcite , @xmath102 , and @xmath103 deg @xcite . quite generally , even the uncertainties in the analytic novikov - thorne model are significantly less important than the uncertainties in @xmath7 , @xmath8 and @xmath9 , as has been shown via grmhd simulations of thin accretion disks ( section 5.1 ) . the model errors in the case of cygnus x-1 are very small because the black hole s spin is extreme and the disk s luminosity is low , only @xmath104% of the eddington limit . @xcite have shown via a detailed analysis that for an inclination of 30 deg ( closely approximating cygnus x-1 s 27 deg inclination ) the novikov - thorne model overestimates the spin parameter by only @xmath105 for spin parameters in the range 0.900.98 . the contribution to the uncertainty in the spin of cygnus x-1 due to the uncertainty in the absolute calibration of the flux is about the same as that due to the 6% uncertainty in the distance . we therefore include in our error budget a 10% uncertainty in flux @xcite by inflating the uncertainty in @xmath7 by the method described in gou11 . the final error we report for @xmath15 therefore includes the uncertainty in the absolute flux calibration as well as the uncertainties in @xmath7 , @xmath8 , @xmath9 . collectively , the uncertainties in these four quantities completely dominate the error budget . ( other , smaller sources of error are discussed in detail in appendix a and section 5 in @xcite , and section 5 in @xcite ) . following precisely the same procedures described in section 6 of gou11 , we determined the error in @xmath15 due to the the combined uncertainties in @xmath7 , @xmath8 and @xmath9 via monte carlo simulations . figure [ fig : sixth ] shows the resultant spin histograms for our six spectra and displays for each spectrum the corresponding lower limits on @xmath15 at a @xmath1 level of confidence . were we to use these six limits to derive a joint constraint on spin , it would be more stringent than any one of the individual limits . we choose instead the conservative approach of adopting the most constraining _ single _ limit for our final result , namely , the limit for spectrum s3 . _ we therefore conclude that @xmath2 at the @xmath1 level of confidence _ obtained for sp1 as our final result because it was the only one of the three spectra whose scattering fraction was @xmath106% . ] . we note the following two caveats : first , we assume that the spin of the black hole is approximately aligned with the angular momentum vector of the binary ( section 5.10 ) . second , we assume that the asynchronous dynamical model is correct ( see section 7.3 in gou11 ) . we first discuss three spin estimates for cygnus x-1 made using the fe - line method , which provide support for an extreme value of spin . we then relate cygnus x-1 to the other members of its distinctive class of black - hole x - ray sources that are persistently bright . three recent measurements of the spin of cygnus x-1 obtained using x - ray reflection spectroscopy , aka the fe line method @xcite , support a high or extreme value of spin . @xcite report @xmath107 . their provisional result is based on an analysis of a single simultaneous observation made using _ xmm - newton _ and _ rxte_. a limitation of their result is that it depends on assuming a single , fixed value of 3 for the emissivity index @xmath48 , which is a canonical value . that is , they assume that the intensity of the flux irradiating the disk varies with radius as @xmath108 . when they allow @xmath48 to vary freely , both the spin parameter and emissivity index are poorly constrained ( see their table 1 ) . in short , their data are unable to determine both the profile of the illuminating radiation and the spin . the result of @xcite is superseded by that of @xcite who report @xmath109 . this result is based on an analysis of a single hard - state _ suzaku _ spectrum . fabian et al . describe this spectrum as `` an average data set '' ( from a collection of 20 similar spectra ) and report that consistent results were obtained for other data sets . the fit over a 1500 kev band gives precise results for a 3-parameter , broken power - law model of the radial profile of the irradiating flux : inside the break radius ( @xmath110 @xmath111 ) , @xmath112 , and outside @xmath113 . most recently , @xcite fitted the fe - k@xmath34 line using _ suzaku _ and _ nustar _ data . cygnus x-1 was in the soft state . their best - fitting model gives @xmath114 ( 90% confidence level ) and all the models that provided a good fit to the spectrum indicate a rapidly rotating bh with @xmath115 . a strength of this work relative to prior studies of cyg x-1 s spin ( including our first paper , gou11 ) is the considerable attention we give here to assessing the effects of a wide range of systematic errors . in doing so , and from a wider breadth of data , our work supplies the strongest evidence for cyg x-1 s extreme spin , confirming the prior leading measurements by gou11 and @xcite . earlier , @xcite reported a near - zero spin for cygnus x-1 , @xmath116 , based on an analysis of two _ xmm - newton _ spectra . neither @xcite nor @xcite offer an explanation for this glaring discrepancy . however , recently an explanation was suggested for the near - zero spin reported by miller et al . in terms of pileup effects ( see section 4.3 in * ? ? ? this example shows that measurements of spin in the literature can be grossly affected by systematic effects , which should be carefully considered in assessing the reliability of spin results . there are five dynamically established black - hole binaries containing wind - fed black holes and o - supergiant or wolf - rayet companions @xcite ; these systems are persistently x - ray bright . in the following , we do not consider the two systems with wolf - rayet companions , ic 10 x-1 and ngc 300 x-1 , because the masses of their black holes are very uncertain and their spins have not been estimated . by contrast , the three remaining systems cygnus x-1 , lmc x-1 and m33 x-7 have well - determined values of both mass and spin . these fundamental data , which provide a _ complete _ description of these three black holes , appear in the two leftmost columns of table 4 . while acknowledging that the sample is small , it appears that wind - fed black holes with supergiant companions are restricted to high spin , @xmath117 , in contrast with the broad distribution of spins observed for roche - lobe - fed black holes with low or intermediate mass companions : four of them have low spins , @xmath118 , two have high spins , @xmath119 , and one has an extreme spin , @xmath0 ( see table 1 in * ? ? ? not only are the persistent black holes all rapidly spinning , they are also relatively massive , @xmath120 ( table 4 ) . by comparison , the masses of the transient black holes are significantly lower , and their mass distribution is remarkably narrow : @xmath121 @xcite . the data in table 4 highlight a sharp and well - known distinction between the persistent systems and the transient systems , namely that the secondary stars in the former are far more massive , @xmath122 ( table 4 ) ; they likewise have much higher temperatures , @xmath123 k @xcite . the masses and temperatures of the secondaries in the transient systems are typically @xmath124 and @xmath125 k ; even in exceptional cases , their masses and temperatures are only @xmath126 and @xmath127 k @xcite . finally , we note that for the persistent systems the radii of the secondaries and orbital periods fall in quite narrow ranges ( table 4 ) , while the radii and periods for the transient systems are very broadly distributed , a distinction that is elegantly illustrated in jerome orosz s schematic sketch of 21 black hole binaries ( see figure 1 in * ? ? ? the persistent black holes were very likely born with their high spins because their host systems are too young for the black holes to have had time to spin up via accretion torques ( see section 7.7 in gou11 for details ) . the ages of cygnus x-1 , lmc x-1 and m33 x-7 are @xmath128 million years , whereas the spin - up times are @xmath129 million years if one assumes the maximum , eddington - limited accretion rate . meanwhile , the spin - up times are likely much longer than 17 million years given that the systems are presently radiating at only @xmath130% of the eddington luminosity ( table 4 ) . the rotational energy of the persistent black holes is enormous , @xmath131 for m33 x-7 and lmc x-1 @xmath132 for cygnus x-1 @xcite billion year lifetime the energy radiated by the sun is @xmath133 . ] . correspondingly , a substantial fraction of the gravitational mass of these black holes is attributable to their rotational energy : @xmath134% for m33 x-7 and lmc x-1 and @xmath135% for cygnus x-1 . in gou11 , while considering a wide range of systematic effects , including uncertainties in the novikov - thorne disk model , we concluded that the spin of the black hole in cygnus x-1 is extreme : @xmath0 ( @xmath1 ) . unfortunately , the result was potentially biased by the relatively strong compton component of emission , the strength of which can be characterized by the fraction @xmath5 of seed photons that are scattered into the power - law component . the three spectra analyzed in gou11 have @xmath136% , while @xmath137% is the upper limit for reliable application of the continuum - fitting method @xcite . subsequently , @xcite employed the independent fe - line method and confirmed that the spin of cygnus x-1 is @xmath0 ( @xmath35 ) ; however , this result is less certain because systematic effects in the model have not been assessed . herein , we present a continuum - fitting analysis of six additional spectra , each of which confirms that @xmath0 ( @xmath1 ) . this confirmation is compelling first because sources of systematic error have been thoroughly addressed ( see section 5 herein ; sections 57 in gou11 ; * secondly , and crucially , five of the spectra , s1s5 , are only moderately comptonized with scattering fractions @xmath138 , a regime where it has been firmly established that continuum - fitting results are reliable . this conclusion is based on studies of two black holes : ( i ) 33 spectra of h1743322 with @xmath139% ( in the spl state ) each gave spins consistent with those obtained for dozens of thermal - state spectra ( @xmath140% ; @xcite ) ; and ( ii ) 25 spectra of xte j1550564 with @xmath141% each likewise gave spins consistent with those obtained for dozens of thermal - state spectra ( @xmath142% ; @xcite ) . in short , these two studies show that moderately comptonized spectra with @xmath143 , like s1s5 , give the same values of spin as spectra that are strongly disk - dominated with @xmath144 . our bottom line is that new and more reliable continuum spectra confirm the findings of gou11 while establishing an even more stringent limit on the extreme spin of cygnus x-1 s black hole : @xmath2 at a confidence level of @xmath1 ( 99.7% ) . we thank an anonymous referee for his / her many constructive comments and criticisms , particularly those concerning pileup . we are grateful for allocations of _ chandra _ , _ rxte _ and _ swift _ observing time granted by director h. tananbaum and project scientists t. strohmayer and n. gehrels , respectively . for help in planning the _ chandra _ observations , we thank m. nowak and n. schulz . we also thank m. nowak , j. wilms and bin - bin zhang for discussions on x - ray data analysis , r. smith for calling the effects of dust scattering to our attention , and s. yamada for reducing the suzaku data and j. g. xiang for reducing the chandra te data . this research has made use of data obtained from the high energy astrophysics science archive research center ( heasarc ) at nasa / goddard space flight center . for technical support in using the odyssey cluster , ljg thanks the harvard fas sciences division research computing group . ljg acknowledges the support of nsfc grant ( y211541001 , 11333005 ) and naoc grant ( y234031001 ) , and is also supported by the strategic priority research program `` the emergence of cosmological structures '' of the chinese academy of sciences , grant no . xdb09000000 , jem acknowledges support from nasa grant nnx11ad08 g , jfs has been supported by nasa hubble fellowship grant hst - hf-51315.01 , and mh acknowledges funding from the bundesministerium fr wirtschaft und technologie under grant number dlr 50or0701 . , k. a. 1996 , in astron . soc . of the pac . , vol . 101 , astronomical data analysis software and systems v , ed . g. h. jacoby & j. barnes , 17 , xspec is available at + http://heasarc.gsfc.nasa.gov/xanadu/xspec/ [ http://heasarc.gsfc.nasa.gov/xanadu/xspec/ ] , g. p. , bautz , m. w. , ford , p. g. , nousek , j. a. , & ricker , jr . , g. r. 2003 , in presented at the society of photo - optical instrumentation engineers ( spie ) conference , vol . 4851 , society of photo - optical instrumentation engineers ( spie ) conference series , ed . j. e. truemper & h. d. tananbaum , 2844 1 & s1 & _ chandra & rxte _ & hetg(cc ) & pca & 0.8 - 8.0 & 2.9 - 50 & 2011 - 01 - 06 14:06:4014:35:44 & 455 & 1744 & 0.52 & 0.24 & 0.32 + 1 & s2 & _ chandra & rxte _ & hetg(cc ) & pca & 0.8 - 8.0 & 2.9 - 50 & 2011 - 01 - 06 15:44:1616:09:52 & 398 & 1536 & 0.61 & 0.44 & 0.33 + 1 & s3 & _ chandra & rxte _ & hetg(cc ) & pca & 0.8 - 8.0 & 2.9 - 50 & 2011 - 01 - 06 17:15:2817:43:44 & 462 & 1696 & 0.57 & 0.33 & 0.35 + 1 & s4 & _ chandra & rxte _ & hetg(cc ) & pca & 0.8 - 8.0 & 2.9 - 50 & 2011 - 01 - 06 18:19:4419:17:52 & 997 & 3488 & 0.38 & 0.26 & 0.36 + 1 & s5 & _ chandra & rxte _ & hetg(cc ) & pca & 0.8 - 8.0 & 2.9 - 50 & 2011 - 01 - 06 19:53:3620:50:08 & 847 & 3392 & 0.38 & 0.22 & 0.37 + 2 & s6 & _ chandra & rxte _ & hetg(te ) & pca & 0.5 - 10.0 & 2.9 - 50 & 2011 - 02 - 05 07:02:0009:37:00 & 3593 & 3600 & 0.58 & 0.25 & 0.64 + 2 & s7 & _ chandra & rxte _ & hetg(te ) & pca & 0.5 - 10.0 & 2.9 - 50 & 2011 - 02 - 05 10:10:0010:31:00 & 929 & 1232 & 0.74 & 0.31 & 0.65 + 3 & s8 & _ swift & rxte _ & xrt(wt ) & pca & 0.5 - 10.0 & 2.9 - 50 & 2011 - 10 - 08 20:03:2820:26:08 & 1355 & 1344 & 0.59 & 0.32 & 0.48 + 3 & s9 & _ swift & rxte _ & xrt(wt ) & pca & 0.5 - 10.0 & 2.9 - 50 & 2011 - 10 - 08 21:40:0022:02:08 & 1326 & 1328 & 0.90 & 0.28 & 0.49 + 4 & s10 & _ swift & rxte _ & xrt(wt ) & pca & 0.5 - 10.0 & 2.9 - 50 & 2011 - 10 - 26 03:28:0004:10:00 & 1454 & 2464 & 0.47 & 0.35 & 0.57 + 1 & kerrbb2 & @xmath145 & @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 + 2 & kerrbb2 & @xmath151 & @xmath152 & @xmath153 & @xmath154 & @xmath155 & @xmath156 + 3 & const & & @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 + 4 & tbabs & @xmath70 & @xmath162 & @xmath163 & @xmath164 & @xmath165 & @xmath166 + 5 & simplr & @xmath167 & @xmath168 & @xmath169 & @xmath170 & @xmath171 & @xmath172 + 6 & simplr & @xmath173 & @xmath174 & @xmath175 & @xmath176 & @xmath177 & @xmath178 + 7 & kerrdisk & @xmath179 & @xmath180 & @xmath181 & @xmath182 & @xmath183 & @xmath184 + 8 & kerrdisk & @xmath185 & @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath190 + 9 & kerrdisk & @xmath191 & @xmath192 & @xmath193 & @xmath194 & @xmath195 & @xmath196 + 10 & kerrdisk & ew & 0.283 & 0.238 & 0.211 & 0.292 & 0.228 + 11 & ireflect & @xmath197 $ ] & @xmath198 & @xmath199 & @xmath200 & @xmath201 & @xmath202 + 12 & ireflect & @xmath203 & @xmath204 & @xmath205 & @xmath206 & @xmath207 & @xmath208 + 13 & & @xmath62 & 0.95(595/628 ) & 1.02(587/573 ) & 0.97(605/625 ) & 1.20(890/745 ) & 1.12(1119/998 ) + 14 & & @xmath42 & 1.60 & 1.62 & 1.62 & 1.61 & 1.61 + 15 & & @xmath209 & 0.022 & 0.022 & 0.022 & 0.020 & 0.019 + 16 & adopted & @xmath145 & @xmath210 & @xmath211 & @xmath212 & @xmath213 & @xmath214 1 & kerrbb2 & @xmath145 & @xmath215 & @xmath216 & @xmath217 & @xmath218 & @xmath219 + 2 & kerrbb2 & @xmath151 & @xmath220 & @xmath221 & @xmath222 & @xmath223 & @xmath224 + 3 & const & & @xmath225 & @xmath226 & @xmath227 & @xmath228 & @xmath229 + 4 & tbabs & @xmath70 & @xmath230 & @xmath231 & @xmath232 & @xmath233 & @xmath234 + 5 & simplr & @xmath167 & @xmath235 & @xmath236 & @xmath237 & @xmath238 & @xmath239 + 6 & simplr & @xmath173 & @xmath240 & @xmath241 & @xmath242 & @xmath243 & @xmath244 + 7 & kerrdisk & @xmath179 & @xmath245 & @xmath246 & @xmath247 & @xmath248 & @xmath249 + 8 & kerrdisk & @xmath185 & @xmath250 & @xmath251 & @xmath252 & @xmath253 & @xmath254 + 9 & kerrdisk & @xmath191 & @xmath255 & @xmath256 & @xmath257 & @xmath258 & @xmath259 + 10 & kerrdisk & ew & 0.190 & 0.141 & 0.176 & 0.146 & 0.187 + 11 & ireflect & @xmath197 $ ] & @xmath260 & @xmath261 & @xmath262 & @xmath263 & @xmath264 + 12 & ireflect & @xmath203 & @xmath265 & @xmath266 & @xmath267 & @xmath268 & @xmath269 + 13 & & @xmath62 & 1.40(491/352 ) & 1.61(323/201 ) & 1.37(484/353 ) & 1.54(612/399 ) & 1.24(416/337 ) + 14 & & @xmath42 & 1.60 & 1.59 & 1.59 & 1.60 & 1.59 + 15 & & @xmath209 & 0.021 & 0.021 & 0.021 & 0.024 & 0.020 + 16 & adopted & @xmath145 & @xmath270 & & & & & @xmath271 & @xmath272 & @xmath273 & @xmath274 & @xmath275 & this work ; @xcite + lmc x-1 & @xmath276 & @xmath277 & @xmath278 & @xmath279 & @xmath280 & @xcite + m33 x-7 & @xmath281 & @xmath282 & @xmath283 & @xmath284 & @xmath285 & @xcite

we have fabricated ambipolar transistors on chemically prepared hydrogen - terminated si(111 ) surfaces , in which a two - dimensional electron system ( 2des ) or a two - dimensional hole system ( 2dhs ) can be populated in the same conduction channel by changing the gate voltage of a global gate applied through a vacuum gap . depending on the gate bias , ion implanted n and p regions function either as ohmic contacts or as in - plane gates , which laterally confine the carriers induced by the global gate . on one device , electron and hole densities of up to and respectively are obtained . the peak electron mobility is/vs , and the peak hole mobility is/vs at 300 mk ; the ratio of about 20 is mainly due to the very different valley degeneracies ( 6:1 ) of electrons and holes on the si(111 ) surface . on another device , the peak electron mobility of/vs is reached at 300 mk . these devices are hexagonal in order to investigate the underlying symmetry of the 2dess , which have a sixfold valley degeneracy at zero magnetic field . three magnetoresistance measurements with threefold rotational symmetry are used to determine the symmetry of the 2dess at different magnetic field . at filling factor , the observed anisotropy can be explained by a single valley pair occupancy of composite fermions ( cfs ) . qualitatively the cfs preserve the valley anisotropy , in addition to the twofold valley degeneracy . at magnetic field up to 35 t , the 2/3 fractional quantum hall state is observed with a well developed hall plateau ; at , the three magnetoresistances show a large anisotropy ( 50:1 ) . we also show that device degradation is not a serious issue for our measurements , if the device is kept in vacuum or a nitrogen gas environment and its time in air is minimized .

1509.03849

investigations of two - dimensional systems ( 2dss ) on si surfaces began with metal - oxide - semiconductor field - effect transistors ( mosfets ) @xcite . however , the amorphous sio@xmath10/si interface severely limits the quality of the 2dss . the development @xcite of hydrogen - terminated silicon ( 111 ) [ h - si(111 ) ] vacuum field effect transistors ( fets ) is based on a simple but extraordinary fact : a wet chemical treatment of a si(111 ) surface with an ammonium fluoride ( nh@xmath11f ) solution can produce an ideal h - passivated si surface , which is both atomically flat and possesses a very low surface state density ( @xmath12 @xmath2 ) @xcite , leading to much higher carrier mobilities . in recent years , both a high quality two - dimensional electron system ( 2des ) with electron mobility of 325,000 @xmath13/vs @xcite and a high quality two - dimensional hole system ( 2dhs ) with hole mobility of 10,000 @xmath13/vs @xcite have been realized in this system . they are comparable to the best si / sige heterostructures @xcite , which are limited to the ( 100 ) oriented surfaces because of much higher threading dislocation densities on the other surface orientations @xcite . in this work we have fabricated ambipolar hexagonal devices , which can switch between a 2des and a 2dhs in the same device by changing a gate voltage . compared to other ambipolar devices @xcite , the 2des on the si(111 ) surface has a sixfold valley degeneracy , while the valley degeneracy of holes is one . the ambipolar devices provide a direct comparison between electrons and holes with very different valley degeneracies in the same conduction channel , which allows a new perspective to the fundamental research , such as the properties of 2d metals @xcite . in addition , the hexagonal devices enable the exploration of the sixfold valley - degenerate electron system . in the h - si(111 ) vacuum fet device , a global si / sio@xmath10 gate piece is used to induce a 2ds , which is further confined into a well - defined 2d region by pn junction isolation and trench isolation . similarly , future nanoelectronic devices , like quantum point contacts ( qpcs ) and quantum dots ( qds ) , may be realized on the si(111 ) surfaces using the in - plane depleting gates based on pn junctions . moreover , because of the sixfold valley degeneracy of the 2des , the valley degree of freedom can provide extra resources for practical applications in the same way as spintronic devices utilize the spin degree of freedom . this will open up new opportunities for another class of devices valleytronic devices @xcite . the remainder of this paper is organized as follows . in sec . [ sec : fab ] , we describe the device structure and operation principles , followed by detailed fabrication processes . in sec . [ sec : characterization ] , we characterize the devices to make sure that they work as intended , including surface topography measurement , a gate leakage check , pn junction isolation and leakage tests , and contact resistance measurements . in sec . [ sec : transport ] , we discuss the transport measurement results on these devices . we determined the 2d carrier mobilities in the experimentally accessible range of densities , which show that electron and hole mobilities are very different . magnetotransport measurements show rich phenomena , including shubnikov - de haas ( sdh ) oscillations , integer quantum hall effect ( iqhe ) , fractional quantum hall effects ( fqhe ) and large anisotropy in three magnetoresistance measurements at @xmath9 . in sec . [ sec : degradation ] , we discuss the device degradation with time and with exposure to ambient air , including the mobility deterioration and the increase of contact resistance . we conclude in sec . [ sec : conclusion ] with a discussion of further improvements and future directions . ( a ) schematic cross section of an ambipolar vacuum fet ( not to scale ) , with a sio@xmath10/si(100 ) remote gate piece and a h - si(111 ) piece contact bonded in a vacuum . the remote gate piece is used to induce 2d carriers on the h - si(111 ) surface by applying a gate voltage on a p@xmath0 gate layer through an encapsulated vacuum cavity . it also has a fully covered sio@xmath10 layer with a thin layer of sio@xmath10 ( @xmath14 30 nm ) left in the cavity . ( b ) six degenerated valleys of the 2des with crystallographic orientations . ( c ) ( d ) in the h - si(111 ) piece , n@xmath0 ( p@xmath0 ) ohmic contacts are used to access a 2des ( 2dhs ) at the center , with p@xmath0 ( n@xmath0 ) regions acting as lateral confinement to restrict current flow between contacts to the hexagonal center 2ds . two different configurations are investigated . ( c ) one has trench isolation , and ( d ) the other does nt have trench isolation . the dash lines in ( c ) show the cross section depicted in ( a ) . ] the vacuum fet device is fabricated from two pieces . one is a sio@xmath10/si(100 ) remote gate piece , and the other is a h - terminated high purity si(111 ) piece , as shown in fig.[fig : schematics](a ) . just like a mosfet , when a positive ( negative ) gate voltage is applied between the remote gate and a n@xmath0 ( p@xmath0 ) source ohmic contact through an encapsulated vacuum cavity , a 2des ( 2dhs ) will be induced on the h - si(111 ) surface . since the 2des has a sixfold valley degeneracy [ fig.[fig : schematics](b ) ] , the hexagonal device is designed to investigate its underlying symmetry , with six n@xmath0 ohmic contacts placed along the major axes of the constant energy ellipses on the si(111 ) surface , labeled as 2,4,@xmath15 , 12 . it also has six p@xmath0 ohmic contacts placed between the n@xmath0 contacts , labeled as 1,3,@xmath16,11 . the remote gate piece with a hexagonal cavity [ fig . [ fig : realdevice ] ( b ) ] is contact bonded on the h - si(111 ) piece in a vacuum chamber . when a positive ( negative ) gate voltage is applied , the n@xmath0 ( p@xmath0 ) ohmic contacts are used to access the 2des ( 2dhs ) at the center , and the p@xmath0 ( n@xmath0 ) contacts act as lateral confinement . figure [ fig : schematics ] ( b)-(d ) show the arrangement of the n@xmath0 and p@xmath0 ohmic contacts relative to crystallographic directions . two different types of devices are investigated . one has shallow trench isolation ( sti ) to confine the center 2ds [ the center blue hexagon in fig . [ fig : schematics](c ) ] ; the other does nt have the sti , and the center 2ds is defined by the hexagonal cavity [ fig . [ fig : schematics](d ) ] . there is typically a @xmath14 100 @xmath17 m misalignment between the remote gate piece and the h - si(111 ) piece when they are contact bonded . because the distance between the trenches and the edges of the hexagonal cavity is more than 200 @xmath17 m , and the trenches are etched on the si(111 ) piece , the misalignment does nt change the center 2ds in the device with the sti ; the misalignment can affect the center 2ds in the device without the sti when the n@xmath0 and p@xmath0 contacts are not concentric and aligned to the hexagonal cavity . however , the measurement results show that there is no significant difference between these two types of devices , because the relative placement of the contacts is fixed , and the bonding alignment is not critical in defining the device with the edges of the cavity far outside the contact perimeters . we have fabricated six devices ( 194 - 199 ) with peak electron mobilities of @xmath7 @xmath5/vs , @xmath18 @xmath5/vs , @xmath19 @xmath5/vs , @xmath20 @xmath5/vs , @xmath4 @xmath5/vs and @xmath21 @xmath5/vs at 300 mk respectively , in which four ( 194 - 197 ) of them have the sti , and the other two ( 198,199 ) do not have it . most data presented here are from two devices , sample 194 ( with the sti ) and sample 198 ( without the sti ) . ( a ) a h - si(111 ) piece with six n@xmath0 contacts ( yellow ) and six p@xmath0 contacts ( purple ) . ( b ) a si / sio@xmath10 remote gate piece which is fully covered by sio@xmath10 , except the two grey crosses where the gate contacts are made . there is a @xmath14 30 nm oxide at the bottom of the hexagonal cavity . ( c ) 3d view of a bonded device . ( d ) a bonded device placed in a customized holder and installed on an 8-pin dual in - line package ( dip ) header without wiring . ] the remote gate piece is fabricated from a si(100 ) wafer ( float zone , si : b , resistivity @xmath22 cm ) . first , a p@xmath0 gate conducting layer is formed by ion implantation with @xmath23 @xmath2 , 15 kev boron ions , as shown in fig . [ fig : schematics](a ) . then a 2 @xmath17 m deep si mesa is dry etched around the edges of each die using reactive ion etching ( rie ) . next , a @xmath14 350 nm thick sio@xmath10 layer is thermally grown on the top : dry oxidation at 1050 @xmath24c for 15 min , followed by wet oxidation at 1050 @xmath24c for 27 min and another dry oxidation at 1050 @xmath24c for 10 min . the dry - wet - dry cycle is used to improve the gate oxide quality , and also activates the implanted dopants . after that , a cavity with a @xmath14 30 nm thick oxide layer left behind is formed by dry etching and wet etching . in order to make sure the oxide thickness is reduced to @xmath14 30 nm in the cavity , a test wafer and the device wafer are loaded into a rie machine and dry etched at the same time . both have the same @xmath14 350 nm thick oxide layer , and the oxide layer is dry etched by @xmath14 270 nm . then the depth of the cavity in the test wafer is measured using a profilometer after removing photoresist . the time of the wet etching to leave the @xmath14 30 nm thick oxide layer is calculated based on the oxide thickness , the depth of the cavity and the wet etching rate . the fully sio@xmath10 encapsulated gate piece with the oxide reduced cavity drastically reduces the likelihood of gate leakage , compared to the gate piece where the doped regions were exposed . we have tested remote gate pieces with either @xmath14 350 nm or @xmath14 200 nm thick thermal oxide , and both of them work well . finally , the wafer is diced into @xmath25 mm@xmath26 remote gate pieces , as shown in fig . [ fig : realdevice ] ( b ) . the h - si(111 ) piece is also @xmath25 mm@xmath26 made from a p@xmath27 high purity si(111 ) wafer ( float zone , resistivity @xmath28 cm ) . first , a 30 nm thick sacrificial sio@xmath10 layer is thermally grown on the top of the si(111 ) substrate by dry oxidation at 950 @xmath24c for 30 min in a cmos grade furnace to reduce the channeling effect during ion implantation and protect the clean si(111 ) surface . next , the six n@xmath0 contact regions are patterned by photolithography with the primary flat of the si(111 ) wafer ( oriented to [ @xmath29 ) aligned ( @xmath30 ) along the long side of the dies . then the wafer is ion implanted with @xmath31 @xmath2 phosphorus ions at 50 kev . alignment marks are formed at the same time . similarly , the six p@xmath0 contact regions are defined by photolithography and ion implantation with @xmath32@xmath2 boron ions at an energy of 15 kev . the ion implantation parameters are based on previous works @xcite and simulation results from suprem - iv process simulator . the parameters are selected so that after thermal annealing , the peak doping concentration is located near the si(111 ) surface , and about one order of magnitude higher than that of the three - dimensional ( 3d ) metal - insulate transition ( mit ) , as shown in fig . [ fig : dopingprofile ] . ( the 3d mit occurs at doping concentration of @xmath33 @xmath34 for phosphorus @xcite and @xmath35 @xmath34 for boron @xcite . ) after each ion implantation , the photoresist is removed by acetone cleaning and piranha cleaning . first the wafer is immersed in boiling acetone for 30 min to remove most of the photoresist . then the wafer is rinsed with isopropyl alcohol ( ipa ) and de - ionized ( di ) water . next , the wafer is immersed in a piranha solution ( 4 h@xmath10so@xmath11(98@xmath36 ) : 1 h@xmath100@xmath10 ( 30@xmath36 ) ) at 100 @xmath24c for 30 min to remove the residual photoresist . right before rapid thermal annealing ( rta ) , the wafer is rca-1 cleaned for 5 min with a h@xmath10o@xmath10/nh@xmath11oh / h@xmath10o solution , then thoroughly rinsed in di water , and spin - dried . the wafer is annealed in a cmos grade rapid thermal annealer at 950 @xmath24c for 60 sec . alignment marks from ion implantations are visible before the rta process , but invisible after it . rca-1 clean etches the ion implanted sio@xmath10 regions before the rta process . they serve as alignment marks to do mesa etching , but the contrast is very low . it is helpful to define global alignment marks before the rta process using cmos compatible processes . next , if sti is desired , the 200 nm deep and 10 @xmath17 m wide shallow trenches are formed by dry etching . after that , a 2 @xmath17 m deep si mesa is dry etched around the edges of each die . this mesa prevents particles from accumulating on the surface while handling the substrate and ensures a clean edge for bonding @xcite . it also defines each die for dicing . finally , the wafer is diced into individual si(111 ) pieces , as shown in fig.[fig : realdevice](a ) . after standard cleaning ( acetone cleaning , ipa rinse , di water rinse , piranha cleaning , and di water rinse ) in clean room , one si(111 ) piece and one si / sio@xmath10 remote gate piece are transferred into an oxygen - free glove box with a typical o@xmath10 concentration of 2 ppm . the si(111 ) piece is put in a diluted solution of hf / h@xmath10o ( 1:20 ) for 2 min to remove the sacrificial sio@xmath10 layer , and then immersed in a 40@xmath36 nh@xmath11f solution for 10 min to create an atomically flat h - passivated si(111 ) surface , with di water rinse after each step . it is crucial that each of these solutions is oxygen - free to ensure the high quality of the final device @xcite . usually they are deoxygenated with a constant agitation from a magnetic stir rod for 72 hours in the glove box . finally , these two pieces are transferred into a vacuum chamber , and pushed against a sapphire boss to apply a pressure between them , sufficient to initiate bonding through van der waals forces . the bonded device is loosely placed in a customized holder machined from polyimide material , which is designed to facilitate handling of the device without applying stress . it is wired in a nitrogen glove box by hand soldering using indium at 320 @xmath24c and installed on a standard 8-pin dual in - line package ( dip ) header , which is compatible with sample holders at the national high magnetic field laboratory ( nhmfl ) . if only the 2des is to be investigated , eight contacts are sufficient , including two gate contacts and six ohmic contacts to the 2des . there are 12 wires in our cryostat . in order to measure both the 2des and the 2dhs , the 8 pin dip header is inserted into another 20 pin dip header , and four ohmic contacts to the 2dhs are directly wired to the 20 pin dip header . two p@xmath0 ohmic contacts 3 , 9 are not wired . the wired sample is transferred from the nitrogen glove box to the cryostat , and the cryostat is pumped to vacuum ( @xmath37 mbar ) in less than one hour . figure [ fig : realdevice](d ) shows one of the final devices without wiring . similar fabrication processes have been discussed elsewhere in detail @xcite . afm images of a h - si(111 ) surface with ( a ) a phosphorus - doped contact edge and ( b ) a boron - doped contact edge . they also show atomic steps in undoped areas and the crystallographic directions . bottom graphs show height profiles along white dash lines in the top graphs . ] because the si(111 ) piece and the si / sio@xmath10 remote gate piece are contact bonded through van der waals forces , successful bonding requires that the surfaces are clean , flat and smooth . atomic force microscopy ( afm ) is used to investigate the topography of the si(111 ) surface , particularly the differences in height and flatness of doped and undoped areas . figure [ fig : afm ] shows that the si(111 ) surface is clean with atomic steps . typically the doped contacts on the si(111 ) surface are at different elevations from the undoped si(111 ) surface after wet chemical cleaning and passivation . specifically , n@xmath0 ohmic contacts ( phosphorus doped ) are @xmath14 1 nm below the undoped region , as shown in fig . [ fig : afm](a ) ; p@xmath0 ohmic contacts ( boron doped ) are @xmath14 1.5 nm above the undoped region , as shown in fig . [ fig : afm](b ) . we have found that the rca-1 clean with a h@xmath10o@xmath10/nh@xmath11oh / h@xmath10o solution etches the sio@xmath10 region exposed to ion implantation much faster than the unimplanted oxide region . once it etches through the sacrificial oxide layer completely , it etches the underlying doped silicon . this results in larger height difference , which may impair the ohmic contact to the 2ds . therefore , the time in the rca-1 clean should be minimized , and under 20 minutes @xcite . piranha clean with a h@xmath10so@xmath11/h@xmath10o@xmath10 solution does not have this issue and thus is preferred when cleaning si(111 ) pieces . gate current @xmath38 vs. gate voltage @xmath39 on sample 198 when @xmath39 is swept between -50 v ( @xmath40 @xmath2 ) and 50 v ( @xmath41 @xmath2 ) at a sweep rate of 0.2 v / s with all wired n@xmath0 , p@xmath0 contacts grounded . arrows indicate the voltage sweep direction . current peaks at @xmath39 @xmath14 -12 v and -3 v are due to the formation or depletion of 2dss . @xmath42 na is from the gate capacitance . ] samples are intended for measurements at low temperature ; consequently even sub - nanoampere gate leakage can potentially affect the data . all electrical measurements made in this section ( sec . [ sec : characterization ] ) were performed at @xmath43 mk in a helium 3 refrigerator ( oxford instruments , helioxvl ) . gate leakage was checked when all ohmic contacts were grounded and the gate voltage v@xmath44 was swept between -50 v ( @xmath40 @xmath2 ) and 50v ( @xmath41 @xmath2 ) at a sweep rate of 0.2 v / sec . the measured gate current was less than 0.5 na at -50 v @xmath45 -20 v and 0 v @xmath45 50 v , as shown in fig.[fig : gateleak ] . ( the capacitance between the gate and the ground was measured to be about 3 nf , so the gate current of 0.6 na was expected . ) the static leakage current was less than 50 pa at @xmath46 v. there is no measurable leakage current for these devices . in fig . [ fig : gateleak ] , the peak current at @xmath39 @xmath14 -12 v and -3 v is due to the formation or depletion of the 2des or 2dhs . ( a ) a test device with four n@xmath0 ohmic contacts ( yellow ) and p@xmath0 isolation lines ( purple ) . a center vertical p@xmath0 isolation line separates the 2des ( shaded area ) into two halves . ( b ) leakage current @xmath47 vs. @xmath48 at @xmath49 @xmath2 with either p@xmath0 isolation lines ( s1 , s2 ) grounded or floated . the leakage due to forward biasing happens at @xmath50 v when s1 , s2 are grounded . arrows show the voltage sweep direction . ] in earlier generation devices , a silicon - on - insulator ( soi ) remote gate piece was used to define the 2ds on the h - si(111 ) surface with a shield layer to restrict the electric field of the gate @xcite . with the introduction of the si / sio@xmath10 global gate , the electric field is all over the h - si(111 ) surface , as shown in fig.1 ( c ) ( d ) . pn junction isolation and trench isolation are used to confine the 2ds . using the suprem - iv process simulator , the depth of the p@xmath0 and n@xmath0 ohmic contacts , defined as the region where doping concentration is higher than that of the 3d mit ( @xmath51 @xmath34 @xcite ) , is determined to be about 100 nm , as shown in fig . [ fig : dopingprofile](a)(b ) . for the trench isolation , 200 nm deep and 10 @xmath17 m wide trenches are used . since the depth of the 2ds is less than 10 nm for 2d carrier density higher than 10@xmath52 @xmath2 in the h - si(111 ) vacuum fet @xcite , both pn junction isolation and trench isolation should work well . pn junction leakage current on sample 198 . ( a ) p@xmath0 contact leakage current @xmath53 at @xmath54 v ( @xmath41 @xmath2 ) when a voltage @xmath55 is applied at each wired p@xmath0 contact , while all n@xmath0 contacts are grounded . ( b ) n@xmath0 contact leakage current @xmath53 at @xmath56 v ( @xmath40 @xmath2 ) when @xmath55 is applied at each @xmath57 contact , while all wired p@xmath0 contacts are grounded . in both cases , the resistance is larger than 100 g@xmath58 at @xmath59 v. ] a test device [ fig . [ fig : pnisolation](a ) ] was fabricated to verify the effectiveness of the pn junction isolation . a 2des [ shaded area in fig . [ fig : pnisolation](a ) ] is separated by a 10 @xmath17 m wide boron ion - implanted line ( the vertical line ) . at an electron density of @xmath60 @xmath2 , the resistance between n@xmath0 contacts c1 and c3 were measured , while p@xmath0 contacts s1 and s2 were either grounded or floated . in both cases , fig.[fig : pnisolation](b ) shows that the resistance between c1 and c3 was larger than 25 g@xmath58 at @xmath61 v , likely dominated by inter - wire leakage , so the pn junction isolation is effective . when s1 and s2 were grounded , the pn junction began to leak at @xmath62 v because of forward bias . for the device under investigation ( sample 198 ) , we also measured the pn junction leakage current . at an electron density of @xmath63 @xmath2 , the leakage current from p@xmath0 contacts 1 , 5 , 7 , 11 to the 2des was measured by applying a voltage @xmath55 at each contact , while all n@xmath0 contacts were grounded . ( here p@xmath0 contacts 3 , 9 were not wired , and thus floated . ) the resistance was larger than 100 g@xmath58 at @xmath59 v , as shown in fig . [ fig : pnleak](a ) . similarly , the leakage current from n@xmath0 contacts 2 , 4 , 6 , 8 , 10 , 12 to the 2dhs was measured by applying a voltage @xmath55 at each n@xmath0 contact , while all wired p@xmath0 contacts were grounded . figure [ fig : pnleak](b ) shows that the resistance was also larger than 100 g@xmath58 at @xmath64 v. in our transport measurements , lateral confining gates were floated with respect to the 2ds , and the source - drain voltage is less than @xmath65 v with the source - drain current equal or larger than 10 na , so the leakage current from the pn junction isolation has only a negligible effect on the measurements . we also measured the conductance between n@xmath0 contacts c1 and c2 while negatively biasing p@xmath0 contacts s1 and s2 . the results show that the conductance decreases with increasing negative bias voltage , which suggests that depletion is occurring and reverse biasing lateral gates to confine carriers should be possible . contacts to the 2dss on h - si(111 ) surfaces have been problematic especially at low temperature ( @xmath66 k ) and low carrier density , which limits the lowest accessible 2d carrier density in experiments . the main issues are large contact resistances ( on the order of 1 m@xmath58 ) and nonlinearity in current - voltage ( iv ) curves @xcite . for the device without sti , the contact problem can be mitigated by the presence of sio@xmath10 over the contact regions . at the center 2ds region , the dielectric is vacuum with a dielectric constant of 1 ; at the contact regions , sio@xmath10 is the dielectric with a dielectric constant of 3.9 . so at the same gate voltage , the contact regions have about four times higher carrier density than the center 2ds region when the threshold voltages are negligible , and the contact resistance is reduced . n@xmath0 contacts on sample 198 measured at 300 mk . ( a ) circuit diagram measuring contact 6 , with current @xmath47 injected from contact 12 to contact 6 and voltage @xmath67 measured between contacts 2 and 6 . ( b ) current - voltage curves of n@xmath0 contacts at @xmath68 v ( @xmath69 @xmath2 ) ) and 6v ( @xmath70 @xmath2 ) . ( c ) contact resistance @xmath71 as a function of gate voltage @xmath39 . at @xmath72 v , @xmath71 is calculated when @xmath73 na ; otherwise , @xmath71 is determined at @xmath74 na . ] the contact resistance was measured at @xmath43 mk , using a homemade voltage amplifier with input bias current @xmath14 1 pa and input impedance @xmath75 g@xmath58 . figure [ fig : ncontact](a ) shows the circuit diagram measuring contact 6 , where current @xmath47 is injected from contact 12 to contact 6 , and voltage @xmath67 is measured between contacts 2 and 6 . the iv curves were nonlinear at @xmath68 v ( @xmath69 @xmath2 ) and became linear at @xmath76 v ( @xmath70 @xmath2 ) , as shown in fig.[fig : ncontact](b ) . so the lowest accessible 2d electron density was about @xmath77 @xmath2 at 300 mk for this device . figure [ fig : ncontact](c ) shows contact resistance @xmath71 as a function of gate voltage @xmath39 . the sheet resistance of the n@xmath0 regions was measured to be 129 @xmath78 at 4.2 k , and this 3d metallic resistance should not change too much at 300 mk because of the lack of phonon scattering with the doping concentration well above the 3d mit density . the n@xmath0 contact resistance is calculated to be about @xmath79 k@xmath58 if only considering this sheet resistance , which is much smaller than the measured contact resistance at @xmath80 v. the commercial software package comsol multiphysics is used to calculate the contact resistance including both the n@xmath0 contact regions and the 2ds region . for example , at @xmath81 v , the sheet resistances of the n@xmath0 contact regions and the 2des region are 129 @xmath78 and 807 @xmath78 respectively , and the calculated contact resistance is about 2.9 k@xmath58 , which is significantly less than measured contact resistance of 4.9 k@xmath58 . the difference is likely related to the highly resistive transition region between the metallic n@xmath0 contact and the 2des , where the doping concentration is less than the 3d mit density ( @xmath82 @xmath34 @xcite ) , as shown in fig . [ fig : dopingprofile](a ) . the 2des in the transition region can undergo a 2d mit and become an insulator at low carrier density and low temperature , resulting in high contact resistance @xcite . p@xmath0 contacts on sample 198 measured at 300 mk . ( a ) current - voltage curves of p@xmath0 contacts at @xmath83 v ( @xmath84 @xmath2 ) and -25 v ( @xmath85 @xmath2 ) . ( b ) contact resistance @xmath71 as a function of gate voltage @xmath39 . at @xmath86 v , @xmath71 is calculated when @xmath73 na ; otherwise , @xmath71 is determined at @xmath74 na . ] the p@xmath0 contact resistance was also measured . figure [ fig : pcontact](a ) shows that the iv curves were nonlinear at @xmath83 v ( @xmath84 @xmath2 ) and became linear at @xmath87 v ( @xmath85 @xmath2 ) , which is much higher than the corresponding gate voltage for the n@xmath0 contacts . the lowest accessible 2d hole density was about @xmath88 @xmath2 at 300 mk . figure [ fig : pcontact](b ) shows contact resistance @xmath71 as a function of gate voltage @xmath39 . we measured the sheet resistance of the p@xmath0 regions to be 103 @xmath78 at 4.2 k , which is comparable to the n@xmath0 regions . if we compare the doping profile of the phosphorus doped region and the boron doped region ( fig . [ fig : dopingprofile ] ) , especially the top @xmath89 nm area where 2dss are populated , there is no significant difference . it is quite puzzling why the p@xmath0 contacts are much worse than the n@xmath0 contacts . 2d doping profile of ( a ) phosphorus - doped region and ( b ) boron - doped region in si using suprem - iv simulator . in ( a ) , @xmath31@xmath2 , 50 kev phosphorus ions are implanted through a 30 nm oxide ; in ( b ) , @xmath32@xmath2 , 15 kev boron ions are implanted through a 30 nm oxide . both are annealed at 950 @xmath24c for 1 min . ions are implanted at 0@xmath24 with vertical photoresist edge at length=0 . ( c ) doping concentrations of phosphorus and boron along the si surface at ( a ) and ( b ) . ] there are probably two reasons . one is that electrons have much stronger coulomb disorder screening strength than holes due to the very different valley degeneracy ( 6:1 ) @xcite ; the other is that there could be very different lateral defect profiles , such as end - of - range ( eor ) dislocation loops @xcite . the very different defect profiles can be caused by two reasons : first , phosphorus and boron ions have very different critical doses for amorphization of silicon @xcite . at room temperature , the critical dose of phosphorus ions is @xmath90 @xmath2 , about twice as high as the dose used here ; while it is @xmath91 @xmath2 for boron ions , which is about 100 times higher than the current dose . it is often easier to regrow the crystal from an amorphous layer via solid state epitaxy ( activation energy @xmath92 ev in silicon ) than it is to anneal out defects ( activation energy @xmath93 ev ) . thus , two schemes for ion implantation are usually used : either perform ion implantation above the critical dose and use low temperature annealing to regrow material or perform ion implantation below the critical dose and use high temperature annealing to get rid of defects @xcite . second , phosphorus and boron ions have different mask edge effects , such as lateral straggle and shadowing effect , resulting in different lateral doping profile @xcite . it is probably beneficial to do the ion implantation at higher dose and cryogenic temperature . we observed improvement of the p@xmath0 contacts when the dose of the boron ion implantation was increased from @xmath94@xmath2 to @xmath23@xmath2 , then the contacts were annealed at 1000 @xmath24c for 10 min @xcite . but there is a possibility that the contact bonding could be difficult , because of the larger height difference between the p@xmath0 contact region and the undoped region after wet chemical etching than that shown in fig . [ fig : afm ] ( b ) . in order to access lower 2d carrier densities , a structure like a bi - layer gate structure using soi , as we will discuss in sec . [ sec : degradation ] , can be used to solve the contact problem . ( a ) electron and hole densities as a function of gate voltage for sample 198 . the electron density is determined from the magnetic field of @xmath95 minima at the filling factor @xmath96 , and the hole density is calculated from the shubnikov - de haas oscillations . ( b ) 2d carrier mobility of sample 194 and 198 vs. carrier density at 300 mk . for the ambipolar device 198 , the ratio between the peak electron and hole mobilities is about 20 . ] longitudinal ( @xmath95 ) and hall ( @xmath97 ) resistances were measured at 300 mk using standard low - frequency ac lock - in techniques with an excitation current of 100 na at 7 hz . sheet resistance was determined by standard van der pauw method . magnetotransport measurements were performed at different gate voltages in a perpendicular magnetic field ( @xmath98 ) up to 12 t. from the sdh oscillations , the 2d hole densities are determined at each gate voltage . the 2d electron densities @xmath99 are calculated from the magnetic field @xmath100 of @xmath95 minima at the filling factor @xmath96 : @xmath101 , where @xmath102 is the electron charge , and @xmath103 is planck s constant . the 2d densities and their linear fits for sample 198 are shown in fig . [ fig : densitymobvg](a ) . for electrons , the slope is @xmath104 @xmath2/v , and the threshold voltage is @xmath105 v. for holes , the slope is @xmath106 @xmath2/v , and the threshold voltage is @xmath107 v. if a parallel plate capacitor model is used , the equivalent depth of the vacuum cavity is @xmath108 nm , which is consistent with the measurement result using a profilometer . the interface trap density in the band gap is determined to be : @xmath109(@xmath2 ) , where the band gap @xmath110 is 1.17 v for silicon at 0 k @xcite . 2d carrier mobilities are calculated from the sheet resistance , and plotted in fig . [ fig : densitymobvg](b ) . for sample 194 , the peak electron mobility was @xmath7 @xmath5/vs at 300 mk . for sample 198 , the peak electron mobility and the peak hole mobility were very different , @xmath4 @xmath5/vs and @xmath6 @xmath5/vs respectively at 300 mk , although the electrons and holes were populated in the same conduction channel . the ratio between the peak electron and hole mobilities is about 20 , which can not be explained by the different effective mass of electrons and holes . in fact , the effective mass is comparable ( @xmath111 for electrons @xcite and @xmath112 for holes @xcite , where @xmath113 is the electron mass ) at the investigated densities . this is quite different from the results of gaas ambipolar devices , in which the ratio of the peak electron mobility to the peak hole mobility is about 5 , consistent with the different effective mass of electrons and holes @xcite . the large difference in the electron and hole mobilities is a direct consequence of the very different coulomb disorder screening strength due to the different valley degeneracies of electrons ( @xmath114 ) and holes ( @xmath115 ) at the si(111 ) surface @xcite . in addition , the electron mobility increases monotonically with the electron density , and the mobility saturation is not observed up to the highest electron density measured ( @xmath116 @xmath2 ) , which is limited by the gate breakdown voltage . from the electron conductivity @xmath117 vs. electron density @xmath118 data , the fitting exponent @xmath119 in the relation of @xmath120 is about 1.3 at @xmath121 @xmath2 in these devices @xcite , which is consistent with the case ( @xmath122 ) where the electron mobility is mainly limited by 3d bulk impurities , while deviating from the case ( @xmath123 ) where 2d interface impurities limit the electron mobility . we have developed a rebonding technique , which can revive a sample by taking it apart , cleaning the two pieces and h - passivating the si(111 ) piece , then rebonding them together . the piranha cleaning and the diluted hf etching can remove the top few nm of silicon from the si(111 ) piece . we tested it on sample 195 with an initial electron mobility of @xmath18 @xmath5/vs at @xmath124 @xmath2 and @xmath43 mk . after repeated piranha cleaning and diluted hf etching three times , the electron mobility increased to @xmath125 @xmath5/vs ( @xmath124 @xmath2 , @xmath43 mk ) , which is likely due to the removal of the surface layers with higher 3d impurity density from earlier processes . all these suggest that the electron mobility is still mainly limited by 3d bulk impurities for current generation devices . magnetoresistance and hall resistance measured at 300 mk for similar ( a ) electron and ( b ) hole densities on sample 198 . the frequency of the sdh oscillations is much higher for holes than electrons owing to different valley degeneracy ( 1:6 ) . the insert in ( a ) clearly shows the 12-fold periodicity from the sixfold valley degeneracy of the 2des and the twofold spin degeneracy . the insert in ( b ) shows a beating pattern with a beating node at the arrow . ] all magnetotransport measurements were carried out in @xmath98 in this paper . figure [ fig : magnetotransport ] shows the magnetoresistance for similar electron and hole densities in the same device ( sample 198 ) at @xmath43 mk . the frequency of the sdh oscillations was much higher for holes than electrons at low field , because electrons and holes have different areas of the fermi surface resulting from the different valley degeneracies . for the 2dhs , the characteristic beating node was observed at @xmath126 t for @xmath127 @xmath2 @xcite , and the iqhe was also observed at high field ( @xmath128 t ) . for the 2des , the sdh oscillations began at @xmath129 t , and @xmath95 minima showed clearly the 12-fold periodicity [ the insert of fig.[fig : magnetotransport](a ) ] from the sixfold valley degeneracy and the twofold spin degeneracy . in contrast to previous investigations of si(111 ) transport on mosfets with peak electron mobility @xmath130 @xmath5/vs which have shown conflicting valley degeneracies of two @xcite and six @xcite , all six current generation h - si(111 ) devices with electron mobility of more than 50,000 @xmath5/vs show exclusively the sixfold valley degeneracy at low magnetic field . ( a ) top view of a device with six electron contacts and six hole contacts in relation to ( b ) the 3 pairs of the valleys , labeled as a , b and c. ( c ) three magnetoresistance measurements on sample 194 at @xmath131 @xmath2 with the insert showing the low field behavior . ( d ) ( e ) three magnetoresistance measurements on sample 198 at @xmath41 @xmath2 . for both samples 194 and 198 , one trace is higher than the other two similar traces at @xmath132 . ] because electrons have six equivalent valleys on the si(111 ) surface , the six n@xmath0 ohmic contacts are placed along the major axes of the constant energy ellipses , labeled as 2,4,@xmath15 , 12 , as shown in fig . [ fig:3magnetor ] ( a)(b ) . the six equivalent valleys form three pairs with opposite momentum , labeled as a , b and c. the degeneracy of valley pairs with @xmath133 symmetry on the si(111 ) surface can not be broken within the effective mass approximation or by a confinement potential @xcite . similarly , longitudinal magnetoresistance @xmath95 , measured with current contacts on opposite sides of the hexagonal devices , also has three different directions defined by the direction of current flow . here four - terminal resistance is defined as @xmath134 ( i , j , l , m = 2,4,@xmath15 , 12 ) , where current @xmath135 is injected from contact i to contact j , and voltage @xmath136 is measured between contacts l and m. if the current flows from 2 to 8 , it is called trace a ( @xmath137 ) . if the current flows from 10 to 4 , it is called trace b ( @xmath138 ) , and if the current flows from 12 to 6 , it is called trace c ( @xmath139 ) . these three magnetoresistances @xmath95 with threefold rotational symmetry can be used to determine the underlying symmetry of the 2des , and help identify the valley occupancies of the 2des . for example , if electrons only occupy valley pair a , trace a will show higher resistance , while trace b and c will show similar lower resistance , due to the different effective mass . figure [ fig:3magnetor](c)-(e ) show the three magnetoresistance measurements on sample 194 and 198 . data were taken at 300 mk and a density of @xmath140 @xmath2 or @xmath1 @xmath2 for samples 194 and 198 respectively . the two simplest cases are @xmath141 and @xmath100 at @xmath142 where composite fermions ( cfs ) experience a zero effective magnetic field . at @xmath141 , both devices show more or less isotropic resistances . for sample 194 , the three resistances are 37.2 @xmath58 , 30.4 @xmath58 , and 30.3 @xmath58 ; for sample 198 , they are 16.0 @xmath58 , 17.9 @xmath58 , and 17.9 @xmath58 , as shown in fig . [ fig:3magnetor](c)(d ) . at @xmath143 , the devices are anisotropic . for sample 194 , the three magnetoresistances are 558.1 @xmath58 , 226.3 @xmath58 , and 232.8 @xmath58 . for sample 198 , they are 1356.9 @xmath58 , 615.4 @xmath58 , and 664.7 @xmath58 . the observed anisotropy can be explained by cfs occupying valley pair a. because of the effective mass anisotropy , trace a shows higher resistance , and traces b and c show similar lower resistance . the ratios between the higher and lower resistances are 2.43 and 2.12 for sample 194 and 198 respectively . in a drude model with noninteracting electrons , assuming an isotropic scattering time , the resistance ratio is equal to the mass ratio @xmath144 , where @xmath145 and @xmath146 @xcite are the longitudinal effective mass and the transverse effective mass respectively . for cfs , all physical quantities are determined by the coulomb interaction @xcite . in an isotropic 2ds , the coulomb interaction is @xmath147 , where @xmath148 and @xmath149 are components of the distance between two electrons . according to ref . , in a system with an anisotropic fermi surface , it can be mapped to a system with an isotropic fermi surface with an anisotropic coulomb interaction @xmath150 , where @xmath151 , @xmath148 and @xmath149 are along the major and minor axes of the constant energy ellipse . if the transport mass of cfs along some direction is determined by the strength of the coulomb interaction along this direction , the mass - anisotropy ratio of cfs is given by @xmath152 . we have calculated two cases using comsol assuming the isotropic scattering time : in the first case , the mass - anisotropy ratio of cfs is assumed to be @xmath144 ; in the second case , it is assumed to be @xmath153 . using a hexagonal geometry similar to fig . [ fig : schematics](c ) and assuming the anisotropic resistance solely from the anisotropic effective mass , the calculated ratios between the higher and lower resistances are 3.58 and 2.07 respectively for these two cases . the measured ratios 2.43 and 2.12 fall between these two cases and are closer to the second case . our results are similar to ref . , but deviate from a theoretical study which predicts almost isotropic cf effective mass @xcite . because the scattering time anisotropy is not well known , the above discussion becomes more complicated . nevertheless , in a simplest picture , electrons occupy the lowest valley pair at filling factor @xmath132 . figure [ fig:3magnetor](c ) and ( e ) show that the corresponding cfs occupy valley pair a , and qualitatively preserve the valley anisotropy , in addition to the twofold valley degeneracy manifested by observed exclusive even numerator fractional quantum hall ( fqh ) states ( 8/5 and 4/3 ) @xcite . moreover , at filling factor @xmath154 between 4 and 6 , the devices also show anisotropic magnetoresistances . for sample 194 , trace c is highest , while trace a is highest for sample 198 . both cases can be qualitatively explained when the fermi level is located at valley pair c or valley pair a respectively . for filling factor @xmath154 between 3 and 4 , the three magnetoresistances are similar for both devices , indicating an isotropic distribution of electrons in the three valley pairs . this can not be explained by a valley occupancy of noninteracting electrons , and may related to the quantum hall nematic phase @xcite . when filling factor @xmath154 is between 2 and 3 , the three magnetoresistances are quite different for sample 194 , but similar for sample 198 . at lower b ( @xmath155 ) , the three magnetoresistances are similar especially for sample 198 , consistent with the approximation that six valleys are equally occupied . the three magnetoresistance measurements indeed provide a powerful tool to investigate the underlying symmetry of the 2des on the si(111 ) surface . in addition , a tilted magnetic field can re - distribute electrons between the three pairs of valleys through the application of an in - plane magnetic field @xcite . if these two techniques are combined , it will further our understanding of this 2des and the nature of the observed anisotropy . ( a ) three magnetoresistances and hall resistance measured on sample 194 at 25 mk with the clearly observed 2/3 fractional quantum hall state . ( b ) the three magnetoresistances are highly anisotropic at @xmath9 , with two similar traces higher than the other one . this may indicate the change of the underlying symmetry . ] sample 194 was further investigated at filling factor @xmath156 in the portable dilution refrigerator ( pdf ) at nhmfl with a magnetic field up to 35 t. the device was measured at @xmath157 mk and density of @xmath140 @xmath2 . as shown in fig.[fig:3magnetor35t](a ) , the 2/3 fqh state is clearly observed with a well developed hall plateau . moreover , fig.[fig:3magnetor35t](b ) shows that the three magnetoresistances are highly anisotropic at @xmath9 . it is less than 1 k@xmath58 along trace b , and about 50 k@xmath58 along traces a and c. this 50:1 anisotropy can not arise solely from the different effective mass . the three magnetoresistances also change from one higher resistance trace with two similar lower resistance traces at @xmath158 , to two similar higher resistance traces with one lower resistance trace at @xmath9 . this may indicate the change of the underlying symmetry of the 2des . arrhenius plots for @xmath159 , 4/3 showing activated behavior on sample 194 . ] at filling factor @xmath154 between 1 and 2 , only even numerator fqh states ( 8/5 and 4/3 ) are observed with no 5/3 fqh state , consistent with twofold valley - degenerate cfs @xcite . it is quite interesting to find out the underling symmetry of the 2/3 fqh state , i.e. whether it also has the twofold valley degeneracy . because valley degenerate 2/3 fqh state may be related to non - abelian states which are being intensively studied for possible applications in intrinsically fault tolerant quantum computation @xcite , the question becomes more important . we have determined the activation energies at @xmath159 and 4/3 from the temperature dependence of magnetoresistance @xmath95 ( fig.[fig : activationenergy ] ) . the activation energies are comparable , 1.64 k for @xmath159 and 1.34 k for @xmath160 . in a twofold valley - degenerate system , 4/3 ( @xmath161 ) state can be related to 2/3 state by using particle - hole symmetry @xcite ; their activation energies are proportional to @xmath162 , and @xmath163 . the measured ratio is about 1.2 , which may imply that the 2/3 fqh state also has the twofold valley degeneracy . the h - si(111 ) surface is encapsulated in a vacuum cavity in the h - si(111 ) vacuum fet [ fig . [ fig : schematics](a ) ] , but the seal is not perfectly hermetic . consequently h - si(111 ) vacuum fets still suffer from device degradation in air owing to degradation of h - terminated surfaces @xcite , which may limit their applications . however , if care is taken to limit the time of the device spent in air , while keeping it mostly in vacuum or a nitrogen gas environment , the device degradation is manageable . electron mobility of sample 194 vs. electron density during five cooldowns , measured at 300 mk except the 4th cooldown ( at 25 mk ) . there was one week of air exposure between the 4th and 5th cooldowns . ] ( a ) electron mobility of sample 198 as a function of electron density during four cooldowns , measured at 300 mk except the 3rd cooldown ( at 25 mk ) . one week air exposure happened between the 3rd and 4th cooldowns . ( b ) hole mobility of sample 198 vs. hole density at 1st and 4th cooldowns , measured at 4.2 k. ] the degradation of the two devices has been recorded . sample 194 was cooled down five times in a period of 246 days , as shown in fig.[fig:194mobdegrade ] . the mobility was mostly measured at 300 mk , except the fourth cooldown ( at 25 mk ) . as described in sec . [ sec : fab ] , after the device was bonded and wired in the nitrogen glove box , it was rapidly loaded into the cryostat at the first cooldown . there was about one hour exposure in air before pumping down the cryostat at each cooldown . after the device was measured for about two weeks , it was warmed up , put into a nitrogen gas filled barrier foil ziplock storage bag with oxygen absorbers , and then the bag was heat sealed . after one week , it was cooled down and measured a second time at day 26 . figure [ fig:194mobdegrade ] shows that the change of the peak electron mobility was about 2@xmath36 . the device was then warmed up and stored in a nitrogen glove box ( o@xmath10 @xmath164 ppm ) for about 6 months . right before transferring the device to the nhmfl , the device was measured a third time . the peak electron mobility increased by about 3@xmath36 , which may be due to the different temperatures or the different contact resistances . so within 211 days , the device mobility did not change much when keeping it mostly in a nitrogen gas environment . however , the mobility fluctuations were pronounced at lower electron densities at the third cooldown , which was caused by the degradation of the contacts . for transport to nhfml , the device was again placed in a nitrogen gas filled storage bag , as described above . after the nhmfl experiment , the device was exposed in air for about one week , and the peak electron mobility decreased by 35@xmath36 at the fifth cooldown . similarly , sample 198 was cooled down four times in 193 days , as shown in fig . [ fig:198mobdegrade ] . after the first measurement , the device was stored in the nitrogen glove box for about 3 months . the peak electron mobility decreased by about 17@xmath36 at the second cooldown , which degraded faster than that of sample 194 . after the same nhmfl experiment , the device was in air for about one week , and the peak electron mobility decreased by another 18@xmath36 at the fourth cooldown . the hole mobility was decreased by about 26@xmath36 between the first and fourth cooldown , measured at 4.2 k [ fig.[fig:198mobdegrade](b ) ] . after @xmath14 200 days , the peak electron mobility decreased by about 35@xmath36 for both devices . after one week exposure in air , the peak electron mobility decreased by about 35@xmath36 and 18@xmath36 for sample 194 and 198 respectively . roughly it can be estimated that the peak electron mobility decreased by @xmath14 0.2@xmath36/day in vacuum or a nitrogen gas environment , and by @xmath14 5@xmath36/day in air . n@xmath0 contact resistance @xmath71 as a function of gate voltage @xmath39 on sample 198 at 300 mk . for 4th cooldown , at @xmath165 v , @xmath71 is calculated when @xmath73 na ; otherwise , @xmath71 is determined at @xmath74 na . compared to 1st cooldown , the contact resistance increased dramatically at low electron density ( @xmath166 v ) . ] in addition to the mobility degradation , the contacts also degrade with time . the contact degradation makes accurate resistance measurements difficult , and shows up as the conductance ( mobility ) fluctuations both for electrons and holes at lower densities @xcite . in sample 198 ( without sti ) , the contact problem had been mitigated , but it resurfaced at the 4th cooldown . the contacts were linear only at above @xmath167 v ( @xmath168 @xmath2 ) rather than @xmath76 v ( @xmath70 @xmath2 ) at the first cooldown , and the contact resistances were also much worse than those at the 1st cooldown ( fig . [ fig : ncontactdegrade ] ) . correspondingly , the mobility fluctuated at @xmath169 @xmath2 for the 4th cooldown in fig . [ fig:198mobdegrade](a ) . although resistance measurements are possible with pure large contact resistances ( @xmath170 m@xmath58 ) using 4-terminal dc method , there probably exist charge trapping and emission processes in these devices like single electron transistors @xcite . the 4-terminal dc method is not effective in this case . in order to solve the contact problem , the bi - layer gate structure using soi , as shown in fig . [ fig : bilayergate ] can be adopted . similar to a split - gate geometry , the bi - layer gate structure which permits high densities around the contacts while allowing independent control of the density of the 2dss under investigation , is quite important going forward . [ ! b ] a vacuum fet with a bi - layer gate structure using soi . the 2d carrier densities around the contacts and at the investigated area can be independently controlled . the gate is also fully covered by a sio@xmath10 layer . , title="fig : " ] we have investigated a process to fabricate high mobility ambipolar h - si(111 ) vacuum fets with electron mobility of @xmath171 @xmath13/vs and hole mobility of @xmath172 @xmath13/vs . from the electron conductivity vs. electron density analysis , the electron mobility is still mainly limited by 3d bulk impurities , which implies that the mobility can be further improved with attention to high temperature steps . at a minimum , electron mobility of @xmath173 @xmath13/vs should be able to realize in the near future , since we have already fabricated a device with this mobility using previous sample fabrication techniques @xcite . combined with a bi - layer gate structure to solve the contact issue , the device quality should be comparable to the best sige quantum well devices . in addition , the 2des has a sixfold valley degeneracy , which will open up many opportunities for fundamental research and practical applications . three magnetoresistance measurements are a very useful tool to determine the underlying symmetry of the 2des . they can be used to detect possible phase transition when the associate symmetry changes @xcite . further experiments with a tilted magnetic field to re - distribute electrons between valleys can dynamically change the underlying symmetry . it will surely improve our understanding of this 2des and the nature of the observed anisotropy . the devices were also investigated at filling factor @xmath156 in a magnetic field up to 35 t. we clearly observed the 2/3 fqh state with a well developed hall plateau , and a high magnetoresistance anisotropy at @xmath9 on both sample 194 and 198 . on sample 194 , there were some hints of possible 3/5 , 4/7 , 5/9 and 6/11 fqh states ( data not shown ) . with better devices , we should be able to verify these additional states . it is quite possible to observe new phases in this unique sixfold valley - degenerate system . because of the distinct device structure , the 2d systems are resident at the surface in the vacuum cavity . atoms , molecules , superconductors and other systems can easily couple to the 2dss and form hybrid systems @xcite , and novel functional devices can be implemented in the future . this work was funded by the laboratory for physical sciences . research was performed in part at the nist center for nanoscale science and technology , and the national high magnetic field laboratory , which is supported by national science foundation cooperative agreement no . dmr-1157490 and the state of florida . we also acknowledge the support of the maryland nanocenter and its fablab . the authors are grateful to dr . neil zimmerman and dr . michael stewart for their help on the rta process . 70ifxundefined [ 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.54.437 [ * * , ( ) ] , \doibase http://dx.doi.org/10.1016/0375-9601(75)90250-9 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/0038-1098(75)90297-5 [ * * , ( ) ] link:\doibase 10.1016/0038 - 1098(74)90566 - 3 [ * * , ( ) ] link:\doibase 10.1016/0038 - 1098(74)91126 - 0 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.2001734 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.99.016801 [ * * , ( ) ] link:\doibase 10.1103/physrevb.80.161310 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4729584 [ * * , ( ) ] link:\doibase 10.1103/physrevb.89.041107 [ * * , ( ) ] @noop * * ( ) @noop * * ( ) @noop * * ( ) link:\doibase 10.1103/physrevlett.57.249 [ * * , ( ) ] http://stacks.iop.org/0268-1242/12/i=12/a=001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.93.156805 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.121307 [ * * , ( ) ] link:\doibase 10.1063/1.112392 [ * * , ( ) ] link:\doibase 10.1063/1.111244 [ * * , ( ) ] @noop * * , ( ) \doibase http://dx.doi.org/10.1063/1.4891463 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.3673837 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4793658 [ * * , ( ) ] @noop * * , ( ) \doibase http://dx.doi.org/10.1063/1.3459136 [ * * , ( ) ] link:\doibase 10.1021/nn505868h [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4898704 [ * * , ( ) ] \doibase http://dx.doi.org/10.1063/1.4919110 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.115.036801 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * ( ) @noop @noop link:\doibase 10.1103/physrevlett.45.1723 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.66.1914 [ * * , ( ) ] @noop * * ( ) _ _ , @noop ph.d . thesis , ( ) _ _ , http://drum.lib.umd.edu/handle/1903/10755[ph.d . thesis ] , ( ) link:\doibase 10.1103/physrevb.75.073301 [ * * , ( ) ] link:\doibase 10.1103/physrevb.87.075306 [ * * , ( ) ] @noop http://stacks.iop.org/0034-4885/67/i=1/a=r01 [ * * , ( ) ] , @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) link:\doibase 10.1103/physrevb.14.2480 [ * * , ( ) ] link:\doibase 10.1103/physrevb.76.241302 [ * * , ( ) ] link:\doibase 10.1103/physrevb.15.5758 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.035439 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.42.595 [ * * , ( ) ] link:\doibase 10.1103/physrev.163.816 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.57.126 [ * * , ( ) ] @noop _ _ ( , ) @noop * * , ( ) link:\doibase 10.1103/physrevb.62.1481 [ * * , ( ) ] link:\doibase 10.1103/physrevb.82.035428 [ * * , ( ) ] http://arxiv.org/abs/1411.3354 [ ( ) ] link:\doibase 10.1103/physrevb.91.205139 [ * * , ( ) ] link:\doibase 10.1103/physrevb.92.035103 [ * * , ( ) ] link:\doibase 10.1103/physrevb.91.245126 [ * * , ( ) ] @noop ( ) link:\doibase 10.1103/physrevlett.55.1606 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/s0038-1098(01)00105-3 [ * * , ( ) ] @noop * * ( ) @noop * * ( ) link:\doibase 10.1021/jp0222752 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.2164 [ * * , ( ) ]

we investigate the visibility of the current and shot - noise correlations of electrical analogs of the optical mach - zehnder interferometer and the hanbury brown twiss interferometer . the electrical analogs are discussed in conductors subject to high magnetic fields where electron motion is along edge states . the transport quantities are modulated with the help of an aharonov - bohm flux . we discuss the conductance ( current ) visibility and shot noise visibility as a function of temperature and applied voltage . dephasing is introduced with the help of fictitious voltage probes . comparison of these two interferometers is of interest since the mach - zehnder interferometer is an amplitude ( single - particle ) interferometer whereas the hanbury brown twiss interferometer is an intensity ( two - particle ) interferometer . a direct comparison is only possible for the shot noise of the two interferometers . we find that the visibility of shot noise correlations of the hanbury brown twiss interferometer as function of temperature , voltage or dephasing , is qualitatively similar to the visibility of the first harmonic of the shot noise correlation of the mach - zehnder interferometer . in contrast , the second harmonic of the shot noise visibility of the mach - zehnder interferometer decreases much more rapidly with increasing temperature , voltage or dephasing rate .

cond-mat0505511

with the advent of mesoscopic physics , it has become possible to experimentally investigate quantum phase coherent properties of electrons in solid state conductors in a controlled way . in particular , in ballistic mesoscopic samples at low temperatures , electrons can propagate up to several microns without loosing phase information . this opens up the possibility to investigate electrical analogs of various optical phenomena and experiments . an investigation of such analogs is of fundamental interest . on the one hand , it allows one to establish similarities between the properties of photons and conduction electrons , a consequence of the wave nature of the quantum particles . on the other hand , it also allows one to investigate the differences between the two types of particles arising from the different quantum statistical properties of fermions and bosons . for many - particle properties , such as light intensity correlations or correspondingly electrical current correlations , noise , the quantum statistical properties are important . @xcite both the wave - nature of the particles as well as their quantum statistics are displayed in a clearcut fashion in interferometer structures . in this work we are concerned with the electrical analogs of two well known optical interferometers , the single - particle mach - zehnder ( mz ) interferometer and the two - particle hanbury brown twiss ( hbt ) interferometer . the mz - interferometer is a subject of most textbooks in optics . @xcite in the framework of quantum optics , considering individual photons rather than classical beams of light , the interference arises due to the superposition of the amplitudes for two different possible paths of a single photon . this leads to an interference term in the light intensity . the mz - interferometer is thus a prime example of a single particle interferometer.@xcite various electronic interferometers with ballistic transport of the electrons have been investigated experimentally over the last decades , as e.g. aharonov - bohm ( ab ) rings@xcite and double - slit interferometers.@xcite detailed investigations of dephasing in ballistic interferometers was carried out in refs . [ ] . only very recently was the first electronic mz - interferometer realized by ji et al.@xcite in a mesoscopic conductor in the quantum hall regime . a high visibility of the conductance oscillations was observed , however the visibility was not perfect . this led the authors to investigate in detail various sources for dephasing . as a part of this investigation , also shot noise was measured . still , some aspects of the experiment are not yet fully understood . theoretically , seelig and one of the authors@xcite investigated the effect of dephasing due to nyquist noise on the conductance in a mz - interferometer . the effect of dephasing on the closely related four - terminal resistance in ballistic interferometers @xcite was investigated as well . dephasing in ballistic strongly interacting systems is discussed by le hur . @xcite following the experimental work of ji et al.,@xcite marquardt and bruder investigated the effect of dephasing on the shot - noise in mz - interferometers , considering dephasing models based on both classical@xcite as well as quantum fluctuating fields . @xcite very recently , frster , pilgram and one of the authors @xcite extended the dephasing model of refs . [ ] to the full statistical distribution of the transmitted charge . the hbt - interferometer @xcite was originally invented for stellar astronomy , to measure the angular diameter of stars . it is an intensity , or two - particle,@xcite interferometer . the interference arises from the superposition of the amplitudes for two different two - particle processes . importantly , there is no single particle interference in the hbt - interferometer . consequently , in contrast to the mz - interferometer there is no interference in the light intensity , the interference instead appears in the intensity - intensity correlations . moreover , the intensity - intensity correlation also display the effect of quantum statistics . photons originating from thermal sources tend to bunch , giving rise to positive intensity cross correlations . for the electronic analog of the hbt - interferometer , it was the corresponding anti - bunching of electrons that originally attracted interest . it was predicted @xcite that the electrical current cross correlations in mesoscopic conductors would be manifestly negative , i.e. display anti - bunching , as a consequence of the fermionic statistics of the electrons . negative current cross correlations were subsequently observed in two independent experiments.@xcite recently , anti - bunching for field emitted electrons in vacuum was also demonstrated.@xcite the two - particle interference in the hbt - experiment has received much less attention . we emphasize that while the bunching of the photons was necessary for obtaining a finite , positive cross correlation signal , it was the two - particle effect that was of main importance to hbt since the angular diameter of the star was determined from the two - particle interference pattern . in electrical conductors , two - particle effects in ab - interferometers were investigated theoretically in refs . only very recently two of the authors and sukhorukov@xcite proposed a direct electronic analog of the optical hbt - interferometer which permits to demonstrate two - particle interference in an unambiguous way . in this work we investigate and compare in detail the current and and zero - frequency noise in electronic mz and hbt interferometers . we consider interferometers implemented in mesoscopic conductors in the integer quantum hall regime , where the transport takes place along single edge states and quantum point contacts ( qpc s ) serve as controllable beam splitters . the effect of finite temperature , applied bias and asymmetry , i.e. unequal length of the interferometer arms , is investigated . the strength of the interference contribution is quantified via the visibility of the phase oscillations . the dependence of the visibility on the beam splitter transparencies as well as on the temperature , voltage and asymmetry is discussed in detail . of interest is the comparison of visibility of the shot - noise correlation of the mz - interferometer and the hbt - intensity interferometer . shot noise correlations in the mz - interferometer exhibit two contributions , one with the fundamental period of @xmath0 and a second harmonic with period @xmath1 . the shot noise correlations in the hbt - interferometer , even though they are due to two particle processes , are periodic with period @xmath0 . thus the aharonov - bohm period can not be used to identify the two particle processes which give rise to the hbt effect . it is therefore interesting to ask whether the hbt two - particle processes have any other signature , for instance in the temperature or voltage dependence of the visibility of the shot - noise correlation . we find that this is not the case . to the contrary , we find that the the shot noise correlations in the hbt intensity interferometer behave qualitatively similar to the @xmath0 shot noise correlation in the mz - interferometer . in contrast the @xmath1 contribution in the shot noise of the mz - interferometer decreases more rapidly with increasing temperature , voltage or dephasing rate than the @xmath0 oscillation in the mz- or hbt - interferometer . we investigate dephasing of the electrons propagating along the edge states by connecting one of the interferometer arms to a fictitious , dephasing voltage probe . in all cases , the current and noise of the mz - interferometer as well as the noise in the hbt - interferometer , the effect of the voltage probe is equivalent to the effect of a slowly fluctuating phase . in the paper we consider implementations of the mz and hbt interferometers in mesoscopic conductors in strong magnetic fields , in the integer quantum hall regime.@xcite the typical system is a two - dimensional electron gas in a semiconductor heterostructure , with the lateral confinement of the electron gas controllable via electrostatic gating . the transport between reservoirs@xcite connected to the conductor takes place along edge states.@xcite the edge states , quantum analogs of classical skipping orbits , are chiral , the transport along an edge state is unidirectional . scattering between edge states is suppressed everywhere in the conductor except at electrostatically controllable constrictions , qpcs.@xcite for a magnetic field that does not break the spin degeneracy of the edge states , each edge state supplies two conduction modes , one per spin . these properties make conductors in the integer quantum hall regime ideal for realizing analogs of optical experiments . first , the edge states correspond to single mode waveguides for the light . the unidirectional motion along the edge states allows for `` beams '' of electrons to be realized . second , the qpc s work as electronic beam splitters with controllable transparency . moreover , due to chirality the beamsplitters are reflectionless , a property essential for the mz and hbt interferometers but difficult to achieve for beam splitters in conductors in weak ( or zero ) magnetic fields.@xcite these properties of conductors in the quantum hall regime have been demonstrated experimentally in a number of works , see e.g. [ ] . theoretically , several works have been concerned with the conductance and noise properties of beam splitters and interferometers in quantum hall systems , for a recent reviews see e.g. refs . [ ] recently , it was proposed to use these appealing properties of edge states in the context of orbital @xcite quasi - particle entanglement in static @xcite and dynamic @xcite systems as well as for quantum state transfer.@xcite it is interesting to note that the edge state description also hold for conductors at even higher magnetic fields , in the fractional quantum hall regime . as examples , the fractional charge has been determined in shot - noise experiments@xcite and the quantum statistical properties of the fractionally charged quasi - particles have been investigated theoretically in beam - splitter@xcite and hbt@xcite geometries . various interferometer structures have also been considered.@xcite very recently , a mz - interferometer in the fractional quantum hall regime was proposed.@xcite in this work we however consider only the integer quantum hall effect , where the quasi - particles are noninteracting and the electrical analogs to optical experiments can be directly realized . this discussion leads us to consider single mode , multi - terminal conductors with noninteracting electrons . the principle aim of this work is a comparison of the mz and hbt - interferometers . in reality in both interferometers interactions ( screening ) play a role both for the voltage and temperature dependence . a non - interacting scattering approach is not gauge invariant but requires a treatment of screening . @xcite however , these effects are expected to be simliar in the two interferometers and will not affect the main conclusions of this work . therefore , below we treat non - interacting qausi - particle interferometers . the conductors are connected to several electronic reservoirs , biased at a voltage @xmath2 or grounded . the current@xcite and the noise@xcite are calculated within the scattering approach for multi - terminal conductors . we first introduce the creation and annihilation operators for ingoing , @xmath3 and @xmath4 , and outgoing , @xmath5 and @xmath6 , particles , at energy @xmath7 in terminal @xmath8 . for simplicity we suppress spin notation . considering a conductor with @xmath9 terminals , the in- and out - going annihilationoperators are related via the @xmath10 scattering matrix , as @xmath11 where @xmath12 is the amplitude to scatter from terminal @xmath13 to terminal @xmath8 . the current operator in the lead @xmath8 has the form@xcite @xmath14t/\hbar ) \nonumber \\ & \times & a_{\beta \gamma } ^{\alpha } ( e , e^{\prime } ) \hat{a}_{\beta } ^{\dagger}(e)\hat{a}_{\gamma } ( e^{\prime}),\end{aligned}\ ] ] with the notation @xmath15 the average current is given by@xcite @xmath16 where the spectral current density is @xmath17 here @xmath18)$ ] is the fermi dirac distribution of terminal @xmath19 , with @xmath20 the corresponding applied voltage . the spectral conductance @xmath21 is given by @xmath22 the zero frequency correlator between current fluctuations in terminals @xmath23 and @xmath19 is defined as @xmath24 where @xmath25 the current correlator is given by@xcite @xmath26 where @xmath27 \label{noisedens}\end{aligned}\ ] ] is the spectral current correlator . there are several physical mechanisms that might lead to dephasing of the electrons propagating along the edge states ( see e.g. the discussion in ref . [ ] ) . in this work we are not interested in any particular mechanism for dephasing but consider instead a phenomenological model , a dephasing voltage probe . the idea of using a voltage probe to induce dephasing was introduced in refs . a voltage probe connected to a mesoscopic sample was considered , leading to a suppression of coherent transport due to inelastic scattering . the probe model , originally considered for the average current , was extended to treat the effect of inelastic scattering on shot noise by bttiker and beenakker@xcite by considering a conservation of current fluctuations at the probe as well . later de jong and beenakker@xcite extended the voltage probe concept and introduced a ( fictitious ) voltage probe which breaks phase but does not dissipate energy . scattering in the voltage probe is ( quasi-)elastic . this is achieved with the help of a distribution function in the voltage probe which conserves not only total current like a real voltage probe , but conserves current in every small energy interval . such a probe provides a model of pure dephasing . the different probe models have been used as qualitative models in a number of works , see refs . [ ] for a review . for an application to quantum hall systems , [ ] . in this work we consider the dephasing voltage probe model , which conserves the current at each energy . the model is based on the assumption that the current is conserved on a time scale @xmath28 , much shorter than the time of the measurement but much longer than the time between injection of individual electrons , here of the order of @xmath29 . one could however consider a more general voltage probe model that takes into account a more complicated dynamics of the probe . a detailed discussion of such a general model in the light of recent work@xcite is however deferred to a later work . here we only note that below we find that the voltage probe in both the mz and hbt - interferometers only gives rise to a suppression of the phase dependent terms in conductance and noise , just as one would naively expect to be the effect of pure dephasing . the condition of zero current into the fictitious probe @xmath30 at each energy is fulfilled by considering a time dependent distribution function of the probe @xmath31 where @xmath32 fluctuates to conserve current on the timescale @xmath28 . as a consequence , the spectral current density at each energy in lead @xmath33 fluctuates in time as @xmath34 where the fluctuations @xmath35 consist of two parts , the intrinsic fluctuations @xmath36 and the additional fluctuations due to @xmath37 . the requirement of zero average current into the probe , @xmath38 , leads to the averaged distribution function at the probe reservoir @xmath30 @xmath39 the average spectral current density @xmath40 is then found from eq . ( [ currspecdens ] ) . the fluctuating part of the distribution function , @xmath41 is obtained from the requirement of zero current fluctuations into the probe , @xmath42 the total current density fluctuation is then given by @xmath43 as a result , in the presence of dephasing the total spectral current correlation @xmath44 is @xmath45 where @xmath46 is the correlation function between the intrinsic current fluctuations , @xmath47 and @xmath48 , of contact @xmath8 and @xmath19 , given by eq . ( [ noisedens ] ) , and @xmath49 is the conductance , given by eq . ( [ speccond ] ) . a schematic of the mz - interferometer is shown in fig . [ mzopt ] . an incident beam of light from source @xmath50 is divided in two parts at the semitransparent beam splitter @xmath51 . the two partial beams are reflected at mirrors and later joined at the second beam splitter b. beams of light going out from @xmath52 are detected in @xmath53 and @xmath54 . the amplitude of the light in an outgoing beam is the sum of the amplitudes for the two partial beams , @xmath55 . this gives an intensity @xmath56)\right\}$ ] . the interference term @xmath57)\right\}$ ] thus contains the difference between the geometrical phases , @xmath58 . importantly , the four terminal geometry together with the reflectionless beam splitters lead to that the incident beam traverses the interferometer only once . this is a defining property of the mz - interferometer . we then turn to the electric analog of the mz - interferometer , shown in fig . as pointed out above , several results for the current and noise are available in the literature.@xcite here we analyze the most general situation possible , with finite voltage , temperature and interferometer arm asymmetry as well as different beam splitters @xmath51 and @xmath52 with arbitrary transparency . when we consider limiting cases for e.g. small temperature , bias or asymmetry , known results are recovered . this detailed analysis of the mz - interferometer is of importance when comparing to the hbt - interferometer below . we first discuss a fully coherent interferometer , the effect of dephasing is investigated below . an electric potential @xmath2 is applied at terminal 1 , all other terminals are kept at zero potential . the injected electrons propagate along single edge states . scattering between the edge states can take place only at the two qpc s , acting as beam splitters with controllable transparency . the beam splitters @xmath59 are characterized by the scattering matrices @xmath60 where @xmath61 and @xmath62 are the transmission and reflection probabilities , respectively . we note that any additional phases of the beam splitters just give rise to a constant phase shift of the oscillations in the interference terms and are therefore not considered . propagating along the edge states , the electrons pick up geometrical phases @xmath63 and @xmath64 as well as phases @xmath65 and @xmath66 due to the ab - flux @xmath67 through the center of the interferometer . for example , the amplitude for scattering from terminal 1 to 4 is given by @xmath68 for the geometrical phases , to be specific we consider the case when the potential landscape @xmath69 of the conductor in fig . [ mzel ] is varying smoothly on the scale of the magnetic length @xmath70 , with @xmath71 the applied magnetic field perpendicular to the plane in fig . [ mzel ] ( the effect of selfconsistency of the potential @xcite is neglected ) . this allows for a semiclassical treatment . @xcite in a high magnetic field the edge states at fermi energy @xmath72 follow equipotential lines determined by @xmath73 where @xmath74 is the cyclotron frequency and @xmath75 the effective electron mass . we are concerned here with the case where there is only one edge state and thus @xmath76 . suppose the @xmath77-axis is a line intersecting quantum point contacts @xmath51 and @xmath52 in fig . [ mzel ] . excluding self - intersections we can express the edge state in terms of functions @xmath78 and @xmath79 for the left and right path of the interferometer . working in the symmetric gauge , the geometric phases can be written @xcite @xmath80 , where @xmath81 and @xmath82 are the locations of the qpc s . importantly , @xmath58 corresponds to the total area @xmath51 enclosed by these two paths divided by the magnetic length squared , or @xmath83 where @xmath84 is the total flux through the enclosed area and @xmath85 the elementary flux quantum . note that the aharonov - bohm flux @xmath67 adds an additional phase @xmath86 and @xmath87 , with @xmath88 , to each of the paths . for the discussion of the temperature and voltage dependence of the current and the noise , we also need to know the energy dependence of the phases . first , instead of parameterizing the edge state through @xmath77 we introduce the parameter @xmath89 which measures directly the path length , i.e. @xmath90 , @xmath91 . in addition at @xmath89 we introduce local coordinates @xmath92 along and @xmath93 perpendicular to the equipotential line . in these coordinates , an edge state that follows the equipotential line at a small energy @xmath7 away from @xmath72 acquires the additional phase @xmath94 with @xmath95 . the potential gradient @xmath96 determines the local electric field @xmath97 at @xmath89 . but @xmath98 where @xmath99 is the drift velocity of the guiding center of the cyclotron orbit at point @xmath89 of the edge state . thus a small increase in energy leads to a phase increment given by @xmath100e$ ] . a rough estimate using a drift velocity which is constant along the edge gives @xmath101 with @xmath102 the length of the edge state @xmath103 . for the phase - difference of the two interfering paths we have @xmath104 with @xmath105 the equilibrium phase difference . formally , higher order terms in energy can be neglected for characteristic energies @xmath106 and @xmath2 much smaller than @xmath107 $ ] . the asymmetry of the two edges thus gives rise to an energy scale @xmath108 - \int ds [ 1/\hbar v_d(s_2)]\}^{-1}$ ] which is due to the mismatch of the edge state path lengths , i.e. @xmath109 with @xmath110 . in principle , for a completely symmetric interferometer one has @xmath111 . given the scattering amplitudes @xmath112 , the spectral current density is found from eqs . ( [ adef ] ) , ( [ currspecdens ] ) and ( [ speccond ] ) . for e.g. terminal 4 , one gets @xmath113 \left [ \mathcal{t}_{a}r_{b}+\mathcal{t}_{b}r_{a}\right . \notag \\ & & \left.+2\sqrt{\mathcal{t}_{a}\mathcal{t}_{b}r_{a}r_{b}}\cos \left ( e / e_{c}+\theta \right ) \right ] , \end{aligned}\ ] ] where we introduce the total , energy independent phase @xmath114 . here @xmath115 is the distribution functions of the grounded terminals 2,3 and 4 and @xmath116 the distribution function of terminal 1 . the current is then given from eq . ( [ avcurr ] ) , as @xmath117 . \notag \\ & & \label{currmz}\end{aligned}\ ] ] current conservation gives @xmath118 . the current consists of two physically distinct parts . the first term in eq . ( [ currmz ] ) is the phase independent , incoherent part , the current in the absence of interference , while the second , phase dependent term is the interference contribution . we note that a bias @xmath2 of the order of the asymmetry energy @xmath119 leads to the phase shifts of the oscillation . the strength of the interference can conveniently be quantified via the visibility as @xmath120}{\left\langle i\right\rangle},\ ] ] which gives for the current in the mz - interferometer @xmath121 the visibility is a product of a term containing the qpc scattering probabilities and a function depending on the energy scales @xmath122 @xmath2 and @xmath119 . the scattering probability term is maximum for identical qpc s , @xmath123 the energy scale dependence is shown in fig . [ figcurrvismz ] where the visibility for identical point contacts is plotted as a function of the normalized temperature , @xmath124 . we note several interesting features from fig . [ figcurrvismz ] and eq . ( [ currvismz ] ) . ( i ) the visibility shows decaying oscillations as a function of voltage @xmath125 for arbitrary temperature . ( ii ) a symmetric mz - interferometer , @xmath126 @xmath2 , has unity visibility ( for @xmath127 ) , i.e. shows perfect interference . ( iii ) the visibility decays monotonically with increasing temperature . for large temperatures , @xmath128 , the visibility decays exponentially with the temperature as @xmath129 it is interesting to compare the calculated visibility to the experimentally measured one in ref . as already shown in ref [ ] , the measured scattering probability dependence of @xmath130 is well reproduced by eq . ( [ currvismz ] ) . for the energy scale dependence , no information about the drift velocity @xmath131 or the asymmetry @xmath132 needed to determine @xmath133 is provided in ref . however , to obtain the order of magnitude of @xmath133 , considering as a rough estimate a typical drift velocity @xcite @xmath134 m / s at a magnetic field @xmath135 and an asymmetry @xmath136 m gives an @xmath133 corresponding to an applied bias @xmath137 or a temperature @xmath138 . these values are typically of the same order of magnitude as the ones considered in the experiment . as a first approximation , one would thus expect asymmetry effects to be of importance . the observed temperature dependence , a strong decrease of the visibility for increased temperature , is also qualitatively described by eq . ( [ currvismz ] ) with an @xmath139 mk . this is however not the case with the voltage dependence . ji et al find a differential visibility , i.e. the visibility of @xmath140 , which decays strongly with applied voltage , while eq . ( [ currmz ] ) predicts a constant , voltage independent differential visibility . there are several possible explanations to why the voltage dependence in contrast to the temperature dependence is not reproduced by the theory . ji et al themselves point out two voltage dependent dephasing mechanism : low frequency noise of @xmath141 type due to moving impurities , induced by a higher current and fast fluctuations of the potential landscape ( and hence of the phase via the enclosed area ) caused by screening of the additional charges injected at higher current . screening might also , for the nonlinear current - voltage characteristics predicted by eq . ( [ currmz ] ) , lead to a voltage dependent renormalization of the transmission probabilities , introducing a voltage dependence in the differential visibility . @xcite we also note that in the model of ref . [ ] , inducing dephasing by coupling the mz - interferometer to a quantum bath , gives a dephasing rate that increases with increasing voltage . clearly , further investigations are needed to clarify the origin of the dephasing in the experiment in ref . [ ] . turning to the noise , we focus on the cross correlator between currents flowing in terminals 3 and 4 ( the auto - correlator can be obtained analogously ) . this allows for a straightforward comparison to the result of the hbt - interferometer , for which the cross correlator was investigated in ref . [ ] . from eqs . ( [ noise ] ) and ( [ noisedens ] ) and the expressions for the scattering amplitudes , we arrive at the noise spectral density @xmath142 ^{2 } \nonumber \\ & \times&\left\ { c_{0}+c_{\theta } \cos \left ( \frac{e}{e_{c}}+\theta \right)+c_{2\theta } \cos \left ( 2\left[\frac{e}{e_{c}}+\theta \right ] \right ) \right\ } , \nonumber \\ & & \label{noisespecdens}\end{aligned}\ ] ] with coefficients @xmath143 performing the energy integrals in eq . ( [ noise ] ) we find for the cross correlator @xmath144 \right\ } \label{mznoisetot}\end{aligned}\ ] ] where we introduce the functions @xmath145 @xmath146 . \label{enfcn3}\end{aligned}\ ] ] and @xmath147 . \label{enfcn2}\end{aligned}\ ] ] containing the dependence on the energy scales @xmath148 @xmath149 and @xmath119 . just as the current in eq . ( [ currmz ] ) , the noise consists of a phase independent , incoherent part and a phase dependent , interference part . however , in contrast to the current , the phase dependent part of the noise contains two terms with different periods in @xmath150 , corresponding to oscillations periodic in @xmath0 and @xmath1 . these terms result from two - particle scattering processes which enclose the ab - flux one and two times respectively . similarly to the current , the phase of the oscillations are shifted for a bias @xmath2 of the order of the asymmetry energy @xmath119 . it is important to note that in the mz ( in contrast to the hbt ) interferometer , two particle and higher order scattering processes are just products of single particle scattering processes . the full distribution of current fluctuations@xcite is thus a function of single particle scattering probabilities only . in particular , the noise spectral density @xmath151 in eq . ( [ noisespecdens ] ) is proportional to @xmath152 , i.e. partition noise@xcite with phase dependent scattering probabilities . as a consequence , the phase independent , incoherent part of the noise can not be understood as partition noise from incoherent single particle processes , i.e. @xmath153/2 $ ] , from two coherent scattering processes , obviously contribute to the phase independent part of the noise . as a consequence , as shown by marquardt and bruder,@xcite a model @xcite with a filled stream of classical particles injected from reservoir @xmath50 correctly reproduces the incoherent part of the current but fails to reproduce the incoherent part of the noise . in contrast , as found in ref . [ ] and further discussed below , the completely dephasing voltage probe model correctly reproduces the incoherent part of both the current and the noise . to quantify the strength of the oscillations we introduce two separate quantities , @xmath154 and @xmath155 , here simply called visibilities , which in close analogy to the current visibility in eq . ( [ currvismz ] ) are defined as the ratio of the amplitudes of the noise oscillations and the average noise . they become @xmath156 and @xmath157 similarly to the current , both visibilities are products of a term containing the scattering probabilities and a function of the energy scales @xmath2 , @xmath149 and @xmath119 . we first focus on the scattering probability dependent term by considering the visibility in the limit of a symmetric interferometer , @xmath158 , @xmath149 , where the energy - scale dependent terms are unity . this gives @xmath159 and @xmath160 the two visibilities are plotted in fig . both visibilities are symmetric under the substitutions @xmath161 and @xmath162 . the visibility @xmath163 is zero for @xmath164 , i.e. for a symmetric setting of any of the qpc s . the visibility increases for increasing qpc asymmetry , reaching a maximum for @xmath165 and @xmath166 ( unity only in the limit @xmath167 ) and then decreases again toward zero at @xmath168 or @xmath169 . interestingly , the visibility @xmath170 shows an opposite behavior . it is maximum , equal to unity , at @xmath171 and then decreases monotonically for increasing qpc asymmetry , reaching zero at @xmath168 or @xmath169 . this different dependence on the scattering probabilities makes it possible to investigate the two oscillations independently by modulating the qpc transparencies . turning to the energy scale behavior , we consider for simplicity @xmath163 in the limit @xmath172 and @xmath173 in the limit @xmath171 where respective scattering probability terms are unity . for a symmetric interferometer , i. e. @xmath174 , both visibilities are unity . considering the situation when the temperature is comparable to the asymmetry energy scale @xmath119 but the voltage is small @xmath175 , we get the visibilities @xmath176\ ] ] and @xmath177.\ ] ] the temperature dependence of the visibilities are shown in fig . [ noisevis2 ] . both visibilities decrease monotonically with increasing temperature . for large temperature @xmath128 , the visibilities decay exponentially as @xmath178 and @xmath179 . the visibility @xmath180 is thus considerably more sensitivity to thermal smearing than @xmath181 . in the opposite limit , for a small temperature but a voltage comparable to @xmath119 , i.e. @xmath182 , we instead get the visibilities @xmath183 and @xmath184 the visibilities as a function of voltage are plotted in fig . [ noisevis2 ] . both visibilities show an oscillating behavior , decaying as a power law @xmath185 with increasing voltage . the period of oscillations , in @xmath2 , is @xmath186 for @xmath187 but @xmath188 for @xmath180 , half the value for @xmath181 . the different voltage dependence gives an additional possibility to investigate the two visibilities independently . in the experiment of ji et al.@xcite the noise was measured in the high voltage regime , with the interference terms in both the current and noise completely suppressed . the dependence of the incoherent noise on the transparencies @xmath189 and @xmath190 was investigated ( @xmath189 was kept at @xmath191 and @xmath190 was varied ) . a good agreement was found with the first , incoherent term in eq . ( [ mznoisetot ] ) . taken the open questions on the effect of decoherence on the average current , a detailed experimental investigation on the phase dependent , interference part of the noise would be of great interest . next we consider the effect of dephasing on the current and noise . as discussed above , dephasing is introduced by connecting one of the two arms of the interferometer to a fictitious , dephasing voltage probe . the interferometer with the probe , denoted terminal @xmath192 , is shown in fig . [ mzdeph ] . the dephasing probe is connected to the edge via a contact described by a scattering matrix @xmath193 where the dephasing parameter @xmath194 varies between @xmath195 ( no dephasing , fully coherent transport ) and @xmath50 ( complete dephasing , fully incoherent transport ) . the presence of the dephasing probe modifies the amplitudes for scattering between the terminal 1 , 2 , 3and 4 . as an example , the scattering amplitude @xmath196 , given in eq . ( [ s41 ] ) in the absence of dephasing , now becomes @xmath197 in addition , amplitudes for scattering into and out from the probe terminal 5 have to be considered . the current is obtained from eqs . ( [ avcurr ] ) , ( [ currspecdens ] ) and ( [ fav ] ) . for the current in terminal 4 , we find @xmath198.\end{aligned}\ ] ] comparison with the result in the absence of dephasing in eq . ( [ currmz ] ) shows that the effect of the dephasing is to suppress the phase - dependent oscillations by multiplying the phase - dependent interference term with a factor @xmath199 . for complete dephasing @xmath200 , the phase dependent term is completely suppressed . the effect of dephasing can thus be simply incorporated in the visibility as @xmath201 where @xmath202 is the visibility of the current oscillations in the absence of dephasing , given by eq . ( [ currvismz ] ) . as is clear from the discussion above , to account for the experimental observations in ref . [ ] , one would have to consider a voltage dependent dephasing parameter @xmath194 . turning to the noise , we obtain the cross correlator between currents in lead 3 and 4 in the presence of dephasing from eqs . ( [ noise ] ) and ( [ dephnoisedens ] ) , giving @xmath203 \right\}.\end{aligned}\ ] ] here the terms @xmath204 @xmath205 @xmath206 and @xmath207 are defined above in eqs . ( [ ccoeff ] ) and ( [ enfcn1 ] ) to ( [ enfcn2 ] ) . similarly to the current , the effect of the dephasing is only to suppress the amplitude of the phase - dependent oscillations . that is what one would naively expect to be the consequence of pure dephasing . the two phase - dependent terms are however affected differently , the @xmath208 term is suppressed by a factor @xmath199 while the @xmath209 term is suppressed by @xmath210 . the @xmath209 oscillations are thus more strongly suppressed . the visibilities of the two oscillations in the presence of dephasing can simply be written @xmath211 and @xmath212 where @xmath181 and @xmath173 are the visibilities for the noise oscillations in the absence of dephasing , given by eqs . ( [ noisevismz1 ] ) and ( [ noisevismz2 ] ) , respectively . importantly , both oscillating terms are fully suppressed for complete dephasing , @xmath200 . complete dephasing within the voltage probe model thus gives a noise expression that only consists of the phase independent , incoherent term in eq . ( [ noisespecdens ] ) . we note already here that the same result is found below for the hbt - interferometer . since quantum interference by definition is excluded from the model , i.e. all scattering phases are neglected , the completely dephasing voltage probe thus constitutes a classical model that correctly reproduces the incoherent part of the noise . as pointed out above , a more detailed discussion of the physics of the voltage probe and a comparison with refs . [ ] is deferred to a later work . it is interesting to note that the effect of dephasing introduced with the voltage probe , both for the current and noise , is for arbitrary dephasing strength identical to a phase averaging . the result in eqs . ( [ dephvismz1 ] ) and ( [ dephvismz2 ] ) can be obtained by averaging the fully coherent expressions in eqs . ( [ noisevismz1 ] ) and ( [ noisevismz2 ] ) with respect to a lorentzian distribution @xmath213 of slow fluctuations of the phase @xmath150 around the average value @xmath214 , as @xmath215 with the lorentzian distribution @xmath216 we note that , as pointed out in ref . [ ] , a gaussian distribution of the phase fluctuations gives a different result , not consistent with the dephasing voltage probe approach for arbitrary dephasing stregth . we emphasize that the results above are independent on to which edge the probe is connected . moreover , we also point out that the effect of the voltage probes , for arbitrary @xmath194 , is multiplicative , i.e. attaching @xmath217 voltage probes at arbitrary places along the arms can be described by renormalizing @xmath218 . writing @xmath219)=\mbox{exp}(-l / l_{\phi})$ ] , with @xmath220 $ ] and @xmath221 with @xmath222 the distance between two probes , we can quite naturally incorporate the effect of a uniform distribution of probes into a dephasing length @xmath223 . the suppression of the visibilities of the @xmath0 and @xmath1 oscialltions due to dephasing in eqs . ( [ dephvismz1 ] ) and ( [ dephvismz2 ] ) are then modified as @xmath224 and @xmath225 the hbt - interferometer is less well known than the mz - interferometer and deserves some additional comments.@xcite the hbt - interferometer was invented as a tool to measure the angular diameter of stars . the first measurement @xcite was carried out on a radio star in 1954 . compared to existing schemes based on michelson interferometers , the hbt - interferometer proved to be less sensitive to atmospheric scintillations , which allowed for a more accurate determination of the angular diameter . after having demonstrated a table - top version of the interferometer in the visual range,@xcite the angular diameter of the visual star sirius was determined.@xcite the experimental results , both the two - particle interference and the positive intensity cross correlations , were successfully explained within a semi - classical framework . soon after the experiments , it was however shown by purcell@xcite that the positive cross correlations could be explained in terms of bunching of individual photons , emerging from the star , a thermal source of light . this bunching was also demonstrated explicitly in subsequent photo counting experiments.@xcite the hbt experiment thus laid the foundations for quantum statistical methods in quantum optics.@xcite the hbt approach has also been of importance in experimental particle physics.@xcite it is interesting to note that positive intensity cross correlations between beams of light emerging from a thermal source , according to some contemporary @xcite `` _ would call for a major revision of some fundamental concepts in quantum mechanics _ '' . purcell,@xcite however , providing an elegant explanation of the bunching phenomena , pointed out that `` _ the hanbury brown twiss effect , far from requiring a revision of quantum mechanics , is an instructive illustration of its elementary principles _ '' . an optical table - top version@xcite of the hbt - interferometer is shown in fig . [ hbtopt ] . a beam of light is emitted from each one of the sources @xmath226 and @xmath53 , completely uncorrelated with each other . the beams are split in two partial beams at the semitransparent beam splitters @xmath227 and @xmath228 respectively . the partial beams acquire phases @xmath229 to @xmath230 before scattering at the second pair of beam splitters @xmath51 and @xmath52 . the resulting beams are collected in detectors at ports @xmath192 to @xmath231 . importantly , there is no interference pattern in the intensities at the detectors @xmath192 to @xmath231 , instead the interference occurs only in the cross correlations between intensities at @xmath232 and @xmath233 . the intensity cross correlations are sensitive to the two - particle amplitudes : the interference is thus between two different two - particle scattering events , e.g. ( i ) one particle from @xmath226 scatters to @xmath192 and one particle from @xmath53 scatters to @xmath231 , with an amplitude @xmath234)$ ] and ( ii ) one particle from @xmath226 scatters to @xmath231 and one particle from @xmath53 scatters to @xmath192 , with an amplitude @xmath235)$ ] . the amplitude to detect one particle in @xmath192 and one in @xmath231 is then the sum of the two two - particle amplitudes . this is the case since both scattering processes have the same initial and final states and can not be distinguished . the ( reducible ) cross correlation between intensities in @xmath192 and @xmath231 is directly related to the corresponding two - particle probability @xmath236)+a_2\mbox{exp}(i[\phi_3+\phi_4])|^2$]= @xmath237)\right\}$ ] . the interference term @xmath238)\right\}$ ] contains the four geometrical phases @xmath229 to @xmath230 . the hbt - interferometer is thus , in contrast to the mz - interferometer , a two - particle interferometer . the electrical analog of the hbt - interferometer , presented in ref . [ ] , is shown in fig . [ hbtel ] . it consists of a ( rectangular ) conductor with a hole in the middle , a corbino geometry . similar to the mz - interferometer , the electrons propagate along single edge states . scattering between the edge states take place only at the beam splitters @xmath51 to @xmath228 . the beam splitters are described by scattering matrices given by eq . ( [ smat ] ) . we first consider the fully coherent case . in contrast to the mz - interferometer , the scattering amplitudes contain the phases @xmath239 and @xmath240 only via multiplicative phase factors . as an example , the amplitude to scatter from terminal @xmath226 to terminal 5 is given by @xmath241 as a consequence , the average currents which depend only on the modulus squared of the scattering amplitudes [ see eqs . ( [ avcurr ] ) and ( [ speccond ] ) ] do not contain any scattering phases . we get the currents at terminals 5 to 8 as @xmath242 turning to the current noise , the correlation between currents in terminals 5,6 and 7,8 is given by eq . ( [ noisedens ] ) . we find for the spectral density for the correlators between terminal 5 and 8 @xmath243 ^{2 } \notag \\ & \times & \left\{c_{0,58}+c_{\theta } \cos \left ( e / e_{c}+\theta \right ) \right\}\end{aligned}\ ] ] with the coefficients @xmath244 and for the correlator between terminal 5 and 7 @xmath245 ^{2 } \notag \\ & \times & \left\ { c_{0,57}+\bar c_{\theta } \cos \left ( e / e_{c}+\theta \right ) \right\}\end{aligned}\ ] ] with the coefficient @xmath246 performing the energy integrals in eq . ( 9 ) , we obtain the corresponding current cross correlators @xmath247 \label{hbtnoise1}\ ] ] and @xmath248 . \label{hbtnoise2}\ ] ] here @xmath249 and @xmath250 are given by eqs . ( [ enfcn1 ] ) and ( [ enfcn3 ] ) . the other two correlators @xmath251 and @xmath252 are given by the substitutions @xmath253 and @xmath254 . here , as for the mz - interferometer we have @xmath255 with @xmath256 and @xmath257 several observation can be made from the results above , put in comparison with the result for the noise correlations for the mz - interferometer in eq . ( [ mznoisetot ] ) . just as for the mz - interferometer , the noise consists of an incoherent , phase independent part , and a coherent , interference part . the phase dependent part of the noise in eqs . ( [ hbtnoise1 ] ) and ( [ hbtnoise2 ] ) however contains only one term . the amplitude of the oscillating term is a product of a scattering probability term and an energy - scale dependent function , just as for the mz - interferometer . this phase dependent term has the same dependence on the phase @xmath150 , the same voltage dependent phase shift as well as the same energy - scale dependence as the second term in eq . ( [ mznoisetot ] ) . this is the case since they both arise from processes which enclose the ab - flux once . despite the fact that in the hbt interferometer the ab - effect results from two - particle processes , the periodicity is determined by the single electron flux quantum @xmath0 . the dependence on the scattering probabilities is however different , a consequence of the mz and hbt interferometer geometries being different . importantly , there is no term in the noise in eqs . ( [ hbtnoise1 ] ) and ( [ hbtnoise2 ] ) that corresponds to the last term in eq . ( [ mznoisetot ] ) , describing processes which enclose the ab - flux twice . we note that the elementary scattering processes in the hbt - interferometer , in contrast to the mz - interferometer , are two - particle processes . an important consequence of this is that the incoherent , phase independent noise term in eqs . ( [ hbtnoise1 ] ) and ( [ hbtnoise2 ] ) can directly be reproduced by a model with filled streams of classical particles incident from reservoirs @xmath226 and @xmath53 . since there is only one phase - dependent term , the visibility of the phase - dependent oscillations can again be directly defined , giving for @xmath258 and @xmath259 @xmath260 since the energy - scale dependence of the visibilities is identical to @xmath261 for the mz - interferometer in eq . ( [ noisevismz1 ] ) , shown in fig . [ noisevis2 ] , we focus here only on the scattering probability terms . we thus consider the limit of a symmetric interferometer , @xmath262 for which the energy - scale dependent part is unity . several symmetries exists , e.g. all visibilities @xmath263 are unchanged by the substitutions @xmath264 and @xmath265 . the visibility @xmath266 is unity for scattering probabilities obeying @xmath267 and similar relations hold for the other visibilities . all visibilities go to zero for any of the transmission probabilities approaching either zero or unity . focusing on the case with @xmath268 ( or equivalently @xmath269 ) , the visibilities are given by @xmath270 and @xmath271 the two different visibilities are plotted in fig . [ hbtnoisevisfig ] as a function of @xmath272 for different @xmath273 . the visibility @xmath266has a maximum equal to unity for @xmath274 , while @xmath275 instead has a maximum equal to unity for @xmath276 . just as in the mz - interferometer , the dephasing in the hbt - interferometer is introduced by connecting a fictitious voltage probe to an edge between any of the two point contacts . the hbt - interferometer with the probe , denoted 9 , is shown in fig . [ hbtdepfig ] . here the probe is connected to the edge between contact c and a , we emphasize that the results discussed below do not depend on to which edge - state the probe is connected . the presence of the probe modifies the amplitudes for scattering from terminals 2 , 3 to terminals 5 to 8 . as an example , the scattering amplitude in eq . ( [ s52 ] ) is modified @xmath277 in addition , we also have to consider amplitudes for scattering into and out from the probe terminal 9 . the average currents in the presence of dephasing , given from eqs . ( [ avcurr ] ) to ( [ speccond ] ) and ( [ fav ] ) , turn out to be given by the same equations as in the absence of dephasing , i.e. eq . ( [ hbtcurr ] ) . this is what one expects , i.e. that dephasing affects only the phase - dependent parts of the observables . turning to the current correlators , given from eqs . ( [ noise ] ) , ( [ noisedens ] ) and ( [ dephnoisedens ] ) , we find for the correlators between terminal 5 and 8 @xmath278 \label{hbtnoisedeph1}\ ] ] and for the correlators between terminals 5 and 7 @xmath279 . \label{hbtnoisedeph2}\ ] ] the two remaining correlators are again given by the substitutions @xmath280 and @xmath281 . we see from eq . ( [ hbtnoisedeph1 ] ) and ( [ hbtnoisedeph2 ] ) that just as for the mz - interferometer , the only effect of dephasing is to suppress the phase - dependent term . the suppression factor is @xmath199 , just the same as for the @xmath282 term in the noise for the mz - interferometer in eq . ( [ mznoisetot ] ) . we can thus directly write the visibilities in the presence of dephasing as @xmath283 this leads to the conclusion that the voltage probe for the hbt - interferometer , just as for the mz - interferometer , just has the same effect as dephasing due to slow fluctuations of the phase @xmath284 , with the distribution of the phase fluctuations obeying the relation in eq . ( [ dephrel ] ) . moreover , the voltage probes have the same multiplicative property as for the mz - interferometer , allowing one to describe the effect of a continuum of probes along the edges ( of total length @xmath285 ) with a dephasing length @xmath223 . the suppression of the visibilities of the @xmath0 oscillations due to dephasing are then modified as @xmath224 , just as for the @xmath0 oscillations of the mz - interferometer . the mz - interferometer is an amplitude interferometer : it exhibits a visibility in the average current with period @xmath0 and exhibits a visibility in the shot noise with periods of both @xmath0 and @xmath1 . in contrast , the hbt interferometer is an intensity interferometer , it exhibits no ab - effect in the current and exhibits only an @xmath0-effect in the shot noise correlations . interestingly , our investigation shows that the shot noise visibility of the hbt interferometer as a function of temperature , voltage and dephasing rate , is qualitatively similar to that of the @xmath0-component of the shot noise of the mz - interferometer . this is contrary to the naive expectation that the visibility of the two particle processes which lead to the hbt effect should be similar to the visibility of the two particle processes in the mz - interferometer , that is the @xmath1 component of the shot noise . instead it is the number of times the ab - flux is enclosed which determines the behavior of the visibility . in this paper we have investigated and compared in detail the voltage , temperature and asymmetry dependence for the current and noise visibilities in the mz and hbt - interferometers . the experimental realization of the hbt - interferometer is of large importance since it allows for an unambiguous demonstration of two - particle interference effects with electrons , to date not demonstrated . moreover , a successful realization of the hbt - interferometer would also enable a first demonstration of orbital entanglement in electrical conductors , a fundamentally important result . the results presented in this work should prove useful for the experimental work aiming to detect the hbt effect in electrical conductors . we thank m. heiblum , i. neder , h. frster and e. sukhorukov for stimulating discussions . this work was supported by the graduate students study abroad program , taiwan national science council and the taiwan nsc93 - 2112-m-009 - 036 , the swedish research council and the swiss national science foundation and the network for materials with novel electronic properties . m. bttiker , phys . b * 46 * , 12485 ( 1992 ) . ya . blanter and m. bttiker , phys . rep . * 336 * 1(2000 ) . m. born and e. wolf , _ principles of optics _ , 7:th ed . ( cambridge university press , uk , 1999 ) . l. mandel , rev . phys . * 71 * , s274 ( 1999 ) . see e.g. s. pedersen , a.e . hansen , a. kristensen , c.b . sorensen and p.e . lindelof , phys . b * 61 * 5457 ( 2000 ) and references therein . e. buks , r. schuster , m. heilblum , d. mahalu and v. umansky , nature * 391 * 871 ( 1999 ) . a.e . hansen , a. kristensen , s. pedersen , c.b . sorensen , and p.e . lindelof , phys . b * 64 * 045327 ( 2001 ) . k. kobayashi , h. aikawa , s. katsumoto and y. iye , j. phys . * 71 * 2094 ( 2002 ) . y. ji , y. chung , d. sprinzak , m. heilblum , d. mahalu and h. shtrikman , nature * 422 * , 415 ( 2003 ) . g. seelig , m. bttiker , phys . rev . b * 64 * , 245313 ( 2001 ) . g. seelig , s. pilgram , a.n . jordan and m. bttiker , phys . b * 68 * , r161310 ( 2003 ) . k. le hur , phys . b * 65 * , 233314 ( 2002 ) . k. le hur , cond - mat/0503652 . f. marquardt and c. bruder , phys . lett . * 92 * , 56805 ( 2004 ) . f. marquardt and c. bruder , phys . b * 70 * , 125305 ( 2004 ) . f. marquardt , cond - mat/0410333 . h. frster , s. pilgram and m. bttiker , cond - mat/0502400 . r. hanbury brown and r.q . twiss , philos . mag . * 45 * , 663 ( 1954 ) . r. hanbury brown and r.q . twiss , nature ( london ) * 177 * , 27 ( 1956 ) . r. hanbury brown and r.q . twiss , nature ( london ) * 178 * , 1046 ( 1956 ) . m. henny , s. oberholzer , c. strunk , t. heinzel , k. ensslin , m. holland and c. schnenberger , science * 284 * , 296 ( 1999 ) . oliver , j. kim , r.c . liu , and y. yamamoto , science * 284 * , 299 ( 1999 ) . h. kiesel , a. renz , and f. hasselbach , nature ( london ) * 418 * , 392 ( 2002 ) . m. bttiker , physica b * 175 * , 199 ( 1991 ) . m. bttiker , phys . . lett . * 68 * , 843 ( 1992 ) . d. loss and e.v . sukhorukov , phys . * 84 * , 1035 ( 1992 ) . p. samuelsson , e.v . sukhorukov , and m. bttiker , phys . lett . * 92 * , 26805 ( 2004 ) . k. v. klitzing , g. dorda , m. pepper , phys . * 45 * , 494 ( 1980 ) . m. bttiker , phys . b * 38 * , 9375 ( 1988 ) . b. i. halperin , phys . b * 25 * , 2185 ( 1982 ) . b. j. van wees , h. van houten , c. w. j. beenakker , j. g. williamson , l. p. kouwenhoven , d. van der marel , and c. t. foxon , phys . lett . * 60 * , 848 ( 1988 ) . d. a. wharam , t. j. thornton , r. newbury , m. pepper , h. ahmed , j. e. f. frost , d. g. hasko , d. c. peacock , d. a. ritchie and g. a. c. jones , j. of phys . c * 21 * , l861(1988 ) . liu , b. odom , y. yamamoto , and s. tarucha , nature * 391 * 263 ( 1998 ) . s. oberholzer , m. henny , c. strunk , c. schonenberger , t. heinzel , k. ensslin , m. holland , physica * 6e * , 314 ( 2000 ) . m. bttiker , p. samuelsson , e.v . sukhorukov , physica e * 20 * , 33 ( 2003 ) . p. samuelsson , e.v . sukhorukov , and m. bttiker , phys . lett . * 91 * , 157002 ( 2003 ) . beenakker , c. emary , m. kindermann , and j.l . van velsen , phys . lett . * 91 * , 147901 ( 2003 ) . beenakker , and m. kindermann , phys . lett . * 92 * , 056801 ( 2004 ) . p. samuelsson , and m. bttiker , cond - mat/0410581 ( phys . b , in press ) . see also related work by c.w.j . beenakker , m. titov , b. trauzettel , phys . lett . * 94 * , 186804 ( 2005 ) ; a. v. lebedev , g. b. lesovik , g. blatter , cond - mat/0504583 . t. m. stace , c. h. w. barnes , and g. j. milburn phys . rev . lett . * 93 * , 126804 ( 2004 ) . l. saminadayar , d. c. glattli , y. jin and b. etienne , phys . lett . * 79 * , 2526.2529 ( 1997 ) . r. de - picciotto , m. reznikov , m. heiblum , v. umansky , g. bunin , d. mahalu , nature * 389 * 162 ( 1997 ) . i. safi , p. devillard , and t. martin , phys . lett . * 86 * , 4628 ( 2001 ) . s. vishveshwara , phys . lett . * 91 * , 196803 ( 2003 ) . s.a . kivelson and v.l . pokrovsky , phys . b * 40 * , r1373 ( 1989 ) . c. de c. chamon , d.e . freed , s.a . kivelson , s.l . sondhi and x.g . wen , phys . b * 55 * , 2331 ( 1997 ) . m.r . geller and d. loss phys . b * 56 * , 9692 ( 1997 ) . t. jonckheere , p. devillard , a. crepieux , th . martin , cond - mat/0503617 . m. bttiker , j. phys . condensed matter * 5 * , 9361 ( 1993 ) . m. bttiker , phys . * 57 * , 1761 ( 1986 ) . m. bttiker , phys . lett . * 65 * , 2901 ( 1990 ) . m. bttiker , phys . b * 33 * , 3020 ( 1986 ) . m. bttiker , ibm j. res . dev . * 32 * , 63 ( 1988 ) . beenakker , and m. bttiker , phys . b * 46 * , 1889 ( 1992 ) . m. j. m. de jong , and c. w. j. beenakker , physica a * 230 * , 219 ( 1996 ) . s. a. van langen , and m. bttiker , phys . b * 56 * , r1680 ( 1997 ) . c. texier and m. bttiker , phys b * 62 * , 7454 ( 2000 ) . * 46 * , 1889 ( 1992 ) . a.a . clerk and a.d . stone , phys . b * 69 * , 245303 ( 2004 ) . c.w.j . beenakker and b. michaelis , cond - mat/0503347 . d. b. chklovskii , b. i. shklovskii and l. i. glazman , phys . b * 46 * , 4026 ( 1992 ) ; n. r. cooper and j. t. chalker , phys . rev . b * 48 * , 4530 ( 1993 ) . h. a. fertig , phys . b * 38 * , 996 ( 1988 ) . see e.g. the discussion in n. c. van der vaart , m. p. de ruyter van steveninck , l. p. kouwenhoven , a. t. johnson , y. v. nazarov , c. j. p. m. harmans , and c. t. foxon , phys . lett . * 73 * , 320 ( 1994 ) and references therein . d. sanchez and m. bttiker , phys . rev . lett . * 93 * , 106802 ( 2004 ) ; t. christen and m. bttiker , phys . rev . b * 53 * 2064 , ( 1996 ) . for an interesting historical account , see r. hanbury brown , _ the intensity interferometer _ ( london : taylor and francis ) 1974 . purcell , nature * 178 * , 1449 ( 1956 ) . twiss , a.g . little , and r. hanbury brown , nature ( london ) * 180 * , 324 ( 1957 ) . g.a . rebka , and r.v . pound , nature ( london ) * 180 * , 1035 ( 1957 ) . see e.g. r. loudon , rep . , * 43 * , 58 ( 1980 ) . g. baym , acta phys . pol . b * 29 * , 1839 ( 1998 ) . e. brannen and h.i.s . ferguson , nature ( london ) * 178 * , 481 ( 1956 ) . r. hanbury brown and r.q . twiss , nature ( london ) * 178 * , 1447 ( 1956 ) . b. yurke and d. stoler , phys . lett . * 68 * , 1251 ( 1992 ) . b. yurke and d. stoler , phys . a * 46 * , 2229 ( 1992 ) .

crystals which have a uniform distribution of defects are endowed with a lie group description which allows one to construct an associated discrete structure . these structures are in fact the discrete subgroups of the ambient lie group . the geometrical symmetries of these structures can be computed in terms of the changes of generators of the discrete subgroup which preserve the discrete set of points . here a classification of the symmetries for the discrete subgroups of a particular class of three - dimensional solvable lie group is presented . it is a fact that there are only three mathematically distinct types of lie groups which model uniform defective crystals , and the calculations given here complete the discussion of the symmetries of the corresponding discrete structures . we show that those symmetries corresponding to automorphisms of the discrete subgroups extend uniquely to symmetries of the ambient lie group and we regard these symmetries as ( restrictions of ) elastic deformations of the continuous defective crystal . other symmetries of the discrete structures are classified as ` inelastic ' symmetries .

1310.0324

in this paper it will be shown how the generalisation of symmetry properties of perfect solid crystals to crystals with certain uniform distributions of defects leads to a classification of the symmetries of discrete defective crystals as elastic or inelastic . we show that , in contrast with the perfect crystal case , some of the symmetries of a discrete defective crystal do not extend uniquely to a symmetry of the continuum model of the defective crystal . this allows us to classify the symmetries which do nt extend as inelastic symmetries of the discrete defective crystal , while those which do extend uniquely are restrictions of elastic deformations of the continuous crystal and we call these symmetries elastic symmetries of the discrete crystal . a starting point for the study of the mechanics of perfect solid crystals is to consider the geometrical symmetries of perfect lattices in @xmath0 : @xmath1 where @xmath2 are defining basis vectors and the summation convention operates on repeated indices . the perfect lattice @xmath3 defines a discrete set of points in @xmath0 and can also be thought of as a discrete subgroup of the continuous lie group @xmath0 with addition as group composition . the discrete structures @xmath3 in @xmath0 have geometrical symmetries @xmath4 given by @xmath5 the symmetries of @xmath3 are bijective and preserve addition as well as the set of points in @xmath0 defined by @xmath3 . moreover , every symmetry @xmath6 of a perfect lattice @xmath3 extends uniquely to a bijection of @xmath0 . here we extend and generalise these properties of the symmetries of perfect crystals to a certain class of defective crystal where the distribution of defects is uniform . we use davini s continuum model of defective crystals @xcite , in which the dislocation density tensor @xmath7 , @xmath8 is defined by @xmath9 where the fields @xmath10 are dual to the smooth _ lattice vector fields _ @xmath11 , which represent the crystal geometry in a region @xmath12 . a crystal with a uniform distribution of defects has dislocation density tensor which is constant in space . ( note that for perfect crystals @xmath13 . ) suppose that @xmath14 is a set of lattice vector fields which are elastically related to @xmath15 in the sense that there exists a smooth invertible mapping @xmath16 such that @xmath17 we say that @xmath18 is an elastic deformation . if @xmath19 is calculated via the analogue of , using fields dual to @xmath20 , then @xmath21 therefore , each component @xmath22 of the dislocation density tensor is an ` elastic ' scalar invariant so that the value of the dislocation density is unchanged by elastic deformations of the crystal . due to this elastic invariance , for a given dislocation density tensor @xmath23 there are infinitely many choices of ( elastically related ) sets of corresponding lattice vector fields . it is a commonly held idea in the elasticity theory of perfect crystals that there is a continuum energy density @xmath24 which depends on the underlying perfect lattice @xmath3 ; that is @xmath25 where @xmath26 denotes the set of vectors @xmath27 . if the basis vectors @xmath28 generate the same lattice @xmath3 ( i.e. @xmath29 in ) then @xmath30 in other words the geometrical symmetries of @xmath3 correspond to the material symmetries of the energy density function @xmath24 . here we use a generalisation of this theory which accounts for the presence of a continuous distribution of defects in the crystal . it is assumed that the strain energy density per unit volume in such a crystal depends on the values of the lattice vector fields @xmath31 , @xmath32 , @xmath33 and dislocation density tensor @xmath34 at some point in @xmath12 . thus @xmath35 where @xmath26 denotes the set of vectors @xmath27 which are the values of the lattice vector fields at some point in @xmath12 , and @xmath23 denotes the dislocation density tensor evaluated at the same point . as we shall see in section [ sec : structure ] , the arguments of the energy density function determine a structure @xmath36 which under certain conditions will be a discrete set of points . if another set of arguments , say @xmath37 , determine the same discrete structure then it is assumed that @xmath38 hence for crystals with defects we are associating that the symmetries of the energy density function with the geometrical symmetries of the structure @xmath36 . therefore the structure @xmath36 is taken to be the defective crystal analogue of the perfect lattice @xmath3 underlying perfect crystals , and it is a central task to determine the geometrical symmetries of @xmath36 . notice that the arguments @xmath39 of @xmath24 give no information regarding gradients of @xmath23 , and we shall assume that they are zero . therefore the dislocation density tensor @xmath23 is constant in space and the crystal has a uniform distribution of defects . it is this assumption that endows the crystal with a lie group structure . suppose that the vector fields @xmath31 , @xmath32 , @xmath33 , defined here and henceforth on @xmath40 , give constant @xmath23 . then according to pontryagin @xcite , the system of partial differential equations @xmath41 has a solution for the function @xmath42 , where @xmath43 denotes the gradient of @xmath42 with respect to its first argument . moreover , the function @xmath44 can be taken to satisfy the properties required for it to be a lie group composition function on @xmath0 , i.e. @xmath45 where @xmath46 is the group identity element and @xmath47 is the unique inverse of the element @xmath48 @xcite . here , the lie group @xmath49 has underlying manifold @xmath0 so that an element @xmath50 can be uniquely specified by @xmath51 where @xmath52 and @xmath53 is a basis of @xmath0 . we will often use the alternative notation @xmath54 . ( note that in the perfect crystal case where @xmath55 , we can take @xmath42 to be addition so that @xmath56 . ) relation expresses the _ right invariance _ of the fields @xmath57 with respect to the lie group @xmath58 . suppose that the fields @xmath59 are elastically related to @xmath57 via @xmath18 as in . then if @xmath60 is defined by @xmath61 then @xmath62 hence the fields @xmath59 are right invariant with respect to the lie group @xmath63 and since @xmath64 is invertible , the groups @xmath65 and @xmath66 are isomorphic . thus , elastically related crystal states have isomorphic corresponding lie groups . recall that the dislocation density tensor @xmath23 is an elastic invariant so there is an infinite choice of elastically related lattice vector fields which have duals which satisfy for a given constant dislocation density . therefore , there is also an infinite number of choices of isomorphic lie groups @xmath49 corresponding to @xmath23 . hence a given constant @xmath23 determines up to lie group isomorphism a lie group @xmath65 . in this paper we make a ` canonical ' choice of the lie group @xmath65 to simplify the computations . we identify material points of the crystal with elements of the lie group . when @xmath65 has a uniform discrete subgroup @xmath36 ( where @xmath67 is uniform if the left coset space @xmath68 is compact ) the material points corresponding to the elements in @xmath36 have a minimum separation distance and form discrete geometrical structures which we take to the defective crystal analogue of the perfect lattice @xmath3 . the requirement that @xmath68 be compact is a generalisation of the fact that in the perfect crystal case that @xmath69 ( the unit cell of the lattice @xmath3 with appropriate identification of boundary points ) is compact . according to auslander , green and hahn @xcite there are precisely three classes of non - abelian , three dimensional lie groups @xmath65 with uniform discrete subgroups . these are a certain class of nilpotent lie groups and two non - isomorphic classes of solvable lie groups . for each of these cases we are interested in the form of the geometrical structures corresponding to the discrete subgroups @xmath36 and the geometrical symmetries of these structures ( i.e. the changes of generators of @xmath36 which preserve the points in the geometrical structure ) . these have already been determined in all three cases : * when the lie group @xmath65 is nilpotent ( with corresponding lie algebra with rational structure constants ) , cermelli and parry @xcite have shown that the corresponding discrete subgroups give either a simple lattice or a 4-lattice ( in pitteri and zanzotto s terminology @xcite ) even though the composition function in @xmath65 is not additive . for such groups , parry and sigrist @xcite construct explicitly all sets of generators of a given discrete subgroup . the formulae that connect different sets of generators generalise the perfect crystal case given by . * auslander et al . @xcite call the two classes of solvable groups @xmath70 and @xmath71 . it has been shown by parry and nicks that in both cases the geometrical structures corresponding to the discrete subgroups of these solvable groups are simple lattices . the changes of generators preserving these structures were also determined ( see @xcite for the @xmath70 case and @xcite for the @xmath71 case ) . in this paper we focus on a property of the symmetries of perfect lattices which does not hold in the generalisation to the symmetries of discrete structures underlying crystals with uniform distributions of defects . recall that the geometrical symmetries of a perfect lattice @xmath3 are bijections @xmath4 as in which preserve addition . each of these symmetries extends uniquely to a bijection @xmath72 defined by @xmath73 thus every symmetry of @xmath3 represents a ( restriction of an ) elastic deformation of the continuum perfect crystal . this is not the case for crystals with constant @xmath74 where the underlying lie group , @xmath58 , is solvable or nilpotent . in these cases there is a difference between the set of all geometrical symmetries of a discrete structure @xmath75 and the subset of these symmetries which preserve the group structure of @xmath36 and extend uniquely to elastic deformations of @xmath0 . this allows us to classify the symmetries of the discrete structures @xmath36 which preserve the group structure as elastic or inelastic depending on whether or not they are restrictions of elastic deformations of the continuum defective crystal . the observation that such a classification can be made is interesting because it indicates a possible link between the inelastic symmetries of the discrete crystal @xmath36 ( which preserve the elastic invariant @xmath23 and the discrete structure ) and observed inelastic processes in crystal behaviour such as slip in particular planes and directions determined by geometry . notice that we consider here only symmetries of discrete structures @xmath36 which additionally preserve the group structure . we do not discuss here symmetries where the discrete structure represents discrete subgroups of different lie groups , isomorphic or not . our simplified task then is to identify which of the geometrical symmetries of discrete subgroups @xmath75 extend uniquely to elastic deformations of @xmath0 . this task breaks down into two stages . first we must determine which of the geometrical symmetries of @xmath36 preserve the group structure of @xmath36 ; that is which of the symmetries will extend to _ automorphisms _ of @xmath36 . secondly we need to determine if these automorphisms of @xmath36 extend uniquely to automorphisms of the ambient lie group @xmath65 . for a geometrical symmetry of @xmath36 to be classified as an elastic symmetry it must extend to an automorphism of @xmath36 and that automorphism must extend uniquely to an automorphism of @xmath65 , since these are requirements that must be satisfied in order that the geometrical symmetry is a restriction of an elastic deformation of the defective crystal . in the cases where the structure @xmath36 is a discrete subgroup of a nilpotent lie group or a solvable lie group in the class @xmath70 such a classification of the geometrical symmetries of @xmath36 has been carried out . the automorphisms of the discrete subgroups @xmath36 have been computed ( see @xcite for the nilpotent case and @xcite for the @xmath70 case ) and it has been observed that theorems of malcev @xcite and gorbatsevich @xcite guarantee that every automorphism of @xmath36 extends uniquely to an automorphism of the ambient lie group @xmath65 . in this paper we will complete the analysis by classifying the geometrical symmetries of discrete subgroups @xmath36 of solvable groups in the class @xmath71 . in this case we must work a little harder since although it remains relatively straightforward to compute which of the geometrical symmetries of @xmath36 correspond to automorphisms of @xmath36 , there is no analogue of the theorems of malcev and gorbatsevich for solvable groups of this class . therefore we must determine directly whether or not automorphisms of @xmath36 extend uniquely to automorphisms of @xmath71 . the difficulties arise for the @xmath71 class due to the fact that the exponential mapping from the corresponding lie algebra @xmath76 to @xmath71 is not one - to - one . we begin by recalling how to construct discrete structures @xmath36 corresponding to a particular set of arguments @xmath39 of the energy density function @xmath24 . these are discrete subgroups of lie groups @xmath65 and we will also recall elements of lie group theory that will be required in this paper , including facts about lie group isomorphisms . in section [ sec : solvable ] , following auslander et al . @xcite and nicks and parry @xcite we introduce the group @xmath71 and the canonical group in the isomorphism class which we will work with . we also introduce the lie algebra @xmath76 of the lie group @xmath71 and calculate the automorphisms of @xmath71 . in section [ sec : discrete ] we discuss the discrete subgroups @xmath36 of @xmath71 , recalling results from nicks and parry @xcite concerning their geometrical symmetries . we next compute the automorphisms of these discrete subgroups @xmath36 which amounts to determining the matrices @xmath77 which commute with a given matrix @xmath78 which is related to the dislocation density . this is a number theoretic problem studied by baake and roberts @xcite and here we summarize their results which are relevant to this work . finally in section [ sec : extensions ] we demonstrate explicitly that each of these automorphisms of @xmath36 extends uniquely to an automorphism of @xmath71 . suppose that we are given a set of arguments of an energy density function for a crystal with a uniform distribution of defects . that is , we are given @xmath39 where @xmath23 is some value of the dislocation density tensor and the ( linearly independent ) vectors @xmath79 , @xmath80 are values of some lattice vector fields @xmath81 evaluated at some point , say @xmath46 , in @xmath0 such that their duals satisfy . furthermore ( since we assume the crystal is uniform ) the fields @xmath81 also satisfy for some group composition function @xmath42 on @xmath0 . the following survey of facts about the lie group @xmath49 follows that given in @xcite , @xcite , @xcite , @xcite , @xcite and is given here for completeness . the reader who is familiar with this background material may omit section [ sec : structure ] and focus on the subsequent new material . the lie group @xmath58 has corresponding lie algebra @xmath82 which is the vector space @xmath0 with the lie bracket operation @xmath83:\mathbb{r}^3 \times \mathbb{r}^3 \to \mathbb{r}^3 $ ] given by @xmath84= c_{ijk } x_j y_k { \mbox{\boldmath{$e$}}}_i , \qquad { \mbox{\boldmath{$x$ } } } , { \mbox{\boldmath{$y$ } } } \in \mathbb{r}^3,\ ] ] with respect to some basis @xmath85 of @xmath0 . here , @xmath86 are the structure constants of the lie algebra and are related to the lie group composition function @xmath42 via @xmath87 where @xmath88 . the connection between the dislocation density tensor @xmath89 , defined via , and the structure constants is @xmath90 where @xmath91 is the permutation symbol and @xmath92 , see elzanowski and parry @xcite . in this paper we shall be concerned with the automorphisms of lie groups @xmath65 which are of course isomorphisms of the lie group to itself ( preserving the group composition function ) . these are related to the automorphisms of the corresponding lie algebra @xmath82 . let @xmath93 and @xmath94 be lie algebras with lie brackets @xmath95_{{\mbox{\boldmath{$\scriptstyle{\mathfrak { g}}$}}}},\;[\cdot,\cdot]_{{\mbox{\boldmath{$\scriptstyle{\mathfrak { g'}}$}}}}$ ] respectively . a lie algebra isomorphism is an invertible linear transformation @xmath96 which satisfies @xmath97_{{\mbox{\boldmath{$\scriptstyle{\mathfrak { g'}}$}}}}=l[{\mbox{\boldmath{$x$}}},{\mbox{\boldmath{$y$}}}]_{{\mbox{\boldmath{$\scriptstyle{\mathfrak { g}}$}}}},\quad { \mbox{\boldmath{$x$ } } } , { \mbox{\boldmath{$y$}}}\in { \mbox{\boldmath{$\mathfrak g$}}}.\ ] ] if @xmath98 are the structure constants for @xmath99 respectively , then implies that @xmath100 where @xmath101 . let @xmath102 and @xmath103 be lie groups with corresponding lie algebras @xmath93 and @xmath94 respectively . a smooth invertible mapping @xmath104 is a lie group isomorphism if @xmath105 it is a fact that if @xmath106 is a lie group isomorphism then @xmath107 is a lie algebra isomorphism from @xmath93 to @xmath94 . conversely , if an invertible linear transformation @xmath3 satisfies , then it is a major result of lie theory that there exists a unique lie group isomorphism @xmath64 such that @xmath108 ( see @xcite ) . let @xmath109 be given real numbers and define the right invariant vector field @xmath110 . define the integral curve of @xmath111 through @xmath112 to be the solution @xmath113 of the differential equation @xmath114 , @xmath115 . note that @xmath116 determines the field @xmath117 by the right invariance of @xmath111 . one can then define the mapping @xmath118 by @xmath119 and the group element @xmath120 by @xmath121 also , note that @xmath122 is called the exponential mapping of the lie algebra to the lie group . it is standard result of lie group theory that @xmath123 and this states that the flow along the integral curves of the lattice vector fields corresponds to group multiplication by the group element @xmath124 . in the case of perfect crystals , choosing @xmath125 for a basis @xmath53 of @xmath0 , iterating the flow along the lattice vector fields ( which in this case is just translation by @xmath126 ) produces a perfect lattice . in the case of @xmath58 the analogue of the perfect lattice is the set of points ( or group elements ) produced by iterating the flow ( from @xmath127 to @xmath128 ) along the lattice vector fields , starting at the origin . by one obtains the subgroup of @xmath65 that is generated by the group elements @xmath129 , @xmath130 , @xmath131 where @xmath132 , @xmath80 . we will be interested in the automorphisms of both @xmath65 and its subgroup @xmath36 generated by @xmath129 , @xmath130 , @xmath131 whenever this is a uniform discrete subgroup . the method we will use to compute the automorphisms of @xmath65 makes use of the fact that the diagram in figure [ fig : commuting ] commutes so that @xmath133 and also the fact that the automorphisms of the corresponding lie algebra @xmath82 can be computed using the fact that they must satisfy with @xmath134 . recall that in this paper we shall be completing the classification of symmetries of discrete structures associated with crystals with uniform distributions of defects . as we have seen in section [ sec : structure ] , the discrete structures are uniform discrete subgroups @xmath36 of three - dimensional lie groups @xmath58 . according to auslander et al @xcite , there are only three classes of non - abelian , connected , simply connected , three - dimensional lie groups @xmath58 which have such uniform discrete subgroups . these are a class of nilpotent lie group and two non - isomorphic classes of solvable lie group which they call @xmath70 and @xmath71 . the symmetries of the discrete subgroups of the nilpotent lie groups and the solvable groups in the class @xmath70 have already been computed and classified . here we complete the analysis by considering the @xmath71 case . we begin with the relevant definitions . let @xmath82 be a lie algebra with corresponding connected lie group @xmath65 . define the following sequence of subalgebras : @xmath135 , \quad \ldots , \quad\mathfrak{g}_k= \left [ \mathfrak{g}_{k-1 } , \mathfrak{g}_{k-1}\right].\ ] ] the lie algebra @xmath82 is solvable if @xmath136 for some integer @xmath137 . let @xmath138 denote the commutator of @xmath139 where group multiplication is represented as juxtaposition . then let @xmath140 denote the commutator ( or derived ) subgroup of @xmath65 generated by all commutators of elements of @xmath65 . if one defines @xmath141 the @xmath65 is solvable if @xmath142 for some integer @xmath137 . the lie algebra of @xmath143 is @xmath144 and @xmath65 is solvable if and only if @xmath82 is solvable . in solvable groups of dimension three we have @xmath145 so that all commutators of elements of @xmath65 commute with each other . furthermore , it can be shown that there are basis vectors @xmath146 , @xmath147 , @xmath148 of @xmath0 such that @xmath149 = { \mbox{\boldmath{$0 $ } } } , \qquad \left [ { \mbox{\boldmath{$f$}}}_1 , { \mbox{\boldmath{$f$}}}_3 \right ] = \alpha { \mbox{\boldmath{$f$}}}_1 + \beta { \mbox{\boldmath{$f$}}}_2 , \qquad \left [ { \mbox{\boldmath{$f$}}}_2 , { \mbox{\boldmath{$f$}}}_3 \right ] = \gamma { \mbox{\boldmath{$f$}}}_1 + \delta { \mbox{\boldmath{$f$}}}_2,\ ] ] where @xmath150 and @xmath151 . we shall be concerned with the three - dimensional solvable lie group @xmath71 which has corresponding lie algebra which we shall denote @xmath76 . we now define this group and give the form of corresponding dislocation density tensor . further details regarding derivation of facts about this group or the related group @xmath70 can be found in nicks and parry @xcite and nicks and parry @xcite respectively . we identify group elements with points @xmath152 , representing them as @xmath153 with respect to some basis @xmath85 of @xmath0 . auslander et al @xcite choose to represent the elements as @xmath154 matrices ( still parameterised by @xmath155 ) and these matrix representations form an isomorphic group @xmath156 where the matrix representation of @xmath157 is @xmath158 , defined by @xmath159 \multicolumn{2}{c|}{\phi(x_3)}\\[0.3 cm ] \hline 0&0\\ 0&0\end{array } \begin{array}{|cc } 0&x_1\\0&x_2\\\hline 1&x_3\\ 0&1\end{array } \right ) , \quad{\mbox{\boldmath{$x$}}}\equiv \left(\begin{array}{c } x_1\\x_2\\x_3\end{array}\right ) \in \mathbb{r}^3 , \quad \phi(x_3)=\left ( \begin{array}{cc}a(x_3)&b(x_3)\\ c(x_3)&d(x_3)\end{array}\right).\ ] ] in , @xmath160 and @xmath161 is a one parameter subgroup of the unimodular group . this implies that @xmath162 and hence @xmath163 , the @xmath164 identity matrix . the one parameter subgroups of @xmath165 which have @xmath166 fall into two classes depending on the eigenvalues of @xmath167 . let us define @xmath168 the eigenvalues , @xmath169 , of @xmath170 are roots of the characteristic polynomial @xmath171 , and are therefore given by @xmath172 where @xmath173 . if @xmath174 then the eigenvalues of @xmath170 are a complex conjugate pair and the group of matrices of the form @xmath175 is isomorphic to @xmath71 . moreover , if @xmath176 then @xmath177 . if @xmath178 then the eigenvalues of @xmath170 are positive , real and distinct and the group of matrices of the form @xmath175 is isomorphic to the class of solvable group @xmath70 which has been considered previously ( see nicks and parry @xcite ) . if @xmath179 then the group of matrices of the form @xmath175 is isomorphic to a nilpotent lie group . for other values of @xmath180 , @xmath170 can not lie on a one parameter subgroup of @xmath165 ( see auslander et al . @xcite ) . differentiating with respect to @xmath181 and evaluating at @xmath182 , and also doing the same for @xmath183 , we see that the one parameter subgroup @xmath184 of @xmath165 satisfies @xmath185 where @xmath186 denotes @xmath187 . we also define @xmath188 so that we have @xmath189 since @xmath190 , we have @xmath191 and differentiating this relation with respect to @xmath183 and setting @xmath192 we see that @xmath193 since @xmath194 . hence @xmath195 and any matrix satisfying this condition has matrix exponential satisfying @xmath196 ( \cos k ) \mathbb{i}_2+\left ( \frac{\sin k}{k}\right ) \mathcal{a},\quad&\text { if } \det ( \mathcal{a})>0,\quad k\equiv \sqrt{\det(\mathcal{a})};\\[0.3 cm ] \mathbb{i}_2+\mathcal{a } , \quad&\text { if } \det ( \mathcal{a})=0 . \end{array}\right.\ ] ] since @xmath193 we have @xmath197 2\cos k , \quad & \text { if } \det ( \mathcal{a})>0;\\[0.3 cm ] 2 , \quad & \text { if } \det ( \mathcal{a})=0,\end{array}\right.\ ] ] where @xmath198 . recall that we are interested in the case where @xmath156 is isomorphic to @xmath71 where @xmath199 . in that case @xmath200 and @xmath201 where@xmath202 when @xmath203 so that @xmath204 , @xmath205 is any traceless @xmath164 matrix with determinant @xmath206 , @xmath207 an odd integer . if @xmath208 then @xmath209 which implies that @xmath210 note that in the cases where @xmath208 the integers @xmath211 and @xmath212 in @xmath205 and @xmath170 are nonzero , since if either were zero then it must be the case that @xmath213 and hence @xmath214 , implying that @xmath215 . noting that the mapping @xmath216 given by is one to one and that matrix multiplication is the group composition function in the matrix group @xmath156 , it follows that the group composition function @xmath42 in @xmath71 satisfies @xmath217 and hence @xmath218 for a given a group composition function @xmath42 it is easy to see that @xmath219 , @xmath80 , is a set of lattice vector fields which is right invariant with respect to the group @xmath220 ; that is is satisfied where @xmath221 . for our group composition function we have @xmath222 recalling that @xmath223 . using the duals of these vector fields we then compute using that the components of the dislocation density tensor are @xmath224 in particular the dislocation density tensor is rank 2 , symmetric and uniquely determined by the matrix @xmath205 . also note that the correspondence between the @xmath164 matrices @xmath205 and @xmath225 is infinitely many to one since @xmath170 determines @xmath226 where @xmath137 may take a countable infinity of values . from the group composition function @xmath42 in @xmath71 given by one can calculate using that the lie bracket on @xmath76 ( the lie algebra of @xmath71 ) is given by @xmath227= ( a^\prime(0 ) { \mbox{\boldmath{$x$ } } } \wedge { \mbox{\boldmath{$y$ } } } \cdot { \mbox{\boldmath{$e$}}}_2 - b^\prime(0){\mbox{\boldmath{$x$ } } } \wedge { \mbox{\boldmath{$y$ } } } \cdot { \mbox{\boldmath{$e$}}}_1 ) { \mbox{\boldmath{$e$}}}_1 + ( c^\prime(0 ) { \mbox{\boldmath{$x$ } } } \wedge { \mbox{\boldmath{$y$ } } } \cdot { \mbox{\boldmath{$e$}}}_2 + a^\prime(0){\mbox{\boldmath{$x$ } } } \wedge { \mbox{\boldmath{$y$ } } } \cdot{\mbox{\boldmath{$e$}}}_1 ) { \mbox{\boldmath{$e$}}}_2 , \ ] ] for @xmath228 , @xmath229 . in particular @xmath230 = { \mbox{\boldmath{$0 $ } } } , \ \ \left [ { \mbox{\boldmath{$e$}}}_1 , { \mbox{\boldmath{$e$}}}_3\right]= -a^\prime(0){\mbox{\boldmath{$e$}}}_1 -c^\prime(0){\mbox{\boldmath{$e$}}}_2 , \ \ \left [ { \mbox{\boldmath{$e$}}}_2 , { \mbox{\boldmath{$e$}}}_3\right]= -b^\prime(0){\mbox{\boldmath{$e$}}}_1 + a^\prime(0){\mbox{\boldmath{$e$}}}_2,\ ] ] so that it is clear from that @xmath76 is solvable since @xmath231 . we now want to compute the automorphisms of this lie algebra . it turns out that these computations are much simplified if we make a change of basis in the lie algebra . suppose that we are given a particular value of @xmath170 with @xmath232 or @xmath204 and make a choice of corresponding @xmath137 ( c.f . ) . consider the basis @xmath233 where @xmath234 for @xmath235 note that @xmath236 for @xmath232 by the remark following and if @xmath237 when @xmath238 then @xmath239 since @xmath240 but this contradicts the fact that @xmath241 . thus the change of basis matrix @xmath242 is invertible since it has @xmath243 . the basis @xmath233 satisfies @xmath244 = { \mbox{\boldmath{$0 $ } } } , \ \ \left [ { \mbox{\boldmath{$f$}}}_1 , { \mbox{\boldmath{$f$}}}_3\right]= k { \mbox{\boldmath{$f$}}}_2 , \ \ \left [ { \mbox{\boldmath{$f$}}}_2 , { \mbox{\boldmath{$f$}}}_3\right]= -k { \mbox{\boldmath{$f$}}}_1.\ ] ] furthermore , with respect to this basis the structure constants of @xmath76 are given by @xmath245 and using with @xmath246 one computes that a linear transformation @xmath3 is an automorphism of @xmath76 with respect to the basis @xmath233 if @xmath3 has the form ( c.f . @xcite ) @xmath247 where @xmath248 , @xmath249 such that @xmath250 . with respect to the basis @xmath85 , the automorphisms of @xmath76 are given by @xmath251 where @xmath252 denotes transpose . here we discuss how to compute the group automorphisms of @xmath71 . as previously discussed we shall use relation to compute these using our knowledge of the lie algebra automorphisms of @xmath76 . this computation is not as straight forward as it may appear at first glance , due to the fact that the exponential mapping @xmath253 is not surjective . again we work with respect to the basis @xmath233 for ease of computation . we begin by giving details of the required functions and mappings with respect to the basis @xmath233 . when changing basis from @xmath254 to @xmath233 , the matrix @xmath205 changes to @xmath255 and , with respect to the basis @xmath233 , @xmath256 thus with respect to the basis @xmath233 the lie group composition function in @xmath71 is given by @xmath257 where @xmath258 , @xmath259 . computing the lattice vector fields @xmath260 one can then use to find that the exponential mapping @xmath261 with respect to the basis @xmath233 is given by @xmath262 and for @xmath263 @xmath264 for @xmath265 , @xmath266 . note that if @xmath267 for some @xmath268 then @xmath269 , the @xmath164 zero matrix , and @xmath270 for any values of @xmath271 and @xmath272 . therefore @xmath261 is not a homeomorphism - it is not surjective because it is not possible to write every element @xmath273 as @xmath274 for some @xmath275 . hence the exponential mapping @xmath261 does not have a well defined inverse . however , one can verify that any @xmath276 may be written as the group composition of two exponentials : @xmath277 recall that there is a one to one correspondence between the lie algebra automorphisms @xmath278 and the lie group automorphisms @xmath279 given by @xmath280 . also the lie group and algebra automorphisms satisfy we now use this relation to compute the lie group automorphisms @xmath279 . suppose that @xmath278 is a lie algebra automorphism and hence has the form , and let @xmath273 . then @xmath281 can be written as in so that lie group automorphisms @xmath279 satisfy @xmath282 where @xmath283 and @xmath284 are as in . hence @xmath285 and @xmath286 so that automorphisms of @xmath71 with respect to the basis @xmath233 are given by @xmath287 the automorphisms of @xmath76 and @xmath71 as computed in previous sections form lie groups @xmath288 and @xmath289 respectively under composition of mappings . in this section we show that each automorphism of @xmath76 or @xmath71 is a composition of automorphisms in various subgroups of @xmath288 or @xmath289 . moreover , the groups @xmath288 and @xmath289 are isomorphic . the elements of @xmath288 with respect to the basis @xmath233 are @xmath290 define the following subgroups of @xmath288 : @xmath291 @xmath292 then , noting that @xmath293 is normal in @xmath288 and @xmath294 , if @xmath295 then @xmath3 can be written uniquely as a product of an element of @xmath296 and an element of @xmath293 . also define the following subgroups of @xmath293 : @xmath297 @xmath298 then @xmath23 is normal in @xmath293 and @xmath299 so that any @xmath300 can be uniquely written as a product of an element of @xmath296 , an element of @xmath252 and an element of @xmath23 or @xmath301 . there is a one to one correspondence between the automorphisms of @xmath76 and the automorphisms of @xmath71 , in fact there is an isomorphism @xmath302 , and the lie group automorphism @xmath303 corresponding to the lie algebra automorphism @xmath304 , @xmath305 , can be uniquely decomposed as @xmath306 where @xmath307 and @xmath308 is a subgroup of @xmath289 , and so on . recall that in this paper we are investigating whether or not automorphisms of discrete subgroups of the lie group @xmath71 extend to automorphisms of the continuous lie group . the automorphisms of a discrete subgroup @xmath36 correspond to a certain subset of the geometrical symmetries of @xmath36 , where the geometrical symmetries of @xmath36 are the changes of generators that preserve the set of points in @xmath36 . in this section we introduce the discrete subgroups @xmath36 of @xmath71 and recall from nicks and parry @xcite@xcite how different sets of generators of @xmath36 are related to each other , thus determining the geometrical symmetries of @xmath36 . according to auslander , green and hahn @xcite , when @xmath309 has @xmath310 the discrete subgroups @xmath311 are isomorphic ( via @xmath312 defined in ) to a discrete subgroup @xmath313 of @xmath156 and @xmath313 is generated by three elements @xmath314 \multicolumn{2}{c}{\theta}\\[0.3 cm ] \hline 0&0\\ 0&0\end{array}\begin{array}{|cc } 0&0\\0&0\\\hline1&1\\0&1\end{array}\right ) , b\equiv r_m ( { \mbox{\boldmath{$e$}}}_1 ) = \left ( \begin{array}{cccc } 1&0&0&1\\ 0&1&0&0\\ 0&0&1&0\\ 0&0&0&1\end{array}\right ) , c\equiv r_m({\mbox{\boldmath{$e$}}}_2 ) = \left ( \begin{array}{cccc } 1&0&0&0\\ 0&1&0&1\\ 0&0&1&0\\ 0&0&0&1\end{array}\right).\ ] ] note that here we are again working with respect to the basis @xmath85 . let @xmath315 denote the commutator of elements @xmath316 , @xmath317 where juxtaposition denotes matrix multiplication . we then note that @xmath318 where @xmath319 is the identity element in @xmath320 . from this one can see that any element of @xmath320 can be expressed as a product of the form @xmath321 for some @xmath322 where @xmath323 , @xmath324 and @xmath325 . a general element @xmath326 , @xmath327 has the representation @xmath328 \multicolumn{2}{c}{\theta^q}\\[0.3 cm ] \hline 0&0\\ 0&0\ \end{array}\begin{array}{|ccc } \begin{array}{c } 0\\0 \end{array } & \theta^q \left(\begin{array}{c } m\\n \end{array}\right)\\\hline1&q\\0&1\end{array}\right ) = r_m({\mbox{\boldmath{$x$ } } } ) , \quad \mbox{where } \ \ { \mbox{\boldmath{$x$}}}= \left ( \begin{array}{c } \theta^q \left(\begin{array}{c } m\\n \end{array}\right ) \\ q\end{array}\right)\in s_2,\ ] ] with respect to the basis @xmath53 . it is then clear that since @xmath309 , @xmath329 . thus the discrete structures which are the analogues of perfect lattices @xmath3 in this case are the lattice @xmath330 with group multiplication @xmath42 given by . the symmetries of the discrete subgroup @xmath311 are the choices of three elements @xmath331 such that the subgroup of @xmath36 generated by the three elements ( which we shall denote @xmath332 ) is in fact equal to @xmath36 . these changes of generators preserve the integer lattice @xmath330 . for the discrete subgroups of @xmath71 , the conditions on @xmath333 that are necessary and sufficient that @xmath334 can be shown to be precisely the conditions that are necessary and sufficient for the commutator subgroups @xmath335 and @xmath336 to be equal ( see nicks and parry @xcite ) . in this section we state these conditions without proof . proofs of the statements below can be found in @xcite or @xcite . let @xmath333 be elements of @xmath36 . that is , @xmath337 is a word in the generators @xmath338 of @xmath313 and we can use the commutator relations to write @xmath339 if @xmath340 then clearly we must have @xmath341 since @xmath342 it must also be the case that @xmath343 if we are to have @xmath334 . due to the particular form of the commutator relations for the generators @xmath344 of @xmath313 this implies that @xmath345 . in that case , the following lemma holds . [ lem : change ] let @xmath346 be given by , let @xmath347 and suppose that @xmath348 . then there is a set of generators of @xmath349 , denoted @xmath350 , such that @xmath351 see nicks and parry @xcite for the proof of this lemma . it is then shown in nicks and parry @xcite that if one defines @xmath352 by @xmath353 where the values of @xmath354 are as in , then conditions necessary and sufficient that @xmath334 are that @xmath355 where the components of @xmath356 , are @xmath357 . from now on we will drop the primes on the @xmath358 , @xmath359 in the definitions of @xmath360 , @xmath361 . in the previous section we gave details of the possible changes of generators of a discrete subgroup @xmath311 . these are the symmetries of @xmath36 . we now consider which of these symmetries extend to automorphisms of the discrete subgroup @xmath36 . here we state a result ( lemma [ lem : auto ] ) that gives necessary and sufficient conditions for a change of generators of @xmath36 to extend to an automorphism of @xmath36 . the proof of the lemma is given in nicks and parry @xcite , and follows from results of johnson @xcite and magnus , karrass and solitar @xcite concerning free substitutions and automorphisms . let @xmath362 as in satisfy the conditions stated in the previous section so that they generate the group @xmath313 ( which is isomorphic to @xmath36 ) . recall that @xmath344 given by also generate @xmath313 . thus @xmath363 and @xmath364 can each be written as a word in @xmath365 and their inverses . moreover , the commutator relations can be expressed as relations in terms of the generators @xmath365 . we define mutually inverse mappings @xmath366 between the sets of generators @xmath367 , @xmath368 by @xmath369 and the following lemma holds . [ lem : auto]let the mappings @xmath6 and @xmath370 be as defined in . these mappings extend to mutually inverse automorphisms @xmath371 , @xmath372 of @xmath313 if 1 . the commutator relations continue to hold when @xmath373 , @xmath374 and @xmath364 are replaced by @xmath375 , @xmath376 and @xmath377 respectively ; and 2 . the relations in terms of the generators @xmath365 obtained from the commutator relations continue to hold when @xmath378 are replaced by @xmath379 respectively . conversely , if @xmath371 and @xmath380 are automorphisms of @xmath313 then conditions ( i ) and ( ii ) hold , and in addition the commutator relations continue to hold when @xmath373 , @xmath374 and @xmath364 are replaced by @xmath381 , @xmath382 and @xmath383 respectively . see nicks and parry @xcite for the proof of this lemma . we now use this result to compute the automorphisms of @xmath313 . suppose that @xmath6 and @xmath370 are the changes of generators of @xmath313 defined in and they satisfy the conditions in section [ sec : symms ] . furthermore , let @xmath384 for @xmath385 , @xmath386 . if @xmath6 is to extend to an automorphism @xmath371 of @xmath313 then by lemma [ lem : auto ] and the third commutator relation of the mapping must satisfy @xmath387 which expresses the fact that @xmath376 and @xmath377 must commute . this implies that we must have @xmath388 in . similarly we deduce that @xmath389 in if @xmath370 extends to an automorphism @xmath372 . since @xmath371 and @xmath372 are to be mutually inverse automorphisms @xmath390 must be the identity and therefore @xmath391 , @xmath392 , @xmath393 , from which we deduce that @xmath394 thus @xmath395 and @xmath396 . finally , from the first two commutator relations of we must have @xmath397 which can be shown to imply that the matrix @xmath398 defined in must satisfy @xmath399 for the given matrix @xmath78 . further details of these computations are given in nicks and parry @xcite . recall that the discrete subgroup @xmath313 depends on a given matrix @xmath309 . the conditions that a change of generators @xmath6 of @xmath313 extends to an automorphism of @xmath313 can be summarized as @xmath400 where @xmath401 , @xmath402 are arbitrary integers and the matrix @xmath398 of the exponents @xmath403 , @xmath404 , defined in , satisfies condition . these conditions are also sufficient that the conditions of lemma [ lem : auto ] hold . hence it remains to determine the matrices @xmath398 satisfying . by the results obtained in section [ sec : auto ] , in order to determine the changes of generators @xmath6 which extend to automorphisms of @xmath313 it remains to compute the @xmath164 matrices of exponents , denoted @xmath398 above , that satisfy condition for a given matrix @xmath309 . we now summarize here the work of baake and roberts @xcite on how to determine the matrices @xmath77 satisfying for a given matrix @xmath309 with @xmath405 . as in @xcite , we define the set of _ symmetries _ of the matrix @xmath309 as @xmath406 this is a subgroup of @xmath407 and is the centralizer of @xmath170 in @xmath407 . if @xmath77 satisfies @xmath408 then we say that @xmath398 is a reversing symmetry of @xmath170 and when such a @xmath398 exists we call @xmath170 reversible . we define the following subgroup of @xmath407 as the reversing symmetry group of @xmath170 , @xmath409 it is a subgroup of the normalizer of the group generated by @xmath170 in @xmath407 and clearly @xmath410 . ( if @xmath411 , the normalizer of @xmath412 in @xmath65 is @xmath413 ) . moreover , @xmath414 is a normal subgroup of @xmath415 . given a matrix @xmath309 with @xmath405 we want to compute @xmath415 . as observed earlier , when @xmath176 , @xmath170 can only lie on a one - parameter subgroup of @xmath165 if @xmath204 . clearly , in this case @xmath415 is @xmath407 . if @xmath232 then there are potentially two possibilities 1 . this occurs in cases where @xmath170 is not reversible ; 2 . the index of @xmath414 in @xmath415 is @xmath417 so that @xmath415 is a @xmath418-extension of @xmath414 ( where @xmath418 is the cyclic group of order @xmath417 ) . for a given matrix @xmath170 with @xmath232 there is a finite algorithm for computing @xmath415 which we briefly summarize following baake and roberts @xcite . clearly @xmath419 for all @xmath420 and we can observe that * if @xmath421 then @xmath422 ; * if @xmath423 then @xmath424 ; * and if @xmath425 then @xmath426 . does @xmath414 contain any other matrices @xmath398 ? when @xmath232 the eigenvalues of @xmath170 , @xmath169 and @xmath427 , are distinct . consequently @xmath170 can be diagonalised by a matrix @xmath428 which has entries in @xmath429 , the smallest field extension of the rationals that contains @xmath169 , so that @xmath430 suppose that @xmath431 . then @xmath432 commutes with @xmath433 and since only diagonal matrices can commute with a diagonal matrix with different diagonal entries , @xmath434 thus @xmath414 contains all matrices @xmath77 which are diagonalised by @xmath428 . since @xmath77 , it has eigenvalues @xmath435 which are algebraic integers . also @xmath436 and hence @xmath437 and @xmath438 are units in @xmath439 , the maximal order of @xmath429 ( i.e the intersection of @xmath429 with the set of algebraic integers ) . thus @xmath414 is isomorphic to a subgroup of the unit group of @xmath439 . * when @xmath423 , @xmath440 and the unit group is isomorphic to @xmath441 . since @xmath170 commutes with each of these matrices , @xmath442 . * when @xmath443 , @xmath444 and the unit group is isomorphic to @xmath445 . since @xmath170 commutes with each of these matrices , @xmath446 ( see @xcite , @xcite ) . in particular , when @xmath232 the group of matrices @xmath414 which commute with @xmath170 is finite . in order to compute @xmath415 for a given matrix @xmath170 we need to determine the matrices @xmath447 such that @xmath448 . recall that either @xmath416 ( if no such @xmath449 exists ) or @xmath415 is a @xmath418-extension of @xmath414 . also , if @xmath170 has a reversing symmetry @xmath449 then all other reversing symmetries of @xmath170 are obtained as @xmath450 where @xmath431 . for any matrix @xmath309 , @xmath451 . moreover it can be shown that when @xmath232 , @xmath170 is conjugate in @xmath407 to @xmath452 that is , there is only one conjugacy class of matrices for each value of @xmath453 and hence @xmath309 must be conjugate in @xmath407 to its inverse . in other words , for any @xmath309 with @xmath453 there exists a matrix @xmath454 such that @xmath448 . hence @xmath170 has a reversing symmetry @xmath449 and @xmath415 is a @xmath418-extension of @xmath414 . this implies that @xmath455 notice that for matrices @xmath309 with @xmath453 there is only a finite number of symmetries and reversing symmetries of @xmath170 but the number of changes of generators @xmath6 which extend to automorphisms of the corresponding discrete subgroup @xmath313 is infinite due to the fact that the choice of the exponents @xmath456 and @xmath457 in is arbitrary . also , note that if @xmath458 for a given @xmath309 with @xmath232 then the entries @xmath459 , @xmath460 in the matrix @xmath398 can be expressed in terms of the entries @xmath461 , @xmath462 and the entries @xmath463 of @xmath170 . we next discuss which of the automorphisms of @xmath36 computed in section [ sec : automorphisms ] extend to automorphisms of the continuous lie group @xmath71 . recall that this enables us to classify the symmetries ( changes of generators ) @xmath6 of @xmath36 given in section [ sec : symms ] as 1 . ` elastic ' if @xmath6 extends to an automorphism @xmath371 of @xmath36 and @xmath371 extends uniquely to an automorphism @xmath464 . these changes of generators are restrictions of elastic deformations of the continuum crystal whose distribution of defects is uniform and has corresponding lie group @xmath71 . 2 . ` inelastic ' if 1 . if @xmath6 does not extend to an automorphism of @xmath36 ; or 2 . if @xmath6 extends to an automorphism @xmath371 of @xmath36 but @xmath371 does not extend uniquely to an automorphism of @xmath71 . + these changes of generators preserve the set of points in @xmath36 but not its group structure . in this section we will show by direct calculation that for crystals with underlying lie group @xmath71 there are no changes of generators @xmath6 in the class 2(a ) above . that is ; all automorphisms of @xmath465 extend uniquely to automorphisms of @xmath71 . this is a computation which has not been required in the analysis of other classes of crystals with uniform distributions of defects , where the underlying lie group if nilpotent or in the solvable class @xmath70 . in those cases results of malcev @xcite and gorbatsevich @xcite respectively guarantee that any automorphism of a uniform discrete subgroup will extend uniquely to an automorphism of the ambient continuous lie group . since no corresponding general result exists for automorphisms of the class of solvable lie group @xmath71 we proceed by direct calculation . we show explicitly how each of the automorphisms of @xmath36 computed in section [ sec : automorphisms ] extends uniquely to an automorphisms of @xmath71 computed in section [ sec : s2auto ] . we should note first that a particular discrete subgroup @xmath36 ( determined by the matrix @xmath309 with @xmath405 ) is a subgroup of infinitely many ( isomorphic ) lie groups @xmath71 corresponding to different choices of @xmath137 where @xmath466 . ( recall that @xmath467 for any @xmath468 where @xmath469 is the value of @xmath137 in with @xmath470 , so there are infinitely many possible values of @xmath137 for a given value of @xmath405 . choosing one of these values of @xmath137 determines the matrix @xmath205 which specifies a particular lie group in the isomorphism class of @xmath71 . ) assume that we are given @xmath170 with @xmath405 and a corresponding fixed value of @xmath137 . thus a particular group @xmath471 and discrete subgroup @xmath472 are determined . here we will show that the automorphisms of @xmath473 extend uniquely to automorphisms of @xmath471 . notice that in section [ sec : s2auto ] we computed the automorphisms of @xmath156 with respect to the basis @xmath233 and the automorphisms of @xmath313 in section [ sec : automorphisms ] are given in terms of changes of generators expressed with respect to the basis @xmath85 . the change of basis is specified by the matrix @xmath242 in which depends on @xmath137 . let @xmath474 be an automorphism and suppose that it extends to an automorphism @xmath464 . we show that this extension exists and is unique . let @xmath475 and @xmath476 for some @xmath477 . thus @xmath478 if @xmath479 extends to @xmath480 then @xmath481 for all @xmath482 . in section [ sec : automorphisms ] we computed all automorphisms @xmath483 of @xmath313 and these automorphisms are in one to one correspondence with the automorphisms of @xmath36 since they satisfy @xmath484 therefore @xmath485 \\ q\zeta\end{array}\right),\end{aligned}\ ] ] with respect to the basis @xmath233 , recalling the definition of the matrix of exponents @xmath398 from and where @xmath486 from the automorphism @xmath480 is uniquely determined by the values of @xmath248 and @xmath150 and satisfies @xmath487 where @xmath488 ( the top left @xmath164 matrix contained in @xmath242 ) and the quantities @xmath489 and @xmath490 are as defined in section [ sec : s2auto ] . here the components are again given with respect to the basis @xmath233 . comparing and we see that if @xmath479 is to extend to @xmath480 , it must be the case that @xmath491 and furthermore , for all @xmath477 @xmath492= \\ & & \qquad \qquad w(\epsilon ) \left(\begin{array}{cc } \alpha & \beta \\ -\beta & \alpha\end{array } \right ) \overline{m}^{-t } \theta^q \left(\begin{array}{c } m \\ n\end{array}\right ) + f(\mathcal{b}(-1)^\epsilon q ) w(\epsilon ) q\left(\begin{array}{c } \gamma \\ \delta \end{array}\right ) . \label{eq : extcond } \end{aligned}\ ] ] thus we must have @xmath493 where @xmath494 in . using the fact that @xmath495 since @xmath479 is an automorphism , from we find @xmath496 and hence the real numbers @xmath497 and @xmath498 are uniquely determined by the values of @xmath499 and @xmath398 . furthermore , let us define @xmath500 then we can note from that if @xmath501 , the matrix @xmath490 has a well defined inverse and we can write @xmath502 ( if @xmath503 then both sides of are zero and we gain no information about @xmath504 and @xmath505 for given values of @xmath456 and @xmath506 . ) it can be observed that the value of the matrix @xmath507 is independent of @xmath501 and is given by @xmath508 since this matrix is nonsingular , the values of @xmath504 and @xmath505 are determined uniquely by the values of the exponents @xmath456 and @xmath506 for a given automorphism of @xmath36 . thus we have shown directly that every automorphisms @xmath474 extends uniquely to an automorphism @xmath464 for a given value of @xmath137 . in particular , we have shown that for a given automorphism @xmath479 of @xmath36 specified by the matrix @xmath398 , @xmath456 , @xmath509 and @xmath248 there are unique values of @xmath510 and @xmath511 which specify the automorphism @xmath464 to which @xmath479 extends . the technical calculations given in this paper have allowed us to complete the classification of symmetries of crystals with uniform distributions of defects , which preserve the lie group structure . let us now briefly summarize the results . the energy density @xmath24 of a crystal with a uniform distribution of defects depends on arguments @xmath512 and @xmath23 , where the vectors @xmath26 specify a set of generators of a discrete subgroup @xmath36 of the lie group associated with the given constant value of the dislocation density tensor @xmath23 . the underlying lie group must be in one of three classes : it is either nilpotent or isomorphic to one of two classes of solvable lie groups denoted @xmath70 and @xmath71 . in this paper we have focused on the case where the lie group is a solvable group in the class @xmath71 . a symmetry of the energy density function is a change of generators of @xmath36 which preserves the set of points in @xmath0 associated with the elements of @xmath36 . these symmetries have been shown to satisfy conditions for subgroups @xmath311 . of these changes of generators , only those which extend to automorphisms @xmath479 of @xmath36 also preserve the group structure . these are the changes of generators which satisfy condition . these symmetries also have been shown in this paper to extend uniquely to automorphisms of the underlying continuous lie group @xmath71 and hence they are ( restrictions of ) elastic deformations of the continuum defective crystal . we call such symmetries of @xmath36 ` elastic ' . the remaining symmetries of @xmath36 which do not extend to automorphisms of @xmath36 ( or @xmath71 ) are classified as inelastic . together with work presented in @xcite and @xcite , this completes the classification of all group preserving symmetries of uniform discrete defective crystals as either elastic or inelastic . a future task will be to understand the properties of the inelastic symmetries with reference to corresponding mechanical problems including exploring any possible correlations between these symmetries and the presence of slip planes in crystalline materials . the author acknowledges the support of the uk engineering and physical sciences research council through grant ep / g047162/1 . the author wishes to thank gareth parry for many useful discussions and his advice concerning this paper .

we use quantum monte carlo simulations to study a finite - temperature dimensional - crossover - driven evolution of spin and charge dynamics in an anisotropic two - dimensional system of weakly coupled hubbard chains with a half - filled band . the low - temperature behavior of the charge gap indicates a crossover between two distinct energy scales : a high - energy one - dimensional ( 1d ) mott gap due to the umklapp process and a low - energy gap which stems from long - range antiferromagnetic ( af ) spin fluctuations . away from the 1d regime and at temperature scales above the charge gap , the emergence of a zero - frequency drude - like feature in the interchain optical conductivity implies the onset of a higher - dimensional metal . in this metallic phase , enhanced quasiparticle scattering off finite - range af spin fluctuations results in incoherent single - particle dynamics . the coupling between spin and charge fluctuations is also seen in the spin dynamical structure factor displaying damped spin excitations ( paramagnons ) close to the af wave - vector and particle - hole continua near 1d momentum transfers spanning quasiparticles at the fermi surface . we relate our results to the charge deconfinement in quasi-1d organic bechgaard - fabre salts .

1412.0287

dimensional crossovers are interesting because there are relevant couplings between the lower dimensional objects which dramatically alter the physical properties of the system . @xcite the renormalization at low energy and long wavelengths is therefore higher - dimensional , but the question of how remnants of the one - dimensional ( 1d ) physics remain visible when the couplings get stronger is largely left open . one would expect the 1d characteristics to survive when probing the system at high energy or when the temperature is larger than the interchain coupling but lower than the typical scales of the 1d problem . below we give three examples of experiments in different fields whose complete description requires a deeper understanding of dimensional crossover phenomena . cold atoms in optical lattices provide a controllable and flexible model realization of correlated quantum systems and offer a clean setup to probe their dynamics . @xcite for example , a variable strength of the interchain coupling in a recent realization of a tunable optical lattice comprising weakly coupled 1d chains , @xcite allows one to study the impact of a dimensional crossover on antiferromagnetic ( af ) spin correlations and stimulated a renewed interest in low - dimensional quantum many - body physics . @xcite other systems to explore the interplay between low - dimensional quantum dynamics and electron correlations are quasi-1d organic bechgaard - fabre salts . @xcite a rich variety of phenomena in their global temperature - pressure phase diagram has been ascribed to a decreasing degree of dimerization , i.e. , the strength of the umklapp process , @xcite and increasing electronic dimensionality with applied pressure which triggers a metal - insulator transition . @xcite whether the nature of this higher - dimensional metallic phase and its low - energy excitations can be accounted for by a fermi liquid ( fl ) theory is the key question in the physics of these compounds . finally , fingerprints of a strongly anisotropic metallic phase have been observed in the proximity to the af ground state in high-@xmath3 cuprate superconductors . @xcite in this case , a dimensional reduction is driven by an inhomogeneous self - organization of doped holes which condense into arrays of parallel stripes separating regions with enhanced af spin correlations . since the charge dynamics occurs mainly along the stripes , it can be effectively described by a quasi-1d model in which the transport across the stripes is incoherent . @xcite from the theoretical point of view , the complexity of a dimensional crossover in coupled 1d hubbard chains comes from single- and two - particle processes generated by the interchain coupling . @xcite on the one hand , the crossover in metallic chains is easily induced by the interchain _ one - particle _ hopping process thus replacing the luttinger liquid ( ll ) behavior with a conventional fl metal . @xcite on the other hand , the umklapp - induced mott gap in a half - filled band makes the problem more difficult due to the enhanced relevance of _ two - particle _ fluctuations : binding of particle - hole pairs from neighboring chains generates a finite af superexchange coupling @xmath4 and may induce an onset of the broken - symmetry spin - density - wave ( sdw ) phase . @xcite in contrast , when the single - particle tunneling @xmath5 overcomes the magnetic coupling @xmath4 , it drives a metal - insulator transition thus leading to the formation of a higher - dimensional fermi surface ( fs ) . @xcite recently , we have studied the effects of interchain coupling between the half - filled hubbard chains in two limiting cases in the effective _ zero_-temperature regime . @xcite first , the dimensional crossover dominated by single - particle tunneling can be examined with a cluster extension of the dynamical mean - field theory ( cdmft ) . @xcite on the one hand , cdmft neglects most of the spatial magnetic fluctuations restricted to the cluster size . on the other hand , it captures well umklapp scattering and thus reproduces the density - driven mott transition in the 1d hubbard model . @xcite the cdmft scenario of the dimensional - crossover - driven metallic phase with a fs broken into pockets is plausible when the low - temperature sdw instability of the nested fs is eliminated , e.g. , by a sufficiently large geometrical frustration ; a similar broken fs has also been found in the spinless fermion model . @xcite second , the two - particle crossover and a dynamical confinement which occurs , for instance , in a system of coupled spin-1/2 heisenberg chains , @xcite can be addressed with lattice quantum monte carlo ( qmc ) simulations by tracking the evolution of elementary spin excitations . @xcite indeed , the interchain af superexchange coupling confines fractionalized excitations ( spinons ) of the individual 1d chains back to form @xmath6 magnons , the goldstone modes of the broken continuous su(2 ) symmetry group . this gives rise to low - frequency spin waves which decay into a two - spinon continuum at high energies seen in the qmc dynamical spin structure factor @xmath1 . @xcite such a dual nature of magnetic excitation spectra has been resolved in the inelastic neutron scattering data on weakly coupled @xmath7 chains of bacu@xmath8si@xmath8o@xmath9 and kcuf@xmath10 . @xcite the present work is aimed at investigating the regime in which the single- and two - particle interchain fluctuations intertwine and contribute to the low - frequency dynamics of the system . in a correlated metal , a mutual interaction between finite - range spin fluctuations and fermionic quasiparticles ( qps ) introduces damping of the collective spin excitations ( paramagnons ) which decay into a continuum of independent electron - hole pairs . @xcite concurrently , dressing of qps with a cloud of spin fluctuations enhances the qp scattering rate and leads to a partial depletion of the single - particle spectral weight at the fermi level . @xcite this non - fermi - liquid behavior is frequently reflected in anomalous transport and thermodynamic properties of nearly af metals and its theoretical description has proven to be a challenge . @xcite to treat the single - particle tunneling and dynamically generated interchain superexchange interaction on equal footing , we use an auxiliary - field qmc algorithm . @xcite we characterize the low - frequency excitations in the quasi-1d metallic phase by looking at both momentum - resolved single - particle spectral function and two - particle quantities : frequency- and polarization dependent optical conductivity as well as spin and charge dynamical structure factors . the paper is organized as follows : after introducing the model and method in sec . [ model ] , we present our results in sec . [ results ] , and then conclude in sec . [ conclusions ] . we consider the hubbard model on a square lattice with an anisotropic hopping at half - filling , @xmath11 where the hopping @xmath12 is @xmath13 ( @xmath5 ) on the intrachain ( interchain ) bonds , extended by a _ diagonal _ next - nearest neighbor hopping @xmath14 . thus , the corresponding single - particle dispersion relation reads , @xmath15 the diagonal hopping @xmath16 brings about geometrical frustration thus reducing fs nesting properties responsible for the onset of the long - range af phase in coupled spin @xmath7 heisenberg chains . @xcite to account for broken particle - hole symmetry and force a half - band filling away from the 1d limit , we adjust the chemical potential @xmath17 . previously , @xcite we have shown that the hubbard model ( [ hubb ] ) with @xmath18 develops nel order in the presence of any infinitesimally small interchain coupling in the effective zero - temperature limit . here , we consider a smaller value of the coulomb interaction @xmath19 and study _ finite_-temperature properties of the emergent quasi-1d ( @xmath20 . ) metallic phase . for this purpose , we use a finite - temperature implementation of the auxiliary - field qmc algorithm ( see ref . and references therein ) which allows one to compute the expectation value of an observable @xmath21 in the grand - canonical ensemble : @xmath22}{{\mbox{tr}}\bigl [ e^{-\beta(h-\mu n)}\bigr ] } .\ ] ] it is based on a path integral formulation of the partition function which maps a quantum system in @xmath23 spatial dimensions onto a @xmath24-dimensional classical problem with an additional imaginary time dimension @xmath25 . the essence of the qmc method is to separate the one - body kinetic @xmath26 and two - body hubbard interaction @xmath27 terms with the help of the trotter decomposition , @xmath28 we have used a fixed small discretization of the temporal axis @xmath29 . this introduces an overall controlled systematic error of order @xmath30 . we have opted for a discrete , ising , hubbard - stratonovitch field coupling to the @xmath31-component of the magnetization . the qmc simulations were performed for lattice sizes ranging from @xmath32 to @xmath33 and in the broad range of temperatures @xmath34 restricted by the qmc minus - sign problem . in addition to equal - time observables , the qmc method provides access to imaginary time displaced correlation functions . we use a stochastic version of the maximum entropy method @xcite to analytically continue the imaginary - time qmc data and to extract the real - frequency single- and two - particle excitation spectra . we proceed now to present our qmc results for the hubbard model ( [ hubb ] ) . we first consider static quantities and then study single- and two - particle dynamical correlation functions . vs. @xmath35 for several values of the interchain hopping @xmath5 found on a @xmath36 lattice . the inset shows the evolution of the qp gap @xmath37 on increasing @xmath5 extracted from a linear fit . , scaledwidth=45.0% ] due to the relevance of umklapp scattering in the 1d limit , the single - particle imaginary time green s function , @xmath38 , follows for large @xmath39 an exponential decay @xmath40 , where @xmath37 is the qp gap . the latter might be extracted from the uniform charge susceptibility , @xmath41 where @xmath42 is the particle number operator . at low temperatures one expects @xmath43 . in order to quantify the behavior of the qp gap upon increasing @xmath5 , we plot in fig . [ c00 ] @xmath44 vs. @xmath35 for several values of the interchain hopping . in this way , the activated behavior is reflected in the slope of the curves given by @xmath37 : a decreased negative slope signals the reduction of the magnitude of @xmath37 . a rough estimate of the qp gap @xmath37 extracted from a linear fit of the data points indicates that it does not close but shows an inflection point around @xmath45 , see the inset in fig . . found on a 16@xmath46 lattice : ( a ) @xmath47 and ( b ) @xmath2 . the inset shows the finite - size scaling of the staggered magnetic moment @xmath48 for @xmath49 at @xmath50 , @xmath51 , and @xmath52 . , title="fig:",scaledwidth=45.0% ] + found on a 16@xmath46 lattice : ( a ) @xmath47 and ( b ) @xmath2 . the inset shows the finite - size scaling of the staggered magnetic moment @xmath48 for @xmath49 at @xmath50 , @xmath51 , and @xmath52 . , title="fig:",scaledwidth=45.0% ] the existence of the inflection point is suggestive of a crossover in the origin of the charge gap . numerical evidence for this conjecture is provided in fig . [ s_t ] showing the fourier transform of equal - time spin - spin correlations , @xmath53 on the one hand , the low-@xmath54 enhancement of @xmath55 for @xmath56 in fig . [ s_t](a ) indicates the predominance of spin - spin correlations along the chains thus confirming the relevance of umklapp scattering . on the other hand , the low-@xmath54 increase of the staggered spin structure factor @xmath57 shown in fig . [ s_t](b ) suggests the formation of significant af spin correlations . to further study the spin - spin correlations , we plot in the inset of fig . [ s_t](b ) the finite - size scaling of the staggered magnetic moment , @xmath58 at @xmath49 . as is apparent , @xmath48 is consistent with a finite value _ below _ our lowest temperature @xmath59 , thus marking the effective zero - temperature regime in our finite - size systems . hence , on the basis of the static quantities , one can conclude in the @xmath60 limit the onset of a higher - dimensional insulating phase gapped out by long - range af spin fluctuations . however , a strong reduction of the qp gap @xmath37 upon increasing @xmath5 implies that for this value of the coulomb repulsion @xmath19 , the onset of the insulating state occurs at lower temperatures in comparison to the 1d mott gap . this opens up a possibility to study finite , but still low - enough to avoid dominant thermal broadening effects , temperature properties of the quasi-1d metallic state in the dimensional crossover . a response function particularly suitable for investigating transport properties in the anisotropic system such as weakly coupled 1d hubbard chains is the frequency- and polarization - dependent optical conductivity @xmath61 : it allows one to resolve a distinct behavior of the charge dynamics along @xmath62 and perpendicular @xmath0 to the chains . noting that , experimentally , the dimensionality is controlled not only by the physical or chemical pressure , which changes the ratio of the interchain transfer integral to the intrachain one , but also by the energy scale used in the measurement , the knowledge of @xmath61 offers a possibility to track the evolution of remnant aspects of the 1d physics in the high - energy part of the spectrum . finally , optical conductivity is a useful experimental @xcite and numerical @xcite response to study the interplay between the qp itineracy and the tendency towards their localization by dressing with af spin fluctuations . a real part of @xmath61 is a measure of the rate at which particle - hole pairs are created by the absorption of photons with a given frequency @xmath63 . it might be calculated from the kubo formula by looking at the qmc imaginary time current - current correlation functions , @xmath64 here , @xmath65 ( @xmath66 ) is the intrachain ( interchain ) component of the current operator : @xmath67 respectively , with @xmath68 being the @xmath69 component of the vector connecting sites @xmath70 and @xmath71 . a dimensional - crossover - driven reconstruction of electronic states as evinced by frequency - dependent intra- @xmath62 and interchain @xmath0 optical conductivities at a given temperature @xmath72 is shown in fig . [ sig_20 ] . at our smallest interchain coupling @xmath73 , both optical conductivities display solely a finite - frequency feature typical of the 1d mott insulating phase . @xcite its location @xmath74 roughly matches _ twice _ the magnitude of the single - particle gap @xmath37 , cf . [ c00 ] , thus corresponding to the particle - hole pair absorption across the lower and upper hubbard bands . while some smearing of this absorption mode becomes apparent , the peak retains its position in the weakly coupled regime with @xmath75 . it reflects the relevance of umklapp processes and indicates charge confinement to the individual chains . a further increase in @xmath5 renders umklapp scattering progressively irrelevant and the system enters a transient regime with a competing interchain single - particle tunneling . as a result , a drude - like response in @xmath62 develops around @xmath76 accompanied by a tiny zero - frequency weight in @xmath0 . still , given a nearly @xmath5-independent position of the finite - frequency absorption , the overall structure of @xmath62 is reminiscent of what is expected for a weakly doped 1d mott insulator . @xcite although there is no actual doping in the system , a non - negligible warping of the fs introduced by @xmath5 might be considered as an effective deviation from the commensurate filling of individual chains which otherwise continue to exhibit a substantial tendency to confine charge carriers reflected in a strongly reduced zero - frequency weight in @xmath0 . . the plot shows : ( a ) intra- @xmath62 and ( b ) interchain @xmath0 optical spectra found on a @xmath36 lattice at @xmath72 . from bottom to top : @xmath73 , 0.1 , 0.15 , 0.2 , 0.25 , and 0.3 . , scaledwidth=45.0% ] at larger coupling @xmath49 , a pronounced drude - like feature in @xmath0 signals the onset of a higher - dimensional metallic phase : single - particle _ interchain _ tunneling requires the recombination of fractionalized excitations characteristic of the 1d regime into electronic qps . the crossover in the effective dimensionality is not restricted to the lowest energies : a high - energy mode in @xmath62 , remnant of the 1d mott gap , is replaced by a low - frequency hump whose evolution tracks the qp energy scale @xmath37 . the hump stems from coupling between charge carriers and short - range af spin fluctuations beyond the 1d framework . a similar piling up of optical weight at finite frequency is also resolved in @xmath0 for @xmath77 and then , in contrast to @xmath62 , the hump merges with the high - frequency tail of the drude mode . we attribute this difference to a spatial anisotropy in the spin - wave velocity ; the _ smaller _ velocity associated with the _ interchain _ spin - wave - like dispersion relation renders the paramagnon decay rate into particle - hole excitations _ higher _ thus giving rise to a fl - like response in @xmath0 . . the plot shows : ( a ) intra- @xmath62 and ( b ) interchain @xmath0 optical spectra at @xmath49 ( solid ) and @xmath78 ( dashed ) . from top to bottom : @xmath50 , @xmath79 , @xmath51 , and @xmath52 . for @xmath78 , a complete optical gap develops at @xmath72 and we expect at lower temperatures qualitatively similar spectra ( not shown ) . , scaledwidth=45.0% ] inspection of the static quantities in sec . [ static ] revealed the formation of a small qp gap @xmath37 associated with the onset of the sdw state in the @xmath80 limit . it is interesting to explore the behavior of this gap as a function of temperature . thermal evolution of @xmath61 at a fixed value of the interchain coupling @xmath49 is shown in fig . [ sig_t ] . at elevated temperature @xmath50 , both intra- @xmath62 and interchain @xmath0 optical conductivities exhibit broad drude - like features exhausting most of the optical weight . this indicates a fl - like metal albeit with anisotropic transport properties due to inequivalent hopping amplitudes . as the temperature is decreased down to @xmath72 , a low - frequency depletion of optical weight separating the narrow drude - like mode from the finite - frequency hump becomes apparent . it follows from a marked enhancement of the af spin correlation length and dressing of qps with a cloud of spin fluctuations . upon further cooling , the af spin correlations extend across the entire lattice and the sdw gap opens up . we now compare the above evolution of @xmath61 with the corresponding redistribution of optical weight at a twice _ smaller _ interchain coupling @xmath78 . the low - temperature increase of the static spin structure factor @xmath55 reveals dominant 1d spin correlations in this regime , see fig . [ s_t](a ) . a pronounced 1d character of the system is also reflected in a different frequency dependence of intra- and interchain optical conductivities . this effect is particularly dramatic at @xmath50 : while a clear finite - frequency depletion of optical weight is visible in @xmath62 , only a narrow drude - like peak is resolved in @xmath0 . in spite of thermal broadening comparable with the interchain bandwidth masking partially warping of the fs , the presence of the zero - frequency mode in @xmath0 suggests that charge carriers retain their electronic qp nature . this indicates that thermal melting of the quasi-1d mott insulator induces at the _ smallest _ energy scales a crossover to the fl characteristics rather than to the ll behavior . furthermore , at the intermediate low temperature @xmath81 , the frequency range with a reduced weight in @xmath62 becomes wider thus signaling proximity to the insulating phase . a reduced mobility of charge carriers also affects @xmath0 ; here , most of the drude - like weight is transferred to a high - energy feature . finally , a complete optical gap driven by umklapp scattering develops already at @xmath72 . our next aim is to address momentum - resolved single - particle spectral properties . in particular , one would like to know ( i ) whether the quasi-1d metallic phase can be described within the fl theory , and ( ii ) topology and qp weight along the emergent fs . another interesting question is whether some spectral features in the higher - dimensional electronic structure can be traced back to those of the 1d regime . @xcite to gain insight into these issues , we compute the momentum - resolved single - particle spectral function @xmath82 which is related to the qmc imaginary time green s function @xmath83 by the spectral theorem : @xmath84 obtained at @xmath72 on : ( a ) 32-site chain with a dummy transverse momentum and ( b),(c ) @xmath36 lattice with : @xmath45 ( middle ) and @xmath85 ( bottom ) . solid line in panel ( c ) gives the paramagnetic hartee - fock band structure . , title="fig:",scaledwidth=43.0% ] + obtained at @xmath72 on : ( a ) 32-site chain with a dummy transverse momentum and ( b),(c ) @xmath36 lattice with : @xmath45 ( middle ) and @xmath85 ( bottom ) . solid line in panel ( c ) gives the paramagnetic hartee - fock band structure . , title="fig:",scaledwidth=43.0% ] + obtained at @xmath72 on : ( a ) 32-site chain with a dummy transverse momentum and ( b),(c ) @xmath36 lattice with : @xmath45 ( middle ) and @xmath85 ( bottom ) . solid line in panel ( c ) gives the paramagnetic hartee - fock band structure . , title="fig:",scaledwidth=43.0% ] the evolution of @xmath82 with increasing interchain coupling @xmath5 is summarized in fig . [ akw ] . in the 1d regime , the relevance of umklapp process generates a gap in the half - filled band at @xmath86 and the equivalent points , see fig . [ akw](a ) . at frequencies above the single - particle gap , the dispersion is approximately linear reflecting aspects of the ll theory . @xcite moreover , the spectrum significantly broadens on approaching the brillouin zone edges reminiscent of the spinon and holon branches . @xcite next , fig . [ akw](b ) provides evidence for highly incoherent single - particle dynamics in the metallic regime with @xmath45 ; a significant depletion of spectral weight at the @xmath87 momentum is accompanied by a backfolding behavior in the dispersion at @xmath88 and @xmath89 points . upon further increase in the interchain coupling strength , the spectral intensity at the @xmath87 point gradually recovers . finally , a single qp peak becomes apparent at the fermi level at @xmath85 and the low - frequency part of @xmath82 follows roughly the paramagnetic hartee - fock band structure , see fig . [ akw](c ) . another noteworthy feature is found in the inverse photoemission @xmath90 part of the spectra , i.e. , the formation of a weakly dispersive qp - like band near the @xmath91 momentum . while the flatness is reminiscent of the 1d nature of the problem , only a broad structure is resolved in the photoemission @xmath92 part around the @xmath93 point . the difference between the inverse- and photoemission parts reflects a broken particle - hole symmetry due to the finite next - nearest - neighbor hopping @xmath16 . ( solid ) and of the purely 1d case ( dashed).,scaledwidth=35.0% ] the evolution of the spectral function @xmath82 is a consequence of dramatic changes in the single - particle green s function @xmath94 . in the mott insulator , the presence of a spectral gap requires that the real part of the zero - frequency green s function @xmath95 changes sign in momentum space by going through a _ zero_. this is accomplished by the singularity in the corresponding self - energy . as one approaches the mott transition , the locus of zeros in @xmath96-space affects the topology of the emergent fs defined by zero - frequency _ poles _ of the green s function . @xcite we address this issue by examining two special momenta : ( i ) nodal @xmath87 , corresponding to a vanishing interchain kinetic energy @xmath97 , and ( ii ) an antinodal @xmath88 one , where the maximum warping of the noninteracting 1d fs occurs , see fig . [ fs ] . we focus first on the nodal @xmath87 point considered in fig . [ g_nod ] . at our smallest @xmath73 , @xmath95 has a negative slope in a broad range of frequencies around the fermi level . the imaginary part of the corresponding self - energy @xmath98 displays a diverging - like behavior on approaching the smallest matsubara frequency @xmath99 thus signaling a zero of @xmath95 , see fig . [ g_nod ] , left inset . the anomalous behavior of the self - energy stems from umklapp scattering and is responsible for a robust mott gap in the single - particle spectral function @xmath82 , right inset of fig . [ g_nod ] . at larger @xmath5 , the umklapp process becomes less effective at low - energy scales . this shrinks the frequency region with a negative slope of @xmath95 and reduces the scattering rate @xmath100 . as a result , some thermally excited single - particle states whose weight is controlled by @xmath5 become apparent at the fermi energy . finally , @xmath95 develops a positive slope at @xmath85 thus forming a pole - like structure as in the fl phase . still , a small kink at @xmath101 signals substantial qp scattering off af spin fluctuations . consequently , a broad qp - like feature is resolved in @xmath82 . at the nodal @xmath87 point . insets show the corresponding : ( left ) low - frequency dependence of the imaginary part of the self - energy and ( right ) single - particle spectral function from bottom to top : @xmath73 , 0.1 , 0.15 , 0.2 , 0.25 , and 0.3 . , scaledwidth=45.0% ] we turn now to the antinodal @xmath88 point . as shown in fig . [ g_anod ] , the zero of @xmath95 and the mott gap in the spectral function @xmath82 remain for small values of @xmath102 nearly pinned at the @xmath88 momentum . hence , at the expense of loss in the _ interchain _ kinetic energy , the interaction renders the fs warping tendency irrelevant . @xcite in contrast , at larger interchain hopping amplitude , a reduced scattering rate @xmath103 indicates that the kinetic energy gain can not be further ignored and the warping effects become discernible . indeed , vanishing @xmath104 at the @xmath88 point requires now a finite frequency @xmath90 thus approaching a pole - like behavior around @xmath105 . the latter produces a faint spectral feature in @xmath82 ; it signals backfolding of the conduction band and as such is a fingerprint of the mott gap slightly off the @xmath88 momentum . finally , the finite - frequency zero and pole - like features annihilate together and only a shallow dip in @xmath95 is resolved at @xmath45 which in turn evolves into a smooth behavior at @xmath49 . obviously , we can not resolve the fermi wave vector @xmath106 across the @xmath107 path on our discrete @xmath96-point mesh . given , however , a comparable slope of the qp energy dispersion near the antinodal @xmath88 and nodal @xmath87 points , see fig . [ akw](c ) , we expect a _ continuous _ warped fs . this should be contrasted with the isotropic two - dimensional hubbard model at half - filling where the flatness of the qp band around the @xmath91 point renormalizes locally the fermi velocity and enhances qp scattering off af spin fluctuations . as a result , the depletion of qp weight starts at different temperatures in different regions of the brillouin zone . @xcite however , this momentum - sector - selective opening of the single - particle spectral gap is suppressed by next - nearest - neighbor hopping @xmath16 . @xcite the latter shifts the flat region in the qp dispersion away from the fermi level thus providing further support for the continuous fs in the quasi-1d regime . but at the antinodal @xmath88 point . , scaledwidth=45.0% ] the evolution of the qmc green s function @xmath94 and the shape of the emergent fs are different from those concluded on the basis of the cdmft approach on the @xmath108 cluster . @xcite in cdmft , the range of af spin fluctuations is restricted to the cluster size . therefore , the evolution of the electronic structure is governed by dominant umklapp process : a finite @xmath5 shifts the _ zero _ of @xmath95 away from the @xmath109 points giving way to a finite - frequency _ pole _ from the conduction band , see fig . 5(a ) in ref . . at a critical value of @xmath5 , the latter crosses the fermi level which opens up elongated _ electron _ pockets centered around the @xmath110 momenta . at the same time , _ hole _ pockets open up at the @xmath111 points and thus a _ broken _ fs emerges with the 1d nesting wave vector @xmath91 . hence , in the absence of finite - range af spin fluctuations , this particular evolution of @xmath95 ensures the relevance of umklapp scattering and extends the stability of the mott phase away from the 1d regime . let us however point out that a larger next - nearest - neighbor hopping @xmath16 in the lattice calculations will frustrate the af spin correlations and could bring our findings in line with the cdmft results . unfortunately , this leads to a sign problem which renders qmc simulations very expensive . competing effects in strongly correlated systems often result in various energy scales which govern the population of excited states . this also holds for weakly coupled 1d hubbard chains with a half - filled band : in the regime where the interaction between spin and electronic degrees of freedom becomes relevant , both single- and two - particle processes contribute to the excited state dynamics . the energy- and momentum - resolved spin and charge structure factors help to separate these contributions which is crucial for the understanding of the low - energy dynamics . in addition , these quantities allow one to identify momentum - resolved energy scales above which spin and charge fluctuations lose their 1d nature . thus , to complement the characteristic of the emergent quasi-1d metallic phase , we consider the spin @xmath1 and charge @xmath112 dynamical structure factors defined as : @xmath113 here , @xmath114 and @xmath115 correspond to the generalized charge and spin susceptibilities . the susceptibilities can be obtained from the imaginary - time displaced two - particle correlation functions , @xmath116 where , @xmath117 with @xmath118 being the average filling level . the elementary excitations of the 1d hubbard model are bosonic collective spin and charge density oscillations with a linear dispersion in the long - wavelength limit @xmath119 . @xcite the relevance of umklapp scattering at half - filling opens up a gap for long - wavelength charge excitations while leaving the spin sector gapless , see fig . [ chain ] . the dynamical charge structure factor @xmath112 in fig . [ chain](a ) shows an overall agreement with the one obtained at a slightly smaller interaction @xmath120 using the time - dependent density matrix renormalization group method in the @xmath60 limit , see fig . 10 in ref . . above the charge gap , one finds aspects of the ll dynamics with low - lying charge excitations located at long wavelengths as well as at @xmath121 . as found in ref . , the latter carry little spectral weight and are thus more difficult to resolve in our finite - temperature qmc simulations . since the charge sector is fully gapped , the spin dynamics can be understood within a spin - only @xmath7 heisenberg chain . in this case , the spin dynamics is characterized by the two - spinon continuum of states bounded from below and above by , @xcite @xmath122 with the lowest - lying excitations carrying dominant spectral weight at low temperatures . @xcite as is apparent in fig . [ chain](b ) , for the considered value of the interaction @xmath19 , the majority of spectral weight is located at the lower bound of the continuum as in the heisenberg model . and ( b ) dynamical spin structure factor @xmath1 obtained on a 32-site chain at @xmath72 . the transverse momentum is a dummy label . , title="fig:",scaledwidth=43.0% ] + and ( b ) dynamical spin structure factor @xmath1 obtained on a 32-site chain at @xmath72 . the transverse momentum is a dummy label . , title="fig:",scaledwidth=43.0% ] + in spatial dimensions greater than one , the spinons bind together and form magnons , the goldstone modes of the broken continuous su(2 ) symmetry group . as a result , the continuum of excitations in the dynamical spin structure factor gives way to well defined spin - wave modes . @xcite the low - frequency magnons in af _ insulators _ and in the @xmath60 limit are well accounted for within the linear spin - wave theory ( lswt ) of the heisenberg model to leading @xmath123 order . @xcite this picture should break down at finite temperatures where cranking up the interchain coupling triggers the crossover to a _ metal_. let us consider first a weakly coupled regime with @xmath45 . as shown in fig . [ g_nod ] , the corresponding single - particle spectral function @xmath82 exhibits two broadened peaks separated by a region with a strongly reduced weight that we refer to as a _ pseudogap_. as is apparent , the system is at the verge of localization near the 1d mott phase . figure [ dyn15](a ) depicts the dynamical charge structure factor @xmath112 . using the continuity equation one finds that upon approaching the long - wavelength limit @xmath119 parallel ( perpendicular ) to the chains , this observable is related to the intrachain ( interchain ) real part of the optical conductivity @xmath61 , respectively : @xcite @xmath124 where @xmath125 is the @xmath69 component of the momentum transfer parallel / perpendicular to the chains . given that the intensity of charge fluctuations rapidly decreases in the long - wavelength limit , it is difficult to pin down the metallic vs. insulating nature of the system . this issue is resolved by looking at the low - frequency part of the optical conductivity @xmath61 , see fig . [ sig_20 ] . on the one hand , zero - frequency weight in the interchain optical conductivity @xmath0 indicates a metallic phase at @xmath45 . on the other hand , the smallness of this weight reveals a large degree of incoherence in the interchain transport . moreover , as shown in fig . [ g_nod ] , the imaginary part of the corresponding self - energy @xmath126 increases at low matsubara frequency contributing to the effective mass enhancement and _ reduced _ mobility of charge carriers . , and ( b ) dynamical spin structure factor @xmath1 obtained on a @xmath36 lattice with @xmath45 at @xmath72 . solid line in panel ( b ) gives the lswt dispersion eq . ( [ magnon ] ) with @xmath127 and @xmath128 . the chosen exchange couplings are fit parameters subject to the constraint of a finite magnetic order parameter in lswt consistent with the long - range af order in the system . , title="fig:",scaledwidth=43.0% ] + , and ( b ) dynamical spin structure factor @xmath1 obtained on a @xmath36 lattice with @xmath45 at @xmath72 . solid line in panel ( b ) gives the lswt dispersion eq . ( [ magnon ] ) with @xmath127 and @xmath128 . the chosen exchange couplings are fit parameters subject to the constraint of a finite magnetic order parameter in lswt consistent with the long - range af order in the system . , title="fig:",scaledwidth=43.0% ] + from this standpoint , we can analyze the magnetic excitation spectrum shown in fig . [ dyn15](b ) . the most striking difference with respect to the 1d regime is a low - frequency dispersive feature along the @xmath129 path accompanied by a broad upward dispersion along the @xmath130 direction . they are signatures of damped af spin fluctuations ( paramagnons ) which are not strong enough to gap out the fs and develop long - range af order but nevertheless can propagate an appreciable distance . to get further insight into the spin dynamics , we consider the spin @xmath7 heisenberg model with nearest neighbor interactions @xmath131 ( @xmath4 ) along the intrachain ( interchain ) bonds , respectively , extended by next - nearest neighbor interaction @xmath132 : @xmath133 as illustrated in fig . [ dyn15](b ) , the lswt dispersion relation of the heisenberg model ( [ hj ] ) , @xcite @xmath134 with two - dimensional structure factors , @xmath135 tracks the low - frequency paramagnon spectrum . clearly , the frequency region with depleted single - particle spectral weight efficiently separates the low - frequency spin - wave - like excitations from the high - frequency particle - hole excitations across the pseudogap in @xmath82 . moreover , a reduced mobility of charge carriers leads to a separation of time scales , a characteristic feature of correlated systems with complex dynamics ; @xcite charge carriers are _ localized _ on the spin time scale and thus a coherent - like precession of individual spins might still take place . consequently , despite broadening and renormalization of the paramagnon dispersion ( softening ) as compared to lswt with localized moments , an approximate heisenberg - like picture turns out to still be applicable below the single - particle pseudogap . but for @xmath85 . , title="fig:",scaledwidth=43.0% ] + but for @xmath85 . , title="fig:",scaledwidth=43.0% ] + we now examine the spin and charge responses at our largest interchain coupling @xmath85 . a closure of the pseudogap in the single - particle spectral function @xmath82 illustrated in fig . [ g_nod ] gives rise to a more pronounced gapless charge mode in the long - wavelength limit @xmath119 , see fig . [ dyn3](a ) . as follows from eq . ( [ sigma_cqw ] ) , this is in turn reflected in a marked increase in the drude - like weight shown in fig . [ sig_20 ] thus providing the evidence of an _ increased _ mobility of charge carriers . consequently , the low - frequency part of the dynamical spin structure factor @xmath1 can not anymore be interpreted solely in terms of spin - wave - like excitations assuming _ localized _ spins . @xcite our next goal is to identify these various components in the spin excitation spectrum . figure [ dyn3](b ) reveals featureless continua in @xmath1 near @xmath136 and @xmath47 momenta . it is natural to assign these incoherent excitations to the continuum of _ independent _ particle - hole pairs , a hallmark of an electron system with mobile charge carriers . indeed , in the presence of electronic qps , a _ single _ spin - flip excitation can be made at arbitrary low energy . hence , the paramagnons can not propagate without exciting unbound particle - hole pairs . the latter contribute to the charge excitation spectrum @xmath112 which features a similar continuum around @xmath136 , cf . [ dyn3](a ) . this similarity confirms that these features stem from the particle - hole bubble of dressed single - particle green s functions . @xcite less clear evidence of the particle - hole excitations in @xmath112 is found at @xmath47 : since most of the spectral weight is exhausted by a high - frequency @xmath137 charge mode , and [ dyn3 ] supports the point of view that the high - frequency features of @xmath1 and @xmath112 at @xmath85 are reminiscent of the 1d isolated - chain limit . ] it is more difficult to resolve the expected particle - hole continuum . however , we believe that a faint low - frequency mode corresponds to a lower bound of the particle - hole continuum . in contrast , the particle - hole interaction vertex remains finite near the af wave - vector @xmath2 . consequently , an excited particle and a hole are _ bound _ together in this part of the brillouin zone . this leads to low - frequency paramagnon excitations , see fig . [ dyn3](b ) . as compared to the weakly coupled regime shown in fig . [ dyn15](b ) , these paramagnons are broadened by scattering off mobile charge carriers and dissolve into a fl - like particle - hole continuum on moving away from the af wave - vector . such a _ localized _ nature of spin fluctuations in momentum space restricted to a narrow range around @xmath2 arises from _ spatial _ af spin correlations rather than from a set of mutually interacting local moments . @xcite independently of the intrinsic interest in finding fingerprints of the ll behavior at elevated temperatures , quasi-1d materials often exhibit low-@xmath54 broken - symmetry ground states . these instabilities occur at a temperature scale at which the system effectively becomes three - dimensional and long - range order can occur at low but finite temperatures . the physics associated with a thermal crossover can be studied in weakly coupled hubbard chains : for a weak interchain superexchange coupling @xmath138 , the energy difference between the broken - symmetry af ground state and excited states is small , thus facilitating their thermal population . this offers the opportunity to analyze the interaction of low - energy electronic qps with finite - range af spin fluctuations in a thermally disordered quasi-1d metal and to probe transient changes in the spin excitation spectrum intimately connected to the onset of the long - range af order in the @xmath80 regime . figure [ sqw_t ] tracks the temperature dependence of the dynamical spin structure factor @xmath1 at a fixed @xmath49 . for this value of the interchain coupling , optical spectra in fig . [ sig_t ] provide evidence of a thermal crossover from a fl - like regime at @xmath50 to the broken - symmetry sdw ground state in the effective zero - temperature limit . obtained on a @xmath36 lattice with fixed @xmath49 : ( a ) @xmath50 , ( b ) @xmath72 , and ( c ) @xmath59 . insets show the single - particle spectral function at @xmath87 ( left ) and low - frequency dependence of the imaginary part of the corresponding self - energy ( right ) . , title="fig:",scaledwidth=43.0% ] + obtained on a @xmath36 lattice with fixed @xmath49 : ( a ) @xmath50 , ( b ) @xmath72 , and ( c ) @xmath59 . insets show the single - particle spectral function at @xmath87 ( left ) and low - frequency dependence of the imaginary part of the corresponding self - energy ( right ) . , title="fig:",scaledwidth=43.0% ] + obtained on a @xmath36 lattice with fixed @xmath49 : ( a ) @xmath50 , ( b ) @xmath72 , and ( c ) @xmath59 . insets show the single - particle spectral function at @xmath87 ( left ) and low - frequency dependence of the imaginary part of the corresponding self - energy ( right ) . , title="fig:",scaledwidth=43.0% ] at elevated temperature @xmath50 , see fig . [ sqw_t](a ) , @xmath1 features solely broad continua associated with single - particle spin - flip excitations prominent features of the fl dynamics . the validity of the fl - like picture is further supported by a well defined qp peak in the single - particle spectral function @xmath82 resolved at the nodal @xmath87 point , see fig . [ sqw_t](a ) . these qp excitations are responsible for a drude - like feature in the corresponding optical conductivity shown in fig . [ sig_t ] . although a drude - like weight signals that the system remains metallic upon cooling down to @xmath72 , the single - particle dynamics becomes _ incoherent _ ; a single qp peak in @xmath82 at the @xmath87 point is replaced by two broadened peaks separated by a shallow pseudogap , see fig . [ sqw_t](b ) . this double peak structure yields a _ finite_-frequency hump in the optical spectra due to particle - hole excitations _ across _ the pseudogap , cf . [ sig_t ] . further evidence of the growing incoherence in the system is brought by the enhancement of the qp scattering rate @xmath139 , cf . insets in figs . [ sqw_t](a ) and [ sqw_t](b ) . as is apparent from the low-@xmath54 growth of the staggered spin structure factor @xmath57 depicted in fig . [ s_t](b ) , the change in the qp damping rate should be traced to the increase in the magnetic correlation length . this has a pronounced effect on the spin excitation spectrum which develops a paramagnon branch near the af wave vector , see fig . [ sqw_t](b ) . as indicated by their broad spectral width , these paramagnons have a short lifetime due to scattering off mobile charge carriers and merge into a particle - hole continuum on moving away from the @xmath2 momentum . obtained on a @xmath36 lattice with fixed @xmath49 : ( a ) @xmath50 , ( b ) @xmath72 , and ( c ) @xmath59 . , title="fig:",scaledwidth=43.0% ] + obtained on a @xmath36 lattice with fixed @xmath49 : ( a ) @xmath50 , ( b ) @xmath72 , and ( c ) @xmath59 . , title="fig:",scaledwidth=43.0% ] + obtained on a @xmath36 lattice with fixed @xmath49 : ( a ) @xmath50 , ( b ) @xmath72 , and ( c ) @xmath59 . , title="fig:",scaledwidth=43.0% ] finally , at our lowest temperature @xmath59 , a further increase in the qp damping rate @xmath140 leads to a strong depletion of single - particle spectral weight , see the inset in fig . [ sqw_t](c ) . despite some thermally excited qp states at the fermi level , a complete suppression of the drude weight in the corresponding optical conductivity shown in fig . [ sig_t ] provides clear evidence of charge localization . hence , the formation of the insulating state appears here as the outcome of a divergent qp mass rather than the disappearance of charge carriers . this insulating phase features rich spin dynamics with a pronounced frequency - dependent damping , see fig . [ sqw_t](c ) . on the one hand , we resolve at our lowest temperature @xmath59 the increase in both the magnetic intensity and stiffness of the spin - wave - like dispersion near the af wave - vector . the latter effect is similar to spin - wave stiffening found in the weakly doped two - dimensional @xmath13-@xmath131 model on reducing the hole doping and approaching the long - range af phase . @xcite moreover , a correlation - induced gap for spin excitations becomes discernible at the @xmath47 momentum . the size of this gap might be considered as a measure of the vertex correction , i.e , binding energy between an excited particle and a hole , a precursor of collective spin excitations . on the other hand , the interaction of electronic single - particle excitations with magnetic modes results in the overdamped spin dynamics at energies above the pseudogap in @xmath82 . figure [ cqw_t ] summarizes the corresponding temperature evolution of the dynamical charge structure factor @xmath112 . the intensity of the low - frequency weight in the long - wavelength limit @xmath119 progressively decreases with decreasing temperature and is completely washed out below the crossover scale marking the onset of the sdw phase . this agrees with a small gap in the optical conductivity resolved at @xmath59 , cf . [ sig_t ] . let us now discuss the relevance of our results to quasi-1d organic bechgaard - fabre salts . at ambient pressure , the fabre ( tmttf)@xmath8x salts are typically mott insulators due to a combined effect of umklapp process and dimerization leading effectively to a half - filled band . upon increasing pressure , the 1d mott phase is replaced by a higher - dimensional metal when the interchain hopping reaches the order of the mott gap . @xcite in contrast , the bechgaard ( tmtsf)@xmath8x salts are quasi-1d metals due to larger interchain hopping . it effectively warps the fs thus introducing deviation from ideal nesting properties of the 1d limit . the optical conductivity spectra of the bechgaard salts are markedly different from those of a simple metal : a narrow drude - like weight is accompanied by a finite - frequency feature exhausting most of the spectral weight . @xcite the latter is usually interpreted in terms of the remnant 1d mott gap . @xcite the drude feature survives up to arbitrarily low temperatures before being disrupted by the onset of a small sdw gap driven by the fs nesting . @xcite on the one hand , recent cdmft studies capturing only short - range af spin correlations have shown evidence of a quasi-1d metallic phase with a fs broken into pockets . @xcite as the interchain coupling was increased , a continuous fs was restored although with a substantial variation of the qp weight in the brillouin zone . a strong momentum dependence of the self - energy offers a simple framework accounting for the unusual frequency dependence of the optical conductivity in ( tmtsf)@xmath8x salts . indeed , it is natural to associate a drude - like feature ( finite - frequency absorption ) to the fs pockets ( gapped regions of the brillouin zone ) , respectively . on the other hand , our qmc simulations yield a continuous fs in the quasi-1d metallic phase . as a result , a two - component optical response follows from the depletion of single - particle spectral weight at the fermi level with a finite - frequency contribution assigned to particle - hole excitations across the pseudogap . one possible reason behind the continuous fs is that the formation of the fs pockets requires simulations in the thermodynamic limit . alternatively , it is known that the breakup of the fs into pockets appears only at sufficiently low temperatures , @xcite i.e. , possibly below the onset of the intervening sdw phase . in this respect , a larger deviation from the fs nesting , e.g. , a larger next - nearest - neighbor hopping @xmath141 , would be required to extend the region of stability of the quasi-1d metallic phase to lower temperatures . finally , we point out a strong renormalization of the qmc drude - like weight with respect to the one found in the chain - dmft studies . @xcite it stems from dressing of mobile charge carriers with a cloud of finite - range af spin fluctuations beyond the chain - dmft approximation . interestingly , a reduced qmc drude response is reminiscent of the experimental data on the ( tmtsf)@xmath8x salts with the zero - frequency mode containing a tiny @xmath142 fraction of the total optical weight . we have presented a thorough auxiliary - field qmc study of a dimensional crossover in weakly coupled 1d hubbard chains with a half - filled band . as a function of the interchain coupling @xmath5 , we interpret our finite - temperature results in terms of the crossover from the 1d mott phase which exhibits spin - charge separation to a higher - dimensional af metal . we have placed emphasis on clarifying the nature of this metallic phase and its low - energy excitations . to this aim , we have examined both the single- and two - particle excitation spectra . on the one hand , the simultaneous emergence of the zero - frequency drude - like response in both intra- @xmath62 and interchain @xmath0 optical conductivities implies the onset of a quasi-1d metal at @xmath45 . this should be contrasted with the anisotropic localization in coupled spinless ll chains leading to metallic transport in some directions and not along the others ( smectic metal ) . @xcite on the other hand , most of the optical weight does not form a coherent drude peak as in a fl metal , but accumulates at finite frequency . this signals unconventional charge dynamics attributed to enhanced qp scattering off finite - range af spin fluctuations . our study reveals a rich structure of the corresponding magnetic excitation spectrum : signatures of the fl dynamics ( particle - hole continua ) and a paramagnon branch near the @xmath2 momentum are apparent . the temperature dependence of the single- and two - particle excitation spectra in the quasi-1d regime with @xmath49 is equally rich . at elevated temperatures , they are suggestive of the formation of a fl - like metal . in contrast , at our lowest temperatures , the disappearance of the drude weight indicates the onset of an insulating ground state gapped out by af spin correlations . as a result , we resolved a spin - wave - like mode in the dynamical spin structure factor @xmath1 spreading out nearly through the entire brillouin zone . finally , we have discussed a possible relationship of our findings to the experiments . on the one hand , the drude weight in quasi-1d organic bechgaard - fabre salts arises from deviations from the strictly 1d half - filled band which lead to warping of the fs . @xcite moreover , there is strong experimental evidence of the importance of higher - dimensional af spin correlations . thus , the extension of the model hamiltonian from 1d to quasi-1d is essential to account for the metallic behavior in these compounds . on the other hand , finite - energy absorption in the optical conductivity spectrum indicates that the interchain coupling becomes ineffective above a certain frequency threshold . in this regime , one recovers the 1d physics with remnant umklapp scattering . instead , our qmc data indicates that the high - frequency 1d mott feature in the optical response is quickly replaced upon coupling the chains by a low - frequency mode associated with the onset of finite - range higher - dimensional af spin correlations . we conclude that the low - temperature optical spectra in quasi-1d correlated metals with a half - filled band and a nearly nested fs are intimately related to dominant af spin fluctuations . however , we expect that the importance of longer - range interchain hopping in the actual electronic band structure of bechgaard - fabre salts @xcite will enhance the frustration effects and thus could have the potential of reproducing the experimentally observed optical spectra in lattice simulations . we thank the lrz - munich and the jlich supercomputing centre for a generous allocation of cpu time . this work was supported by the fp7/erc starting grant no . 306897 and fp7/marie - curie grant no . 321918 . f. f. a. acknowledges support from the dfg grant as120/8 - 2 ( for1346 ) . 111ifxundefined [ 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/physrevb.62.3422 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.80.885 [ * * , ( ) ] link:\doibase 10.1038/nphys2259 [ * * , ( ) ] link:\doibase 10.1126/science.1236362 [ * * , ( ) ] link:\doibase 10.1103/physreva.88.063629 [ * * , ( ) ] link:\doibase 10.1103/physrevb.87.224513 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.113.185301 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.115301 [ * * , ( ) ] link:\doibase 10.1103/physreva.89.023605 [ * * , ( ) ] link:\doibase 10.1021/cr030647c [ * * , ( ) ] link:\doibase 10.1103/physrevb.44.2905 [ * * , ( ) ] link:\doibase 10.1126/science.281.5380.1181 [ * * , ( ) ] link:\doibase 10.1103/physrevb.81.125109 [ * * , ( ) ] link:\doibase 10.1080/00018730903122242 [ * * , ( ) ] @noop ( ) , @noop _ _ ( , , ) link:\doibase 10.1103/physrevlett.72.316 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.74.968 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.74.2997 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.83.128 [ * * , ( ) ] http://jetp.ac.ru/cgi-bin/e/index/e/62/6/p1340?a=list [ * * , ( ) ] link:\doibase 10.1103/physrevlett.87.276405 [ * * , ( ) ] link:\doibase 10.1103/physrevb.65.115117 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.99.126404 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.045112 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.193103 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.109.126404 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.085120 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.68.13 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.87.186401 [ * * , ( ) ] link:\doibase 10.1103/physrevb.67.075110 [ * * , ( ) ] link:\doibase 10.1103/physrevb.69.195105 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.97.136401 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.77.2790 [ * * , ( ) ] link:\doibase 10.1103/physrevb.56.11001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.4799 [ * * , ( ) ] link:\doibase 10.1103/physrevb.64.054422 [ * * , ( ) ] link:\doibase 10.1038/nmat1327 [ * * , ( ) ] link:\doibase 10.1103/physrevb.14.1165 [ * * , ( ) ] link:\doibase 10.1007/978 - 3 - 642 - 82499 - 9 [ _ _ ] ( , , ) link:\doibase 10.1103/physrevb.48.7183 [ * * , ( ) ] link:\doibase 10.1103/physrevb.41.6399 [ * * , ( ) ] link:\doibase 10.1103/physrevb.42.7967 [ * * , ( ) ] link:\doibase 10.1080/0001873021000057123 [ * * , ( ) ] link:\doibase 10.1088/0034 - 4885/66/8/202 [ * * , ( ) ] link:\doibase 10.1088/0034 - 4885/71/2/026501 [ * * , ( ) ] link:\doibase 10.1080/00018732.2014.940227 [ * * , ( ) ] link:\doibase 10.1103/physrevd.24.2278 [ * * , ( ) ] link:\doibase 10.1088/0305 - 4470/27/22/009 [ * * , ( ) ] link:\doibase 10.1088/0305 - 4470/29/11/003 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.83.3069 [ * * , ( ) ] link:\doibase 10.1103/physrevb.62.6378 [ * * , ( ) ] in @noop _ _ , , vol . , ( , , ) pp . @noop ( ) , link:\doibase 10.1103/revmodphys.77.721 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.83.471 [ * * , ( ) ] link:\doibase 10.1103/physrevb.72.045113 [ * * , ( ) ] link:\doibase 10.1103/physrevb.80.161105 [ * * , ( ) ] link:\doibase 10.1103/physrevb.81.174536 [ * * , ( ) ] link:\doibase 10.1103/physrevb.83.125108 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.125115 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.085128 [ * * , ( ) ] link:\doibase 10.1103/physrevb.89.155126 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.3910 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.86.680 [ * * , ( ) ] link:\doibase 10.1016/s0921 - 4526(96)00768 - 5 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.84.4673 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.076401 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.035111 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.92.256401 [ * * , ( ) ] link:\doibase 10.1103/physrevb.73.165119 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.73.732 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.84.522 [ * * , ( ) ] link:\doibase 10.1103/physrevb.70.235107 [ * * , ( ) ] link:\doibase 10.1103/physrevb.72.075136 [ * * , ( ) ] link:\doibase 10.1103/physrevb.74.125110 [ * * , ( ) ] link:\doibase 10.1103/physrevb.75.104503 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.102.056404 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.035102 [ * * , ( ) ] link:\doibase 10.1103/physrevb.67.205111 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.155109 [ * * , ( ) ] link:\doibase 10.1103/physrevb.80.245102 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1103/physrevb.85.165132 [ * * , ( ) ] link:\doibase 10.1103/physrev.128.2131 [ * * , ( ) ] link:\doibase 10.1103/physrevb.48.10227 [ * * , ( ) ] link:\doibase 10.1103/physrevb.55.14953 [ * * , ( ) ] link:\doibase 10.1103/physrevb.75.205128 [ * * , ( ) ] link:\doibase 10.1103/physrevb.79.245101 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.111.137205 [ * * , ( ) ] link:\doibase 10.1038/nphys2652 [ * * , ( ) ] link:\doibase 10.1103/physrev.86.694 [ * * , ( ) ] link:\doibase 10.1103/physrev.87.568 [ * * , ( ) ] link:\doibase 10.1103/physrevb.66.094431 [ * * , ( ) ] link:\doibase 10.1103/physrevb.78.155124 [ * * , ( ) ] link:\doibase 10.1103/physrevb.81.054444 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.74.5124 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.112.247002 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.024404 [ * * , ( ) ] link:\doibase 10.1103/physrevb.45.12357 [ * * , ( ) ] link:\doibase 10.1103/physrevb.47.463 [ * * , ( ) ] link:\doibase 10.1103/physrevb.54.15483 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.77.398 [ * * , ( ) ] link:\doibase 10.1103/physrevb.58.1261 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.76.3838 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.85.2160 [ * * , ( ) ] link:\doibase 10.1103/physrevb.87.155139 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.125128 [ * * , ( ) ]

we identified unresolved source candidates in the hubble ultra deep field , down to . unresolved objects were identified using a parameter , which measures the deviation from the curve - of - growth of a point source . extensive testing of this parameter was carried out , including the effects of decreasing signal - to - noise and of the apparent motions of stars , which demonstrated that stars brighter than could be robustly identified . low resolution grism spectra of the objects brighter than identify m and later stellar type dwarfs , 2 candidate l - dwarfs , qsos , and 4 white dwarfs . using the observed population of dwarfs with spectral type m4 or later , we derive a galactic disk scale height of for m and l stars . the local white dwarf density is computed to be as high as . based on observations taken 73 days apart , we determined that no object in the field has a proper motion larger than 0.027/year ( detection limit ) . no high velocity white dwarfs were identified in the hudf , and all four candidates appear more likely to be part of the galactic thick disk . the lack of detected halo white dwarfs implies that , if the dark matter halo is 12 gyr old , white dwarfs account for less than 10% of the dark matter halo mass .

astro-ph0412097

studying the stellar content of our galaxy down to the faintest possible magnitudes is necessary to study the structures of the galactic disk and population ii halo ( referred to as galactic halo thereafter ) and stellar population @xcite . the nature of the galactic dark matter halo has in addition been the subject of some debate since the macho project @xcite observed lensing events in the direction of the large magellanic cloud ( lmc ) and claimed that 20% of the dark matter halo mass could be in the form of low mass objects with masses smaller than @xmath5 @xcite . while there are alternative explanations for the observed lensing events ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) , the detection of faint , blue unresolved objects with high proper motion has been used as evidence for the existence of a large population of old white dwarfs in the galactic dark matter halo . @xcite estimated that about 2% of the dark matter halo mass could be accounted for by old halo white dwarfs . there has since been some debate as to whether the former result might have been caused by a dynamically warmer thick disk component @xcite and the latter might have been affected by the misidentification of faint blue extra - galactic sources ( @xcite , also see @xcite for a complete review ) . the existence of ultra cool white dwarfs ( @xmath6 ) has however been confirmed in recent sloan data @xcite . + a search for distant white dwarfs using deeper observations provides another opportunity to firmly examine the existence of a significant population of old white dwarfs in the galactic halo and the dark matter halo . as shown by @xcite , deep broad band imaging can probe a significant portion of the galactic dark matter halo and should identify galactic dark halo white dwarfs in the process . ideally , one would like to identify white dwarf candidates spectroscopically , instead of relying on broadband colors only , and to determine their galactic disk or halo memberships using sensitive proper motion measurements . + the advanced camera for survey ( acs ) hst observations of the hubble ultra deep field ( hudf , * ? ? ? * ) and their unprecedented depths offer a new opportunity to detect faint stars , such as white dwarfs , at high galactic latitude . additionally , the grapes survey ( grism acs program for extragalactic science . pi : malhotra , see description in * ? ? ? * ) provides low resolution spectra of most sources in the hudf . these spectra can be used to confidently determine spectral types of stars in the hudf , down to @xmath7 mag . more importantly , these spectra allow one to differentiate unresolved , blue , faint extra - galactic objects ( e.g. qso ) from white dwarfs and other stars . + section [ sindexsec ] outlines how unresolved sources were identified in the hudf , while the spectroscopic identification of unresolved sources is discussed in section [ fitting ] . section [ discussion ] describes an analysis of the cumulative number distribution of the late - type dwarfs identified in the hudf and its implication for the galactic disk structure ( section [ mdwarfs ] ) . section [ discussion ] also includes a description of the white dwarf content of the galactic disk and halo ( section [ wd ] ) . the hudf was imaged using the acs and the f435w ( ) , f606w ( ) , f775w ( ) and f850lp ( ) bands at different epochs . the observations were carried out by cycling through each filter while using a small dither pattern to mask out the acs inter - chip gap and other cosmetic problems . the hudf was observed during 4 different visits , rotating the field of view to a different position angle each time ( 0 , 8 , 85 , 91 with respect to the first visit , * ? ? ? the individual acs images had an integration time of 1200 seconds , a size of 4096x4096 pixels and a pixel scale of 0.050/pixel . these images were combined together by drizzling them onto a finer pixel grid with a pixel scale of 0.030/pixel , while geometrically correcting these images . the limiting ab magnitudes ( s / n=10 ) of the and images are 29.0 and 28.7 mag , respectively . in this paper , images taken from 09/24/2003 to 10/28/2003 and from 12/04/2003 to 01/15/2004 are referred to as epoch 1 and 2 images , respectively . the elapsed time between the mean observing dates of the two epochs is 73 days . images from epoch 1 and epoch 2 were taken at position angles that are nearly 90 apart . + while the deep , combined acs hudf images described above were used to determine which objects are unresolved ( section [ sindexsec ] ) , eight special partial image stacks were also assembled to take advantage of the fact that the data were taken over a period of several months . four partial image stacks were created for each hudf epoch , using the exact same method used to create the original combined hudf images . these partial image stacks can be used to determine the variability @xcite as well as measure the proper and parallax motions of the objects in the field . the eight partial image stacks have the same physical properties as the combined acs hudf images but with one eighth of the integration time . note that the cosmic ray rejection step was done separately while assembling each partial image stack so that any strongly varying or moving object would not be removed by the cosmic ray rejection process . the astrometric registration of the combined hudf images and the eight partial image stacks is estimated to be better than 0.1 acs pixel ( 0.005 or 0.16 hudf image pixel ) @xcite . + spectroscopic observations of the udf using the acs slitless grism mode ( g800l ) were carried out as part of the grapes project . the reader is referred to @xcite for an in - depth description of these spectra . these spectra are low resolution spectra ( r=100 ) , span a wavelength range of @xmath8 , and can be used to spectroscopically identify objects in the field . + the hudf public catalog 1.0 , containing objects , was used to identify objects in the field and all objects are referred to by their i d in this catalog ( uid ) . the magnitudes of these objects were re - measured using matched aperture photometry with apertures defined using sextractor @xcite and the band hudf image . the photometric acs magnitude zero - points ( 25.65288 , 26.49341 , 25.64053 , 24.84315 in the , , , bands respectively ) were taken from the goods survey @xcite . we identified unresolved objects in the hudf using a single - index ( ) method that is based on a simple analysis of the curve of growth of the light distribution of an object . the iraf task radprof was used to produce these light curves of growth . this task first computes the intensity weighted center of the object ( in a 3x3 pixel box ) , and then performs aperture photometry using increasing aperture sizes ( up to a radius of 10 pixels ) . these photometric measurements are then fit ( using eight 3 @xmath9 rejection cycles ) to a cubic spline of order 5 . radprof was applied to stamp images of objects which were masked using sextractor segmentation maps . these segmentation maps were used to avoid contamination from neighboring objects . the radprof parameters described here were selected because , following extensive initial testing , they were found to result in curves of growth that are not too sensitive to the presence of faint nearby objects while consistently remaining sensitive to the presence of faint , diffuse extended emission around objects . the stellarity of an object is defined as @xmath10 where @xmath11 and @xmath12 denote the curve of growth of light , normalized to 1.0 at a distance of @xmath13 , of the object and of a point source . a value of 10 . was used for @xmath13 . the index is then simply the square root of the sum of the squared differences between the two curves . unresolved objects have values that are close to 0.0 while increasingly resolved objects produce increasingly larger ( unbound ) values of . the detailed calibration of was done in several stages , each time refining the reference curve of growth that was used . this is described in some detail below . + first , a few obvious , bright , unresolved and unsaturated sources were pre - selected for further study . this was done in both the and band images separately . the grapes spectra of these sources were examined and a main - sequence star was selected ( uid 911 ) . the curves of growth of this object were measured separately in the and bands and were used to compute @xmath14 and @xmath15 values for all objects in the hudf . a new set of bright point sources , well distributed over the whole acs field , were then carefully selected based on their low @xmath14 values ( arbitrarily chosen to be @xmath16 ) . the grapes spectra of these objects were visually inspected to ensure that these objects were stars . stamp images of these objects were individually drizzled @xcite and carefully re - centered to produce a stack of images which were then combined together to produce empirical and psf images . a sigma - based rejection algorithm was used to avoid the effect of any faint neighbor in the final combined images . final and band reference curves of growth were generated using these empirical psf estimates and all of the @xmath14 and @xmath15 values were re - computed . figure [ fig1 ] shows the distributions of the values in both the and bands , as a function of magnitude . there are clearly two distinct groups of object in the hudf images : one group consists of object that are completely or partially resolved , forming a large distribution around the value of 0.2 and @xmath17 mag . a second group consists of unresolved objects with low values at all magnitudes . bright stars in figure [ fig1 ] are observed to have values which are consistently lower than 0.05 . the exception to this are bright stars near @xmath18 mag . which are saturated and have artificially high values of ( @xmath19 ) . the effect of bright star saturation on the measured values of were investigated and values of @xmath20 are expected if the center of a star is saturated or missing in the hudf images . nevertheless , the populations of resolved and unresolved objects appear to be well separated at brighter magnitudes in figure [ fig1 ] . + unresolved sources in the hudf were selected using the criterion that @xmath21 mag . four bright saturated stars were added to the final object list manually . the final list of unresolved udf sources contains objects , spanning the magnitude ranges of @xmath22 mag and @xmath23 mag . table [ table1 ] summarizes some of the properties of these objects . the ability to distinguish between resolved and unresolved objects using values was extensively tested , using a subset of the hudf object catalog containing the 369 brightest ( @xmath24 mag ) objects in the field . simulated and band images of these objects made 0.0 , 0.75 , 2.0 , 3.0 , 4.0 , 5.0 , 6.0 , 7.0 , 8.0 , and 9.0 magnitudes fainter were created and values measured using each simulated image , exactly as described in section [ sindexsec ] . the value of each of these 369 test objects was thus measured 10 times , each measurement corresponding to a different apparent magnitude . figure [ fig2 ] shows how the value remains largely unchanged as an object is made fainter and noisier , all the way down to @xmath25 mag . the vertical pattern seen in this figure is a result of the 10 magnitude steps used in this simulation . this test demonstrates a very small fraction of misclassification : not a single extended object was misclassified as a star ( e.g. no object with @xmath26 originally in the hudf images was measured to have @xmath27 as the object was made fainter ) , and fewer than 5% of stars were misclassified down to @xmath25 mag . while some objects shown in table [ table1 ] are fainter than this , only objects significantly brighter than this limit ( @xmath28 mag ) are actually included in the analysis described below and these are expected to be bona fide unresolved objects . some of the stars in the hudf objects could have a high tangential - velocity ( @xmath29 ) and a correspondingly large proper motion . such objects would be blurred in the combined hudf images and would appear to be resolved . similarly , nearby objects would suffer from a significant amount of parallax , again causing these objects to be blurred and mis - identified . the index proved relatively insensitive to a small amount of object motion . this is not unexpected since was tuned to the detection of faint extended components , the curves of growth produced by radprof were computed by summing , using several sigma rejection steps , the pixel intensities inside annuli of increasing radii . the effect of a nearby source is strongly diminished as its separation from the main source increases and the number of unaffected pixels increases . in addition , the use of a fifth order spline imposes some smoothness on the curve of growth which further lowers the effect of a nearby object . still , the effect of object motion on measured values of was carefully investigated by generating images of the unresolved hudf objects that were selected as described in section [ sindexsec ] . these were appropriately shift - and - added , accounting for the orientation and pointing used for each individual hudf observation while allowing objects to move as a function of time , to simulate the effect of proper motion . as larger values of proper motion were introduced , the values increased only slightly and was found to be mostly unaffected by shifts smaller than 2.5 hudf pixel ( @xmath30 ) . 70% of objects were still properly identified as point sources when shifts 2.5 and 5.0 hudf pixels were applied ( @xmath31 ) . the spectral type of unresolved objects brighter than @xmath1 could be determined robustly using template fitting of their grapes spectra @xcite . the template set included @xcite main - sequence templates , l and t dwarf spectra @xcite , da and db white dwarf spectra @xcite obtained from the sdss archive , and some low temperature ( 3000k , 5000k , and 8000k ) model spectra of white dwarfs @xcite . the quality of each spectral fit was evaluated visually and the results of these fits are given in table [ table2 ] . objects too faint to have reliable fits are indicated . the grapes spectra and the two best - fits listed in table [ table2 ] are shown in figures [ all1 ] , [ all2 ] , and [ all3 ] . unresolved extra - galactic sources are labelled as gal in table [ table2 ] and are discussed in more details in section [ qso ] . + of the point sources detected in the hudf , stars , all brighter than @xmath1 mag were identified . as shown in table [ table2 ] , the majority of stars in the hudf ( out of ) are found to be m dwarfs . figures [ vbv ] and [ ivi ] show these objects in vs. - and vs. - plots . such plots have in the past shown a lack of faint red objects in the region @xmath32 mag and @xmath33 @xcite . objects found in this region of the plot , appear to be either extra - galactic or l - type dwarf spectra ( uid 443 , 366 ) . further nir observations will be required to verify the nature of these 2 l - dwarf candidates by confirming that they both have the expected blue nir colors of this type of object . + the hudf field is at high galactic latitude ( b=223 , l=54 ) and the effects of extinction and reddening are small ( @xmath34 , * ? ? ? * ) . the distances to the hudf stars , which are shown in table [ table2 ] , were determined using their observed magnitudes , their estimated stellar types , and the dwarf star main - sequence absolute johnson v magnitudes from @xcite . the latter were converted to the filter bandpass and ab magnitudes to be consistent with the data at hand . distance estimates obtained under the assumption that these objects are white dwarfs with intrinsic magnitudes of @xmath35 are also given . the galactic halo is isotropic and has no or little overall rotation . halo objects are therefore expected to lag behind the rapidly rotating galactic disk component ( and the local standard of rest ) with an average velocity of @xmath36 . as done in @xcite , it is assumed here that galactic halo objects have a velocity that is more than @xmath37 above the velocity expected for an object that is in the galactic disk ( i.e. @xmath38 * ? ? ? + the hudf image stacks ( section [ thedata ] ) were subtracted from one another and the results were examined to see if any object , whether classified to be extended or not based on its @xmath39 value , had moved significantly ( e.g. more than 2.5 udf pixel ) from one image to the next . no such object was found . + the motion of objects in the hudf was further investigated by carefully measuring the position of objects separately in each of the eight available band hudf partial image stacks . these positions were measured using the idl task mpfit2dpeak , using a fixed width ( fwhm=2.5 pixel ) gaussian fit . initially , the positions of the most compact sources ( @xmath40 and @xmath41 mag ) were examined . an average position and a standard deviation of the mean were computed for each object in the field using epoch 1 and epoch 2 hudf image stacks separately . average epoch 1 and epoch 2 positions were subtracted from one another to compute the observed shift of each object . these are shown ( scaled up by a factor of 500 ) in the left panel of figure [ figpm ] . repeated measurements of the position of sources in the field using hudf image stacks taken within the same epoch were always within 0.1 hudf pixel , or 0.003 . still , and as shown in figure [ figpm ] , there is a systematic , field dependent disagreement between the epoch 1 and epoch 2 measurements . this is likely caused by epoch 1 and epoch 2 images having been taken at position angles that are 90 apart , and/or the effect of telescope breathing , and/or very small amount of residual image distortion that is not taken into account in the current model of the acs distortion maps . to produce a @xmath4 detection with these data , an object must have a measured proper motion larger than 0.54 hudf pixel or @xmath42 . no object in table [ table1 ] was observed to have such a large proper motion . + the situation was however improved using a simple third order two dimensional polynomial fit to the observed distribution shown in the left panel of figure [ figpm ] . this fit , when applied as a correction , results in significantly lower systematics across the field , as shown in the right panel of figure [ figpm ] . the accuracy of the corrected shifts ( of compact objects ) between epoch 1 and epoch 2 is improved by a factor of 3 , with a @xmath4 detection now corresponding to a shift of 0.18 hudf pixel or @xmath43 . + the search for high proper motion objects in the hudf was extended to all objects in the field by measuring and correcting ( using the correction determined above ) the observed shifts of these objects . only four sources were found to have moved significantly ( but with a significance slightly lower than @xmath4 ) from epoch 1 to epoch 2 . these objects are uid 443 , 9020 , 911 , and 7525 and they are all unresolved objects that are listed in table [ table1 ] . their measured shifts are 0.17 , 0.17 , 0.14 , and 0.15 hudf pixels respectively , or approximately 0.02/year , which is smaller than our imposed @xmath4 limit . two unresolved , extra - galactic sources with @xmath28 were identified ( uid 6732 , 9397 ) . the grapes spectrum of these objects show prominent emission lines and the redshift of these objects is estimated to be z=3.2 and z=3.0 , respectively @xcite . identification of such objects using grapes spectra could easily be carried out down to magnitudes fainter than = 27.0 mag , as demonstrated by the fact that several fainter extragalactic objects with @xmath44 were easily identified ( e.g. uid 4120 , 8157 ) . without spectral information , these objects would have been misidentified as stars ( e.g. uid 9397 would be a good candidate for a cool white dwarf at a distance of @xmath45 1 kpc ) . the issue of contamination by faint , blue extra - galactic objects in previous studies @xcite has been discussed in the past @xcite and figures [ vbv ] and [ ivi ] illustrate how these objects could be mistaken for a variety of stellar objects ranging from m and l dwarfs to white dwarfs . spectroscopic confirmation is therefore crucial to accurately determine stellar counts in deep fields such as the hudf . there are extra - galactic objects amongst the point sources with @xmath28 mag , which corresponds to a 7% contamination level at these galactic coordinates . + the extra - galactic objects ( uid 6732 , 9397 ) were used to check the accuracy of the proper motion measurements described in section [ pm ] . the measured shifts of these objects between epoch 1 and epoch 2 are smaller than 0.04 hudf pixel ( 0.0012 ) , which is what one would expect based on the proper motion error estimate quoted in section [ pm ] ) . + there are a total of 14 faint ( @xmath44 ) , blue ( @xmath46 , @xmath47 ) objects listed in table [ table1 ] that were not included in the analysis above because they are either too faint to have useable grapes spectra or robust based classifications . some of these sources might well be faint white dwarfs in either the galactic disk or halo but are hard to distinguish from extra - galactic sources without higher signal - to - noise spectra or more accurate proper motion measurements . objects were identified to be m ( or early l ) dwarf main - sequence stars . two objects were found to be best fit by early l dwarf templates @xcite . both of these objects are red enough to be outside of the color range shown in figures [ vibv ] and [ izvi ] with @xmath48 and @xmath49 . deep infrared observations of these two objects will be required to confirm that they are l dwarfs . nicmos observations of the hudf did not detect these objects in either the j or h bands . all other very red objects are best fit using m dwarf templates . the broad spectral features of m dwarfs listed in table [ table2 ] were well fitted using m dwarf @xcite templates , and for objects brighter than @xmath1 mag we believe the identification to be secure . + because these late - type sources can be seen to great distances , they can be used to probe the structure of the galactic disk and halo as traced by the lowest - mass stellar components of these populations . this was explored using monte carlo mass function ( mf ) simulations based on those developed by @xcite . assuming power - law representations of the mf , dn / dm @xmath50 m@xmath51 , for masses 0.005 to 0.2 m@xmath52 , luminosity ( lf ) and effective temperature ( t@xmath53 ) distributions were created using evolutionary models from @xcite over the range 100 @xmath54 t@xmath53 @xmath54 4000 k. for all simulations , @xmath55 was fixed at 1.13 for @xmath56 @xcite but allowed to vary between 0.51.5 for @xmath57 ; number densities are normalized to the empirical value at 0.1 m@xmath52 from the 8 pc sample @xcite . two separate populations were simulated : a disk population with a flat age distribution spanning 10 myr to 10 gyr ( the majority of sources have age greater than 1 gyr ) , and a halo population with uniformly sampled ages between 9 - 10 gyr and a relative number density of 0.25% @xcite . the resulting bolometric lfs were converted to @xmath58 ( cousins ) lf s using a polynomial fit to empirical data ( spanning @xmath59 ; i.e. , m4 to t8 ) from @xcite and @xcite . note that this fit is based on measurements for solar metallicity field dwarfs , and may not be appropriate for a subsolar metallicity halo subdwarf population . apparent @xmath58-band distributions were determined by first assuming a constant density population out to a limiting magnitude m@xmath60 = 27.5 ( taking into account the difference between hudf ab - magnitudes and vega magnitudes from the empirical data ) and then applying a correction for the vertical distribution of sources . for the disk population , we assumed a density distribution @xmath61 @xcite , where @xmath62 , @xmath63 is the source distance , @xmath64 is the galactic latitude of the hudf , and @xmath65 is the 1/e disk scale height , assumed to range between 200 and 500 pc . for the halo population , a galactocentric oblate spheroid distribution as given in @xcite was assumed , using the values derived there and an axial ratio q = c / a = 0.7 . figure [ mlf2 ] presents the results of these simulations in the form of cumulative number distributions as a function of magnitude down to the limiting magnitude of our sample . these distributions show that variations in the disk scale height are far more pronounced than those from the different mf s assumed , and this analysis is limited to the former parameter . the cumulative distribution of the six dwarfs in table [ table2 ] with spectral types m4 and later ( consistent with the mass constraints of the simulations ) is also shown . the observed distribution matches that of the @xmath66 disk simulation very well , particularly out to @xmath67 mag where the halo population makes negligible contribution . because the simulated distributions are rather sensitive to the disk scale height , particularly at fainter apparent magnitudes , one can conservatively constrain @xmath65 to @xmath68100 pc assuming poisson uncertainties . this is in good agreement with the estimate of @xmath69 from @xcite which was computed using acs hudf parallel fields i - drops . for @xmath70 mag , the observed distribution is slightly greater than the disk population alone but does not increase as sharply as the combined disk+halo distribution . indeed , the shape of the observed number distribution implies few if any halo stars in the hudf down to @xmath7 mag , consistent with the lack of significant proper motion sources in this sample . this suggests that either the number density ratio of halo to disk stars , or the adopted axial ratio for the halo density distribution , or both , may be smaller than assumed here . with no late - type halo subdwarf detections in this sample , we can not usefully constrain these possibilities . objects brighter than @xmath7 were individually examined and distances were estimated under various assumptions of what the exact nature of each object might be : ( 1 ) a main sequence star ; ( 2 ) a young white dwarf ; ( 3 ) an old white dwarf . the reddest objects , which are all well - fitted by m - and - later type templates , as well as the few extra - galactic objects identified in section [ qso ] were excluded from this analysis . the remaining objects were all initially considered to potentially be white dwarfs . + assuming that an object is a white dwarf , one can use the measured @xmath71 color of that object ( table [ table1 ] ) and the cooling curves of @xcite ( after accounting for passbands and zeropoints differences ) , to derive an absolute band magnitude , an age , and a distance to that object . as shown in figure 2 of @xcite , blue white dwarfs can either be young , hot objects ( @xmath72 , t @xmath73k , @xmath74 mag ) or older , cooler , intrinsically dimmer ones ( age @xmath75 10 gyr , t @xmath76 , @xmath77 mag ) . this results in several white dwarf distance estimates for each object . + proceeding via a process of elimination , the distinction between main sequence stars , young white dwarfs , old white dwarfs , and disk or halo white dwarfs is possible . assuming that a star is a main sequence star , which is intrinsically much more luminous than a white dwarf , can cause the distance estimate for that object to be unreasonably large . assuming that a star is a young white dwarf ( @xmath78 ) implies that this object is less likely to be part of the galactic halo since the latter is composed of much older objects . also , and based on the lack of proper motion detection discussed in section [ pmm ] , one can define an upper limit to the tangential velocity ( @xmath79 ) of each object once its distance is determined . the motion of an object moving by more than 0.18 hudf pixel ( 0.027 / yr ) should have been detected at the @xmath4 significance level as discussed in section [ pm ] . as discussed above , objects in the galactic halo are expected to have velocities around 100 - 200 km / s . even if projection effects should result in lower values of @xmath80 , one would expect at least some of them to have @xmath80 values larger than 30 - 60 km / s . a final clue to help narrow down the nature of a particular object is provided by the fact that an object that is nearby would suffer from a significant , easily measurable , parallax during the 73 days interval between epoch 1 and epoch 2 observations . calculating the parallax vector for the hudf exposures shows that an object 30 pc away would produce a parallax of about 0.060 or 2 hudf pixel between epoch 1 and epoch 2 observations . objects that are closer than 200 pc are therefore expected to produce a parallax that would be detected at more than @xmath4 level . + examining table [ table4 ] , objects 4322 , 4839 , 7768 , 9020 are unlikely to be main - sequence stars as this would make them very distant objects . the magellanic stream @xcite is too far away from the hudf to possibly explain the existence of main - sequence stars at such large distances . objects 1147 , 3166 , and 5921 which do not ( this is true of all objects listed in table [ table4 ] ) have measurable parallax or proper motion , are not likely to be white dwarfs . being either young or old white dwarfs would place these two objects less than 200 pc away and would have resulted in measurable parallax in the hudf image stacks . another object unlikely to be a white dwarf is object 9230 which was observed to have a fainter , unresolved companion 0.5 away . this companion , for which we have no grapes spectrum , has a band magnitude of @xmath81 mag and the colors @xmath82 mag . based on these colors , this object should be an early m dwarf that is @xmath83 away . this places both object 9230 and its companion at the same distance ( within error bars ) . the likelihood of any two stars being within 0.5 in the hudf , and at nearly the same distance , is very small ( @xmath84 ) and these two objects are likely to be part of a binary system , with object 9230 being a k type main - sequence star . independently of this , object 9230 can not be an old white dwarf since this would place this object at distance of 34 pc where its parallax motion would have been very easily detected . + four objects ( 4322 , 4839 , 7768 , 9020 ) remain as white dwarf candidates . using a simple 1/@xmath85 @xcite analysis , the white dwarf number density in the direction of the hudf can be computed . the detection of 4 white dwarfs in the hudf implies a local density of @xmath86 . the upper and lower limits on the density are computed using old or young white dwarfs respectively , while the errors reflect the large uncertainties in the intrinsic luminosities of young and old white dwarfs ( @xmath87 mag ) . note that this estimate is actually not affected by whether or not object 9230 was included as a hudf white dwarf . the contribution of this object to the 1/@xmath85 is negligible . for comparison , @xcite previously determined the local white dwarf density to be @xmath88 , while @xcite found a value of @xmath89 , and @xcite derived a density of white dwarfs in the thick disk of @xmath90 . the white dwarf density derived here is consistent with most of the previous density estimates . properly determining the disk and halo membership of the 4 white dwarfs identified in the hudf will be needed before a stronger conclusion can be made . re - observing the hudf field in the band in 1.5 years would allow ( assuming that a 0.18 hudf pixel shift corresponds to a @xmath4 level detection ) one to unambiguously detect the proper motion of halo white dwarfs with @xmath91 km / s at distances up to 1000 pc . + the above density estimate assumed that all four hudf white dwarfs were galactic disk objects . could some of these objects be in the galactic halo ? at least two of these objects ( 4839 , 7768 ) have @xmath92 values that are lower than 60 km / s . the probability that this is caused by a projection effect is under 10% ( assuming a relatively low velocity of @xmath93 km / s for halo objects ) . the last two white dwarfs listed in table [ table4 ] ( 4322 , 9020 ) are at slightly larger distances which average to 583 pc , and more importantly have larger values of @xmath92 . the average distance of these objects still seems a bit low for objects which would be part of a galactic halo whose density increases all the way up to 7 kpc @xcite , while white dwarfs with absolute magnitudes of @xmath94 mag @xcite should be detected all the way out to 1000 pc . unless these objects are intrinsically much dimmer , one would expect the average distance to these two objects to be larger than 583 pc and much closer to our limit of 1000pc . the @xmath92 values inferred for objects 4322 and 9020 , while higher than those of objects 4839 and 7768 , fall substantially short of expected typical halo velocities . the probability of observing each of the objects with the @xmath92 values listed in table [ table4 ] is between 1420% . as a group , the probability that objects 4322 and 9020 are halo objects is under 3% . similarly , the hypothesis that all 4 detected white dwarfs in the hudf are halo rather than disk objects is excluded at the 99.9986 % level . overall , the distances to the four white dwarf candidates identified here is more consistent with them being part of a galactic thick disk . @xcite derived a disk scale height of 400600 pc using a study of low velocity white dwarfs . this value was somewhat larger than previous estimates of 250350 pc but it is interesting to note that the m - dwarf disk scale height derived in section [ mdwarfs ] , as well as the distance estimates of the white dwarfs listed in table [ table4 ] all appear to agree with this value . + it would be interesting to attempt to set an upper limit on the halo white dwarf density based on the finding of up to two white dwarfs in the hudf . under this assumption , the upper limit to the galactic halo white dwarf density is computed to be @xmath95 . this result is consistent with the previous work of @xcite using the combined observations of the hubble deep fields north and south ( @xmath96 ) . + a population ii halo white dwarf population that is older than 12 - 13 gyr would have remained undetected here ( an intrinsically fainter than @xmath97 mag white dwarf population would also help explaining the low average distances of the white dwarfs listed in table [ table4 ] ) . as discussed above , the white dwarfs identified in this work are more likely to be within the galactic thick disk . it is to be noted that @xcite recently observed a disproportionately large fraction of disk to halo white dwarfs in the hubble deep field north ( hdf - n ) . @xcite found no evidence of any white dwarfs with tangential velocities larger than 30 km / s down to @xmath98 mag and the authors concluded that the blue objects they saw were all part of the galactic disk . similarly , the hudf observations presented here successfully identified 4 white dwarf candidates with @xmath99 mag in the hudf , none very likely to be in the galactic halo while 23 detection would have been expected to be consistent with previous studies . + while dynamical studies of the galaxy predict the dark matter white dwarf density as high as @xmath100 ( assuming that 100% of the dark matter halo mass is in the form of @xmath101 white dwarfs ) , the failure to detect a significant population of high velocity white dwarfs in the hudf points to a dark matter halo devoid of a significant white dwarf population . following the methodology of @xcite , one can estimate the amount of dark matter halo mass that is probed by the hudf images , and directly compute the fraction of the dark matter halo mass that could be contained in a white dwarf population . assuming a white dwarf absolute @xmath102 magnitude of 17.0 mag@xcite and a limiting magnitude of @xmath1 mag , the maximum probed distance is 1000 pc . in the corresponding volume , and in the direction of the hudf , the included dark matter halo mass is 3 @xmath103 . a population of dark matter halo white dwarfs could possibly have remained undetected in this study for several reasons : ( 1 ) the method used in this paper may be inefficient at properly identifying halo white dwarfs in table [ table4 ] . in this case , if one assumes that the hudf contains 4 halo white dwarfs ( @xmath28 ) , these would account for @xmath104 , a significant fraction of the expected dark matter halo mass . this would however increase the ratio of halo to disk white dwarfs to unrealistically high levels ( i.e. @xmath105 ) , and would assume that all the white dwarfs seen in the hudf are from the dark matter halo and not from the pop ii halo , which is highly unrealistic . the lack of proper motion detections also makes this scenario unlikely ; ( 2 ) white dwarfs do not contribute significantly to the galactic dark matter halo mass . from microlensing experiements towards the magellanic clouds , the contribution of machos to dark matter has been estimated to be anywhere between 20% to less than 2% ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) . furthermore , @xcite showed how an increased thick disk scale height of 400 - 600 pc makes it unlikely that white dwarfs could be the macho objects ; ( 3 ) white dwarfs could contribute significantly to the mass of the galactic dark matter halo but are too faint to be detected in this study . @xcite showed that an increase in the age of the white dwarf halo population from 10 gyr to 16 gyr is expected to reduce the number of white dwarf detection ( down to @xmath106 ) by a factor of 7 . the lack of high velocity white dwarf detections in the hudf puts upper limits on the contribution of an hypothetical white dwarf population to the galactic dark matter halo . based on the brightest objects in the hudf for which we have spectra ( i.e. @xmath28 ) , if the dark matter halo is as old as about 12 gyr @xcite then dark matter halo white dwarfs with @xmath107 ( @xmath108 ) contribute less than 20% to the dark matter . if the age of the dark matter halo is 10 gyr , white dwarfs ( @xmath109 ) contribute less than 2% to the dark matter . the possibility that there is a large population of faint white dwarfs that has remained unidentified in this study and which would account for a significant fraction of the dark matter halo can be investigated a little further . recall that in section [ qso ] , 14 sources with ( @xmath44 ) and blue ( @xmath46 , @xmath47 ) colors were identified . not all of these sources were expected to be bona - fide stars and some are likely to be extra - galactic sources , but as discussed in section [ test ] , it is estimated that 95% of the stars in the hudf were properly identified , so that this number should be considered to be an upper limit on the number of faint starts in the hudf . following the same methodology used above , but now using a limiting magnitude of @xmath110 , the hudf images probe through 79 @xmath103 of the dark matter halo . even under the extreme assumption that all 14 unresolved candidates in the udf are high velocities white dwarfs , this implies that , when reaching down to = 29.5 , faint white dwarfs in the galactic dark matter halo contributing less than 10% to the total dark matter halo mass for a dark matter halo age of 12 gyr . future observations of the hudf would allow more sensitive measurements of the proper motion of these faint objects ( section [ disk ] ) . setting tighter constraints on the maximum tangential velocities of these faint hudf unresolved objects would allow to exclude some of these objects from high velocity white dwarfs and would allow to further lower the maximum fraction of the dark matter halo mass that could be accounted for by white dwarfs . a systematic search for unresolved objects in the hudf identified objects with @xmath110 mag . using the grapes spectra of these objects , objects with @xmath28 mag were spectroscopically identified , including m and later dwarfs ( including 2 unconfirmed l dwarf candidates ) , and qsos . the m dwarf luminosity function of m4 and later - type stars was computed and compared to predictions based on monte - carlo simulations . assuming a simple analytical model of the galactic disk , the number of detected m dwarfs was shown to be large enough to constraint the scale height of the m and l dwarf galactic disk to be @xmath111 . out of 8 stars that were found to not be m or later - type stars , four stars were determined to be old white dwarf candidates . not a single object was found to have a proper motion that is larger than 0.027/year making them likely galactic thick disk objects . further imaging of the hudf , using a time span larger than 1.5 year would provide more sensitive proper motion measurements . more sensitive proper measurements are required to positively place unresolved objects in the galactic disk or halo . it would also allow to search for high tangential velocity objects amongst the 14 fainter ( @xmath44 ) unresolved sources identified in the hudf . excluding halo membership of objects down to @xmath0 based on proper motion measurements would further constraint the maximum white dwarf contribution to the dark matter halo mass . the currently available observations of the hudf , spanning 73 days , show the absence of a significant population of high velocity white dwarfs down to @xmath7 , and a relatively small number of unresolved , faint blue sources in the field down to @xmath0 . this is interpreted as a consequence of white dwarfs accounting for less than 10% of the dark matter halo , assuming that the dark matter halo is 12 gyr old . we would like to thank i. n. reid for helpful discussions . this work was supported by grant go -09793.01-a from the space telescope science institute , which is operated by aura under nasa contract nas5 - 26555 . afonso c. et al . 2003 , a&a , 404 , 145 alcock , c. , et al . 2000 , apj , 542 , 281 allen , c. w. 2000 , astrophysl quantities 4th ed . ( athlone press , london ) , p388 beckwith s. , et al . 2004 , in preparation bertin , e. arnouts , s. 1996 , a&as , 117 , 393 burgasser , a. j. , kirkpatrick , j. d. , liebert , j. & burrows , a. 2003 , apj , 594 , 510 burgasser , a. j. 2004 , apjs , in press ( astro - ph/0407624 ) burrows , a. et al . 1997 , apj , 491 , 856 chaboyer , b. , demarque , p. , kernan , p.j . , krauss , l.m . j. 494:96 - 110 chiba , m. & beers , t. c. , aj , 199 , 2843 cohen , s. h. , ryan , r. e. , hathi , n. , windhorst , r. a. , koekemoer , a. , xu , c. , pirzkal , n. , malhotra , s. & strolger , l. 2004 , apj in preparation dahn , c. c. et al . 2002 , aj , 124 , 1170 digby , a. p. , hambly , n. c. , cooke , j. a. , reid , i. n. , & cannon , r. d. 2003 , mnras , 344 , 583 evans , n. w. , belokurov , v. 2004 , in press ( astroph/0411222 ) fruchter , a. s. hook , r. n. 2002 , pasp , 114 , 144f gates , i. e. & gyuk , g. 2001 , apj , 547 , 786 gates , i. e. et al . 2004 , in press ( astroph/0405566 ) giavalisco , h. c. et al . 2004 , apj , l93 golimowski , d. a. et al . 2004 , aj , 127 , 3516 harris , h. c. et al . 2001 , apj , 549 , l109 harris , h. c. et al . 2003 , aj , 126 , 1023 hansen , b. m. s. , et al . 2002 , aj , 574 , l155 ibata , r. a. , richer , h. b. , gilliland , r. l. & scott , d. 1999 , apj , 524 , l95 ibata , r. a. , irwin , m. , bienaym , o. , scholz , r. & guibert , j. 2000 , apj , 532 , l41 kawaler , s. , d. 1996 , aj , l61 kilic , m. , von hippel , t. , mndez , r. a. & winget , d. e. 2004 , apj , 609 , 766 kirkpatrick et al . 2000 , aj , 120 , 447 liebert , j. , dahn , c. , c. & monet , d. g. 1988 , apj , 332 , 891 majewski , s. r. , siegel , m. h. , 2002 , apj , 569 , 432 mathewson , d. , s. & ford , v. , l. 1984 , iaus , 108 , 125 mndez , r. a. , minniti , d. , de marchi , g. , baker , a. & couch , w. j. 1996 , mnras , 283 , 666 mndez , r. a. & minniti , d. 2000 , apj , 529 , 911 mndez , r. a. 2002 , a&a , 395 , 779 oppenheimer , b. r. , hambly , n. c. , digby , a. p. , hodgkin , s. t. & and saumon , d. 2001 , science , 292 , 698 pirzkal n. , et al . 2004 , apjs , 154 pickles , a. j. 1998 , pasp , 110 , 863 reid , i. n. & majewski , s. r. 1993 , apj , 409 , 635 reid , i. n. et al . 1999 , apj , 521 , 613 reid , n. & hawley , s. l. 2000 , new light on dark stars ( chichester : praxis ) reid , i. n. , sahu , k. , c. & hawley , s. l. 2001 , apj , 559,942 reid , i. n. , ara&a , 43 , in press richer , h. b. , hansen , b. , limongi , m. , chieffi , a. , straniero , o. & fahlman , g. c. 2000 , apj , 529 , 318 robin , a. c. , reyl , c. , derrire , s. & picaud , s. 2003 , a&a , 409 , 523 ryan , r. e. , hathi , n. p. , cohen , s. h. , & windhorst , r. a. , apj in preparation sahu , k. c. 2003 , space telescope science institute symposium series , vol . 15 , cambridge university press , p. 14 - 23 ( astro - ph/0302325 ) sahu , k. c. 1994 , nature , 370 , 275 schlegel , d. j. , finkbeiner , d. p. & davis , m. 1998 , apj , 500 , 525 schmidt , m. 1968 , apj , 151 , 393 tinney , c. g. , reid , i. n. & mould , j. r. 1993 , apj , 414 , 254 xu , c. et al . 2004 , apjs , in preparation cccccccccc 19 & 53.1623248 & -27.8269460 & 26.58 & 1.91 & 1.01 & 0.30 & 0.015 & 0.004 + 366 & 53.1753062 & -27.8198904 & 29.99 & 2.20 & 3.05 & 1.27 & 0.009 & 0.030 + 443 & 53.1583988 & -27.8189953 & 31.25 & 0.86 & 3.47 & 1.51 & 0.016 & 0.028 + 834 & 53.1648156 & -27.8143652 & 27.67 & 2.35 & 2.00 & 0.76 & 0.013 & 0.011 + 911 & 53.1460739 & -27.8123237 & 25.21 & 2.09 & 1.80 & 0.65 & 0.007 & 0.002 + 1147 & 53.1834639 & -27.8067589 & 19.58 & -0.16 & 0.52 & 0.13 & 0.091 & 0.002 + 1343 & 53.1398080 & -27.8100295 & 27.36 & -0.23 & -0.14 & -0.04 & 0.045 & 0.046 + 2150 & 53.1767483 & -27.7996684 & 22.60 & 1.99 & 1.43 & 0.80 & 0.147 & 0.003 + 2368 & 53.1676057 & -27.8037923 & 29.07 & 1.18 & 0.08 & -0.20 & 0.046 & 0.050 + 2457 & 53.1616105 & -27.8027738 & 26.98 & 1.91 & 0.98 & 0.30 & 0.004 & 0.008 + 2977 & 53.1726388 & -27.8006672 & 27.94 & -0.18 & -0.21 & 0.18 & 0.042 & 0.049 + 3166 & 53.1583425 & -27.7949215 & 20.80 & 0.96 & 0.80 & 0.37 & 0.086 & 0.007 + 3561 & 53.1483050 & -27.7977856 & 29.24 & 1.98 & 0.91 & 0.24 & 0.018 & 0.012 + 3794 & 53.1474097 & -27.7965561 & 28.89 & 2.14 & 0.98 & 0.29 & 0.010 & 0.007 + 3940 & 53.1498728 & -27.7961691 & 27.06 & -0.48 & -0.21 & -0.18 & 0.033 & 0.038 + 4120 & 53.1839107 & -27.7954147 & 28.04 & 0.03 & 0.12 & 1.31 & 0.023 & 0.012 + 4322 & 53.1349058 & -27.7944480 & 26.83 & 0.06 & -0.06 & -0.07 & 0.067 & 0.043 + 4643 & 53.1848922 & -27.7933971 & 31.73 & 1.52 & 0.74 & 0.25 & 0.019 & 0.050 + 4839 & 53.1883621 & -27.7923200 & 27.22 & 1.20 & 0.40 & 0.07 & 0.018 & 0.015 + 4945 & 53.2003709 & -27.7898451 & 25.06 & 2.06 & 1.74 & 0.67 & 0.020 & 0.013 + 5317 & 53.1257144 & -27.7904647 & 29.06 & 0.49 & -0.00 & -0.30 & 0.045 & 0.039 + 5441 & 53.1625253 & -27.7896458 & 27.81 & 1.75 & 1.27 & 0.39 & 0.016 & 0.019 + 5921 & 53.1322881 & -27.7828516 & 20.57 & 0.77 & 1.19 & 0.51 & 0.094 & 0.070 + 5992 & 53.1498432 & -27.7874448 & 27.52 & 1.63 & 0.69 & 0.16 & 0.014 & 0.005 + 6334 & 53.1782000 & -27.7861688 & 27.71 & -0.27 & -0.16 & -0.18 & 0.049 & 0.049 + 6442 & 53.1416300 & -27.7856726 & 28.16 & -0.19 & 0.04 & -0.07 & 0.044 & 0.030 + 6461 & 53.1984259 & -27.7848674 & 26.80 & 1.77 & 1.00 & 0.30 & 0.010 & 0.010 + 6620 & 53.1428439 & -27.7849376 & 31.27 & 3.19 & 0.25 & -0.14 & 0.041 & 0.049 + 6732 & 53.1784892 & -27.7840395 & 25.73 & 1.13 & -0.04 & 0.01 & 0.010 & 0.022 + 7113 & 53.1707114 & -27.7826303 & 30.83 & 2.49 & 0.65 & 0.20 & 0.026 & 0.024 + 7194 & 53.1860335 & -27.7822207 & 26.91 & -0.12 & -0.13 & -0.06 & 0.048 & 0.045 + 7357 & 53.1692789 & -27.7813943 & 27.93 & 0.10 & -0.21 & -0.16 & 0.024 & 0.031 + 7525 & 53.1318579 & -27.7820076 & 28.70 & 2.22 & 1.08 & 0.30 & 0.011 & 0.007 + 7768 & 53.1472186 & -27.7714786 & 27.12 & 1.66 & 0.68 & 0.20 & 0.014 & 0.005 + 7894 & 53.1831085 & -27.7802451 & 27.46 & 0.03 & -0.15 & -0.12 & 0.047 & 0.039 + 8081 & 53.1580838 & -27.7701503 & 27.92 & -0.32 & -0.40 & -0.30 & 0.038 & 0.020 + 8157 & 53.1649526 & -27.7736713 & 30.85 & 2.16 & 0.40 & 0.07 & 0.042 & 0.039 + 8186 & 53.1867094 & -27.7735490 & 27.72 & -0.13 & -0.14 & -0.34 & 0.044 & 0.029 + 9006 & 53.1853179 & -27.7799976 & 27.46 & -0.05 & -0.07 & 0.24 & 0.039 & 0.024 + 9020 & 53.1685421 & -27.7805214 & 27.00 & 0.28 & -0.01 & -0.16 & 0.025 & 0.034 + 9212 & 53.1485421 & -27.7701387 & 25.50 & 1.82 & 0.96 & 0.29 & 0.011 & 0.003 + 9230 & 53.1580164 & -27.7691869 & 21.18 & 0.76 & 0.38 & 0.10 & 0.037 & 0.007 + 9331 & 53.1638484 & -27.767128 & 27.95 & 2.20 & 1.84 & 0.67 & 0.025 & 0.005 + 9351 & 53.1781506 & -27.7691241 & 26.21 & 2.03 & 1.32 & 0.42 & 0.015 & 0.004 + 9397 & 53.162852 & -27.7671662 & 21.44 & 0.26 & 0.11 & 0.02 & 0.043 & 0.033 + 9959 & 53.1611698 & -27.7555109 & 29.12 & 2.42 & 0.91 & 0.29 & 0.022 & 0.021 + ccccccc 19 & m1-m2 & no & @xmath112 & @xmath113 & 3.129 & 2.529 + 366 & l0-l1 & no & @xmath114 & @xmath115 & 3.750 & 3.150 + 443 & l0-m6 & no & @xmath116 & @xmath117 & 4.274 & 3.674 + 834 & m4-m5 & no & @xmath118 & @xmath119 & 3.258 & 2.658 + 911 & m4-m5 & no & @xmath118 & @xmath120 & 2.818 & 2.218 + 1147 & f6-f8 & no & @xmath121 & @xmath122 & 2.142 & 1.542 + 1343 & a0-k7 & yes & @xmath123 & @xmath124 & 3.713 & 3.113 + 2150 & m3-m4 & no & @xmath125 & @xmath126 & 2.315 & 1.715 + 2368 & o9-m6 & yes & @xmath127 & @xmath128 & 3.774 & 3.174 + 2457 & m0-m1 & no & @xmath129 & @xmath130 & 3.209 & 2.609 + 2977 & a0-m4 & yes & @xmath131 & @xmath132 & 3.818 & 3.218 + 3166 & k4-k5 & no & @xmath133 & @xmath134 & 2.164 & 1.564 + 3561 & m0-m1 & no & @xmath129 & @xmath135 & 3.647 & 3.047 + 3794 & m0-m1 & no & @xmath129 & @xmath136 & 3.544 & 2.944 + 3940 & o5-m3 & yes & @xmath137 & @xmath138 & 3.702 & 3.102 + 4120 & gal & yes & & & & + 4322 & a7-k5 & no & @xmath139 & @xmath140 & 3.548 & 2.948 + 4643 & a0-m6 & yes & @xmath141 & @xmath142 & 4.237 & 3.637 + 4839 & g5-k2 & no & @xmath143 & @xmath144 & 3.400 & 2.800 + 4945 & m4-m5 & no & @xmath145 & @xmath146 & 2.796 & 2.196 + 5317 & a0-m6 & yes & @xmath147 & @xmath148 & 3.910 & 3.310 + 5441 & m2-m3 & no & @xmath149 & @xmath150 & 3.407 & 2.807 + 5921 & k4-k5 & no & @xmath133 & @xmath151 & 2.155 & 1.555 + 5992 & k7-m0 & no & @xmath152 & @xmath153 & 3.373 & 2.773 + 6334 & f8-l0 & yes & @xmath154 & @xmath155 & 3.791 & 3.191 + 6442 & a0-g5 & yes & @xmath156 & @xmath157 & 3.865 & 3.265 + 6461 & m0-m1 & no & @xmath129 & @xmath158 & 3.202 & 2.602 + 6620 & b8-m4 & yes & @xmath159 & @xmath160 & 3.812 & 3.212 + 6732 & gal & no & & & & + 7113 & g8-k0 & yes & @xmath161 & @xmath162 & 3.861 & 3.261 + 7194 & o5-k5 & yes & @xmath163 & @xmath164 & 3.601 & 3.001 + 7357 & o9-m5 & yes & @xmath165 & @xmath166 & 3.760 & 3.160 + 7525 & m1-m2 & no & @xmath167 & @xmath168 & 3.491 & 2.892 + 7768 & k5-k7 & no & @xmath169 & @xmath170 & 3.286 & 2.686 + 7894 & b8-k7 & yes & @xmath171 & @xmath172 & 3.680 & 3.080 + 8081 & o5-a3 & yes & @xmath173 & @xmath174 & 3.843 & 3.243 + 8157 & gal & yes & & & & + 8186 & b0-k7 & yes & @xmath175 & @xmath176 & 3.764 & 3.164 + 9006 & b9-a3 & yes & @xmath177 & @xmath178 & 3.697 & 3.097 + 9020 & f0-f2 & no & @xmath179 & @xmath180 & 3.540 & 2.940 + 9212 & m0-m1 & no & @xmath129 & @xmath181 & 2.929 & 2.329 + 9230 & k2-k3 & no & @xmath182 & @xmath183 & 2.279 & 1.679 + 9331 & m5-m6 & no & @xmath184 & @xmath185 & 3.345 & 2.745 + 9351 & m1-m2 & no & @xmath186 & @xmath187 & 3.031 & 2.431 + 9397 & gal & no & & & & + 9959 & m0-m1 & no & @xmath129 & @xmath188 & 3.535 & 2.935 + 1147 & ms f6-f8 & 3.7 - 4.0 & @xmath189 & & & f6-f8 + 1147 & young wd & 14.0 - 15.0 & 110 @xmath68 24 & 1.4 - 3.9 & 14.1 & & + 1147 & old wd & 17.4 - 18.3 & 24 @xmath190 5 & 10.3 - 12.1 & 3.0 & & + 3166 & ms k4-k5 & 6.8 - 7.1 & @xmath191 & & & k4-k5 + 3166 & young wd & 14.8 - 15.8 & 80@xmath190 18 & 2.5 - 5.7 & 10.2 & & + 3166 & old wd & 17.2 - 18.2 & 27 @xmath68 6 & 9.9 - 11.5 & 3.5 & & + 4322 & ms a7-k5&2.3 - 7.1 & @xmath192 & & + 4322 & young wd & 12.1 - 13.1 & @xmath193 & 0.4 - 1.4 & 1545.6 & + 4322 & old wd & 17.5 - 18.4 & 590 @xmath68 126 & 10.7 - 12.9 & 75.5 & disk or halo & @xmath194 + 4839 & ms g5-k2 & 5.0 - 6.3 & @xmath195 & & + 4839 & young wd & 12.5 - 13.5 & @xmath196 & 1.1 - 3.3 & 885 . & disk & @xmath197 + 4839 & old wd & 17.4 - 18.3 & 417 @xmath68 88 & 10.4 - 12.3 & 53.4 & disk or halo & @xmath198 + 5921 & ms k4-k5 & 6.8 - 7.1 & @xmath199 & & & k4-k5 + 5921 & young wd & 14.5 - 15.5 & @xmath200 & 1.9 - 4.7 & 11.9 & & + 5921 & old wd & 17.3 - 18.2 & @xmath201 & 10.1 - 11.7 & 3.3 & & + 7768 & ms k5-k7 & 7.1 - 7.6 & @xmath202 & & & + 7768 & young wd & 14.5 - 15.5 & @xmath203 & 1.9 - 4.7 & 160.2 & disk & @xmath204 + 7768 & old wd & 17.3 - 18.2 & 345 @xmath68 70 & 10.1 - 11.7 & 44.1 & disk or halo & @xmath205 + 9020 & ms f0-f2 & 2.7 - 3.6 & @xmath206 & & + 9020 & young wd & 13.3 - 13.3 & @xmath207 & 0.5 - 1.6 & 793.9 & + 9020 & old wd & 17.5 - 18.4 & 576 @xmath68 123 & 10.7 - 12.9 & 73.7 & disk or halo & @xmath208 + 9230 & ms k2-k3 & 6.3 - 6.7 & @xmath209 & & & k2-k3 + 9230 & young wd & 13.6 - 14.6 & 187 @xmath68 41 & 1.1 - 3.2 & 23.9 & & + 9230 & old wd & 17.5 - 18.4 & 32 @xmath68 7 & 10.4 - 12.3 & 4.0 & & + ( green ) and @xmath39 ( red ) values for the hudf objects . the stellarity index was defined ( see section [ sindexsec ] ) so that higher values correspond to objects which are increasingly resolved in the images . we selected unresolved objects fainter than @xmath210 mag to be unresolved if @xmath2110.05 , and @xmath2110.15 for brighter objects which are saturated.[fig1],width=480 ] values for the brightest objects in the field . each object appears 10 times in this figure , with increasingly lower signal to noise ( increasingly darker dots ) . the observed increase in values as a function of magnitude at @xmath212 mag is similar to the one observed in figure [ fig1 ] where the real distribution of the values from the hudf objects is shown . this demonstrates the robustness of to distinguish between point sources and extended objects down to faint magnitudes . at fainter magnitudes , images of stars become increasingly dominated by noise which tends to increases the measured values of . the same test was performed in the band and yielded an identical behavior . [ fig2],width=480 ] mag and @xmath217 mag . solid red squares are m dwarfs . empty red squares are the two l dwarf candidates . solid green triangles represent the non m or l dwarfs stars which could be main - sequence stars . the empty blue circles are white dwarf candidates ( table [ table4 ] ) . the solid yellow circles represent the qsos . [ vbv],width=480 ] , unresolved , spectroscopically identified objects in the hudf . m and l dwarfs are shown using red squares with photometric error bars . main - sequence stars are shown in green thick circles . extra - galactic objects , including those dimmer than @xmath1 mag , are shown using large blue stars . the 5 white dwarf candidates are shown using large blue crosses . pickles main - sequence objects are also shown using small black circles . the ( - ) locations of the o , a , f , g , k , and m stellar type are shown for reference . hot white dwarfs are shown in large blue squares . cool white dwarfs ( ( 3000k , 4000k , 5000k ) , * ? ? ? * ) are shown using labeled , empty circles . finally , the solid black line shows the qso track , from redshift of z=4.0 ( top right ) to z=0.1 ( bottom left)[vibv],width=480 ] values of 200 , 300 , 400 and 500 pc ( bottom curves to top curves respectively ) . for each value of @xmath65 , mass functions corresponding to @xmath218 and 1.5 are shown ( solid and dashed lines respectively ) . the observed hudf distribution is shown using a solid black line which is best fit by a disk with a scale height of @xmath219 while reasonably excluding values of @xmath65 = 200,300 , and 500.[mlf2],width=480 ]

the-vector is an important enumerative invariant of a flag simplicial homology sphere . it has been conjectured by gal that this vector is nonnegative for every such sphere and by reiner , postnikov and williams that it increases when is replaced by any flag simplicial homology sphere which geometrically subdivides . using the nonnegativity of the-vector in dimension 3 , proved by davis and okun , as well as stanley s theory of simplicial subdivisions and local-vectors , the latter conjecture is confirmed in this paper in dimensions 3 and 4 .

1106.4520

this paper is concerned with the face enumeration of an important class of simplicial complexes , that of flag homology spheres , and their subdivisions . the face vector of a homology sphere ( more generally , of an eulerian simplicial complex ) @xmath1 can be conveniently encoded by its @xmath0-vector @xcite , denoted @xmath3 . part of our motivation comes from the following two conjectures ( we refer to section [ sec : sub ] for all relevant definitions ) . the first , proposed by gal ( * ? ? ? * conjecture 2.1.7 ) , can be thought of as a generalized lower bound conjecture for flag homology spheres ( it strengthens an earlier conjecture by charney and davis @xcite ) . the second , proposed by postnikov , reiner and williams ( * ? ? ? * conjecture 14.2 ) , is a natural extension of the first . [ gal @xcite ] [ conj : gal ] for every flag homology sphere @xmath1 we have @xmath4 . [ postnikov , reiner and williams @xcite ] [ conj : monotone ] for all flag homology spheres @xmath1 and @xmath5 for which @xmath5 geometrically subdivides @xmath1 , we have @xmath6 . the previous statements are trivial for spheres of dimension two or less . conjecture [ conj : gal ] was proved for 3-dimensional spheres by davis and okun ( * ? ? ? * theorem 11.2.1 ) and was deduced from that result for 4-dimensional spheres in ( * ? ? ? * corollary 2.2.3 ) . conjecture [ conj : monotone ] can be thought of as a conjectural analogue of the fact ( * ? ? ? * theorem 4.10 ) that the @xmath2-vector ( a certain linear transformation of the face vector ) of a cohen - macaulay simplicial complex increases under quasi - geometric simplicial subdivision ( a class of topological subdivisions which includes all geometric simplicial subdivisions ) . the main result of this paper proves its validity in three and four dimensions for a new class of simplicial subdivisions , which includes all geometric ones . [ thm:34 ] for every flag homology sphere @xmath1 of dimension 3 or 4 and for every flag vertex - induced homology subdivision @xmath5 of @xmath1 , we have @xmath7 . the previous result naturally suggests the following stronger version of conjecture [ conj : monotone ] . [ conj : newmonotone ] for every flag homology sphere @xmath1 and every flag vertex - induced homology subdivision @xmath5 of @xmath1 , we have @xmath8 . the following structural result on flag homology spheres , which may be of independent interest , will also be proved in section [ sec : pl ] . it implies , for instance , that conjecture [ conj : newmonotone ] is stronger than conjecture [ conj : gal ] . throughout this paper , we will denote by @xmath9 the boundary complex of the @xmath10-dimensional cross - polytope ( equivalently , the simplicial join of @xmath10 copies of the zero - dimensional sphere ) . [ thm : main ] every flag @xmath11-dimensional homology sphere is a vertex - induced ( hence quasi - geometric and flag ) homology subdivision of @xmath9 . the proof of theorem [ thm:34 ] relies on the theory of face enumeration for simplicial subdivisions , developed by stanley @xcite . given a simplicial complex @xmath1 and a simplicial subdivision @xmath5 of @xmath1 , the @xmath2-vector of @xmath5 can be expressed in terms of local contributions , one for each face of @xmath1 , and the combinatorics of @xmath1 ( * ? ? ? * theorem 3.2 ) . the local contributions are expressed in terms of the key concept of a local @xmath2-vector , introduced and studied in @xcite . when @xmath1 is eulerian , this formula transforms to one involving @xmath0-vectors ( proposition [ prop : gammaformula ] ) and leads to the concept of a local @xmath0-vector , introduced in section [ sec : gamma ] . using the davis - okun theorem @xcite , mentioned earlier , it is shown that the local @xmath0-vector has nonnegative coefficients for every flag vertex - induced homology subdivision of the 3-dimensional simplex . theorem [ thm:34 ] is deduced from these results in section [ sec : gamma ] . the proof of theorem [ thm : main ] is motivated by that of ( * ? ? ? * theorem 1.2 ) , stating that the graph of any flag simplicial pseudomanifold of dimension @xmath12 contains a subdivision of the graph of @xmath9 . we now briefly describe the content and structure of this paper . sections [ sec : sub ] and [ sec : enu ] provide the necessary background on simplicial complexes , subdivisions and their face enumeration . the notion of a homology subdivision , which is convenient for the results of this paper , as well as those of a flag subdivision and vertex - induced ( a natural strengthening of quasi - geometric ) subdivision , are introduced in section [ subsec : sub ] . section [ sec : monolocal ] includes a simple example ( see example [ ex : counter ] ) which shows that there exist quasi - geometric subdivisions of the simplex with non - unimodal local @xmath2-vector . section [ sec : pl ] proves theorem [ thm : main ] and another structural result on flag subdivisions ( proposition [ prop : balltosphere ] ) , stating that every flag vertex - induced homology subdivision of the @xmath11-dimensional simplex naturally occurs as a restriction of a flag vertex - induced homology subdivision of @xmath9 . these results are used in section [ sec : gamma ] . local @xmath0-vectors are introduced in section [ sec : gamma ] , where examples and elementary properties are discussed . it is conjectured there that the local @xmath0-vector has nonnegative coordinates for every flag vertex - induced homology subdivision of the simplex ( conjecture [ conj : main ] ) . this statement can be considered as a local analogue of conjecture [ conj : gal ] . it is shown to imply both conjectures [ conj : gal ] and [ conj : newmonotone ] and to hold in dimension three . section [ sec : gamma ] concludes with the proof of theorem [ thm:34 ] . section [ sec : evidence ] discusses some special cases of conjecture [ conj : main ] . for instance , the conjecture is shown to hold for iterated edge subdivisions ( in the sense of ( * ? ? ? * section 5.3 ) ) of the simplex . this section reviews background material on simplicial complexes , in particular on their homological properties and subdivisions . for more information on these topics , the reader is referred to @xcite . throughout this paper , @xmath13 will be a field which we will assume to be fixed . we will denote by @xmath14 the cardinality , and by @xmath15 the set of all subsets , of a finite set @xmath16 . all simplicial complexes we consider will be abstract and finite . thus , given a finite set @xmath17 , a _ simplicial complex _ on the ground set @xmath17 is a collection @xmath1 of subsets of @xmath17 such that @xmath18 implies @xmath19 . the elements of @xmath1 are called _ the dimension of a face @xmath20 is defined as one less than the cardinality of @xmath20 . the dimension of @xmath1 is the maximum dimension of a face and is denoted by @xmath21 . faces of @xmath1 of dimension zero or one are called _ vertices _ or _ edges _ , respectively . a _ facet _ of @xmath1 is a face which is maximal with respect to inclusion . all topological properties of @xmath1 we mention in the sequel will refer to those of the geometric realization @xmath22 of @xmath1 ( * ? ? ? * section 9 ) , uniquely defined up to homeomorphism . for example , we say that @xmath1 is a simplicial ( topological ) ball ( respectively , sphere ) if @xmath22 is homeomorphic to a ball ( respectively , sphere ) . the _ open star _ of a face @xmath19 , denoted @xmath23 , is the collection of all faces of @xmath1 which contain @xmath20 . the _ closed star _ of @xmath19 , denoted @xmath24 , is the subcomplex of @xmath1 consisting of all subsets of the elements of @xmath23 . the _ link _ of the face @xmath19 is the subcomplex of @xmath1 defined as @xmath25 . the _ simplicial join _ @xmath26 of two collections @xmath27 and @xmath28 of subsets of disjoint ground sets is the collection whose elements are the sets of the form @xmath29 , where @xmath30 and @xmath31 . if @xmath27 and @xmath28 are simplicial complexes , then so is @xmath26 . the simplicial join of @xmath1 with the zero - dimensional complex @xmath32 is denoted by @xmath33 and called the _ cone over @xmath1 _ on the ( new ) vertex @xmath34 . a simplicial complex @xmath1 is called _ flag _ if every minimal non - face of @xmath1 has two elements . the closed star and the link of any face of a flag complex , and the simplicial join of two ( or more ) flag complexes , are also flag complexes . in particular , the simplicial join of @xmath10 copies of the zero - dimensional complex with two vertices is a flag complex ( in fact , a flag triangulation of the @xmath11-dimensional sphere ) , which will be denoted by @xmath9 . explicitly , @xmath9 can be described as the simplicial complex on the @xmath35-element ground set @xmath36 whose faces are those subsets of @xmath37 which contain at most one element from each of the sets @xmath38 for @xmath39 . let @xmath1 be a simplicial complex of dimension @xmath12 . we call @xmath1 a _ homology sphere _ ( over @xmath13 ) if for every @xmath19 ( including @xmath40 ) we have @xmath41 where @xmath42 denotes reduced simplicial homology of @xmath43 with coefficients in @xmath13 . we call @xmath1 a _ homology ball _ ( over @xmath13 ) if there exists a subcomplex @xmath44 of @xmath1 , called the _ boundary _ of @xmath1 , so that the following hold : * @xmath44 is a @xmath45-dimensional homology sphere over @xmath13 , * for every @xmath19 ( including @xmath46 ) we have @xmath47 the _ interior _ of @xmath1 is defined as @xmath48 , if @xmath1 is a homology sphere , and @xmath49 , if @xmath1 is a homology ball . for example , the simplicial complex @xmath50 with a unique vertex @xmath34 is a zero - dimensional homology ball ( over any field ) with boundary @xmath51 and interior @xmath52 . if @xmath1 is a homology ball of dimension @xmath12 , then @xmath44 consists exactly of the faces of those @xmath45-dimensional faces of @xmath1 which are contained in a unique facet of @xmath1 . [ rem : homjoin ] it follows from standard facts ( * ? ? ? * ( 9.12 ) ) on the homology of simplicial joins that the simplicial join of a homology sphere and a homology ball , or of two homology balls , is a homology ball and that the simplicial join of two homology spheres is again a homology sphere . moreover , in each case the interior of the simplicial join is equal to the simplicial join of the interiors of the two complexes in question . we will adopt the following notion of homology subdivision of an abstract simplicial complex . this notion generalizes that of topological subdivision of ( * ? ? ? * section 2 ) . we should point out that the class of homology subdivisions of simplicial complexes is contained in the much broader class of formal subdivisions of eulerian posets , introduced and studied in ( * ? ? ? * section 7 ) . [ def : sub ] let @xmath1 be a simplicial complex . a ( finite , simplicial ) _ homology subdivision _ of @xmath1 ( over @xmath13 ) is a simplicial complex @xmath5 together with a map @xmath53 such that the following hold for every @xmath19 : ( a ) the set @xmath54 is a subcomplex of @xmath5 which is a homology ball ( over @xmath13 ) of dimension @xmath55 ; and ( b ) @xmath56 consists of the interior faces of @xmath57 . such a map @xmath58 is said to be a _ subdivision if the complex @xmath57 is homeomorphic to a ball of dimension @xmath55 for every @xmath59 . let @xmath53 be a homology subdivision of @xmath1 . from the defining properties , it follows that the map @xmath58 is surjective and that @xmath60 for every @xmath61 . given faces @xmath62 and @xmath19 , the face @xmath63 of @xmath1 is called the _ carrier _ of @xmath64 ; the subcomplex @xmath57 is called the _ restriction _ of @xmath5 to @xmath20 . the subdivision @xmath58 is called _ quasi - geometric _ * definition 4.1 ( a ) ) if there do not exist @xmath61 and face @xmath19 of dimension smaller than @xmath65 , such that the carrier of every vertex of @xmath64 is contained in @xmath20 . moreover , @xmath58 is called _ geometric _ ( * ? ? ? * definition 4.1 ( b ) ) if there exists a geometric realization of @xmath5 which geometrically subdivides a geometric realization of @xmath1 , in the way prescribed by @xmath58 . clearly , if @xmath53 is a homology ( respectively , topological ) subdivision , then the restriction of @xmath58 to @xmath57 is also a homology ( respectively , topological ) subdivision of the simplex @xmath66 for every @xmath67 . moreover , if @xmath58 is quasi - geometric ( respectively , geometric ) , then so are all its restrictions @xmath57 for @xmath19 . as part ( c ) of the following example shows , the restriction of @xmath58 to a face @xmath19 need not be a flag complex , even when @xmath5 and @xmath1 are flag complexes and @xmath58 is quasi - geometric . = 2.2 in [ ex : nonflagrest ] consider a 3-dimensional simplex @xmath68 , with @xmath69 , and set @xmath70 . \(a ) let @xmath43 be the simplicial complex consisting of the subsets of @xmath71 and the subsets of @xmath72 and let @xmath73 be the subdivision ( considered in part ( h ) of ( * ? ? ? * example 2.3 ) ) which pushes @xmath43 into the simplex @xmath68 , so that the face @xmath20 of @xmath43 ends up in the interior of @xmath68 and @xmath74 ends up in the interior of @xmath66 . formally , for @xmath75 we let @xmath76 , if @xmath77 , we let @xmath78 , if @xmath64 contains @xmath20 and otherwise we let @xmath79 . then @xmath43 is a flag complex and the restriction @xmath80 of @xmath58 is the cone over the boundary of @xmath66 ( with new vertex @xmath74 ) , shown in figure [ fig : nonregtri ] ( a ) , which is not flag . \(b ) let @xmath81 be the simplicial complex consisting of the faces of the simplex @xmath68 and those of the cone , on a vertex @xmath34 , over the boundary of the simplex with vertex set @xmath72 ( note that @xmath81 is not flag ) . consider the subdivision @xmath82 which satisfies @xmath83 for every face @xmath84 containing @xmath34 and otherwise agrees with the subdivision @xmath58 of part ( a ) . then @xmath85 is quasi - geometric and its restriction @xmath86 is again the non - flag complex shown in figure [ fig : nonregtri ] ( a ) . \(c ) let @xmath87 be the simplicial complex on the ground set @xmath88 whose faces are @xmath20 and those of the simplicial subdivision of @xmath66 , shown in figure [ fig : nonregtri ] ( b ) . let @xmath89 consist of the faces of @xmath68 and those of the cone over @xmath87 on a new vertex @xmath34 . we leave to the reader to verify that @xmath89 is a flag simplicial complex and that it admits a quasi - geometric subdivision @xmath90 ( satisfying @xmath91 ) for which the restriction @xmath92 is the non - flag simplicial complex shown in figure [ fig : nonregtri ] ( b ) . the previous examples suggest the following definitions . [ def : flagsub ] let @xmath93 be simplicial complexes and let @xmath53 be a homology subdivision . * we say that @xmath58 is _ vertex - induced _ if for all faces @xmath61 and @xmath19 the following condition holds : if every vertex of @xmath64 is a vertex of @xmath57 , then @xmath94 . * we say that @xmath58 is a _ flag subdivision _ if the restriction @xmath57 is a flag complex for every face @xmath19 . for homology ( respectively , topological ) subdivisions we have the hierarchy of properties : geometric @xmath95 vertex - induced @xmath95 quasi - geometric . the subdivision @xmath43 of example [ ex : nonflagrest ] is not quasi - geometric , while @xmath81 and @xmath89 are quasi - geometric but not vertex - induced ( none of the three subdivisions is flag ) . thus the second implication above is strict . an example discussed on @xcite shows that the first implication is strict as well . we also point out here that if @xmath53 is a vertex - induced homology subdivision and the simplicial complex @xmath5 is flag , then @xmath58 is a flag subdivision . * joins and links . * the notion of a ( vertex - induced , or flag ) homology subdivision behaves well with respect to simplicial joins and links , as we now explain . let @xmath96 and @xmath97 be homology subdivisions of two simplicial complexes @xmath27 and @xmath28 on disjoint ground sets . the simplicial join @xmath98 is naturally a homology subdivision of @xmath26 with subdivision map @xmath99 defined by @xmath100 for @xmath101 and @xmath102 . indeed , given faces @xmath30 and @xmath103 , the restriction of @xmath98 to the face @xmath104 is equal to @xmath105 which , by remark [ rem : homjoin ] , is a homology ball of dimension equal to that of @xmath29 . moreover , @xmath106 and hence @xmath56 is the interior of this ball . similarly , let @xmath53 be a homology subdivision and let @xmath20 be a common face of @xmath1 and @xmath5 ( such as a vertex of @xmath1 ) which satisfies @xmath107 . an easy application of part ( ii ) of lemma [ lem : homology ] shows that @xmath108 is a homology subdivision of @xmath109 with subdivision map @xmath110 defined by @xmath111 . we will refer to this subdivision as the link of @xmath58 at @xmath20 ; its restriction to a face @xmath112 satisfies @xmath113 . the following statement is an easy consequence of the relevant definitions ; the proof is left to the reader . [ lem : joinlink ] the simplicial join of two vertex - induced ( respectively , flag ) homology subdivisions is also vertex - induced ( respectively , flag ) . the link of a vertex - induced ( respectively , flag ) homology subdivision is also vertex - induced ( respectively , flag ) . * stellar subdivisions . * we recall the following standard way to subdivide a simplicial complex . given a simplicial complex @xmath1 on the ground set @xmath17 , a face @xmath19 and an element @xmath34 not in @xmath17 , the _ stellar subdivision _ of @xmath1 on @xmath20 ( with new vertex @xmath34 ) is the simplicial complex @xmath114 on the ground set @xmath115 , where @xmath116 . the map @xmath53 , defined by @xmath117 for @xmath61 , is a topological ( and thus a homology ) subdivision of @xmath1 . we leave to the reader to check that if @xmath1 is a flag complex and @xmath59 is an edge , then the stellar subdivision of @xmath1 on @xmath20 is again a flag complex . this section reviews the definitions and main properties of the enumerative invariants of simplicial complexes and their subdivisions which will appear in the following sections , namely the @xmath2-vector of a simplicial complex , the @xmath0-vector of an eulerian simplicial complex and the local @xmath2-vector of a simplicial subdivision of a simplex . some new results on local @xmath2-vectors are included . a fundamental enumerative invariant of a @xmath11-dimensional simplicial complex @xmath1 is the @xmath2-polynomial , defined by @xmath118 the @xmath2-vector of @xmath1 is the sequence @xmath119 , where @xmath120 . the number @xmath121 is the reduced euler characteristic of @xmath1 and is denoted by @xmath122 . the polynomial @xmath123 satisfies @xmath124 ( * ? ? ? * section 3.14 ) if @xmath1 is an _ complex , meaning that @xmath125 holds for every @xmath19 . for the simplicial join of two simplicial complexes @xmath27 and @xmath28 we have @xmath126 . for a homology ball or sphere @xmath1 of dimension @xmath12 we set @xmath127 and recall the following well - known statement ( see , for instance , theorem 7.1 in ( * ? ? ? * chapter ii ) and ( * ? ? ? * section 2.1 ) for additional references ) . [ lem : homformal ] let @xmath1 be a @xmath11-dimensional simplicial complex . if @xmath1 is either a homology ball or a homology sphere over @xmath13 , then @xmath128 . let @xmath129 be a vector with real coordinates and let @xmath130 be the associated real polynomial of degree at most @xmath10 . we say that _ @xmath131 has symmetric coefficients _ , and that the vector @xmath2 is _ symmetric _ , if @xmath132 holds for @xmath133 . it is easy to check ( * ? ? ? * proposition 2.1.1 ) that @xmath131 has symmetric coefficients if and only if there exists a real polynomial @xmath134 of degree at most @xmath135 , satisfying @xmath136 in that case , @xmath137 is uniquely determined by @xmath131 and called the _ @xmath0-polynomial _ associated to @xmath131 ; the sequence @xmath138 is called the _ @xmath0-vector _ associated to @xmath2 . we will refer to the @xmath0-polynomial associated to the @xmath2-polynomial of an eulerian complex @xmath1 as the _ @xmath0-polynomial of @xmath1 _ and will denote it by @xmath139 . similarly , we will refer to the @xmath0-vector associated to the @xmath2-vector of an eulerian complex @xmath1 as the _ @xmath0-vector of @xmath1 _ and will denote it by @xmath3 . we now recall some of the basics of the theory of face enumeration for subdivisions of simplicial complexes @xcite ( * ? ? ? * section iii.10 ) . the following definition is a restatement of ( * ? ? ? * definition 2.1 ) for homology ( rather than topological ) subdivisions of the simplex . [ def : localh ] let @xmath71 be a set with @xmath10 elements and let @xmath43 be a homology subdivision of the simplex @xmath68 . the polynomial @xmath140 defined by @xmath141 is the _ local @xmath2-polynomial _ of @xmath43 ( with respect to @xmath71 ) . the sequence @xmath142 is the _ local @xmath2-vector _ of @xmath43 ( with respect to @xmath71 ) . the following theorem , stated for homology subdivisions , summarizes some of the main properties of local @xmath2-vectors ( see theorems 3.2 and 3.3 and corollary 4.7 in @xcite ) . [ thm : stalocal ] * for every homology subdivision @xmath5 of a simplicial complex @xmath1 we have @xmath143 * the local @xmath2-vector @xmath144 is symmetric for every homology subdivision @xmath43 of the simplex @xmath68 . * the local @xmath2-vector @xmath144 has nonnegative coordinates for every quasi - geometric homology subdivision @xmath43 of the simplex @xmath68 . parts ( i ) and ( iii ) follow from the proofs of theorems 3.2 and 4.6 , respectively , in @xcite . moreover , lemma [ lem : homformal ] implies that every homology subdivision of a simplicial complex is a formal subdivision , in the sense of ( * ? ? ? * definition 7.4 ) . thus , parts ( i ) and ( ii ) are special cases of corollary 7.7 and theorem 7.8 , respectively , in @xcite . [ ex : counter ] the local @xmath2-polynomial of the subdivision in part ( a ) of example [ ex : nonflagrest ] was computed in @xcite as @xmath145 . this shows that the assumption in theorem [ thm : stalocal ] ( iii ) that @xmath43 is quasi - geometric is essential . for part ( b ) of example [ ex : nonflagrest ] we can easily compute that @xmath146 . since @xmath81 is quasi - geometric , this disproves ( * ? ? ? * conjecture 5.4 ) ( see also ( * ? ? ? * section 6 ) @xcite ) , stating that local @xmath2-vectors of quasi - geometric subdivisions are unimodal . the previous example suggests the following question . [ que : unimodal ] is the local @xmath2-vector @xmath144 unimodal for every vertex - induced homology subdivision @xmath43 of the simplex @xmath68 ? we now show that local @xmath2-vectors also enjoy a locality property ( this will be useful in the proof of proposition [ prop : edgesub ] ) . [ prop : localformula ] let @xmath73 be a homology subdivision of the simplex @xmath68 . for every homology subdivision @xmath81 of @xmath43 we have @xmath147 where @xmath148 for @xmath75 . by assumption , @xmath149 is a homology subdivision of @xmath80 for every @xmath150 . thus , using the defining equation ( [ eq : deflocalh ] ) for @xmath151 and ( [ eq : hformula ] ) to expand @xmath152 for @xmath150 , we get @xmath153 and the proof follows . [ rem : relative ] we call the polynomial @xmath154 , defined by ( [ eq : deflocalhrel ] ) , the _ relative local @xmath2-polynomial _ of @xmath43 ( with respect to @xmath71 ) at @xmath64 . this polynomial reduces to @xmath155 for @xmath156 and shares many of the important properties of @xmath155 , established in @xcite . for instance , using ideas of @xcite and their refinements in @xcite , one can show that @xmath154 has symmetric coefficients , in the sense that @xmath157 for every homology subdivision @xmath43 of @xmath68 and @xmath75 and that @xmath158 has nonnegative coefficients for every quasi - geometric homology subdivision @xmath43 of @xmath68 and @xmath75 . the following monotonicity property of local @xmath2-vectors is a consequence of the latter statement and ( [ eq : localformula ] ) : for every quasi - geometric homology subdivision @xmath43 of @xmath68 and every quasi - geometric homology subdivision @xmath81 of @xmath43 , we have @xmath159 . since these results will not be used in this paper , detailed proofs will appear elsewhere . this section proves theorem [ thm : main ] , as well as a result on flag subdivisions of the simplex ( proposition [ prop : balltosphere ] ) which will be used in section [ sec : gamma ] . the following lemma gives several technical properties of homology balls and spheres ( over the field @xmath13 ) . we will only sketch the proof , which is fairly straightforward and uses standard tools from algebraic topology . [ lem : homology ] * if @xmath1 is a homology sphere ( respectively , ball ) of dimension @xmath12 , then @xmath109 is a homology sphere of dimension @xmath160 for every @xmath19 ( respectively , for every interior face @xmath19 ) . * if @xmath1 is a homology ball of dimension @xmath12 and @xmath161 is a boundary face , then @xmath109 is a homology ball of dimension @xmath160 with interior equal to @xmath162 . * if @xmath1 is a homology sphere ( respectively , ball ) , then the cone over @xmath1 is a homology ball whose boundary is equal to @xmath1 ( respectively , to the union of @xmath1 with the cone over the boundary of @xmath1 ) . * let @xmath27 and @xmath28 be homology balls of dimension @xmath10 . if @xmath163 is a homology ball of dimension @xmath12 which is contained in ( respectively , is equal to ) the boundary of both @xmath27 and @xmath28 , then @xmath164 is a homology ball ( respectively , sphere ) of dimension @xmath10 . * let @xmath1 be a homology sphere of dimension @xmath12 . if @xmath43 is a subcomplex of @xmath1 which is a homology ball of dimension @xmath12 , then the complement of the interior of @xmath43 in @xmath1 is also a homology ball of dimension @xmath12 whose boundary is equal to that of @xmath43 . we first observe that for all faces @xmath19 and @xmath165 , the link of @xmath64 in @xmath109 is equal to @xmath166 . moreover , if @xmath1 is a homology ball and @xmath20 is an interior face , then so is @xmath167 . part ( i ) follows from these facts and the definition of homology balls and spheres . part ( ii ) is an easy consequence of part ( i ) and the relevant definitions . part ( iii ) is an easy consequence of the relevant definitions and the fact that cones have vanishing reduced homology . part ( iv ) follows by an easy application of the mayer - vietoris long exact sequence @xcite . for the last part , we let @xmath168 denote the complement of the interior of @xmath43 in @xmath1 and note that the pairs @xmath169 and @xmath170 are compact triangulated relative homology manifolds which are orientable over @xmath13 . applying the lefschetz duality theorem @xcite and the long exact homology sequence @xcite to these pairs shows that @xmath168 has trivial reduced homology over @xmath13 . similar arguments work for the links of faces of @xmath168 . the details are omitted . [ rem : pl ] although not all parts of lemma [ lem : homology ] remain valid if homology balls and spheres are replaced by topological balls and spheres , they do hold for the subclasses of pl balls and pl spheres ( we refer the reader to ( * ? ? ? * section 4.7 ( d ) ) for this claim , for a short introduction to pl topology and for additional references ) . thus , the results of this paper remain valid when homology balls and spheres are replaced by pl balls and spheres and the notion of homology subdivision is replaced by its natural pl analogue . the following lemma will also be essential in the proof of theorem [ thm : main ] . a similar result has appeared in ( * ? ? ? * lemma 3.2 ) . [ lem : starunion ] let @xmath1 be a flag @xmath11-dimensional homology sphere . for every nonempty face @xmath20 of @xmath1 , the subcomplex @xmath171 is a homology @xmath11-dimensional ball whose interior is equal to @xmath172 . we set @xmath173 , @xmath174 and @xmath175 and proceed by induction on the cardinality @xmath176 of @xmath20 . for @xmath177 , the complex @xmath168 is the cone over @xmath178 on the vertex @xmath179 . since @xmath1 is a homology sphere , the result follows from parts ( i ) and ( iii ) of lemma [ lem : homology ] . suppose that @xmath180 . we will also assume that @xmath181 , since the result is trivial otherwise ( we note that the assumption that @xmath1 is flag is essential in the case @xmath182 ) . since , by lemma [ lem : homology ] ( i ) , links of flag homology spheres are also flag homology spheres , the complex @xmath183 is a flag homology sphere of dimension @xmath184 and @xmath185 is a nonempty face of @xmath43 . thus , by the induction hypothesis , the union @xmath186 is a homology ball of dimension @xmath184 . let @xmath87 denote the boundary of @xmath187 and let @xmath188 denote the complement of the interior of @xmath187 in @xmath43 . thus @xmath87 is a homology sphere of dimension @xmath189 and , by part ( v ) of lemma [ lem : homology ] , @xmath188 is a homology ball of dimension @xmath184 whose boundary is equal to @xmath87 . consider the union @xmath190 and the cones @xmath191 and @xmath192 . it is straightforward to verify that @xmath193 and that @xmath194 . we note that @xmath195 is a homology ball of dimension @xmath12 , by the induction hypothesis , and that @xmath196 and @xmath197 are homology balls of dimension @xmath12 and @xmath184 , respectively , by part ( iii ) of lemma [ lem : homology ] . by the induction hypothesis , the interior of @xmath187 is equal to @xmath198 . therefore , none of the faces of @xmath87 contains any of @xmath199 and hence the same holds for @xmath197 . since , by the induction hypothesis , the interior of @xmath195 is equal to @xmath200 , we conclude that @xmath197 is contained in the boundary of @xmath195 . moreover , @xmath197 is also contained in the boundary of @xmath196 , since @xmath87 is contained in the boundary of @xmath188 . it follows from the previous discussion and lemma [ lem : homology ] ( iv ) that @xmath168 is a homology @xmath11-dimensional ball . we now verify that the interior of @xmath168 is equal to @xmath201 . this statement may be derived from the previous inductive argument , since the interior of @xmath168 is equal to the union of the interiors of @xmath195 , @xmath196 and @xmath197 . we give the following alternative argument . since @xmath168 is a homology ball , its boundary consists of all faces of the @xmath45-dimensional faces of @xmath168 which are contained in exactly one facet of @xmath168 . the validity of the statement for @xmath177 implies that these @xmath45-dimensional faces of @xmath168 are precisely those which do not contain any of the @xmath202 and which are not contained in more than one of the subcomplexes @xmath203 . however , since @xmath1 is @xmath11-dimensional and flag , no @xmath45-dimensional face of @xmath1 may be contained in more than one of the @xmath203 . thus , the boundary of @xmath168 consists precisely of its faces which do not contain any of the @xmath202 and the proof follows . let @xmath1 be a flag simplicial complex of dimension @xmath12 and @xmath9 be the simplicial join of the zero - dimensional spheres @xmath204 for @xmath39 . we fix a facet @xmath205 of @xmath1 and for @xmath206 we define @xmath207 clearly , @xmath63 can not contain any of the sets @xmath38 . thus we have @xmath208 for every @xmath206 and hence we get a map @xmath209 . we will prove that this map is a homology subdivision of @xmath9 , if @xmath1 is a homology sphere . given a face @xmath210 , we need to show that @xmath211 is a subcomplex of @xmath1 of dimension @xmath55 which is a homology ball with interior @xmath56 . we denote by @xmath16 the subset of @xmath205 consisting of all vertices @xmath212 for which @xmath213 and distinguish two cases : case 1 : @xmath214 . we may assume that @xmath215 for some @xmath216 . setting @xmath217 , the defining equation ( [ eq : mainsigmadef ] ) shows that @xmath211 is equal to the intersection of @xmath218 with the complement of @xmath219 in @xmath1 and that @xmath56 consists of those faces of @xmath211 which contain @xmath220 and do not belong to any of the @xmath221 for @xmath222 . consider the complex @xmath223 and let @xmath168 denote the complement of the union @xmath224 in @xmath43 . since links of homology spheres are also homology spheres ( see part ( i ) of lemma [ lem : homology ] ) , the complex @xmath43 is a homology sphere of dimension @xmath160 . by lemma [ lem : starunion ] and part ( v ) of lemma [ lem : homology ] , @xmath168 is a homology ball of dimension @xmath160 whose interior is equal to the set of those faces of @xmath168 which do not belong to any of the @xmath225 for @xmath222 . from the above we conclude that @xmath211 is equal to the simplicial join of the simplex @xmath226 and @xmath168 and that @xmath56 is equal to the simplicial join of @xmath227 and the interior of @xmath168 . the result now follows from part ( iii ) of lemma [ lem : homology ] and the previous discussion . case 2 : @xmath228 . then @xmath211 is contained in @xmath229 . as a result , replacing @xmath1 by @xmath229 reduces this case to the previous one . finally , we note that ( [ eq : mainsigmadef ] ) may be rewritten as @xmath230 , where @xmath231 for every vertex @xmath232 of @xmath1 . this implies that for every @xmath206 , the carrier of @xmath64 is equal to the union of the carriers of the vertices of @xmath64 . as a result , @xmath58 is vertex - induced and the proof follows . [ cor : main1 ] given any flag homology sphere @xmath1 of dimension @xmath12 , there exist simplicial complexes @xmath80 , one for each face @xmath210 , with the following properties : ( a ) @xmath80 is a flag vertex - induced homology subdivision of the simplex @xmath66 for every @xmath210 ; and ( b ) we have @xmath233 this statement follows by applying ( [ eq : hformula ] ) to the subdivision of @xmath1 guaranteed by theorem [ thm : main ] and noting that for every @xmath234 , the restriction @xmath80 of this subdivision to @xmath20 has the required properties and that @xmath235 . [ rem : hvecmin ] it follows from ( [ eq : corpl ] ) and theorem [ thm : stalocal ] ( iii ) that @xmath236 for every flag @xmath11-dimensional homology sphere @xmath1 . this inequality was proved , more generally , for every flag @xmath11-dimensional doubly cohen - macaulay simplicial complex @xmath1 in ( * ? ? ? * theorem 1.3 ) . we now fix a @xmath10-element set @xmath237 and a homology subdivision @xmath43 of @xmath68 , with subdivision map @xmath73 . we let @xmath238 be a @xmath10-element set which is disjoint from @xmath71 and consider the union @xmath1 of all collections of the form @xmath239 , where @xmath240 and @xmath241 are subsets of @xmath242 and @xmath71 , respectively , and @xmath243 ranges over all ordered pairs of disjoint subsets of @xmath244 . clearly , @xmath1 is a simplicial complex which contains as a subcomplex @xmath43 ( set @xmath245 ) and the simplex @xmath246 ( set @xmath247 ) . we let @xmath9 be as in the proof of theorem [ thm : main ] and define the map @xmath248 by @xmath249 for all @xmath250 and @xmath251 such that @xmath252 . the second result of this section is as follows . [ prop : balltosphere ] under the established assumptions and notation , the following hold : * the complex @xmath1 is a @xmath11-dimensional homology sphere . * endowed with the map @xmath253 , the complex @xmath1 is a homology subdivision of @xmath9 . * if @xmath43 is flag and vertex - induced , then @xmath1 is a flag simplicial complex and a flag , vertex - induced homology subdivision of @xmath9 . we first verify ( ii ) . we consider any face @xmath254 , so that @xmath255 for some @xmath250 and @xmath256 , and recall that @xmath257 is a homology ball of dimension @xmath258 . by definition of @xmath253 we have @xmath259 and @xmath260 . thus , it follows from part ( iii ) of lemma [ lem : homology ] that @xmath261 is a homology ball of dimension @xmath262 and that its interior is equal to @xmath263 . part ( i ) may be deduced from part ( ii ) as follows . let @xmath264 be a linear ordering of the facets of @xmath9 such that @xmath265 implies @xmath266 . thus we have @xmath267 , @xmath268 and @xmath269 . by assumption , @xmath270 is a @xmath11-dimensional homology ball . moreover , @xmath271 is equal to the simplicial join of a face of @xmath246 with the restriction of @xmath43 to a face of @xmath68 , for @xmath272 , and hence a @xmath11-dimensional homology ball by part ( iii ) of lemma [ lem : homology ] , and @xmath273 is equal to the simplicial join of the boundary of this face with the same restriction of @xmath43 . it follows from part ( iv ) of lemma [ lem : homology ] by induction on @xmath274 that @xmath275 is a @xmath11-dimensional homology ball , for @xmath276 and a @xmath11-dimensional homology sphere , for @xmath277 . this proves ( i ) , since @xmath278 . to verify ( iii ) , assume that @xmath43 is flag and vertex - induced . it is clear from the definition of @xmath253 that the subdivision @xmath1 is also vertex - induced . since the restriction of @xmath1 to any face of @xmath9 is the join of a simplex with the restriction of @xmath43 to a face of @xmath68 , the subdivision @xmath1 is flag as well . to verify that @xmath1 is a flag complex , let @xmath167 be a set of vertices of @xmath1 which are pairwise joined by edges , where @xmath279 for some @xmath280 and @xmath20 consists of vertices of @xmath43 . we need to show that @xmath281 . we set @xmath282 and @xmath241 and note that the elements of @xmath20 are vertices of @xmath257 , by definition of @xmath1 . since the elements of @xmath20 are pairwise joined by edges in @xmath43 , our assumptions that @xmath43 is vertex - induced and flag imply that @xmath283 . therefore @xmath167 belongs to @xmath239 , which is contained in @xmath1 , and the result follows . [ rem : vineeded ] the conclusion in proposition [ prop : balltosphere ] that @xmath1 is a flag complex does not hold under the weaker hypothesis that @xmath43 is quasi - geometric , rather than vertex - induced . for instance , let @xmath43 be the simplicial complex consisting of the subsets of @xmath284 and @xmath285 and let @xmath73 be the subdivision which pushes @xmath43 into @xmath68 , so that the face @xmath286 of @xmath43 ends up in the interior of @xmath68 and @xmath287 ends up in the interior of @xmath66 . then @xmath43 is quasi - geometric and flag but the simplicial complex @xmath1 is not flag , since it has @xmath288 as a minimal non - face . this section defines the local @xmath0-vector of a homology subdivision of the simplex , lists examples and elementary properties , discusses its nonnegativity in the special case of flag subdivisions and concludes with the proof of theorem [ thm:34 ] . this proof comes as an application of the considerations and results of the present and the previous section . [ def : localgamma ] let @xmath71 be a set with @xmath10 elements and let @xmath43 be a homology subdivision of the simplex @xmath68 . the polynomial @xmath289 defined by @xmath290 is the _ local @xmath0-polynomial _ of @xmath43 ( with respect to @xmath71 ) . the sequence @xmath291 is the _ local @xmath0-vector _ of @xmath43 ( with respect to @xmath71 ) . thus @xmath292 is the @xmath0-polynomial associated to @xmath293 and @xmath294 is the @xmath0-vector associated to @xmath144 , in the sense of section [ subsec : gamma ] . all formulas in the next example follow from corresponding formulas in ( * ? ? ? * example 2.3 ) , or directly from the relevant definitions . [ ex : localgamma ] ( a ) for the trivial subdivision @xmath295 of the @xmath11-dimensional simplex @xmath68 we have @xmath296 \(b ) let @xmath291 , where @xmath43 and @xmath71 are as in definition [ def : localgamma ] . assuming that @xmath297 , we have @xmath298 and @xmath299 , where @xmath300 is the number of interior vertices of @xmath43 . assuming that @xmath301 , we also have @xmath302 , where @xmath303 is the number of interior edges of @xmath43 and @xmath304 is the number of vertices of @xmath43 which lie in the relative interior of a @xmath45-dimensional face of @xmath68 . \(c ) suppose that @xmath305 . as a consequence of ( b ) we have @xmath306 for every homology subdivision @xmath43 of @xmath68 , where @xmath307 is the number of interior vertices of @xmath43 . \(d ) let @xmath43 be the cone over the boundary @xmath308 of the simplex @xmath68 ( so @xmath43 is the stellar subdivision of @xmath68 on the face @xmath71 ) . then @xmath309 and hence @xmath310 is negative for @xmath301 . for instance , we have @xmath311 for @xmath312 . \(e ) for the subdivisions of parts ( b ) and ( c ) of example [ ex : nonflagrest ] we can compute that @xmath313 and hence that @xmath314 . the following proposition shows the relevance of local @xmath0-vectors in the study of @xmath0-vectors of subdivisions of eulerian complexes . [ prop : gammaformula ] let @xmath1 be an eulerian simplicial complex . for every homology subdivision @xmath5 of @xmath1 we have @xmath315 since @xmath1 is eulerian , so is @xmath109 for every @xmath19 . thus , applying ( [ eq : defgamma ] ) to the @xmath2-polynomial of @xmath109 we get @xmath316 where @xmath317 . using this and ( [ eq : defxi ] ) , equation ( [ eq : hformula ] ) may be rewritten as @xmath318 the proposed equality now follows from the uniqueness of the @xmath0-polynomial associated to @xmath319 . the following statement is the main conjecture of this paper . [ conj : main ] for every flag vertex - induced homology subdivision @xmath43 of the simplex @xmath68 we have @xmath320 . parts ( d ) and ( e ) of example [ ex : localgamma ] show that the conclusion of conjecture [ conj : main ] fails under various weakenings of the hypotheses . we do not know of an example of a flag quasi - geometric homology subdivision of the simplex for which the local @xmath0-vector fails to be nonnegative . we now discuss some consequences of theorem [ thm : main ] and proposition [ prop : gammaformula ] , related to conjecture [ conj : main ] . [ cor : main12 ] for every flag homology sphere @xmath1 of dimension @xmath12 we have @xmath321 where @xmath80 is as in corollary [ cor : main1 ] for each @xmath210 . in particular , the validity of conjecture [ conj : main ] for homology subdivisions @xmath43 of dimension at most @xmath12 implies the validity of conjecture [ conj : gal ] for homology spheres @xmath1 of dimension at most @xmath12 . setting @xmath322 in ( [ eq : corpl ] ) and changing the order of summation , results in ( [ eq : corpl2 ] ) . alternatively , one can apply ( [ eq : gammaformula ] ) to the subdivision guaranteed by theorem [ thm : main ] and note that @xmath323 for every @xmath210 . the last sentence in the statement of the corollary follows from ( [ eq : corpl2 ] ) . [ cor : main2 ] the validity of conjecture [ conj : main ] for homology subdivisions @xmath43 of dimension at most @xmath12 implies the validity of conjecture [ conj : newmonotone ] for homology spheres @xmath1 and subdivisions @xmath5 of dimension at most @xmath12 . we observe that the term corresponding to @xmath46 in the sum of the right - hand side of ( [ eq : gammaformula ] ) is equal to @xmath324 . thus , the result follows from ( [ eq : gammaformula ] ) , corollary [ cor : main12 ] and the fact that the link of every nonempty face of a flag homology sphere is also a flag homology sphere of smaller dimension . [ prop : main3d ] conjecture [ conj : main ] holds for subdivisions of the 3-dimensional simplex . let @xmath43 be a flag vertex - induced homology subdivision of the @xmath11-dimensional simplex @xmath68 and let @xmath1 be the homology subdivision of @xmath9 considered in proposition [ prop : balltosphere ] . applying ( [ eq : gammaformula ] ) to this subdivision and noting that @xmath323 for every @xmath210 , we get @xmath325 by definition of @xmath1 , the restriction @xmath326 is a cone over the restriction of @xmath1 to a proper face of @xmath20 for every @xmath210 which is not contained in @xmath71 . since every such subdivision has a zero local @xmath2-vector @xcite , the previous formula can be rewritten as @xmath327 assume now that @xmath312 , so that @xmath328 for some integers @xmath329 . since @xmath298 and @xmath330 by part ( b ) of example [ ex : localgamma ] , it suffices to show that @xmath331 . for that , we observe that the only contribution to the coefficient of @xmath332 in the right - hand side of ( [ eq : corpl3 ] ) comes from the term with @xmath333 . as a result , @xmath310 is equal to the coefficient of @xmath332 in @xmath139 . since , @xmath1 is a 3-dimensional flag homology sphere ( by proposition [ prop : balltosphere ] ) , this coefficient is nonnegative by the davis - okun theorem ( * ? ? ? * theorem 11.2.1 ) and the result follows . _ proof of theorem [ thm:34]_. for 3-dimensional spheres the result follows from proposition [ prop : main3d ] and corollary [ cor : main2 ] . assume now that @xmath1 and @xmath5 have dimension 4 . then we can write @xmath334 and @xmath335 . since @xmath336 and @xmath337 , where @xmath338 and @xmath339 is the number of vertices of @xmath1 and @xmath5 , respectively , it is clear that @xmath340 . as the computation in the proof of ( * ? ? ? * corollary 2.2.2 ) shows , we also have @xmath341 where @xmath342 is the set of vertices of @xmath1 . similarly , we have @xmath343 where we may assume that @xmath344 . since @xmath345 is a flag vertex - induced homology subdivision of @xmath346 for every @xmath347 , by lemma [ lem : joinlink ] , we have @xmath348 by the 3-dimensional case , treated earlier , for every such vertex @xmath34 . since @xmath349 is a 3-dimensional flag homology sphere , we also have @xmath350 by the davis - okun theorem for every @xmath351 . hence @xmath352 and the result follows . [ que : monotone ] does @xmath7 hold for every flag homology sphere @xmath1 and every flag homology subdivision @xmath5 of @xmath1 ? this section provides some evidence in favor of the validity of conjecture [ conj : main ] other than that provided by proposition [ prop : main3d ] . * simplicial joins . * let @xmath43 be a homology subdivision of the simplex @xmath68 and @xmath81 be a homology subdivision of the simplex @xmath353 , where @xmath71 and @xmath354 are disjoint finite sets . then @xmath355 is a homology subdivision of the simplex @xmath356 and given subsets @xmath150 and @xmath357 , the restriction of @xmath355 to the face @xmath358 of this simplex satisfies @xmath359 . since @xmath360 , the defining equation ( [ eq : deflocalh ] ) and a straightforward computation show that @xmath361 this equation and ( [ eq : defxi ] ) imply that @xmath362 from the previous formula and lemma [ lem : joinlink ] we conclude that if @xmath43 and @xmath81 satisfy the assumptions and the conclusion of conjecture [ conj : main ] , then so does @xmath355 . * edge subdivisions . * following ( * ? ? ? * section 5.3 ) , we refer to the stellar subdivision on an edge of a simplicial complex @xmath43 as an _ edge subdivision_. as mentioned in section [ subsec : gamma ] , flagness of a simplicial complex is preserved by edge subdivisions . the following statement describes a class of flag ( geometric ) subdivisions of the simplex with nonnegative local @xmath0-vectors . [ prop : edgesub ] for every subdivision @xmath43 of the simplex @xmath68 which can be obtained from the trivial subdivision by successive edge subdivisions , we have @xmath363 . let @xmath43 be a subdivision of @xmath68 and @xmath81 be the edge subdivision of @xmath43 on @xmath364 . thus we have @xmath365 , where @xmath34 is the new vertex added and @xmath366 . by appealing to ( [ eq : localformula ] ) and noticing that the right - hand side of this formula vanishes except when @xmath367 ( or by direct computation ) , we find that @xmath368 thus it suffices to prove the following claim : if the @xmath0-polynomial corresponding to @xmath154 has nonnegative coefficients for every face @xmath75 of positive dimension ( meaning that @xmath154 can be written as a linear combination of the polynomials @xmath369 with nonnegative coefficients for every @xmath370 ) , then the same holds for @xmath81 . we consider a face @xmath84 of positive dimension and distinguish the following cases ( we note that @xmath64 can not contain @xmath74 and that if @xmath75 , then the carrier @xmath371 of @xmath64 is the same , whether considered with respect to @xmath43 or @xmath81 ) : case 1 : @xmath372 . the links @xmath373 and @xmath374 are then combinatorially isomorphic for every @xmath150 which contains the carrier of @xmath64 ( these two links are equal if @xmath375 ) and the defining equation ( [ eq : deflocalhrel ] ) implies that @xmath376 . case 2 : @xmath377 . for @xmath150 which contains the carrier of @xmath64 , the link @xmath373 is equal to either @xmath374 or to the edge subdivision of @xmath374 on @xmath74 , in case @xmath20 does not or does contain the carrier of @xmath74 , respectively . it follows from this and ( [ eq : hformula ] ) that ( see also ( * ? ? ? * proposition 2.4.3 ) ) @xmath378 and then from ( [ eq : deflocalhrel ] ) that @xmath379 . case 3 : @xmath380 . then we must have @xmath381 and , in particular , @xmath382 . we distinguish two subcases : suppose first that @xmath64 intersects @xmath74 and set @xmath383 . then @xmath384 for every @xmath150 which contains the carrier of @xmath64 in @xmath81 ( the latter coincides with the carrier of @xmath385 in @xmath43 ) and hence @xmath386 . suppose finally that @xmath387 and set @xmath383 . then @xmath388 for every @xmath150 which contains the carrier of @xmath64 in @xmath81 . therefore we have @xmath389 for every such @xmath20 and hence @xmath390 . the expressions obtained for @xmath391 and our assumption on @xmath43 show that indeed , the corresponding @xmath0-polynomial has nonnegative coefficients in all cases . * barycentric and cluster subdivisions . * as a special case of proposition [ prop : edgesub ] , the ( first ) barycentric subdivision of the simplex @xmath68 has nonnegative local @xmath0-vector . several combinatorial interpretations for its entries are given in @xcite . similar results appear there for the simplicial subdivision of a simplex defined by the positive part of the cluster complex , associated to a finite root system . the following special case of conjecture [ conj : main ] might also be interesting to explore . the notion of a cw - regular subdivision can be defined by replacing the simplicial complex @xmath5 in the definition of a topological subdivision ( definition [ def : sub ] ) by a regular cw - complex ; see @xcite . [ que : bary ] does conjecture [ conj : main ] hold for the barycentric subdivision of any cw - regular subdivision of the simplex ? the author wishes to thank an anonymous referee for helpful suggestions and mike davis and volkmar welker for useful discussions . xxx c.a . athanasiadis , _ some combinatorial properties of flag simplicial pseudomanifolds and spheres _ , ark . mat . * 49 * ( 2011 ) , 1729 . athanasiadis , _ cubical subdivisions and local @xmath2-vectors _ , arxiv:1007.3154 , ann . comb . ( to appear ) . athanasiadis and c. savvidou , _ the local @xmath2-vector of the cluster subdivision of a simplex _ , sm . * 66 * ( 2012 ) , article b66c , 21pp ( electronic ) . star clusters in independence complexes of graphs _ , arxiv:1007.0418 . a. bjrner , _ topological methods _ , in _ handbook of combinatorics _ ( r.l . graham , m. grtschel and l. lovsz , eds . ) , north holland , amsterdam , 1995 , pp . 18191872 . a. bjrner , m. las vergnas , b. sturmfels , n. white and g.m . ziegler , oriented matroids , encyclopedia of mathematics and its applications * 46 * , cambridge university press , cambridge , 1993 ; second printing , 1999 . c. chan , _ on subdivisions of simplicial complexes : characterizing local @xmath2-vectors _ , discrete comput . * 11 * ( 1994 ) , 465476 . r. charney and m. davis , _ the euler characteristic of a nonpositively curved , piecewise euclidean manifold _ , pacific j. math . * 171 * ( 1995 ) , 117137 . m. davis and b. okun , _ vanishing theorems and conjectures for the @xmath392-homology of right - angled coxeter groups _ , geom . * 5 * ( 2001 ) , 774 . gal , _ real root conjecture fails for five- and higher - dimensional spheres _ , discrete comput . * 34 * ( 2005 ) , 269284 . munkres , elements of algebraic topology , addison - wesley , reading , ma , 1984 . a. postnikov , v. reiner and l. williams , _ faces of generalized permutohedra _ , doc . math . * 13 * ( 2008 ) , 207273 . stanley , _ subdivisions and local @xmath2-vectors _ , j. amer . soc . * 5 * ( 1992 ) , 805851 . stanley , enumerative combinatorics , vol . 1 , wadsworth & brooks / cole , pacific grove , ca , 1986 ; second printing , cambridge studies in advanced mathematics * 49 * , cambridge university press , cambridge , 1997 . stanley , combinatorics and commutative algebra , second edition , birkhuser , basel , 1996 .

given a finite class of functions , the problem of aggregation is to construct a procedure with a risk as close as possible to the risk of the best element in the class . a classical procedure ( pac - bayesian statistical learning theory ( 2004 ) paris 6 , _ statistical learning theory and stochastic optimization _ ( 2001 ) springer , _ ann . statist . _ * 28 * ( 2000 ) 7587 ) is the aggregate with exponential weights ( aew ) , defined by where is called the temperature parameter and is an empirical risk . in this article , we study the optimality of the aew in the regression model with random design and in the low - temperature regime . we prove three properties of aew . first , we show that aew is a suboptimal aggregation procedure in expectation with respect to the quadratic risk when , where is an absolute positive constant ( the low - temperature regime ) , and that it is suboptimal in probability even for high temperatures . second , we show that as the cardinality of the dictionary grows , the behavior of aew might deteriorate , namely , that in the low - temperature regime it might concentrate with high probability around elements in the dictionary with risk greater than the risk of the best function in the dictionary by at least an order of . third , we prove that if a geometric condition on the dictionary ( the so - called `` bernstein condition '' ) is assumed , then aew is indeed optimal both in high probability and in expectation in the low - temperature regime . moreover , under that assumption , the complexity term is essentially the logarithm of the cardinality of the set of `` almost minimizers '' rather than the logarithm of the cardinality of the entire dictionary . this result holds for small values of the temperature parameter , thus complementing an analogous result for high temperatures .

1303.5180

in this note we study the problem concerning the optimality of the aew in the regression model with random design . to formulate the problem , we need to introduce several definitions . let @xmath7 and @xmath8 be two measure spaces , and set @xmath9 and @xmath10 to be @xmath11 i.i.d . random variables with values in @xmath7 . from a statistical standpoint , @xmath12 is the set of given data at our disposal . the _ risk _ of a measurable real - valued function @xmath13 defined on @xmath8 is given by @xmath14 where @xmath15 is a non - negative function , called the _ loss function _ and @xmath16 is the set of all real - valued measurable functions defined on @xmath8 . if @xmath17 is a statistic constructed using the data @xmath18 , then the risk of @xmath17 is the random variable @xmath19.\ ] ] throughout this article , we restrict our attention to functions @xmath13 , loss functions @xmath20 , and random variables @xmath9 for which @xmath21 almost surely . ( note that some results have been obtained in the same setup for unbounded loss functions in @xcite , and @xcite . ) the loss function on which we focus throughout most of the article is the quadratic loss function , defined when @xmath22 by @xmath23 . in the aggregation framework , one is given a finite set @xmath0 of real - valued functions defined on @xmath8 , usually called a _ dictionary_. the problem of _ aggregation _ ( see , e.g. , @xcite , and @xcite ) is to construct a procedure , usually called an _ aggregation procedure _ , that produces a function with a risk as close as possible to the risk of the best element in @xmath0 . keeping this in mind , one can define the _ optimal rate of aggregation _ @xcite , which is the smallest price , as a function of the cardinality of the dictionary @xmath24 and the sample size @xmath25 , that one has to pay to construct a function with a risk as close as possible to that of the best element in the dictionary . we recall the definition for the `` expectation case ; '' a similar definition for the `` probability case '' can be formulated as well ( see , e.g. , @xcite ) . [ def : definition - optimality ] let @xmath26 . we say that @xmath27 is an optimal rate of aggregation in expectation when there exist two positive constants , @xmath28 and @xmath5 , depending only on @xmath29 , for which the following holds for any @xmath30 and @xmath31 : 1 . there exists an aggregation procedure @xmath32 such that for any dictionary @xmath0 of cardinality @xmath24 and any random variable @xmath9 satisfying @xmath21 almost surely for all @xmath33 , one has @xmath34 2 . for any aggregation procedure @xmath35 , there exists a dictionary @xmath0 of cardinality @xmath24 and a random variable @xmath9 such that @xmath21 almost surely for all @xmath33 and @xmath36 in our setup , one can show ( cf . @xcite ) that in general , an optimal rate of aggregation ( in the sense of @xcite [ optimality in expectation ] and of @xcite [ optimality in probability ] ) is lower - bounded by @xmath37 . thus , procedures satisfying an exact oracle inequality like ( [ eq : exact - oracle - inequality])that is , an oracle inequality with a factor of 1 in front of @xmath38with a residual term of @xmath39 are said to be optimal . only a few aggregation procedures have been shown to achieve this optimal rate , including the exponential aggregating schemes of @xcite , the the `` empirical star algorithm '' in @xcite , and the `` preselection / convexification algorithm '' in @xcite . for a survey on optimal aggregation procedures , see the hdr dissertation of j .- y . audibert . our main focus here is on the problem of the optimality of the aggregation procedure with exponential weights ( aew ) . this procedure originate from the thermodynamic standpoint of learning theory ( see @xcite for the state of the art in this direction ) . aew can be viewed as a relaxed version of the trivial aggregation scheme , which is to minimize the empirical risk @xmath40 in the dictionary @xmath0 . a procedure that minimizes ( [ eq : r - n ] ) is called _ empirical risk minimization _ ( erm ) . it is well known that erm generally can not achieve the optimal rate of @xmath41 , unless one assumes that the given class @xmath0 has certain geometric properties , which we discuss below ( see also @xcite ) . to have any chance of obtaining better rates , one has to consider aggregation procedures that take values in larger sets than @xmath0 . the most natural set is the convex hull of @xmath0 . aew is a very popular candidate for the optimal procedure , and it was one of the first procedures to be studied in the context of the aggregation framework @xcite . it is defined by the following convex sum : @xmath42 for the dictionary @xmath43 . the parameter @xmath2 is called the _ temperature_. can be seen as a gibbs measure with temperature @xmath44 on the dictionary @xmath0 . ] thus far , there have been three main results concerning the optimality of the aew . the first of these is that the progressive mixture rule is optimal in expectation for @xmath44 larger than some parameters of the model ( see @xcite and @xcite ) , and under certain convexity assumption on the loss function @xmath20 . this procedure is defined by @xmath45 where @xmath46 is the function generated by aew ( with a common temperature parameter @xmath44 ) associated with the dictionary @xmath0 and constructed using only the first @xmath47 observations @xmath48 . ( see @xcite for more details and for other procedures related to the progressive mixture rule . ) second , the optimality in expectation of aew was obtained by @xcite for the regression model @xmath49 with a deterministic design @xmath50 with respect to the risk @xmath51 ( with its empirical version being @xmath52 ) . that is , it was shown that for @xmath53 , where @xmath54 is the variance of the noise @xmath55 , @xmath56 finally , @xcite , and @xcite proved that in the high - temperature regime , aew can achieve the optimal rate @xmath37 under the bernstein assumption , recalled below in definition [ def : bernstein ] in expectation and in high probability . this result is discussedin more detail later . despite the long history of aew , the literature contains no results on the optimality ( or suboptimality ) of aew in the regression model with random design in the general case ( when the dictionary does not necessarily satisfy the bernstein condition ) . in this article , we address this issue and complement the results ( assuming the bernstein condition ) of @xcite for the low - temperature regime by proving the following : 1 . aew is suboptimal for low temperatures @xmath4 ( where @xmath5 is an absolute positive constant ) , both in expectation and in probability , for the quadratic loss function and a dictionary of cardinality @xmath57 ( theorem [ tha ] ) . aew is suboptimal in probability for some large dictionaries ( of cardinality @xmath58 ) and small temperatures @xmath59 ( theorem [ thb ] ) . 3 . aew achieves the optimal rate @xmath37 for low temperatures under the bernstein condition on the dictionary ( theorem [ thc ] ) . together with the high - temperature results of @xcite and @xcite , this proves that the temperature parameter has almost no impact ( as long as @xmath60 ) on the performance of the aew under this condition , with a residual term of the order of @xmath61 for every @xmath2 . [ tha]there exist absolute constants @xmath62 for which the following holds . for any integer @xmath63 , there are random variables @xmath64 and a dictionary @xmath65 such that @xmath66 almost surely for @xmath67 , for which the quadratic risk of the aew satisfies the following : 1 . if @xmath4 and @xmath25 is odd , then @xmath68 2 . if @xmath69 , then , with probability greater than @xmath70 , @xmath71 theorem [ tha ] proves that aew is suboptimal in expectation in the low - temperature regime and suboptimal in probability in both the low- and high - temperature regimes , since it is possible to construct procedures that achieve the rate @xmath72 with high probability @xcite and in expectation @xcite in the same setup as for theorem [ tha ] . it should be noted that the problem of the optimality in probability of the progressive mixture rule ( and other related procedures ) was studied by @xcite , who proved that , for a loss function @xmath20 satisfying some convexity and regularity assumption ( e.g. , the quadratic loss used in theorem [ tha ] ) , the progressive mixture rule @xmath73 defined in ( [ eq : progressive - mixture ] ) satisfies that for any temperature parameter , with probability greater than an absolute constant @xmath74 , @xmath75 in addition , it is important to observe that the suboptimality in probability does not imply suboptimality in expectation for the aggregation problem , or vice versa . this property of the aggregation problem was first noted by @xcite , who found the progressive mixture rule ( and other related aggregation procedures ) to be suboptimal in probability for dictionaries of cardinality two but , on the other hand , to be optimal in expectation ( @xcite and @xcite ) . this peculiar property of the problem of aggregation comes from the fact that an aggregate @xmath17 is not restricted to the set @xmath0 , which allows @xmath76 to take negative values . @xcite showed that for the progressive mixture rule @xmath73 , these negative values do compensate on average for larger values , but there is still an event of constant probability on which @xmath77 takes values greater than @xmath78 . the proof of theorem [ tha ] shows that a dictionary consisting of two functions is sufficient to yield a lower bound in expectation in the low - temperature regime and in probability in both the small temperature regime , @xmath79 , and the large temperature regime , @xmath80 . in the following theorem , we study the behavior of aew for larger dictionaries . to the best of our knowledge , negative results on the behavior of exponential weights based aggregation procedures are not known for dictionaries with more than two functions , and we show that the behavior of the aew deteriorates in some sense as the cardinality of the dictionary increases . [ thb ] there exist an integer @xmath81 and absolute constants @xmath5 and @xmath82 for which the following holds . for every @xmath83 , there are random variables @xmath64 and a dictionary @xmath43 of cardinality , @xmath84 , for which the quadratic loss function of any element in @xmath0 is bounded by @xmath57 almost surely , and for every @xmath85 , if @xmath86 , then with probability at least @xmath87 , @xmath88 moreover , if @xmath89 denotes the optimal function in @xmath0 with respect to the quadratic loss ( the oracle ) , then there exists @xmath90 with an excess risk greater than @xmath91 and for which the weight of @xmath92 in the aew procedure satisfies @xmath93 theorem [ thb ] implies that the aew procedure might cause the weights to concentrate around a `` bad '' element in the dictionary ( i.e. , an element whose risk is larger than the best in the class by at least @xmath94 ) with high probability . in particular , theorem [ thb ] provides additional evidence that the aew procedure is suboptimal for low temperatures . the analysis of the behavior of aew for a dictionary of cardinality larger than two is considerably harder than in the two - function case and requires some results on rearrangement of independent random variables that are almost gaussian ( see proposition [ prop : shahar - proposition ] below ) . fortunately , not all is lost as far as optimality results for aew go . indeed , we show that under some geometric condition , aew can be optimal and in fact can even adapt to the `` real complexity '' of the dictionary . intuitively , a good aggregation scheme should be able to ignore the elements in the dictionary whose risk is far from the optimal risk in @xmath0 , or at least the impact of such elements on the function produced by the aggregation procedure should be small . thus , a good procedure is one with a residual term of the order of @xmath95 , where @xmath96 is a complexity measure that is determined only by the richness of the set of `` almost minimizers '' in the dictionary . this leads to the following question : [ qu : adaptivitytocomplexityofdictionary ] is it possible to construct an aggregation procedure that adapts to the real complexity of the dictionary ? this question was first addressed by the pac - bayesian approach . @xcite and @xcite showed that in the high - temperature regime , aew satisfies the requirements of question [ qu : adaptivitytocomplexityofdictionary ] , assuming that the class has a geometric property , called the bernstein condition . [ def : bernstein ] we say that a function class @xmath0 is a @xmath97-bernstein class ( @xmath98 and @xmath99 ) with respect to @xmath9 if every @xmath100 satisfies @xmath101 and @xmath102 there are many natural situations in which the bernstein condition is satisfied . for instance , when @xmath20 is the quadratic loss function and the regression function is assumed to belong to @xmath0 , the excess loss function class @xmath103 satisfies the bernstein condition with @xmath104 , where @xmath105 is the minimizer of the risk in the class @xmath0 . another generic example is when the target function @xmath106 is far from the set of targets with `` multiple minimizers '' in @xmath0 and @xmath107 satisfies the bernstein condition with @xmath104 . ( see @xcite for an exact formulation of this statement and related results . ) the bernstein condition is very natural in the context of erm because it has two consequences : that the empirical excess risk has better concentration properties around the excess risk , and that the complexity of the subset of @xmath0 consisting of almost minimizers is smaller under this assumption . consequently , if the class @xmath107 is a @xmath97-bernstein class for @xmath108 , then the erm algorithm can achieve fast rates ( see , e.g. , @xcite and references therein ) . as the results below show , the same is true for aew . indeed , under a bernstein assumption , @xcite and @xcite proved that if @xmath109 is a convex risk function and if @xmath0 is such that @xmath110 almost surely for any @xmath33 , then for every @xmath111 and @xmath112 , with probability greater than @xmath113 , @xmath114 although the pac - bayesian approach can not be used to obtain ( [ eq : pac - bound ] ) in the low - temperature regime ( @xmath115 ) , such a result is not surprising . indeed , because fast error rates for the erm are expected when the underlying excess loss functions class satisfies the bernstein condition , and because aew converges to the erm when the temperature @xmath44 tends to 0 , it is likely that for `` small values '' of @xmath44 , aew inherits some of the properties of erm , such as fast rates under a bernstein condition . we show this in theorem [ thc ] , proving that aew answers question [ qu : adaptivitytocomplexityofdictionary ] for low temperatures under the bernstein condition . before formulating theorem [ thc ] , we introduce the following measure of complexity . for every @xmath116 , let @xmath117 where @xmath118 denotes the cardinality of the set @xmath119 . observe that @xmath120 is a weighted sum of the number of elements in @xmath0 that assigns smaller and smaller weights to functions with a relatively large excess risk . [ thc ] there exist absolute constants @xmath28 , @xmath121 , and @xmath122 for which the following holds . let @xmath0 be a class of functions bounded by @xmath29 such that the excess loss class @xmath107 is a @xmath123-bernstein class with respect to @xmath9 . if the risk function @xmath109 is convex and if @xmath124 , then for every @xmath125 , with probability at least @xmath113 , the function @xmath126 produced by the aew algorithm satisfies @xmath127 where @xmath128 . in particular , @xmath129 in other words , the scaling factor @xmath130 that we use is proportional to @xmath131 , and if the class is regular ( in the sense that the complexity of @xmath0 is well spread and not concentrated just around one point ) , then @xmath132 is roughly the cardinality of the elements in @xmath0 with risk at most latexmath:[$\sim\!(b+b)(\log observe that for every @xmath116 , @xmath134 for a suitable absolute constant @xmath135 . thus , if @xmath44 is reasonably small ( below a level proportional to @xmath136 ) , then the resulting aggregation rate is the optimal one , proportional to @xmath137 with probability @xmath113 , and proportional to @xmath138 in expectation . thus , theorem [ thc ] indeed gives a positive answer to question [ qu : adaptivitytocomplexityofdictionary ] in the presence of a bernstein condition and for low temperatures . although the residual terms in theorem [ thc ] and in ( [ eq : pac - bound ] ) are not the same , they are comparable . indeed , the contribution of each element in @xmath0 in the residual term depends exponentially on its excess risk . theorem [ thc ] together with the results for high temperatures from @xcite and @xcite show that the aew is an optimal aggregation procedure under the bernstein condition as long as @xmath60 when @xmath24 and @xmath25 tend to infinity . in general , the residual term obtained is on the order of @xmath139 , and it can be proven that the optimal rate of aggregation under the bernstein condition is proportional to @xmath37 using the classical tools in @xcite . finally , a word about the organization of the article . in the next section we present some comments about our results . the proofs of the three theorems follow in the subsequent sections . throughout , we denote absolute constants or constants that depend on other parameters by @xmath5 , @xmath82 , etc . ( of course , we specify when a constant is absolute and when it depends on other parameters . ) the values of constants may change from line to line . we write @xmath140 if there are absolute constants @xmath135 and @xmath141 such that @xmath142 , and write @xmath143 if @xmath144 . although from a theoretical standpoint , whether aew is an optimal procedure in expectation and for high temperatures in the regression model with random design remains to be seen , from a practical standpoint , we believe that exponential aggregating schemes simply should not be used in the setup of this article , because of the following reasons ( see also the comments in @xcite ) : 1 . for any temperature @xmath145 , there is an event of constant probability on which aew performs poorly ( this is the second part of theorem [ tha ] ) . 2 . if the temperature parameter is chosen to be too small , then the aew can perform poorly even in expectation ( the first part of theorem [ tha ] ) . another consequence of the lower bounds stated in theorem [ tha ] is that aew can not be an optimal aggregation procedure both in expectation and in probability at low temperatures for two other aggregation problems : the problem of _ convex aggregation _ , in which one wants to mimic the best element in the convex hull of @xmath0 , and the problem of _ linear aggregation _ , where one wishes to mimic the best linear combination of elements in @xmath0 . indeed , clearly @xmath146 moreover , the optimal rates of aggregation for the convex and linear aggregation problems for dictionaries of cardinality two are of the order of @xmath147 ( see @xcite ) , whereas the residual terms obtained in theorem [ tha ] are on the order of @xmath148 for such a dictionary . thus aew is suboptimal for these two other aggregation problems in the low - temperature regime . we end this section by comparing two seemingly related assumptions , the margin assumption of @xcite and the bernstein condition of @xcite . note that in the proof of theorem [ thc ] , we have restricted ourselves to the case @xmath104 simply to make the presentation as simple as possible . a very similar result , with the residual term @xmath149 for the exact oracle inequality in probability and @xmath150 for the exact oracle inequality in expectation , holds if one assumes a bernstein condition for any @xmath151 , and the proof is identical to that in the case where @xmath104 . this makes the discussion about @xmath152-bernstein classes relevant here . recall the definition of the margin assumption : we say that @xmath0 has margin with parameters @xmath97 ( @xmath98 and @xmath99 ) if for every @xmath153 , @xmath154 where @xmath155 is defined such that @xmath156 , and the minimum is taken with respect to all measurable functions @xmath13 on the given probability space . although the margin condition appears similar to the bernstein condition , they are in fact very different , and have been introduced in the context of different types of problems . in the first of these , the `` classical '' statistical setup , one is given a function class @xmath0 ( the _ model _ ) with an upper bound on its complexity and an unknown target function @xmath155 , the minimizer of the risk over _ all _ measurable functions . one usually assumes that @xmath155 belongs to @xmath0 , and the aim is to construct an estimator @xmath157 for which the risk @xmath158 tends to 0 quickly as the sample size tends to infinity . in this setup , the margin assumption can improve this rate of convergence because of a better concentration of empirical means of @xmath159 around its mean @xcite . the margin assumption ( ma ) for @xmath104 compares the performance of each @xmath153 with the _ best possible measurable function _ , but it has nothing to do with the geometric structure of @xmath0 . the margin is determined for every @xmath13 separately , because @xmath155 does not depend on the choice of @xmath0 . in the second type of problem , the `` learning theory '' setup , one does not assume that the target function @xmath155 belongs to @xmath0 . the aim is to construct a function @xmath17 with a risk as close as possible to that of the best element @xmath160 . assuming that the excess loss class @xmath107 satisfies the bernstein condition ( bc ) , the error rate can be improved ( see , e.g. , @xcite ) . at a first glance , ma and bc ( for @xmath104 ) share very strong similarities . indeed , saying that @xmath107 is a @xmath123-bernstein class means that for every @xmath33 , @xmath161 but nevertheless they are different . indeed , as mentioned earlier , ma is only a matter of concentration ( and classical statistics questions are mostly a question of the trade - off between concentration and complexity ) . on the other hand , bc involves a lot of geometry of the function class @xmath0 , because @xmath162 might change significantly by adding a single function to @xmath0 or by removing a function . in fact , the difficulty of learning theory problems is determined by the trade - off between concentration and complexity , _ and _ the geometry of the given class , since one measures the performance of the learning algorithm relative to the best _ in the class_. assuming that @xmath163 , as is usually done in classical statistics , exempts one from the need to consider the geometry of @xmath0 , but one does not have that freedom in the aggregation framework . indeed , since in the aew algorithm the estimator is determined by the empirical means @xmath164 , this is a learning problem rather than a problem in classical statistics , despite the fact that it has been used in statistical frameworks to construct adaptive estimators ( see , e.g. , @xcite ) . therefore , given their nature , aggregation procedures like the aew are more natural under a bc assumption than under the ma . ( a by - product of theorem [ tha ] is that the ma can not improve the performance of aew since in the setup of theorem [ tha ] , it is easy to check that ma is satisfied with the best possible margin parameter @xmath104 . ) our starting point is the berry essen theorem on gaussian approximation . let @xmath165 be a sequence of i.i.d . , mean-0 random variables with variance @xmath166 , set @xmath167 to be a standard gaussian variable , and write @xmath168 [ theo : berry - esseen]there exists an absolute constant @xmath169 such that for every integer @xmath25 , @xmath170-{\mathbb{p}}[g\leq x]|\leq\frac{a{\mathbb{e}}|w_1|^3}{\sqrt{n}}.\ ] ] from here on , we let @xmath119 denote the constant appearing in theorem [ theo : berry - esseen ] . when the tail behavior of the @xmath171 has a subexponential decay , the gaussian approximation can be improved . indeed , recall that a real - valued random variable @xmath172 belongs to @xmath173 for some @xmath174 if there exists @xmath175 such that @xmath176 the infimum over all constants @xmath135 for which ( [ eq : psi - alpha ] ) holds defines an orlicz norm , which is called the @xmath177 norm and is denoted by @xmath178 . ( for more information on orlicz norms , see , e.g. , @xcite and @xcite . ) [ prop : petrov ] for every @xmath179 , there exist constants @xmath180 , and @xmath82 that depend only on @xmath181 for which the following holds . if @xmath182 , then for any @xmath183 , such that @xmath184 , @xmath185={\mathbb{p}}[g\geq x]\exp\biggl(\frac{x^3 { \mathbb{e}}w^3 } { 6\sqrt { n}}\biggr)\biggl[1+\mathrm{o}\biggl(\frac{x+1}{\sqrt{n}}\biggr)\biggr]\ ] ] and @xmath186={\mathbb{p}}[g\leq - x]\exp\biggl(-\frac{x^3{\mathbb{e}}w^3 } { 6\sqrt { n}}\biggr)\biggl[1+\mathrm{o}\biggl(\frac{x+1}{\sqrt{n}}\biggr)\biggr],\ ] ] where by @xmath187 we mean that @xmath188 . in particular , if @xmath189 and @xmath190 , then @xmath191-{\mathbb{p}}[g\leq x]| \leq c_2\bigl(n^{-1/2}\exp(-x^2/2)\bigr).\ ] ] from here on , we let @xmath192 denote the constant appearing in proposition [ prop : petrov ] . before presenting the proof of theorem [ tha ] , we introduce the following notation . given a probability measure @xmath193 and @xmath194 selected independently according to @xmath193 , we set @xmath195 the empirical measure supported on @xmath194 . we let @xmath196 denote the expectation @xmath197 . we assume that @xmath198 and recall that @xmath25 is an odd integer . let @xmath199 and define @xmath200 by @xmath201=1/2-n^{-1/2}$ ] and @xmath202=1/2+n^{-1/2}$ ] . let @xmath203}$ ] and @xmath204}$ ] , and consider the dictionary @xmath65 . it is easy to verify that the best function in @xmath0 ( the oracle ) with respect to the quadratic risk is @xmath205 , and that the excess loss function of @xmath206 , @xmath207 , satisfies that @xmath208 to simplify notation , set @xmath209 and @xmath210 . an important parameter that lies at the heart of this counterexample is the bernstein constant ( which is very bad in this case ) , @xmath211 straightforward computation shows that aew on @xmath0 with temperature @xmath44 is given by @xmath212 and that for @xmath213 defined for all @xmath214 $ ] , @xmath215&=&{\mathbb{e}}[1-{\widehat}\theta _ 1-\alpha{\widehat}\theta_1(1-{\widehat}\theta_1)]p{\mathcal { l}}_2={\mathbb{e}}[1-h({\widehat}\theta _ 1)]p{\mathcal { l}}_2 \nonumber\\ & = & \biggl[1-\int_0^\infty h^\prime(t){\mathbb{p}}[{\widehat}\theta_1\geq t]\ , \mathrm{d}t\biggr]p{\mathcal { l}}_2 \nonumber \\[-8pt ] \\[-8pt ] \nonumber & = & \biggl[1+\int_0 ^ 1\bigl(2\alpha t-(1+\alpha)\bigr){\mathbb{p}}[{\widehat}\theta _ 1\geq t]\,\mathrm{d}t\biggr]p{\mathcal { l}}_2 \\ & = & \biggl[1+\int_0 ^ 1\bigl(2\alpha t-(1+\alpha)\bigr){\mathbb{p } } [ p_n{\mathcal { l}}_2\geq\gamma(t ) ] \ , \mathrm{d}t\biggr]p{\mathcal { l}}_2,\nonumber\end{aligned}\ ] ] where @xmath216 is an increasing function defined for any @xmath217 by @xmath218 in particular , @xmath219=[i_1+i_2]p{\mathcal { l}}_2\ ] ] for @xmath220\,\mathrm{d}t+1\ ] ] and @xmath221\,\mathrm{d}t.\ ] ] first , we bound @xmath222 from below . to that end , we note the following facts . first , for every @xmath223 , @xmath224 and @xmath225 second , if we set @xmath226 , then for @xmath227 , @xmath228 . in particular , this holds under our assumption that @xmath229 . moreover , because @xmath230 is increasing , for @xmath231 , @xmath232 . therefore , @xmath233\,\mathrm{d}t+1 \\ & = & \int_0^{\alpha^{-1}}\bigl(2\alpha t-(1+\alpha)\bigr)\bigl({\mathbb{p}}[p_n{\mathcal { l}}_2\geq \gamma ( t)]-1\bigr)\,\mathrm{d}t \\ & \geq&\int_{(1+e)^{-1}}^{\alpha^{-1}}(1+\alpha-2\alpha t){\mathbb{p}}[p_n{\mathcal { l}}_2<\gamma(t)]\,\mathrm{d}t \\ & \geq&\int_{(1+e)^{-1}}^{\alpha^{-1}}(1+\alpha-2\alpha t)\,\mathrm{d}t \cdot{\mathbb{p}}\bigl[\bigl(\sqrt{n}/\sigma\bigr)(p_n{\mathcal { l}}_2-p{\mathcal { l}}_2)<\bigl(\sqrt{n}/\sigma \bigr)(-2p{\mathcal { l}}_2)\bigr ] \\ & \geq&\int_{(1+e)^{-1}}^{\alpha^{-1}}(1+\alpha-2\alpha t)\,\mathrm { d}t\bigl({\mathbb{p}}[g\leq-8]-a/\sqrt{n}\bigr)\geq c_0>0,\end{aligned}\ ] ] where in the last step we used the berry essen theorem , with @xmath234 and @xmath235)^2 $ ] , implying that @xmath236 . we turn to a lower bound for @xmath237 . applying a change of variables @xmath238 in the second term of @xmath237 , it is evident that @xmath239\,\mathrm{d}t \\ & & { } + \int_{({\alpha+1})/({2\alpha})}^1\bigl(2\alpha t-(1+\alpha)\bigr){\mathbb{p}}[p_n{\mathcal { l}}_2\geq \gamma(t)]\,\mathrm{d}t \\ & = & \int_{\alpha^{-1}}^{({\alpha+1})/({2\alpha})}\bigl(2\alpha t-(1+\alpha ) \bigr){\mathbb{p}}[\gamma(t)\leq p_n{\mathcal { l}}_2<\gamma(1+\alpha^{-1}-t)]\,\mathrm{d}t = i_3+i_4\end{aligned}\ ] ] for @xmath240\,\mathrm{d}t\ ] ] and @xmath241\ , \mathrm{d}t.\ ] ] to estimate @xmath242 , note that @xmath243 for @xmath244 $ ] , and thus @xmath245 for our choice of @xmath246 . the final step of the proof is to bound @xmath247 and in particular to show that for small values of @xmath44 , @xmath248 . for any @xmath249 , consider the intervals @xmath250 , and set @xmath251 , which is the number of integers in @xmath252 . because @xmath253 , @xmath254 = { \mathbb{p}}\biggl[\sum_{i=1}^n - x_i\in i_t(t)\biggr]={\mathbb{p}}_t(t).\ ] ] recall that @xmath255 , and thus @xmath256={\mathbb{p}}[\sum_{i}-x_i\in i_t(t ) \cap{\mathbb{z}}]$ ] . because @xmath257 is increasing and non - negative for @xmath258 , then if @xmath259 , it follows that @xmath260 , provided that @xmath198 . thus , for such values of @xmath261 , @xmath262 , implying that @xmath263 . on the other hand , if @xmath264 , then @xmath265 . in particular , if @xmath266 , then @xmath267 , and since @xmath25 is odd , then @xmath268=0 $ ] . otherwise , @xmath269 , which implies that @xmath270 , where @xmath271 is the length of @xmath252 , given by @xmath272 therefore , for every @xmath261 in our range , @xmath273\leq 2\delta_t(t)\max_{k\in{\mathbb{z}}}{\mathbb{p}}\biggl[\sum_{i=1}^nx_i = k\biggr].\ ] ] since @xmath274 for every @xmath275 , it is evident that @xmath276 \cdot\int _ { ( 1+c_0/4)\alpha^{-1}}^{({\alpha+1})/({2\alpha } ) } \bigl(2\alpha t-(1+\alpha)\bigr)\log\biggl(\frac{(1-t)(\alpha+1-\alpha t)}{t(\alpha t-1)}\biggr ) \,\mathrm{d}t.\ ] ] it can be shown that @xmath277 $ ] is on the order of @xmath148 either by a direct computation or by the berry essen theorem . moreover , for any @xmath278 , one has @xmath279 , and thus , @xmath280 therefore , combining the two observations with a change of variables @xmath281 for @xmath282 , it is evident that there are absolute constants @xmath121 for which @xmath283 thus , there is an absolute constant @xmath122 such that if @xmath284 , then @xmath285 , implying that @xmath219\geq\frac{c_0}{3\sqrt{n}},\ ] ] and proving the first part of theorem [ tha ] . to prove the second part of the theorem , note that by the berry ess ' een theorem , for every @xmath286 , with probability greater than @xmath287 - 2a/\sqrt{n}$ ] , @xmath288 thus , if @xmath25 is large enough to ensure that @xmath289 - 2a/\sqrt{n}\geq{\mathbb{p}}[g\leq-4]/2=c_4 $ ] , and taking @xmath290 , then with probability at least @xmath70 , @xmath291 . in that case , @xmath292 , which yields that @xmath293 provided that @xmath294 . the first step in the proof of theorem [ thb ] involves a general statement regarding a monotone rearrangement of independent random variables that are close to being gaussian . let @xmath172 be a mean 0 , variance 1 random variable that is absolutely continuous with respect to the lebesgue measure . further assume that @xmath295 has a finite third moment ( in fact , the random variables in which we are interested are bounded ) and set @xmath296 , where @xmath119 is the constant appearing in the berry essen theorem ( theorem [ theo : berry - esseen ] ) . let @xmath297 be independent random variables distributed as @xmath172 and set @xmath298 . let @xmath299 be @xmath300 independent copies of @xmath301 , and put @xmath302 to satisfy that @xmath303=1-\frac{1}{n}.\ ] ] note that such a @xmath304 exists because @xmath172 has a density with respect to the lebesgue measure . throughout the proof of theorem [ thb ] , we require the following simple estimates on @xmath304 . [ lemma : gamma-1 ] there exist absolute constants @xmath305 for which the following hold : 1 . if @xmath306 , then @xmath307 \leq1-c_1\frac { \log n}{\ell}.\ ] ] 2 . if @xmath300 and @xmath25 are such that @xmath308 $ ] , then @xmath309 . 3 . if @xmath309 and @xmath310 , then @xmath311 before we present the proof of lemma [ lemma : gamma-1 ] , recall that for every @xmath312 , @xmath313\leq\frac{1}{\sqrt{2\uppi}}\frac{\exp(-x^2/2)}{x}.\ ] ] proof of lemma [ lemma : gamma-1 ] to prove the first part , note that by independence and because @xmath314 , @xmath315={\mathbb{p}}\bigl[\min_{1 \leq j \leq\ell}\bar x_j>\gamma_1\bigr]^{{1}/{\ell}}=\biggl(\frac{1}{n}\biggr)^{1/\ell } \geq1-\frac{\log n}{\ell}.\ ] ] the reverse inequality follows in an identical fashion , because @xmath316 if @xmath317 . turning to the second part , if @xmath318 , then @xmath319 \geq{\mathbb{p}}\bigl[\min_{1 \leq j \leq\ell } \bar x_j \leq-2\bigr ] = 1-({\mathbb{p}}[\bar x > -2])^\ell,\ ] ] implying that @xmath320 \leq(\log n)/\ell$ ] . on the other hand , by the berry essen theorem , @xmath320 \geq { \mathbb{p}}[g \leq-2 ] - \beta(w)/\sqrt{n}$ ] , which is impossible under the assumptions of ( 2 ) . finally , to prove ( 3 ) , we use the berry essen theorem combined with the lower and upper estimates on the gaussian tail ( [ eq : gaussian - tail - estimate ] ) and ( [ eq : def - gamma1 - 2 ] ) . thus , @xmath321\leq{\mathbb{p}}[\bar x<\gamma_1]+\frac{\beta(w)}{\sqrt{n}}\leq \frac{\beta(w)}{\sqrt{n}}+c_1\frac{\log n}{\ell},\ ] ] and @xmath322 from which both parts of the third claim follow . [ prop : shahar - proposition ] there exist constants @xmath323 , and @xmath70 that depend only on @xmath324 for which the following holds . let @xmath325 , and assume that @xmath190 and that @xmath326 . then @xmath327 \\ & & \quad \geq 1-\frac{1}{n}-c_2\biggl(\frac{1}{\sqrt{n}}+\delta\biggr)(\log n)^2\sqrt{\log m},\end{aligned}\ ] ] provided that @xmath328 . for every @xmath329 , let @xmath330 the events @xmath331 for @xmath329 are disjoint , and thus @xmath332\\ & & \quad={\mathbb{p}}\biggl[\bigcup_{j=2}^m\omega_j\biggr]=(m-1){\mathbb{p}}[\omega_2].\end{aligned}\ ] ] since the variables @xmath333 are independent , we have @xmath334=\int_{-\infty}^{\gamma_1}f_{\bar x}(z)\biggl(\int_{z+\delta}^\infty f_{\bar x}(t)\,\mathrm{d}\mu(t)\biggr)^{m-2}\,\mathrm{d}\mu(z),\ ] ] where @xmath335 is a density function of @xmath301 with respect to the lebesgue measure @xmath336 . on the other hand , for any @xmath337 , @xmath338>0 $ ] because of ( [ eq : def - gamma1 - 2 ] ) . thus , for every @xmath337 , @xmath339 note that for every @xmath317 , @xmath340 , and applied to ( [ eq : proba - omega2 - 2 ] ) , @xmath341&\geq&\int_{-\infty}^{\gamma_1}f_{\bar x}(z)\biggl(\int _ z^\infty f_{\bar x}(t)\,\mathrm{d}\mu(t)\biggr)^{m-2}\,\mathrm{d}\mu(z ) \\[-2pt ] & & { } -(m-2)\int_{-\infty}^{\gamma_1 } f_{\bar x}(z)\biggl(\int_{z}^{\infty } f_{\bar x}(t)\,\mathrm{d}\mu(t)\biggr)^{m-3 } \biggl(\int_{z}^{z+\delta}f_{\bar x}(t)\,\mathrm{d}\mu(t)\biggr)\,\mathrm { d}\mu(z ) \\[-2pt ] & \geq & { \mathbb{p}}[\bar x_2\leq\gamma_1 \mbox { and } \bar x_k\geq\bar x_2 , \mbox{for every } k \geq3]-t_2 \\[-2pt ] & = & \frac{1}{m-1}{\mathbb{p}}\bigl[\min_{2\leq j\leq m}\bar x_j\leq\gamma_1\bigr]-t_2,\end{aligned}\ ] ] where @xmath342 recall the if @xmath343 are independent mean-0 random variables and @xmath344 are real numbers , then @xmath345 , where @xmath135 is an absolute constant @xcite . thus , @xmath346 , and for any @xmath347 , @xmath348\leq2\exp(-t^2/c^2\|{w}\|^2_{\psi_2}).\ ] ] let @xmath349 be such that @xmath350 thus , @xmath351 note that if @xmath352 , then our claim follows . indeed , because @xmath353 = 1-n^{-1}$ ] , we have @xmath334 \geq\frac{1}{m-1}\biggl(1-\frac{1}{n}\biggr ) - \delta\frac { \sqrt { \log(m-1)}}{m-1}.\ ] ] otherwise , we split the interval @xmath354=(-\infty , t_0)\cup [ t_0,\gamma_1]$ ] , and to upper bound @xmath355 , it remains to control the integral on the second interval @xmath356 $ ] . recall that @xmath357 and that @xmath190 . therefore , by proposition [ prop : petrov ] , it is evident that if @xmath358 and @xmath359 satisfy that @xmath360 and @xmath361 , then @xmath362 \nonumber \\[-9pt ] \\[-9pt ] \nonumber & \leq&{\mathbb{p}}[z\leq g\leq z+\delta]+\frac{b_1}{\sqrt{n}}\exp(-z^2/2),\end{aligned}\ ] ] where @xmath192 and @xmath363 are constants that depend only on @xmath364 . in addition , for every @xmath365 , @xmath366\leq\frac{1}{\sqrt{2\uppi}}\exp(-z^2/2 ) \int_0^\delta\exp(-zt)\,\mathrm{d}t\leq\frac{\delta}{\sqrt { 2\uppi}}\exp(-z^2/2).\ ] ] if @xmath367 , then @xmath368 . combining ( [ eq : delta - part ] ) and ( [ eq : upper - bound - gaussian1 ] ) with the definition of @xmath355 , we have @xmath369 \\ & & \quad\leq ( m-2)\biggl(\frac{b_1}{\sqrt{n}}+\frac{\delta}{\sqrt{2\uppi}}\biggr ) \exp(-\gamma_1 ^ 2/2)\frac{\log n}{m-1},\end{aligned}\ ] ] where the last inequality follows from ( [ eq : def - gamma1 - 2 ] ) . by lemma [ lemma : gamma-1 ] , and since @xmath370 , @xmath371 for some constant @xmath372 , from which our claim follows . we next describe the construction needed for the proof of theorem [ thb ] . let @xmath64 and @xmath43 be defined by @xmath373 where @xmath374 are @xmath24 independent random variables with density @xmath375}(u)$ ] for @xmath376 to be fixed later . note that for this choice of density function , @xmath377 is uniformly distributed on @xmath378 $ ] , and the best element in @xmath0 with respect to the quadratic risk is @xmath205 . let @xmath379 be a family of independent random variables distributed as @xmath380 . thus , for every @xmath381 , @xmath382 for every @xmath383 and @xmath384 . for every @xmath385 , set @xmath386 and observe that if @xmath387 , then @xmath172 is a mean 0 , variance @xmath166 random variable that is absolutely continuous with respect to the lebesgue measure and @xmath388 and satisfies that @xmath190 . these properties allow us to apply proposition [ prop : shahar - proposition ] to the random variables @xmath389 . let @xmath390 ( to be named later ) , and set @xmath391-\sqrt{12}\lambda(2-\lambda)\sqrt{n},\ ] ] and @xmath392.\ ] ] consider the system of inequalities @xmath393 and recall that for each @xmath394 @xmath395 denotes the weight of @xmath92 in the aew procedure . [ prop : large - weights ] there exist absolute constants @xmath5 and @xmath82 for which the following holds . let @xmath396 and @xmath329 . if the system ( [ eq : system - cj ] ) is satisfied , then @xmath397 moreover , if @xmath398 , then the quadratic risk of the function produced by the aew procedure satisfies @xmath399 let @xmath329 , and assume that ( [ eq : system - cj ] ) is satisfied . recall that @xmath400 is the empirical risk of @xmath13 , and note that for any @xmath401 , @xmath402 = \frac{\bar r_k-\bar r_j}{\sqrt{n } } \nonumber \\[-8pt ] \\[-8pt ] \nonumber & \geq&\frac{\delta}{\sqrt{n } } = \frac{-t}{n}\log\biggl[\frac{\rho}{2(m-2)(1-\rho)}\biggr].\end{aligned}\ ] ] in addition , since @xmath403 almost surely for any @xmath404 , @xmath405 \nonumber\\ & = & \frac{\bar r_1-\bar r_j}{\sqrt{n}}-\sqrt{12}\biggl(\lambda^2+\frac { 2\lambda } { n}\sum_{i=1}^n{\mathcal { u}}_1^{(i)}\biggr ) \\ & \geq&\frac{\bar r_1-\xi(\bar r_1)}{\sqrt{n}}-\sqrt{12}\lambda(2-\lambda)\geq \frac{-t}{n}\log\biggl[\frac{\rho}{2(1-\rho)}\biggr].\nonumber\end{aligned}\ ] ] combining ( [ eq : empirical - risk - kj ] ) and ( [ eq : empirical - risk-1j ] ) , it is evident that @xmath406}\\ & \geq & \frac{1}{1+(m-2){\rho}/(2(m-2)(1-\rho))+{\rho}/(2(1-\rho))}= 1-\rho.\end{aligned}\ ] ] since the functions @xmath407 are independent in @xmath408 and @xmath409 , @xmath410 and there is an absolute constant @xmath28 for which @xmath411 . thus , @xmath412 provided that @xmath398 , giving @xmath413 as claimed . next , we formulate a general statement , from which theorem [ thb ] follows immediately . there exist absolute constants @xmath414 and an integer @xmath81 for which the following holds . for any @xmath83 , @xmath415 , @xmath416 , and @xmath417 , let @xmath418 , @xmath419 , and @xmath420 . set @xmath0 to be the class of functions defined above with those parameters . then , with probability at least @xmath421 there exists @xmath422 such that @xmath423 in particular , with the same probability and if @xmath424 , @xmath425 set @xmath426,\ ] ] and , by proposition [ prop : large - weights ] , @xmath427={\mathbb{p}}_1.\ ] ] let @xmath428 be defined by @xmath429=1-n^{-1}$ ] , and observe that @xmath304 is well defined and satisfies all three parts of lemma [ lemma : gamma-1 ] for @xmath430 . set , @xmath431 and @xmath432 since the functions @xmath433 are independent , we have @xmath434 { \mathbh{1}}_{\omega_0 } ] \geq{\mathbb{p}}[b ] { \mathbb{p}}[\omega_0].\ ] ] applying proposition [ prop : shahar - proposition ] , we then have @xmath435 \geq1-\frac{1}{n } -c_2\biggl(\frac{1}{\sqrt{n}}+\delta\biggr)(\log n)^2\sqrt{\log m},\ ] ] provided that @xmath328 . to lower bound @xmath436 $ ] , note that @xmath437={\mathbb{p}}\biggl[\bar r_1\geq \gamma_1-\frac{t}{\sqrt{n}}\log\biggl(\frac{\rho}{2(1-\rho)}\biggr ) + \sqrt{12}\lambda(2-\lambda)\sqrt{n}\biggr].\ ] ] fix @xmath438 and assume that @xmath439 and @xmath44 are such that @xmath440 by the berry essen theorem and ( [ eq : gaussian - tail - estimate ] ) , @xmath441&\geq&{\mathbb{p}}[\bar r_1\geq(1 - 2\varepsilon)\gamma_1 ] = 1-{\mathbb{p}}[\bar r_1<(1 - 2\varepsilon)\gamma_1 ] \\ & \geq & 1-{\mathbb{p}}[g\leq ( 1 - 2\varepsilon)\gamma_1]-\frac{2\beta(w)}{\sqrt{n } } \\ & \geq & 1-\frac{1}{\sqrt{2\uppi}(1 - 2\varepsilon)|\gamma_1|}\exp \bigl(-(1 - 2\varepsilon ) ^2\gamma_1 ^ 2/2\bigr ) -\frac{2a}{\sqrt{n}},\end{aligned}\ ] ] and by lemma [ lemma : gamma-1 ] , @xmath442 therefore , @xmath443 provided that @xmath444 . to complete the proof , we need to chose @xmath445 and @xmath446 for which ( [ eq : def - theta - rho ] ) holds . by lemma [ lemma : gamma-1 ] , @xmath447 and thus ( [ eq : def - theta - rho ] ) holds for @xmath445 and @xmath446 for which @xmath448^{1/2}\quad \mbox{and}\quad \rho\geq 2\exp\biggl[\frac{-c_9\varepsilon\sqrt{n}}{t}\log^{1/2}\biggl(\frac{m}{\log n}\biggr)\biggr].\ ] ] in particular , when we take @xmath449 , @xmath450 , and @xmath451 , @xmath446 satisfies the required condition as long as @xmath452 and @xmath453 , as assumed . moreover , @xmath454 implying that @xmath455 the lower bound on the risk of the aew procedure now follows from proposition [ prop : large - weights ] . in this section we prove theorem [ thc ] , which we reformulate below . from here on , we assume that the dictionary @xmath0 is finite , consisting of @xmath24 functions , and that the functions are indexed according to their risk in an increasing order . thus , @xmath456 . in addition , we denote @xmath457 , and thus @xmath458 . for every @xmath116 , recall that @xmath459 which serves as a measure of complexity for the class @xmath0 . the first component needed in the proof of theorem [ thc ] is the level @xmath460 with the following property : with probability at least @xmath113 , @xmath461 is equivalent to @xmath462 if @xmath463 . this `` isomorphism '' constant was introduced by @xcite . to formulate the exact properties that we need , first recall the following definitions and notation . if @xmath464 is the excess loss functions class @xmath465 , then let @xmath466 is the star - shaped hull of @xmath467 and @xmath468 . set @xmath469 , that is , the set of functions in the star - shaped hull of @xmath470 and @xmath468 , with expectation @xmath471 . let @xmath472 where , as always , @xmath473 denotes the empirical mean and @xmath196 is the mean according to the underlying probability measure of @xmath9 . [ thm : isomorphic - coordinate - proj ] there exists an absolute constant @xmath135 for which the following holds . let @xmath0 be a class of functions bounded by @xmath29 , such that @xmath107 is a @xmath123-bernstein class . for every @xmath112 and an integer @xmath25 , let @xmath474 then , with probability at least @xmath113 , for every @xmath153 with @xmath475 , @xmath476 let @xmath477 , where @xmath478 is an absolute constant to be named later . recall that functions in @xmath0 are indexed according to their risk in an increasing order . let @xmath479 , and set @xmath480 as its complement . define the sets @xmath481 and , for @xmath482 , @xmath483 ( note that some of the sets @xmath484 may be empty . ) set @xmath485 and let @xmath486 . from theorem [ thm : isomorphic - coordinate - proj ] , it follows that for every @xmath487 and every @xmath488 , @xmath489 . this is because @xmath490 by the definition of @xmath480 , and @xmath491 . the key factor in the proof of theorem [ thc ] is theorem [ thm : optimal - agg ] . [ thm : optimal - agg ] there exist absolute constants @xmath5 and @xmath82 for which the following holds . let @xmath0 be a class of functions bounded by @xmath29 , such that @xmath107 is a @xmath123-bernstein class with respect to a convex risk function @xmath492 . then , with probability at least @xmath113 , if @xmath126 is produced by the aew algorithm and @xmath493 , then @xmath494 where @xmath460 is as defined in ( [ eq : lambda ] ) . let @xmath495 be the weights of the aew algorithm , and set @xmath496 to be the aggregate function . because @xmath492 is a convex function , @xmath497 note that for every @xmath498 , @xmath499 . in particular , because @xmath500 , @xmath501 on the other hand , with probability at least @xmath113 , for every @xmath502 and every @xmath488 , @xmath503 applying the definition of the weights in the aew algorithm and given that @xmath504 , @xmath505 from the definition of @xmath506 , it is evident that for every @xmath507 , @xmath508 , and thus if @xmath509 and @xmath478 is sufficiently large , then @xmath510 indeed , this follows because for that choice of @xmath44 , @xmath511 , with @xmath70 an absolute constant . thus , with probability at least @xmath113 , @xmath512 as claimed . the next step in the proof of theorem [ thc ] requires several simple facts regarding the empirical process indexed by a localization of the star - shaped hull of a bernstein class . first , it is simple to verify that the star - shaped hull of a @xmath123-bernstein class is a @xmath123-bernstein class as well . second , if @xmath513 and @xmath514 , then @xmath515 in particular , @xmath516 [ lemma : localized - estimates ] there exists an absolute constant @xmath135 for which the following holds . if @xmath107 is a @xmath123-bernstein class with respect to @xmath9 , then for every @xmath471 and @xmath517 , @xmath518 fix @xmath116 and @xmath517 , and let @xmath519 note that every @xmath520 satisfies that @xmath521 for some @xmath33 , and for which @xmath522 . therefore , using the bernstein condition on @xmath107 , @xmath523 moreover , @xmath524 thus , by the gin zinn symmetrization theorem and a contraction argument ( see , e.g. , @xcite and @xcite ) , @xmath525 where the last inequality is evident by the sub - gaussian properties of the rademacher process ( cf . @xcite ) . since @xmath526 , it follows that @xmath527 implying that @xmath528 thus , again using a symmetrization argument and the sub - gaussian properties of the rademacher process , we have @xmath529 [ cor : est - for - r^ * ] there exist absolute constants @xmath5 and @xmath82 for which the following holds . let @xmath0 be a finite class consisting of @xmath24 functions bounded by @xmath29 , such that the excess loss class @xmath107 is a @xmath123-bernstein class . if we set @xmath530 , then @xmath531 observe that for every @xmath116 , @xmath532 where we define @xmath533 . let @xmath534 . since @xmath535 for every @xmath536 , we have @xmath537 and thus @xmath538 moreover , the functions of @xmath471 , @xmath539 and @xmath540 are increasing , and thus for any @xmath541 , @xmath542 and @xmath543 thus , if we consider @xmath544 for appropriate constants @xmath122 and @xmath70 , then @xmath545 . thus , @xmath546 and , therefore , @xmath547 finally , because @xmath548 and @xmath549 , we have @xmath550 . proof of theorem [ thc ] the proof of theorem [ thc ] follows from estimates of @xmath460 and @xmath551 . from corollary [ cor : est - for - r^ * ] , it is evident that @xmath552 where @xmath5 is an absolute constant to be identified later . ( note that @xmath96 is an increasing function . ) next , by the definition of @xmath506 , @xmath553 . therefore , using the notation of theorem [ thm : optimal - agg ] , @xmath554 and , in particular , @xmath555 for an appropriate choice of constant @xmath5 . the second part of theorem [ thc ] follows from a standard integration argument . this article was written while g. lecu was visiting the department of mathematics , technion , and the centre for mathematics and its applications , australian national university . the authors thank both of these institutions for their hospitality . they also thank pierre alquier and olivier catoni for useful discussions . g. lecu was supported by french agence nationale de la recherche anr grant `` prognostic '' anr-09-jcjc-0101 - 01 . s. mendelson was supported in part by the centre for mathematics and its applications , the australian national university , canberra , act 0200 , australia , by an australian research council discovery grant dp0559465 , dp0986563 and by the european community s seventh framework programme ( fp7/2007 - 2013 ) , erc grant agreement 203134 .

in this paper , we study the stochastic optimization of cloud radio access networks ( c - rans ) by joint remote radio head ( rrh ) activation and beamforming in the downlink . unlike most previous works that only consider a static optimization framework with full traffic buffers , we formulate a dynamic optimization problem by explicitly considering the effects of random traffic arrivals and time - varying channel fading . the stochastic formulation can quantify the tradeoff between power consumption and queuing delay . leveraging on the lyapunov optimization technique , the stochastic optimization problem can be transformed into a per - slot penalized weighted sum rate maximization problem , which is shown to be non - deterministic polynomial - time hard . based on the equivalence between the penalized weighted sum rate maximization problem and the penalized weighted minimum mean square error ( wmmse ) problem , the group sparse beamforming optimization based wmmse algorithm and the relaxed integer programming based wmmse algorithm are proposed to efficiently obtain the joint rrh activation and beamforming policy . both algorithms can converge to a stationary solution with low - complexity and can be implemented in a parallel manner , thus they are highly scalable to large - scale c - rans . in addition , these two proposed algorithms provide a flexible and efficient means to adjust the power - delay tradeoff on demand . cloud radio access networks ( c - rans ) , lyapunov optimization , penalized weighted minimum mean square error ( wmmse ) , lagrangian dual decomposition .

1603.01659

the fifth - generation ( 5 g ) wireless networks are expected to provide ubiquitous services to a larger number of simultaneous mobile devices with device density far beyond the current wireless communication systems . to cope with these challenges , ultra - dense low power nodes and cloud computing are regarded as two of the most promising techniques@xcite . leveraged on low power node and cloud computing , the cloud radio access network ( c - ran ) , first proposed in@xcite , is expected to revolutionize the architecture and operations of future wireless systems , and it has attracted considerable amount of attentions in both academia and industry@xcite@xcite . as shown in fig . [ cran_rchitecture ] , a large number of remote radio heads ( rrhs ) are densely deployed in the space domain for c - rans . each rrh is configured only with the front radio frequency ( rf ) components and some basic transmission / reception functionalities . the rrhs are connected to the baseband unit ( bbu ) pool through high - bandwidth and low - latency fronthaul links to enable real - time cloud computing . the c - rans can act as a platform for the practical implementation of coordinated multi - point ( comp ) transmission concepts@xcite . specifically , the bbu pool computes the beamforming weight coefficients for different rrhs , and sends the precoded data to various active rrhs . then the active rrhs cooperatively transmit the precoded data to different ues . the signals observed at each ue are superpositions of signals from multiple active rrhs . the beamforming weight coefficients are designed to steer the data to their intended receivers in the spatial domain . that is , for a given ue , the desired signals are combined coherently yet the interfering signals are combined out - of - phase . here the joint beamforming aims to improve the signal - to - interference - plus - noise ratio ( sinr ) in order to significantly improve the spectral efficiency of c - rans . various beamforming designs have been studied for comp in c - rans with different optimization objectives and constraints recently . in@xcite and@xcite , the number of active front haul links is minimized under a sinr constraint for each user equipment ( ue ) and a power constraint for each rrh . the problem is solved by minimizing an approximate of the original combinatorial objective function . the works in@xcite and @xcite aim to jointly optimize the set of rrhs serving each user and the corresponding beamformers , under the constraint of front haul capacity . specifically , power minimization is studied in@xcite and sum rate maximization is considered in@xcite . the design problems in@xcite@xcite fundamentally differ from that in@xcite@xcite in that they explicitly consider the fronthaul capacity . the problem of minimizing the overall power consumption and comp operational costs by joint user association and downlink beamforming was studied in@xcite , where the problem was addressed using a mixed integer second - order cone program framework . similar problem has been studied in@xcite , where the user association and beamforming were considered in both downlink and uplink c - rans . as existing solutions considering only the downlink like@xcite can not be modified in a straightforward way to solve the problem in@xcite , efficient algorithms were proposed utilizing the uplink - downlink duality result . the dense deployment of rrhs imposes new technical challenges for the design and implementation of large scale c - rans . with the centralized processing at the bbu pool , the power consumed by the fronthaul links that provide high - capacity connections with bbu pool becomes comparable to that for transmission@xcite . therefore , in order to reduce power consumption of the entire network , we can reduce the number of active rrhs by putting some of the rrhs into the sleep mode . the fronthaul links of sleeping rrh will also be turned off to save power consumption . therefore , the scheduling of rrh activation plays a critical role in the development of energy - efficient c - rans . related problem has been studied in@xcite , where two efficient group sparse beamforming algorithms were proposed to obtain the subset of active rrhs and their corresponding beamformers . the works in@xcite compares the energy efficiencies of two different downlink transmission strategies in c - rans by taking the rrh transmission power , rrh activation power , and load - dependent fronthaul power into considerations . compared with the optimal exhaustive search method , the computational complexity of the algorithms in@xcite can be significantly reduced , which , however , can still be very intensive for large - scale c - rans . this is due to the fact that a series of convex problems ( e.g. sdp , socp ) have to be solved centrally using standard cvx solvers . furthermore , the aforementioned literatures are typically based on snapshot - based static models , which indicates that the stochastic and time - varying features are not considered into the formulations . therefore , only the physical layer performance metrics such as power and throughput are optimized and the resulting control policy is only adaptive to channel state information ( csi ) . in practice , delay is also a key metric to measure the quality - of - service ( qos ) , which has also been neglected in these literatures . intuitively , there is a fundamental tradeoff between power consumption and queuing delay@xcite , thus it is important to jointly consider power consumption and delay to balance their tradeoff and to meet various performance requirement in c - rans . in practice , the stochastic control and delay analysis are usually investigated from the queue stability perspective in a time - varying system using the lyapunov optimization technique . many existing literatures have focused on stochastic optimization for time - varying wireless networks . a fundamental approach to stochastic resource allocation and routing for heterogeneous data networks was presented in@xcite , where the flow control is crucial to ensure no network resources are wasted whenever the traffic rates are inside or outside the capacity region . the authors of@xcite investigated stochastic control for wireless networks with finite buffers , where the joint flow control , routing , and scheduling algorithms can achieve high network utility and deterministically bounded backlogs inside the network . the delay analysis was conducted in@xcite for suboptimal scheduling in one - hop wireless networks with general interference set constraints and time - correlated traffic arrivals . there also have been lots of works that focus on optimizing power under queue stability and interference constraints@xcite@xcite . however , these works generally adopt highly simplified physical - layer models for wireless channels , such as interference avoidance constraint or simple channel - rate mapping function . they do not consider the complex non - linear relationship between signal power , interference power , and system throughput as in practical systems , which makes the system design much more challenging . this paper fundamentally differs from previous works in that we consider a power minimization problem by designing queue - aware joint optimization algorithms for c - rans where both rrh activation set and beamforming vector are adaptive to queue state information and channel state information . in addition , the non - linear impacts of interence are explicitly quantified during the system design . therefore , existing solutions can not be readily applied to the c - ran setting considered in this paper . motivated by these facts , we propose to study dynamic joint rrh activation and beamforming for c - rans by considering random traffic arrivals , queuing delays , and the time - varying fading channels . to the best of our knowledge , this is the first such work in c - rans . unlike the static optimization problems studied in the literatures , the problems considered in this paper are formulated as stochastic optimizations , which are notoriously difficult to solve but are important for practical systems . the main contributions of this paper are summarized as follows : the stochastic optimization of joint rrh activation and beamforming is developed for practical c - rans . a wide range of detailed operations and constraints , such as beamforming , rrh activation , time - varying channel , random traffic arrivals , and fronthaul power consumption , are considered in the formulation . to the best of our knowledge , this is the first paper that achieves stochastic control of power and delay by considering realistic system operations across multiple protocol layers . to tackle the np - hardness of the optimization problem , two low - complexity algorithms are proposed using the group sparse beamforming ( gsb ) approach and the relaxed integer programming ( rip ) approach , respectively . both algorithms can be implemented in a parallel manner with closed - form expressions , thus they are scalable to large - scale c - rans . the delay and power performance of the two proposed algorithms are numerically evaluated . significant performance gains are achieved by the proposed algorithms due to the fact that they are adaptive to the queue state information . the proposed algorithms can provide a flexible and efficient means to control the delay - power tradeoff on demand . all these make the proposed algorithms attractive and useful in practical applications . the remainder of this paper is organized as follows . in section ii , we introduce the system model and formulate the stochastic optimization problem . in section iii , the lyapunov optimization is introduced and utilized to obtain a penalized weighted sum rate maximization problem for each slot . the optimization problem is solved by two efficient algorithms based on gsb approach and rip approach in section iv and section v , respectively . numerical results are presented in section vi . finally , we conclude our paper in section vii . the acronyms used in this paper are summarized in table i. m1.1cm|m5.4 cm * acronym * & * description * + c - ran & cloud radio access network + bbu & baseband unit + rrh & remote radio head + mse & mean square error + mmse & minimum mean square error + wmmse & weighted minimum mean square error + gsb & group sparse beamforming + rip & relaxed integer programming + bcd & block coordinate descent + lasso & least absolute shrinkage and selection operator + admm & alternating direction method of multipliers + fjp & full joint processing + throughout this paper , lower - case bold letters denote vectors and and upper - case bold letters denote matrices . @xmath0 denotes identity matrix . @xmath1 denotes complex domain and the complex gaussian distribution with mean @xmath2 and covariance matrix @xmath3 is represented by @xmath4 . @xmath5 $ ] and @xmath6 represent expectation and determinant operators , respectively . @xmath7 is the real part operator . @xmath8 denotes @xmath9-norm of a vector . the inverse , transpose , conjugate transpose operators are denoted as @xmath10 , @xmath11 , @xmath12 , respectively . we consider a downlink c - ran with @xmath13 rrhs and @xmath14 ues , where each rrh is equipped with @xmath15 antennas and each ue has @xmath16 antennas . let @xmath17 and @xmath18 denote the set of rrhs and the set of ues , respectively . the bandwidth of the system is @xmath19 . we also assume that the network operates in slotted time with time dimension partitioned into decision slots indexed by @xmath20 with slot duration @xmath21 . let @xmath22 denote the channel state information ( csi ) matrix from rrh @xmath23 to ue @xmath24 at slot @xmath25 , let @xmath26 } \in { \mathbb{c}^{n \times mk}}$ ] denote the csi matrix from all rrhs to ue @xmath24 at slot @xmath25 , and let @xmath27 \in \mathbb{c}^{n\times mik}$ ] denote the network csi at slot @xmath25 . the channel is assumed to follow quasi - static block fading , where each element of @xmath28 keeps constant for the duration of a slot , but is identically and independently distributed ( i.i.d . ) across different slots . let @xmath29 denote the beamforming vector at rrh @xmath23 for ue @xmath24 at slot @xmath25 , let @xmath30^t \in { \mathbb{c}^{{mk } \times 1}}$ ] denote the aggregated beamformer for ue @xmath24 at slot @xmath25 , let @xmath31^t \in { \mathbb{c}^{mi \times 1}}$ ] denote the aggregated beamformer used by rrh @xmath23 at slot @xmath25 , and let @xmath32^t \in { \cal c}^{mik \times 1}$ ] denote the aggregated beamformer of the entire network at slot @xmath25 . assume that each ue has its own data stream . let @xmath33 denote the data message for ue @xmath24 at slot @xmath25 . without loss of generality , we further assume that @xmath34 = 1 $ ] and each @xmath33 is i.i.d . among ues . with linear beamforming operated centrally in the bbu pool , the baseband signal to be transmitted by rrh @xmath23 at slot @xmath25 is @xmath35 the encoded baseband signal @xmath36 is delivered to rrh @xmath23 for radio transmission through corresponding fronthaul link . it is worth noting that as we focus on the issue of power - delay tradeoff in this paper , we assume that the fronthaul links are provisioned with sufficiently high capacity and negligible latency . the signal observed by each ue is the superposition of signals from all rrhs . the received signal at ue @xmath24 is given by @xmath37 where @xmath38 is the additive white gaussian noise ( awgn ) at slot @xmath25 with distribution @xmath39 . we assume that all the ues adopt single user detection and the interference is treated as noise . the achievable data rate in the unit of bps / hz of ue @xmath24 is given by @xmath40 in c - rans , the extensive use of high - capacity low - latency fronthaul links makes the fronthaul power consumption comparable to the transmission power of rrhs@xcite . here we consider the passive optical network to provide the effective high - capacity fronthaul connections between the rrhs and the bbu pool . the passive optical network consists of optical network units and an optical line terminal that connects a set of associated optical network units through a single optical fiber@xcite . from the perspective of energy saving , some rrhs and their associated optical network units can be switched into sleep mode with negligible power consumption , but the optical line terminal with constant power consumption @xmath41 can not go into sleep mode as it plays the roles of distributer , aggregator and arbitrator of the transport network . here we ignore @xmath41 because it is a constant and will not affect the scheduling and optimization results . let @xmath42 denote the constant power consumed by the optical network unit associated with active rrh @xmath23 . besides , due to the real - time a / d and d / a processing at each rrh , static circuit power @xmath43 is also consumed . thus , the amount of static power consumption associated with rrh @xmath23 during active mode is @xmath44 . when rrh @xmath23 and its corresponding fronthaul link are switched into the sleep mode , there is no static power consumption . let @xmath45 denote the set of active rrhs at slot @xmath25 . the network power consumption at slot @xmath25 is given by @xmath46 where @xmath47 is the drain efficiency of rf power amplifier at rrh @xmath23 . note that the load - dependent fronthaul power consumption model has been considered in@xcite , while its impact on our formulation will be left for future study . the network power consumption is a random process , in that it depends on the policy of rrh activation set and corresponding beamforming vectors , which is dynamically determined with the observation of traffic queues and channel conditions at each slot . the bbu pool maintains @xmath14 traffic queues for the random traffic arrivals towards @xmath14 ues . let @xmath48 $ ] be the vector of stochastic traffic data arrivals ( bits ) at the end of slot @xmath25 . we assume that the traffic arrival @xmath49 is independent w.r.t . @xmath24 and i.i.d . over slots according to a general distribution with mean @xmath50 = { \lambda _ i}$ ] . let @xmath51 $ ] denote the vector of queue state information ( qsi ) ( bits ) for the @xmath14 ues at the beginning of slot @xmath25 . therefore , the queue dynamic for ue @xmath24 is given by @xmath52 + { a_i}(t),\label{qdynamic}\ ] ] where the amount of traffic departure at slot @xmath25 is given by @xmath53 . to model the impacts of joint rrh activation and beamforming policy on average queue delay and average network power consumption , we first present the definitions of queue stability , stability region and throughput optimal policy as follows@xcite . _ a discrete time queue @xmath54 is strongly stable if @xmath55 } < \infty.\ ] ] furthermore , a network of queues is stable if all individual queues of the network are stable . _ ( _ stability region and throughput - optimal policy _ ) : _ the stability region @xmath56 is the closure of the set of all the arrival rate vectors @xmath57 that can be stabilized in a c - ran . a throughput - optimal resource optimization policy is a policy that stabilizes all the arrival rate vectors @xmath58 within the stability region @xmath56 . _ the objective is to simultaneously maintain the network queue stability and minimize the network power consumption , by using joint rrh activation and beamforming . the problem can be formulated as the following stochastic optimization problem : @xmath59}\\ { \rm{s.t . } } & { \rm{c1 } : } { \rm{queue~}}{q_i}(t){\rm{~is~strongly~stable,~}}\forall i , \\ & { \rm{c2}:}||{{{\bf{\tilde w}}}_k}||_2 ^ 2 \le { p_k } , \\ \end{array}\label{stochasticproblem}\ ] ] where the expectation @xmath60 is taken with respect to the distribution of network power consumption , which depends on the random rrh activation set and beamforming vectors . c1 is the network stability constraint to guarantee a finite queue length for each queue . c2 is the constraint on the instantaneous per - rrh power consumption . in practical c - rans , the random traffic arrivals and the time - varying channel conditions are generally unpredictable . the stochastic nature of the channel conditions and traffic arrivals makes it impractical to calculate the optimal solution in an offline manner . to address this problem , we will resort to lyapunov optimization , which can transform the stochastic optimization problem ( [ stochasticproblem ] ) into a deterministic one at each slot . the queue stability constraint is used to depict and control the average delay . according to definition 1 , the queue stability is guaranteed if the average queue length is finite . note that average delay is proportional to average queue length for a given traffic arrival rate from little s theorem . as suggested later in section iii , the average queue length can be arbitrarily bounded by choosing an appropriate control parameter . in this section , we will exploit the framework of lyapunov optimization to solve the stochastic optimization problem in ( [ stochasticproblem ] ) . define the quadratic lyapunov function as @xmath61 , which serves as a scalar metric of queue congestion in the c - ran . to keep the system stable by persistently pushing the lyapunov function towards a lower congestion state , the one - step conditional lyapunov drift is defined as @xmath62,\ ] ] where @xmath60 is the conditional expectation taken with respect to the distribution of lyapunov drift given queue state @xmath63 . the lyapunov drift - plus - penalty function is defined as @xmath64,\label{driftpenalty}\ ] ] where @xmath60 is the conditional expectation taken with respect to the distribution of network power consumption given queue state @xmath63 , and @xmath65 represents an arbitrary control parameter . the parameter @xmath66 can be used to control the power - delay tradeoff . a larger @xmath66 means more emphasis will be put on power minimization during the optimization . on the other hand , when @xmath66 is small , queue stability carries more weight during the optimization . suppose that the expectation of the penalty process @xmath67 is deterministically bounded by some finite constant @xmath68 , @xmath69 , i.e. @xmath70 \le { p_{\max } } $ ] . let @xmath71 denote the theoretical optimal value of ( [ stochasticproblem ] ) , then the relationship between the lyapunov drift - plus - penalty function and queue stability is established in theorem 1@xcite . _ suppose there exist positive constants @xmath72 , @xmath73 and @xmath66 such that for all slots @xmath74 and all possible values of @xmath63 , the lyapunov drift - plus - penalty function satisfies : @xmath75 \le b + v{p^ * } - \epsilon \sum\limits_{i = 1}^i { { q_i}(t ) } , \ ] ] then all queues @xmath76 are strongly stable . the average queue length satisfies @xmath77 } } \le \frac{{b + v({p^ * } - { p_{min } } ) } } { \epsilon } , \ ] ] and the average penalty of power consumption satisfies @xmath78 } \le { p^ * } + \frac{b}{v}.\ ] ] _ the proof can follow that for theorem 4.2 in@xcite . the results in theorem 1 motivate us to minimize the lyapunov drift - plus - penalty in ( [ driftpenalty ] ) to achieve the maximum queue stability region and obtain throughput - optimal policy . rather than directly minimize ( [ driftpenalty ] ) , our policy actually seeks to minimize the upper bound of ( [ driftpenalty ] ) , which is given by the following lemma [ lemma 4.6 of 23 ] . _ under any control policy , the drift - plus - penalty has the following upper bound for all @xmath25 , all possible values of @xmath79 and all parameters @xmath65 , @xmath80 \le b + \\v\mathbb{e } [ p(\mathcal { a}(t ) , { \bf{w}}(t))|{\bf{q}}(t ) ] + \sum\limits_{i \in \mathcal { i } } { q_i}(t)\mathbb{e}[{a_i}(t ) - \mu _ i(t)|{\bf{q}}(t ) ] , \end{array}\label{driftpenaltybound}\ ] ] where @xmath72 is a positive constant and for all slot @xmath25 satisfies @xmath81}$ ] . _ the proof is in appendix a. by the principle of _ _ opportunistically minimizing an expectation__@xcite , the policy that minimizes @xmath82 $ ] is the one that minimizes @xmath83 with the observation of @xmath63 . besides , neither @xmath84 nor @xmath72 in ( [ driftpenaltybound ] ) will be affected by the policy at slot @xmath25 . therefore , the optimization problem can be simplified to @xmath85 the following theorem justifies the throughput optimality by solving problem ( [ p1 ] ) optimally . the rrh activation @xmath86 and beamforming @xmath87 given by solving ( [ p1 ] ) optimally achieves the maximum stability region @xmath56 in c - rans . the proof is in appendix b. however , the weighted sum rate term in ( [ p1 ] ) is nonconvex and is shown to be np - hard in wireless networks with interference@xcite . it is thus extremely difficult , if not impossible , to get the globally optimal solution to ( [ p1 ] ) through efficient algorithms in polynomial time . rather than seeking global optimality , we will focus on developing low - complexity algorithms that produce suboptimal solutions to ( [ p1 ] ) . the following theorem characterizes the performance of ( [ p1 ] ) under suboptimal solutions . let @xmath88 and @xmath89 be constants such that @xmath90 and @xmath91 . suppose there is an @xmath92 , such that @xmath93 if the suboptimal solution makes ( possibly randomized ) decisions every slot to satisfy @xmath94 } \ge \phi \left ( { \max \sum\limits_{i = 1}^i { { q_i}(t){\mu _ i } } } \right ) - c,\label{scaledproblem}\ ] ] then the network is strongly stable . the proof can follow that for theorem 6.3 in@xcite . theorem 3 suggests that the suboptimal solutions that satisfy ( [ scaledproblem ] ) can provide stability whenever the traffic arrival rates are interior to a @xmath88-scaled version of the stability region . in this paper , we will develop suboptimal solutions by relaxing and reformulating the optimization problem in ( [ p1 ] ) , while it is extremely difficult to quantify the @xmath88 and @xmath89 that the algorithms can achieve . the stability region analysis for our proposed algorithms remains challenging and is left for future work . for the general case that the arrival rate vector is outside the stability region @xmath56 or the possible reduced one @xmath95 , congestion controls are need to constrain the arrival rate vector into the stability region . in this case , the problem can be decomposed into a congestion control subproblem and joint rrh activation and beamforming subproblem by following the framework in@xcite . by doing so , we can have the separate congestion control subproblem and joint rrh activation and beamforming subproblem , and the deterministic worst - case delay bound can be guaranteed for each traffic queue . in this section , we will use the gsb approach to solve the optimization problem in ( [ p1 ] ) . since only a subset rrhs will be active , we can solve the problem by exploiting the group sparse structure of the aggregated beamforming vector @xmath96^t \in { \cal c}^{mik \times 1}$ ] , where the coefficients in @xmath97 form a group@xcite . when the rrh @xmath23 is switched off , all the coefficients in the vector @xmath98 are 0 , which results in the group sparse structure . the mixed @xmath99-norm is shown to be effective to induce group sparsity and has attracted lots of attentions@xcite . in this subsection , we try to construct a convex relaxation of ( [ networkpower ] ) , resulting in a weighted mixed @xmath100-norm . specifically , we first calculate the _ tightest positively homogeneous lower bound _ of @xmath101 with the definition @xmath102 , which is still nonconvex . we then calculate the fenchel conjugate to provide its convex envelope @xmath103 , which is called as the _ tightest convex positively homogeneous lower bound _ of @xmath101 and is given by the following proposition . _ the tightest convex positively homogeneous lower bound of ( [ networkpower ] ) is given by @xmath104 _ which is a weighted mixed @xmath105-norm . the proof is in appendix c. the above proposition indicates that the mixed @xmath100-norm can provide a convex relaxation for the cost function ( [ networkpower ] ) , thus it can further introduce group sparsity to @xmath106 , that is , many sub - vectors , @xmath107 will be 0 , which corresponds to inactive rrhs . while the active set of rrhs corresponds to the non - zero sub - vectors in @xmath108 . by minimizing the weighted mixed @xmath105-norm of @xmath106 , the zero entries of @xmath106 will be made to align to the same group @xmath107 , such that the corresponding rrh is forced to switch off . the weight for each group embraces additional system parameters . intuitively , the rrh with a higher static power consumption and a lower rf power amplifier drain efficiency will have a high priority being forced to switch off . using the weighted mixed @xmath105-norm as a surrogate objective function in ( [ p1 ] ) , we finally have the following queue - aware group sparse beamforming problem : @xmath109 in the above formulation , the slot index @xmath25 is skipped to simplify the notation . the objective function in ( [ relaxp ] ) is a convex relaxation to the original problem ( [ p1 ] ) using the group sparsity inducing norm . it has been shown in proposition 1 that ( [ relaxp ] ) is the tightest convex positively homogeneous lower bound of ( [ p1 ] ) , that is , among all convex positively homogeneous functions , ( [ relaxp ] ) has the smallest gap with ( [ p1 ] ) . it is very challenging to quantify the exact performance gap , which normally requires specific prior information , e.g. , in compressive sensing , the sparse signal is assumed to obey a power law ( see eq . ( 1.8 ) in @xcite ) . however , our problem fundamentally differs from the existing compressive sensing problems in that we do not have any prior information about the optimal solution . the optimality analysis of the queue - aware group sparse beamforming algorithm will be left to our future work . next we will design a computationally efficient algorithm that produces a stationary solution to ( [ relaxp ] ) by introducing an equivalent formulation . the equivalence between weighted sum rate maximization problem and wmmse problem is first established in@xcite for multiple - input and multiple - output ( mimo ) broadcast channel and generalized to mimo interfering channel in@xcite . by extending the equivalence in@xcite@xcite , the penalized weighted sum rate maximization problem is equivalent to the following penalized wmmse problem , @xmath110 where @xmath111 is the set of non - negative mean squared error ( mse ) weights , @xmath112 is the mse for estimating @xmath113 , @xmath114 is the collection of the receiving vectors for all ues , and @xmath115 is the parameter that will affect the number of active rrhs . it is worth noting that problem ( [ p2 ] ) is not jointly convex in @xmath116 , while it is convex with respect to each of the individual optimization variables when fixing the others . to this end , the block coordinate descent ( bcd ) method is utilized to obtain the stationary point of problem ( [ p2 ] ) . as proven in@xcite , once the iterative process converges to a fixed point of problem ( [ p2 ] ) , the fixed point is also a stationary point of the problem ( [ relaxp ] ) . it should be noted that the stationary point of problem ( [ relaxp ] ) or ( [ p2 ] ) might not be globally optimal . under fixed @xmath117 and @xmath118 , minimizing the weighted sum - mse leads to the well - known mmse receiver : @xmath119 with the mmse receiver , the mse @xmath120 can be written as @xmath121 under fixed @xmath117 and @xmath122 , the closed - form @xmath118 can be obtained as follows according to the first - order optimality conditions : @xmath123 under fixed @xmath124 and @xmath118 , the optimal @xmath117 can be obtained by solving the following convex problem : @xmath125 where @xmath126 and @xmath127 . the algorithm is summarized in algorithm [ algorithm1 ] . for each slot @xmath25 , observe the current qsi @xmath63 and csi @xmath28 , then make the queue - aware joint rrh activation and beamforming according to the following steps : * initialize * @xmath117 , @xmath124 and @xmath118 ; fix @xmath117 , compute the mmse receiver @xmath124 according to ( [ ummse ] ) and corresponding mse @xmath120 ; update the mse weight @xmath118 according to ( [ alphammse ] ) ; calculate the optimal beamformer @xmath117 under fixed @xmath124 and @xmath118 by solving ( [ problemw]);certain stopping criteria is met ; * update * the traffic queue @xmath128 according to ( [ qdynamic ] ) . [ algorithm1 ] the objective function in ( [ problemw ] ) contains two parts : the quadratic part @xmath129 , and the @xmath130-norm part @xmath131 . unlike the objective of _ standard _ group least absolute shrinkage and selection operator ( lasso ) problem , the two parts are functions of different variables , i.e. , @xmath132 and @xmath107 , rather than the same variable . therefore , existing computationally efficient algorithms developed for group lasso in@xcite can not be readily applied to solve our _ modified _ group lasso problem ( [ problemw ] ) . this fact motivates us to find a new approach to solve the problem in ( [ problemw ] ) . our approach is based on the famous admm algorithm , which will be briefly reviewed below . the admm algorithm , originally introduced in the 1970s , is a simple but powerful algorithm that is well suited to distributed convex optimization , and arbitrary - scale convex optimization . specifically , the admm is designed to solve the following structured convex problem@xcite @xmath133 where @xmath134 , @xmath135 , @xmath136 , @xmath137 and @xmath138 are convex functions , and @xmath139 and @xmath140 are non - empty convex sets . correspondingly , the partial augmented lagrangian function is given by @xmath141 where @xmath142 is the vector of lagrangian dual variables associated with the linear equality constraint , and @xmath143 is a constant . the admm algorithm consists of the following iterations : @xmath144 @xmath145 @xmath146 it can be easily seen that the admm algorithm takes the form of a decomposition - coordination procedure , in which the solutions to small local sub - problems are coordinated through dual variable update ( [ dualupdate ] ) to find a solution to a large global problem . furthermore , the convergence is established in the following theorem@xcite . _ assume that @xmath147 and @xmath148 are invertible , and the optimal solution to ( [ admmform ] ) exists . then the updated sequence @xmath149 is bounded and the converged @xmath150 is an optimal solution of ( [ admmform ] ) . _ in this subsection , based on splitting the variables @xmath108 , the tightly coupled large - sized problem in will be decomposed into several sub - problems and solved efficiently by the admm algorithm with closed - form solutions . to account for the difference between @xmath132 and @xmath151 in , we introduce a copy @xmath152 for the original beamformer @xmath107 , and define @xmath153^t \in { \mathbb{c}^{mki \times 1}}$ ] . the optimization problem in can be equivalently expressed as @xmath154 the partial augmented lagrangian function of the above problem is given by @xmath155 where @xmath156^t$ ] , with @xmath157^t \in \mathbb{c}^{mi \times 1}$ ] being the vector of lagrangian dual variables for the equality constraints in ( [ varsplit ] ) , and @xmath158 is some constant . the main steps of the admm algorithm are summarized in algorithm [ algorithm2 ] . * initialize * all primal variables @xmath159 , @xmath160 and all dual variables @xmath161 . solve the following problem and obtain @xmath162 , @xmath163 solve the following problem and obtain @xmath164 , @xmath165 update the multipliers @xmath166 by @xmath167 certain stopping criteria is met . [ algorithm2 ] before obtaining the closed - form expressions for each iteration in the above algorithm , the convergence of algorithm 2 is first discussed . _ every limit point @xmath168 and @xmath169 generated by algorithm 2 is an optimal solution of problem ( [ problemw ] ) . _ the proof is in appendix d. in algorithm 2 , given @xmath108 and @xmath170 , the step to obtain @xmath171 can be further decomposed into @xmath13 sub - problems , each of which is associated with a rrh and can be solved in a parallel manner . by completing the squares in ( [ p3 ] ) , the @xmath23-th convex sub - problem for rrh @xmath23 can be simplified to @xmath172 the corresponding karush - kuhn - tucker ( kkt ) conditions are given by @xmath173 @xmath174 where @xmath175 , @xmath176 is the optimal lagrangian multiplier associated with the power constraint , and @xmath177 represents the subdifferential of @xmath130-norm @xmath178 at the point @xmath179 , which can be expressed as follows@xcite @xmath180 therefore , we have @xmath181 whenever @xmath182 . when @xmath183 , we have @xmath184 furthermore , according to the complementary conditions , @xmath185 if @xmath186 . otherwise , we should have @xmath187 , which indicates that @xmath188 . in a summary , the closed - form solution for updating @xmath171 is given by @xmath189 as for the step to obtain @xmath117 , it can be further decomposed into @xmath14 sub - problems , each of which is associated with a ue and can be solved in a parallel manner . by completing the squares in ( [ p3 ] ) , the @xmath24-th unconstrained convex sub - problem for ue @xmath24 can be simplified as @xmath190 since the summation term in is the summation of squared @xmath191-norms , it can be alternatively expressed as @xmath192 . differentiating the objective with respect to @xmath193 and setting the result equal to zero , the optimal @xmath193 is given by the following closed - form expression : @xmath194 where @xmath195^t } \in { \mathbb{c}^{mk \times 1}}$ ] , with @xmath196 being the @xmath24-th block of @xmath197 . to further decrease the number of active rrhs by enhancing the group - sparsity for the beamformer , the reweighting procedure that adaptively reweights the coefficient @xmath198 in ( [ problemw ] ) can be utilized . specifically , this can be done in step 6 of algorithm 1 by solving problem ( [ problemw ] ) and updating the coefficient @xmath198 iteratively , see section v - c in@xcite for details . by doing so , the number of active rrhs is expected to be smaller than that obtained by solving problem ( [ problemw ] ) only once . however , the enhanced algorithm 1 with the reweighting procedure will involve two loops , which results in a high computational complexity . as a result , the tradeoff between complexity and accuracy should be carefully considered in the implementation . each step of algorithm 1 and algorithm 2 can be carried out in a parallel manner . except for the mse weights @xmath118 and the lagrangian dual variables @xmath199 , the computation for beamformer @xmath200 and beamformer @xmath201 can be performed in the parallel computing units of bbu pool for each rrh and each ue , respectively , without any information exchange . after that , both @xmath118 and @xmath199 are updated with outputted @xmath200 and @xmath201 . once the update is done , @xmath200 and @xmath201 can be calculated simultaneously again in the parallel computing units of bbu pool . the proposed solution is highly efficient as each step of algorithm 1 and algorithm 2 is in closed - form . specifically , the main computational complexity is related to the matrix inversion in ( [ ummse ] ) and ( [ optimalw_i ] ) , which have computational complexity in the order of @xmath202 and @xmath203 , respectively . compared with the standard interior point method , which has a computational complexity in the order of @xmath204 , the proposed gsb - based solution has a lower computational complexity , especially for large - scale c - rans . in this section , we will use the rip approach to solve the optimization problem in ( [ p1 ] ) . a set of binary variables @xmath205 is first introduced to indicate the active / sleeping states of each rrh , where @xmath206 when rrh @xmath23 is activated , and @xmath207 when rrh @xmath23 is asleep . with @xmath208 , the network power consumption model ( [ networkpower ] ) can be rewritten as @xmath209 replacing the penalty term of ( [ p1 ] ) with the power consumption model defined in ( [ eqpowernew ] ) , we have the following penalized weighted sum rate maximization problem : @xmath210 where @xmath211 . as can be seen , the nonlinear cross - multiplication terms ( ncmts ) in the cost function impose a great challenge on algorithm design . inspired by the fact that @xmath107 is equal to @xmath212 if rrh @xmath23 is turned off , we can cancel all the ncmts in the objective function and constraint as follows : @xmath213 the above problem is a mixed integer nonlinear programming , and some standard algorithms have been developed to solve it , e.g. , the branch - and - bound ( bnb ) algorithm@xcite . however , the computational complexity of bnb is prohibitive for a large - scale c - ran . for the worst case , @xmath214 iterations is required , thus their computational complexity is approximated as @xmath215 , which grows exponentially with the number of rrhs and can not be applied in practice . efficient algorithm will be studied in the following subsection . utilizing the established equivalence between penalized weighted sum rate maximization and penalized wmmse in section iv , and relaxing the binary variable @xmath216 to take continuous value in @xmath217 $ ] , the problem can be transformed as : @xmath218.\\ \end{array}\label{ripp6}\ ] ] the above problem is strongly convex for individual variable @xmath118 and @xmath124 when fixing the rest . correspondingly , the unique optimal solutions @xmath219 and @xmath220 are given by ( [ ummse ] ) and ( [ alphammse ] ) , respectively . furthermore , when fixing @xmath118 and @xmath124 , the optimization problem for @xmath221 and @xmath106 is convex and is given by @xmath222.\\ \end{array}\label{ripbcd}\ ] ] therefore , a stationary solution can be obtained with the bcd method by iteratively optimizing over three block variables @xmath118 , @xmath124 and @xmath223 . the algorithm is summarized in algorithm [ algorithm3 ] . for each slot @xmath25 , observe the current qsi @xmath63 and csi @xmath28 , then make the queue - aware joint rrh activation and beamforming according to the following steps : * initialize * @xmath224 , @xmath117 , @xmath124 and @xmath118 ; fix @xmath224 and @xmath117 , compute the mmse receiver @xmath124 according to ( [ ummse ] ) and the corresponding mse . update the mse weight @xmath118 according to ( [ alphammse ] ) ; obtain the rrh activation decision @xmath221 and the beamformer @xmath117 under fixed @xmath124 and @xmath118 by solving ( [ ripbcd]);certain stopping criteria is met ; * update * the traffic queue @xmath128 according to ( [ qdynamic ] ) . [ algorithm3 ] the lagrangian function of problem ( [ ripbcd ] ) is given by@xcite @xmath225 where @xmath226 \succeq { \bf{0}}$ ] is the vector of dual variables associated with the network power consumption constraints . correspondingly , the lagrange dual function is given by @xmath227 , \\ \end{array}\ ] ] and the dual optimization problem is formulated as @xmath228 the lagrangian function @xmath229 is linear with @xmath230 for any fixed @xmath221 and @xmath117 , while the dual function @xmath231 is the maximum of these linear functions . therefore , the dual optimization problem is always concave . given lagrangian dual variables , the problem of rrh activation and beamforming can be decomposed and solved separately . for the problem of rrh activation , it can be further decomposed into @xmath13 independent problems and solved in a parallel manner . the activation problem for rrh @xmath23 is given by @xmath232 it can be easily seen that the optimal solution of @xmath221 is given by @xmath233 it is worth noting that after the integer relaxation on @xmath234 , we can still obtain optimal solution that is binary . therefore , the integer relaxation does not introduce performance loss and there is no gap with integer relaxation . given lagrangian dual variables , the problem to derive beamforming vectors is given by @xmath235 which can be further rewritten as @xmath236 where @xmath237 \otimes { { \bf{1}}_m})$ ] is a diagonal matrix , @xmath238 denotes the kronecker product of two vectors , and @xmath239 denotes a length @xmath15 all - one vector . the problem ( [ ripwi ] ) is an unconstrained convex problem , and it can be further decomposed into @xmath14 independent problems , each corresponding to an ue and solved in a parallel manner . thus , according to the first - order optimality condition , the optimal beamforming vector for ue @xmath24 is given by @xmath240 as the dual problem is always concave w.r.t . @xmath241 , we can adopt the subgradient projection method to solve it@xcite . in particular , it is easy to prove that the subgradient of the dual function is obtained by @xmath242 where @xmath243 is the optimal rrh activation according to ( [ rrhonoff ] ) in the @xmath244-th iteration given @xmath245 , and @xmath246 can be obtained from the optimal beamformer for each ue in the @xmath244-th iteration given @xmath245 . hence , with the subgradient projection method , the update equation for the dual variable @xmath247 in the @xmath248-th iteration is given by @xmath249 where @xmath250 is a sufficiently small positive step size . the whole procedure to solve ( [ ripbcd ] ) is summarized in algorithm [ algorithm4 ] . * initialize * @xmath221 , @xmath106 , @xmath241 , @xmath251 ; make the rrh activation decision @xmath221 according to ( [ rrhonoff ] ) ; calculate the beamforming vector @xmath193 for each ue according to ( [ beamformingrip ] ) ; update the dual variables @xmath241 according to ( [ dualrip ] ) ; set @xmath252 ; @xmath253 or @xmath254 . [ algorithm4 ] similarly , each step of algorithm 3 and algorithm 4 can be carried out in a parallel manner . except for the mse weights @xmath118 and the lagrangian dual variables @xmath230 , the computation for rrh activation @xmath255 and beamformer @xmath201 can be performed in the parallel computing units of bbu pool for each rrh and each ue , respectively , without any information exchange . after that , both @xmath118 and @xmath241 are updated with outputted @xmath255 and @xmath201 . once the update is done , @xmath255 and @xmath201 can be calculated simultaneously again in the parallel computing units of bbu pool . the proposed solution is highly efficient as each step of algorithm 3 and algorithm 4 is in closed - form . specifically , the most computationally intensive operation is the matrix inversion in ( [ ummse ] ) and ( [ beamformingrip ] ) , which have computational complexity in the order of @xmath202 and @xmath203 , respectively . compared with the standard interior point method , which has a computational complexity in the order of @xmath204 in our setting , the proposed rip - based solution has a lower computational complexity , especially for large - scale c - rans . in this section , we use simulation and numerical results to evaluate and compare the performance of the proposed algorithms . the pathloss model is @xmath256 with @xmath257 ( km ) being the propagation distance . the fast fading is modeled as independent complex gaussian random variable distributed according to @xmath258 and the noise power is -112 dbm . we assumed that the rrhs are configured with 2 antennas , the ues are configured with 1 antennas , and they are uniformly and independently distributed in the square region [ -500 500 ] @xmath259 [ -500 500 ] meters . besides , we assume that the mean arrival rate is the same for all the ues , i.e. , @xmath260 . we fix the power budget of each rrh as @xmath261 w. each point of the the simulation results is averaged over 4000 slots in the simulations . our simulations mainly compare the proposed algorithms with the full joint processing algorithm ( fjp ) . in fjp algorithm , all the rrhs are active , and only the transmission power of rrhs is minimized by solving a penalized weighted sum rate maximization beamforming design problem based on the lyapunov optimization . the fjp algorithm can achieve the highest cooperative beamforming gain with all the rrhs active , and the results from the fjp algorithm can serve as a delay performance lower bound of the proposed algorithms . we first consider a c - ran with @xmath262 rrhs and @xmath263 ues . to indicate the heterogeneous power consumption of different rrhs and the fronthaul links , we set the static power consumption as @xmath264 w and set the drain efficiency of each rrh as @xmath265 . in fig . [ fig_delay ] and [ fig_power ] , we evaluate the average queue delay and the average network power consumption against the control parameter @xmath66 when the mean arrival rate @xmath266 is 1.25 mbits / slot and 1.75 mbits / slot , respectively . for all the algorithms , a larger traffic mean arrival rate always results in a longer average delay and higher network power consumptions . this can be explained by the fact that more power is needed in order to timely transmit larger amount of traffic arrivals . under a given mean arrival rate , the average network power consumption is a monotonically decreasing function in @xmath66 . the rate of power decreasing starts to diminish with excessive increase of @xmath66 . on the other hand , a larger @xmath66 can adversely affect the delay performance because the average queue length grows linearly with @xmath66 . this is due to the fact that the system with a larger @xmath66 will emphasize less on delay performance but more on the network power consumption performance . therefore the parameter @xmath66 features the tradeoff between power consumption and delay performance . from both figures , it is observed that gsb - based wmmse algorithm always outperforms rip - based wmmse algorithm , but only by a small margin . both algorithms achieve significant power saving compared to the fjp algorithm . ] ] fig . [ fig_rrhs ] shows the average number of sleeping rrhs for the two proposed algorithms against different values of the control parameter @xmath66 . the average number of sleeping rrhs demonstrates a similar trend as the average network power consumption performance . to reduce network power consumption for a large @xmath66 , it is necessary to turn off as many rrhs and the corresponding fronthaul links as possible , at the cost of longer average queue length . ] the power - delay tradeoffs of various algorithms are compared in fig . [ fig_tradeoff ] . the different tradeoff points are obtained by varying the control parameter @xmath66 . the average network power consumption is decreasing and convex in the average delay . when the delay is small , slightly increasing the delay requirement can achieve a significant amount of power saving . when the delay is excessively large , increasing the delay further results in only a very small power saving . [ fig_tradeoff ] shows that gsb- and rip - based algorithm provide significantly better power - delay tradeoff than the fjp algorithm . in addition , when the average delay decreases , the power consumption gap between the proposed algorithms and the fjp algorithm becomes smaller . this means that more rrhs need to be turned on with a stricter delay requirement . in the extreme case when all the rrhs are active , all the algorithms will perform fjp algorithm , yielding the same network power consumption . in the small delay regime , the proposed algorithms provide a flexible and efficient means to balance the power - delay tradeoff , given that a slight loosening of the delay requirement contributes to significant energy savings . to achieve a certain power - delay tradeoff , all we need to do is choosing an appropriate control parameter @xmath66 . ] to compare the average network power consumption with different static power consumption , we set the static power consumption to be the same for all rrhs . [ fig_staticpower ] shows the average power consumption as a function of the static power consumption when the control parameter is @xmath267 . as expected , the average network power consumptions for all algorithms are increasing functions of the static power consumption . in addition , the proposed algorithms significantly outperform the fjp algorithm , especially in the high static power consumption regime . when @xmath268 w , the gsb - based algorithm achieves a power saving of 24.6% and 43.5% , at @xmath269 mbits / slot and 1.25 mbits / slot , respectively . on the other hand , the performance gap between the proposed algorithms and the fjp algorithm decreases as @xmath270 decreases . when @xmath271 , all algorithms require almost the same network power consumption . when the static power consumption is smaller , more rrhs will be activated to achieve a higher beamforming gain . [ fig_rate_delay ] compares the average delay with different mean traffic arrival rates for the two proposed algorithms and the backward greedy selection ( bgs)-based algorithm , which iteratively switches off one rrh at each step , while re - optimizing the fjp beamformer for the remaining active rrh set . the bgs - based algorithm has been shown to often yield optimal or near - optimal solutions for rrh activation@xcite . we set the control parameter as @xmath272 . it is observed that , when the mean traffic arrival rate is relatively low , the average delays of the two proposed algorithms are slightly larger than that of the bgs - based algorithm . furthermore , the average delays for all algorithms increase sharply and tend to infinity as the mean arrival rates are beyond certain thresholds ( i.e. , the stability regions ) . specifically , the bgs - based algorithm achieves the biggest stability region , followed by the gsb - based algorithm and the rip - based algorithm . therefore , congestion controls should be adopted to guarantee the queue stability when the network is with traffic load exceeding current stability regions . [ fig_convergence ] shows the average number of outer bcd iterations and the average number of inner iterations required by the gsb - based wmmse algorithm and rip - based wmmse algorithm , respectively , with respect to the network scale factor @xmath273 . here the network scale factor indicates that the considered c - ran is with @xmath274 rrhs and @xmath275 ues distributed uniformly in the squared region @xmath276 \times [ -500\sqrt \theta , 500\sqrt \theta]$ ] meters . we can observe that both gsb - based algorithm and rip - based algorithm can converge fairly fast under different network scale , thus both algorithms are highly scalable to large - scale c - rans . it is observed that more iterations are required by the rip - based solutions compared to that of gsb - based solutions . meanwhile , as can be seen in fig . 5 , the gsb - based solution slightly outperforms the rip - based solution . thus , the gsb - based solution is more preferable in practice . we have developed a joint rrh activation and beamforming algorithm for a downlink slotted c - ran , by considering random traffic arrivals and time - varying channel fadings . the algorithm can achieve flexible and efficient tradeoff between network power consumption and delay by adjusting a single parameter . the stochastic optimization problem of joint rrh activation and beamforming has been transformed into a penalized weighted sum rate maximization problem based on the lyapunov optimization technique . both gsb and rip approaches have been used to reformulate the penalized weighted sum rate maximization problem . the corresponding algorithms for both approaches have been proposed , and they were derived based on the equivalence between the weighted sum rate maximization problem and the wmmse problem . the algorithms are guaranteed to converge to a stationary solution . the solutions do not require any prior - knowledge of stochastic traffic arrivals and channel statistics , and can be implemented in a parallel manner . finally , the efficiency and the efficacy of the proposed algorithms have been confirmed by the numerical simulations . for future works , it would be interesting to consider the queue - aware energy - efficient joint rrh activation and beamforming algorithms for c - rans with imperfect channel state information ( csi ) or capacity - limited fronthaul links . by leveraging on the fact that @xmath277 + c)^2 } \le { a^2 } + { b^2 } + { c^2 } - 2a(b - c),\forall a , b , c \ge 0 $ ] and squaring eq . ( [ qdynamic ] ) , we have @xmath278 let @xmath280}$ ] . finally , the upper bound in ( [ driftpenaltybound ] ) can be obtained by adding @xmath281 $ ] to both sides of ( [ driftbound ] ) . suppose that the traffic arrivals with mean arrival rate @xmath282 is strictly interior to the stability region @xmath56 ( definition 2 ) such that @xmath283 . since channel conditions are i.i.d . over slots , according to theorem 4.5 in@xcite , there exists a stationary randomized control policy that is independent of @xmath63 and yields @xmath284 = \mathbb{e}[{\mu _ i}(t ) ] \ge { \lambda _ i } + \epsilon , \forall i , \\ \mathbb{e}[p(\mathcal { a}(t),{\bf{w}}(t))|{\bf{q}}(t ) ] = \mathbb{e}[p(\mathcal { a}(t),{\bf{w}}(t ) ) ] = \bar p(\epsilon ) . \\ \end{array}\label{ranpolicy}\ ] ] as the stationary randomized control policy is simply a particular control policy , it certainly satisfies ( [ driftpenaltybound ] ) in lemma 1 . in addition , since ( [ p1 ] ) is obtained by minimizing the right - hand - side ( r.h.s . ) of ( [ driftpenaltybound ] ) among all feasible policies ( including the stationary randomized control policy ) , by combining ( [ ranpolicy ] ) with ( [ driftpenaltybound ] ) , we have @xmath80 \le b + \\v\mathbb{e } [ p(\mathcal { a}(t ) , { \bf{w}}(t))|{\bf{q}}(t ) ] + \sum\limits_{i \in \mathcal { i } } { q_i}(t)\mathbb{e}[{a_i}(t ) - \mu _ i(t)|{\bf{q}}(t)]\\ \le b + v\bar p(\epsilon ) - \varepsilon \sum\limits_{i \in \mathcal { i } } { { q_i}(t)}. \end{array}\ ] ] using the results in the proof of theorem 1 , it follows that @xmath285 , which proves that solving ( [ p1 ] ) optimally stabiles all the queues . let @xmath286 and define the index set @xmath287 , then we have @xmath288 $ ] . define the set @xmath289 as a partion of @xmath290 , then we have @xmath291 $ ] . furthermore , define the support of beamformer @xmath106 as @xmath292 , then the power consumption model can be rewritten as @xmath293 where @xmath294 is an indicator function with value 1 if the event @xmath295 is true and 0 otherwise . to simplify notation , let @xmath296 , and let @xmath297 . it can been seen that @xmath298 is combinatorial in @xmath108 and is non - convex . we will obtain a convex relaxation of the combinatorial objective function . we first construct the tightest positively homogeneous lower bound of @xmath101 , which is given by@xcite @xmath299 the last equality in ( [ lowerbound ] ) is obtained by solving @xmath300 . however , @xmath301 is still non - convex in @xmath108 . we next calculate the convex envelope of @xmath301 . define diagonal matrices @xmath302 , @xmath303 with the @xmath23-th diagonal block being @xmath304 and @xmath305 , respectively . the fenchel conjugate of @xmath301 is given by @xmath306 where @xmath307 is the @xmath308-dimensional vector formed with the entries of @xmath309 indexed by @xmath310 ( similarly for @xmath106 ) , and @xmath311 is the @xmath312-dimensional matrix formed with both the rows and the columns of @xmath313 indexed by @xmath314 ( similarly for @xmath315 ) . consider the norm @xmath103 whose dual norm is defined as @xmath316 in eq . ( [ fen ] ) . according to proposition 2 in@xcite , @xmath103 is the convex envelope of @xmath101 . therefore , the tightest convex positively homogenous lower bound of @xmath317 has the following inequality : @xmath318 which is obtained using the norm properties . actually , the above inequality always holds with equality . specifically , let @xmath319 such that @xmath320 , then we have @xmath321 which is obtained using the definition of convex envelope . therefore , we finally have @xmath322 , i.e. , ( [ l1_l2 ] ) . by comparing problem ( [ admmform ] ) and problem ( [ varsplit ] ) , when @xmath323 and @xmath324 , we can observe that @xmath325 since @xmath326 and @xmath327 are invertible , and both @xmath328 and @xmath140 are convex sets , then according to theorem 3 , we can conclude that every limit point @xmath168 and @xmath169 generated by algorithm 2 is an optimal solution of problem ( [ problemw ] ) . p. rost , c. j. bernardos , a. d. domenico , m. d. girolamo , m. lalam , a. maeder , d. sabella , and d. w@xmath329bben , `` cloud technologies for flexible 5 g radio access networks , '' _ ieee commun . mag . 68 - 76 , may 2014 . c. wang , f. haider , x. gao , x. you , y. yang , d. yuan , h. m. aggoune , h. haas , s. fletcher , and e. hepsaydir , `` cellular architecture and key technologies for 5 g wireless communication networks , '' _ ieee commun 2 , pp . 122 - 130 , feb . 2014 . y. cheng , m. pesavento , and a. philipp , `` joint network optimization and downlink beamforming for comp transmissions using mixed integer conic programming , '' _ ieee trans . signal process . 3972 - 3987 , aug . s. luo , r. zhang , and t. j. lim , `` downlink and uplink energy minimization through user association and beamforming in cloud ran , '' _ ieee trans . wireless commun . 1 , pp . 494 - 508 , jan . s. tombaz , p. monti , k. wang , a. v@xmath330stberg , m. forzati , and j. zander , `` impact of backhauling power consumption on the deployment of heterogeneous mobile networks , '' in _ proc . ieee global commun ( globecom ) _ , houston , tx , usa , dec . 2011 , pp . 1 - 5 . y. cui , v. k. n. lau , r. wang , h. huang , and s. zhang , `` a survey on delay - aware resource control for wireless systems - large deviation theory , stochastic lyapunov drift and distributed stochastic learning , '' _ ieee trans . inf . theory _ , vol . 1677 - 1701 , mar . 2012 . h. ju , b. liang , j. li , and x. yang , `` dynamic power allocation for throughput utility maximization in interference - limited networks , '' _ ieee wireless commun . 22 - 25 , jan . 2013 . s. park , o. simeone , o. sahin , and s. shamai , `` joint precoding and multivariate backhaul compression for the downlink of cloud radio access networks , '' _ ieee trans . signal process . 5646 - 5658 , nov . g. auer , v. giannini , c. desset , i. godor , p. skillermark , m. olsson , m. a. imran , d. sabella , m. j. gonzalez , o. blume , and a. fehske , `` how much energy is needed to run a wireless network ? '' _ ieee wireless commun . _ , vol . 18 , pp . 40 - 49 , oct . 2011 . y. a. sambo , m. z. shakir , k. a. qaraqe , e. serpedin , m. a. imran , `` expanding cellular coverage via cell - edge deployment in heterogeneous networks : spectral efficiency and backhaul power consumption perspectives , '' _ ieee commun . mag . 52 , no . 6 , pp . 140 - 149 , jun . 2014 . j. li , j. wu , m. peng , w. wang , and v. k. n. lau , `` queue - aware joint remote radio head activation and beamforming for green cloud radio access networks , '' in _ proc . ieee global commun . ( globecom ) _ , san diego , ca , usa , dec . 2015 . s. s. christensen , r. agarwal , e. carvalho , and j. m. cioffi , `` weighted sum - rate maximization using weighted mmse for mimo - bc beamforming design , '' _ ieee trans wireless commun . _ , vol . 7 , no . 12 , pp . 4792 - 4799 , dec . 2008 . q. shi , m. razaviyayn , z. luo , and c. he , `` an iteratively weighted mmse approach to distributed sum - utility maximization for a mimo interfering broadcast channel , '' _ ieee trans . signal process . 4331 - 4340 , sep . 2011 . s. boyd , n. parikh , e. chu , b. peleato , and j. eckstein , `` distributed optimization and statistical learning via the alternating direction method of multipliers , '' _ found . trends mach . 1 , pp . 1 - 122 , 2011 . f. bach , r. jenatton , j. mairal , and g. obozinski , `` convex optimization with sparsity - inducing norms , '' in _ optimization for machine learning _ , s. sra , s. nowozin , and s. j. wright , eds . cambridge , ma : mit press , 2011 .

the abundances of alkali elements ( li , na , and k ) were determined from the li i 6708 , na i 5682/5688 , and k i 7699 lines by taking into account the non - lte effect for 24 sharp - lined a - type stars ( km s , 7000 k 10000 k , many showing am peculiarities to different degrees ) , based on high - dispersion and high - s / n spectral data secured at boao ( korea ) and oao ( japan ) . we found a significant trend that(na ) tightly scales with(fe ) irrespective of , which means that na becomes enriched similarly to fe in accordance with the degree of am peculiarity . regarding lithium ,(li ) mostly ranges between and ( i.e. , almost the same as or slightly less than the solar system abundance of 3.3 ) with a weak decreasing tendency with a lowering of at k , though several stars exceptionally show distinctly larger depletion . the abundances of potassium also revealed an apparent-dependence in the sense that(k ) in late - a stars tends to be mildly subsolar ( possibly with a weak anti - correlation with(fe ) ) systematically decreasing from ( k ) to ( k ) , while those for early - a stars remain near - solar around 5.05.2 . these observational facts may serve as important constraints for any theory aiming to explain chemical anomalies of a - type stars .

1111.1603

while elemental abundances in the photosphere of a - type stars , which are of particular interest because of the chemical anomalies of various types , have been extensively studied by a number of investigators so far , those of the alkali elements ( li , na , and k ) are barely known even nowadays , despite of their importance in stellar spectroscopy . this is presumably due to the fact that these elements ( with one valence electron weakly bound ) are characterized by fairly low ionization potential ( @xmath12 45 ev ) and most of them are ionized in the atmospheric condition of early - type stars , leaving only a tiny fraction as neutral . since detecting spectral lines of the dominant ionized species is almost hopeless ( because of the closed - shell configuration requiring vary large excitation energy ) , one is obliged to measure the very weak lines of neutral species ( li i , na i , k i ) . thus , the necessity of using sufficiently high - s / n spectra must have hampered past researchers from investigating the abundances of these elements . still , sodium has been comparatively well studied ( e.g. , lane & lester 1987 ; varenne & monier 1999 ; gebran et al . 2008 , 2010 ; gebran & monier 2008 ; fossati et al . 2007 , 2008 ) mainly by using subordinate lines ( na i 5682/5688 or 6154/6160 ) of moderate strengths ( i.e. , neither too strong nor too weak ) , which are less affected by the non - lte effect . however , na abundance determinations in a - type stars conducted so far are restricted to sharp - lined late - a stars ( @xmath13 k ) as well as f - type stars , and thus na compositions of early - a through late - b stars ( @xmath14 k ) remain essentially unexplored , except for our previous non - lte investigations for the very bright stars of sirius ( takeda & takada - hidai 1994 ) and vega ( takeda 2008 ; hereinafter refereed to as paper i ) , and the recent lte studies of fossati et al . ( 2009 ) for two sharp - lined late - b stars ( 21 peg and @xmath15 cet ) . recently , takeda et al . ( 2009 ; hereinafter referred to as paper ii ) carried out an extensive non - lte analysis of na i resonance d lines at 5890/5896 @xmath16 for a large ( 122 ) sample of a - type stars ( 7000 k @xmath17 k ) including rapid as well as slow rotators , with an aim to establish their na abundance behaviors in general . unfortunately , however , since the d lines are so strong and the resulting abundances are quite sensitive to the microturbulent velocity @xmath18 , it turned out that ambiguities in the choice of @xmath18 ( which is likely to be depth - dependent and difficult to assign a value relevant for the high - forming strong d lines ) prevented from precise sodium abundance determinations . almost the same situation holds for lithium : determinations of li abundances in a - type stars have been essentially restricted to late - a stars with @xmath13 k ( e.g. , burkhart & coupry 1991a , 1991b , 1995 , 2000 ; burkhart et al . 2005 ; north et al . 2005 ) , while early - a stars ( @xmath14 k ) remain practically untouched . admittedly , this is a very difficult and challenging task , as the li i 6708 line considerably weakens as @xmath4 becomes higher . the work of paper i may be counted as an exceptional case , where the li abundance of vega could somehow be evaluated from the very weak li i 6708 line ( @xmath19 m@xmath16 ) based on a spectrum of ultra - high s / n ratio . coupry and burkhart ( 1992 ) reported the @xmath20(li i 6708 ) of @xmath21 peg ( a well - known very sharp - lined early - a star ) to be 1.3 m@xmath16 , which was used in paper i ( cf . appendix a therein ) to argue that the li abundance of @xmath21 peg is almost consistent with the solar - system composition . however , as will be shown in this paper ( cf . subsection 4.2 ) , we consider that their measurement is an erroneous overestimation and li is definitely underabundant in @xmath21 peg . this example well illustrates the enormous difficulty of li abundance derivation in hot a - type stars . investigations of potassium abundances are even more lacking . as for late - a stars ( @xmath13 k ) , it is ( to our knowledge ) only the study of fossati et al . ( 2007 ) that reported the k abundances of late - a and am stars in the praesepe cluster , though their neglect of the non - lte effect ( which may be significant for such a resonance line as k i 7699 ) prevented from obtaining reliable results ( cf . subsection 4.4 therein ) . meanwhile , almost nothing has been reported so far regarding the potassium abundances in early - a stars , except for paper i where vega s k abundance could be successfully established . it is thus evident that our current knowledge on the behaviors of alkali - element abundances in a - type stars is considerably insufficient . given this situation , we decided to carry out a systematic spectroscopic study aiming at establishing the photospheric abundances of na , li , and k for an homogeneous sample of a - dwarfs ( 7000 k @xmath17 k ) based on the spectral data of high quality . since measurements of very weak lines are concerned in the present case ( especially for early - a stars ) , we had to confine ourselves only to sharp - lined stars ( @xmath0 km s@xmath1 ) . this inevitably leads to the inclusion of a number of am stars ( metallic - lined a stars ) , which are characterized by deficiencies in comparatively light elements ( c , n , o , ca , sc ) as well as excesses of heavier elements ( such as fe - group or neutron - capture elements ) , since am peculiarities are preferably observed in slow rotators ( such as frequently seen in binaries ) . accordingly , the purpose of this paper is to clarify and discuss the behaviors of na , li , and k abundances in slowly - rotating a - type stars based on the results of our analysis , while paying attention to the possible connection ( if any exists ) between these alkali elements and the appearance / degree of am anomalies . we selected 21 sharp - lined stars uma ; magnetic ap star ) and hd 40932 ( @xmath22 ori ; double - lined binary ) because of their apparently unusual spectra , despite that these two stars have @xmath23 values less than 50 km s@xmath1 . ] satisfying @xmath0 km s@xmath1 from 122 a - type stars , studied in paper ii , where na i 5890/5896 d lines were analyzed based on the spectra ( @xmath24 and s / n ratios of a few hundreds ) obtained by using boes ( bohyunsan observatory echelle spectrograph ) attached to the 1.8 m reflector at bohyunsan optical astronomy observatory ( boao ) . we call these 21 stars as `` boao sample . '' see section 2 of paper ii ( and electronic table e therein ) for more details about these observational data . besides , in order to augment the spectral data for early - a stars for which very high - quality is required to measure very weak lines of neutral alkalis , we also used the spectra of 7 bright sharp - lined stars ( @xmath15 dra , @xmath25 lyr , @xmath25 cma , @xmath26 gem , 15 vul , 68 tau , and @xmath21 peg ) , which were observed with the hides spectrograph ( high dispersion echelle spectrograph ; izumiura 1999 ) attached to the 1.88 m reflector at okayama astrophysical observatory ( oao ) . while the data of @xmath25 lyr were taken from the digital atlas ( based on the observations carried out on 2006 may 14 ) published by takeda , kawanomoto , and ohishi ( 2007 ) as in paper i , those for the remaining 6 stars were secured by the observations on 2008 october 4 ( 15 vul , 68 tau , @xmath21 peg ) , october 7 ( @xmath25 cma , @xmath26 gem ) , and october 8 ( @xmath15 dra ) . these 2008 oao data , which cover the wavelength range of 41007800 @xmath16 with three mosaicked ccds of 2k@xmath274k pixels , have high spectral resolution ( @xmath28 ) and very high s / n ratio ( typically @xmath29 , or even higher for the very bright @xmath25 cma and @xmath26 gem ) . we call these 7 stars as `` oao sample . '' note that the following 4 stars are common to both of the boao and oao samples : hd 172167=@xmath25 lyr , hd 48915=@xmath25 cma , hd 47105=@xmath26 gem , and hd 27962=68 tau . accordingly , the actual ( net ) number of the targets is 24 ( @xmath30 ) . the list of these program stars is presented in table 1 . regarding the effective temperature ( @xmath4 ) and the surface gravity ( @xmath31 ) of each program star , we simply adopted the values used in paper ii for 21 boao sample stars ( including 4 oao samples in common ) . as for the remaining 3 oao stars , those for 15 vul and @xmath21 peg were determined photometrically from strmgren s @xmath32 , @xmath33 , and @xmath34 colors along with the @xmath35 index by using napiwotzki et al.s ( 1993 ) calibration ( as in paper ii ; cf . subsection 3.1 therein ) , while the values derived by adelman ( 1996 ) were adopted for @xmath15 dra ( since color data were not available for this star ) . the final values of @xmath4 and @xmath31 are summarized in table 1 . as can be seen in figure 1 , where the targets are plotted on the theoretical hr diagram , the mass values of the program stars range between @xmath121.5 @xmath36 and @xmath373 @xmath36 . the model atmospheres for each of the stars were constructed by two - dimensionally interpolating kurucz s ( 1993 ) atlas9 model grid in terms of @xmath4 and @xmath31 , where we exclusively applied the solar - metallicity models as in takeda et al . ( 2008a ) as well as paper ii . ( 50mm,50mm)figure1.eps since we need to clarify the characteristics ( e.g. , parameters such as rotation or atmospheric turbulence , degree of abundance peculiarities ) of our sample stars , a synthetic spectrum fitting analysis based on takeda s ( 1995 ) best - fit solution search algorithm was applied to the 61466163 @xmath16 region , in order to establish the projected rotational velocity ( @xmath23 ) , microturbulent velocity dispersion ( @xmath18 ) , and the abundances of iron and oxygen [ @xmath3(fe ) and @xmath3(o ) ; both are good indicators of am anomaly ; cf . takeda and sadakane ( 1997 ) ] ( along with that of na as a by - product ) , similarly to what was done in takeda et al . ( 2008a ; cf . section iv therein ) and paper ii ( cf . subsection 3.2 therein ) . actually , we varied seven free parameters @xmath18 , @xmath23 , @xmath3(o ) , @xmath3(na ) , @xmath3(si ) , @xmath3(ca ) and @xmath3(fe ) to accomplish the best fit , where the non - lte effect was taken into account for o ( as in takeda et al . 2010 ) as well as for na ( as in paper ii ) . the adopted atomic parameters of important o , na , and fe lines relevant to this region are given in table 2 . the instrumental broadening corresponding to the spectral resolving power was also taken into account , since sharp - lined stars are concerned here . since we could not find a reasonable @xmath18 solution for the following five boao cases ( i.e. , stabilized at an apparently unreasonable value , or any convergence could not be accomplished ) , we assumed an appropriate @xmath18 value as fixed and repeated the iterations until convergence . @xmath38 km s@xmath1 ( adelman 1996 ; monier 2005 ) for hd 95418 , @xmath39 km s@xmath1 ( caliskan & adelman 1997 ) for hd 43378 , @xmath40 km s@xmath1 ( sadakane 2006 ) for hd 218396 , @xmath41 km s@xmath1 ( from the analytical formula derived by takeda et al . 2008 ) for hd 204188 , and @xmath42 km s@xmath1 ( @xmath18 solution from the oao spectrum ) for hd 48915 . the finally adopted results of @xmath18 , @xmath23 , @xmath43(fe ) @xmath43(o ) , are presented in table 1 as well as in the electronic table e ( where @xmath43(na ) is also given ) . how the theoretical and observed spectra match each other with the converged solutions of these parameters is shown in figure 2 . ( 50mm,200mm)figure2.eps in figure 3 , the resulting @xmath3(o ) and @xmath3(fe ) values are plotted against @xmath23 and @xmath4 , and their mutual correlation is also displayed . we can see from these figures that the underabundance of o and the overabundance of fe ( both are the characteristics of am peculiarities ) are surely anti - correlated ( figure 3c ) and their weak @xmath23-dependence is recognized ( figures 3a , b ) even in such a small range of @xmath23 ( i.e. , larger anomaly for slower rotators ; cf . takeda & sadakane 1997 ) . we also notice that two stars ( hd 218396 and hd 172167=vega ) are apparently metal - deficient ( [ fe / h]@xmath44 ) , showing the abundance charateristics of vega - like stars ( cf . sadakane 2006 ) , which are presumably related to @xmath45 boo stars ( see , e.g. , paunzen 2004 and the references therein ) . it is interesting that these two appear to follow the o vs. fe anti - correlation trend exhibited by am stars , despite that interaction with interstellar clouds ( cf . kamp & paunzen 2002 ; paunzen et al . 2002 ) is becoming a promising mechanism for explaining the @xmath45 boo phenomenon ( instead of the diffusion process being considered as the standard theory for am stars ) . regarding the microturbulence , the @xmath18 vs. @xmath4 relation depicted in figure 3e implies that the @xmath18 values resulting from the 61466163 @xmath16 fitting are consistent with the analytical formula proposed by takeda et al . ( 2008a ) ( which was invoked in paper ii ) . ( 80mm,200mm)figure3.eps in the next section ( section 3 ) we will derive not only the abundances ( of na , li , and k ) but also the relevant abundance errors in response to the possible ambiguities in the adopted atmospheric parameters , where we estimate @xmath46 k , @xmath47 dex , and @xmath48% as the typical uncertainties in the absolute values of @xmath4 , @xmath31 , and @xmath18 , respectively . while the internal errors of our color - based @xmath4 and @xmath31 are @xmath49 k ( 2.5% ) and @xmath50 dex according to napiwotzki et al . cf . section 5 therein ) , we assumed somewhat larger ambiguities than these in view of the comparison with various published values ( cf . section iv - c in takeda et al . the uncertainty in @xmath18 was estimated from the dispersion 30% ) is regarded as a typical ambiguity in @xmath18 , which can also be recognized by comparing various @xmath18 values published so far ( see , e.g. , figure 1 in coupry & burkhart 1992 ; figure 2 in gebran & monier 2007 ) . yet , apart from this intrinsic dispersion , our @xmath18 values tend to be somewhat larger than the previously reported results ( usually based on @xmath20 values of fe lines of various strengths ) . for example , our figure 3f suggests that the maximum @xmath18 attained at mid - a type is @xmath12 45 km s@xmath1 , while it is @xmath5 km s@xmath1 in figure 2 of coupry and burkhart ( 1992 ) ; also , our @xmath18 results are systematically higher by @xmath12 1 km s@xmath1 with those of landstreet et al . ( 2009 ) , where 5 stars ( 68 tau , @xmath26 gem , @xmath25 cma , 15 vul , and @xmath21 peg ) are in common with our sample . we suspect that this may be due to a depth - dependence of @xmath18 decreasing with height ( cf . section 5 in paper ii ) . that is , our @xmath18-determination ( based on 61466163 @xmath16 fitting ) tends to reflect the physical condition in deeper layers ( larger @xmath18 ) because of being mainly controled by the deep - forming o i 61558 triplet lines of high excitation , while the conventional method using @xmath20s of high - forming strong lines may yield comparatively lower @xmath18 . ] in the @xmath18 vs. @xmath4 relation ( figure 3f ) . since resonance lines of alkali elements with low - ionization potentials are concerned in the present case , it is comparatively easy to understand ( at least qualitatively ) how the abundance is affected by changing each parameter . that is , since almost all atoms are in the ground state of the first - ionized ( closed - shell ) stage , and only a small fraction of them remain neutral , the number population of the ground level of the neutral atoms ( @xmath51 , which is proportional to the line opacity @xmath52 ) is expressed as @xmath53 according to saha s equation ( @xmath54 is the abundance , @xmath55 , @xmath56 is the ionization potential ( @xmath12 45 ev ) , and @xmath57 is the electron density ) . then , the dependence of the line - opacity ( @xmath52 ) upon @xmath58 ) and @xmath59 may be written as @xmath60 where we put @xmath61 and used the approximation @xmath62 ( @xmath63 1/32/3 depending on @xmath4 ; see , e.g. , gray 2005 ) . consequently , the abundance ( @xmath54 ) resulting from a given equivalent width is strongly dependent upon @xmath64 as @xmath65 because of the exponential factor ( i.e. , @xmath54 increases as @xmath4 becomes higher ) . the @xmath31-dependence is somewhat more complicated because the effect of the continuum opacity ( @xmath66 ) has to be taken into consideration . for late a - type stars of comparatively lower @xmath4 ( @xmath12 70008000 k ) , h@xmath67 opacity is still important and @xmath66 has almost the same @xmath59-dependence as @xmath52 ; thus the abundances turn out rather independent on @xmath59 because of the cancellation in the @xmath68 ratio determining the line strength ( cf . takeda et al . meanwhile , for early a stars of higher @xmath4 where paschen continuum opacity of neutral hydrogen is dominant , @xmath59-dependence in @xmath52 directly reflects the line strength and the abundance scales as @xmath69 ( i.e. , @xmath54 decreases as @xmath31 becomes higher ) . finally , regarding the effect of @xmath18 ( @xmath54 tends to decrease with an increase in @xmath18 ) , the abundance results are practically inert to any change in @xmath18 , since weak lines in the linear part of the curve of growth are concerned in most of the present cases ( though only the na abundances derived from na i 5688 line for late a stars with @xmath70 8000 k exceptionally show some appreciable @xmath18-dependence ; cf . figure 5f ) . we will show in section 3 that these expected trends can be actually confirmed by actual calculations ( cf . figures 5 , 8 , and 10 ) . the abundances of na ( @xmath43(na ) ) were already obtained from the fitting analysis in the 61466163 @xmath16 region ( including na i 6154/6160 lines ) as described in subsection 2.3 , which are given in electronic table e. however , while the na i 6154/6160 lines are quite suitable for na abundance determination for late - a stars , they become considerably weak as @xmath4 becomes higher and the accuracy of @xmath43(na ) may be comparatively low at early - a stars , especially for the non - metal - rich case ( actually , @xmath43(na ) of hd 172167=vega in the boao sample could not be determined ; cf . note 1 in electronic table e ) . we therefore decided to determine the na abundances from the spectra in the 56805690 @xmath16 region including the na i 5682/5688 lines ( stronger than na i 6154/6160 while not too strong to be seriously sensitive to @xmath18 ) . as done for the 61466163 @xmath16 region , we applied the fitting procedure to the 56805690 @xmath16 portion of the spectrum , where the telluric lines had been removed by dividing by the spectrum of a rapid rotator ( cf . paper i , paper ii ) . we varied five abundance parameters of @xmath3(na ) , @xmath3(si ) , @xmath3(sc ) , @xmath3(fe ) and @xmath3(ni ) to obtain the best fit , while taking into account the non - lte effect for na as in paper ii ( cf . table 2 for the atomic parameters of na i 5682/5688 ) . how the theoretical and observed spectra match each other with the converged solutions of these parameters is shown in figure 4 . the resulting sodium abundances , @xmath71(na ) , are given in table 1 . besides , as analysis - related quantities , the equivalent widths ( @xmath20 ) inversely computed from the abundance , the corresponding non - lte corrections ( @xmath72 ) , and the abundance variations ( @xmath73 ) in response to each of the perturbations in @xmath4 ( @xmath46 k ) , @xmath31 ( @xmath47 dex ) , and @xmath18 ( @xmath74 ) , which were computed for the na i 5688 line following the procedure described in paper ii ( cf . subsections 4.3 and 4.4 therein ) , are plotted against @xmath4 in figure 5 . the ( @xmath20 , @xmath72 , @xmath73 ) results for both na i 5682 and 5688 lines are also presented along with @xmath71(na ) in electronic table e. ( 50mm,200mm)figure4.eps ( 60mm,160mm)figure5.eps the comparison of the two sodium abundances [ @xmath43(na ) and @xmath71(na ) ] derived from different regions / lines is displayed in figure 6 . we can see from this figure that the agreement is good especially for late - a stars ( red circles ) . the star showing the largest departure ( @xmath12 0.4 dex ) is vega ( oao sample ) where @xmath43(na ) = 5.63 and @xmath43(na ) = 6.05 were derived , which is due to the low accuracy of @xmath43 caused by the extreme weakness of na i 6154/6160 lines in this star ( cf . figure 3 in paper i).(na ) = 5.63 ) based on the 61466163 @xmath16 region fitting is considerably different from the @xmath20-based value of 6.26 ( @xmath75 ; cf . table 1 in paper i ) , despite that both are based essentially on the same lines and the same non - lte corrections . we understand that the reason for this marked discrepancy is the erroneous underestimation of the 61466163 @xmath16 fitting - based result ( 5.63 ) in this study . namely , while the synthetic fitting method applied to a spectrum extending over some range ( e.g. , @xmath76 @xmath16 ) is generally useful for simultaneous determinations of several elements at a time , it is not suitable for such a case where _ extremely _ weak line features are under question . that is , even a slight fluctuation of the continuum level can be a crucial source of error , because an appropriate adjustment of the local continuum is impossible in such a global spectrum synthesis . accordingly , when one has to derive an abundance from an extremely weak line , the classical line - by - line @xmath20 analysis ( adopted in paper i ) should be preferable . this is actually the reason why we adopted the classical @xmath20-based abundance determination for li and k , because the lines of these elements become very weak in early - a stars . ] we hereafter adopt @xmath71(na ) as the representative sodium abundances which we will discuss in subsection 4.1 . ( 50mm,50mm)figure6.eps we derived the abundance of li from the @xmath20 of the merged li i doublet feature at @xmath77 @xmath16 . since very weak lines are generally concerned , @xmath20 measurements were carried out by the gaussian fitting while comparing the spectrum on the computer screen with the theoretical spectrum synthesized with the known @xmath23 as well as appropriately varied @xmath3(li ) , by which we tried to minimize the possibility of erroneous measurement . in case we could not detect any suitable feature at the expected position , we gave up determination of @xmath20 . actually , we were unable to specify @xmath20(6708 ) for about @xmath78% of the targets ( 13 out of the 21 boao sample and 2 out of the 7 oao sample ) . in such cases , only the upper limit values were derived by the formula @xmath79 ( cf . takeda & kawanomoto 2005 ) . here , fwhm was estimated from the root - sum - square of ( i ) the rotational broadening @xmath80 ( cf . footnote 12 of takeda et al . 2008b ) , ( ii ) the approximate separation of the doublet @xmath81 , and ( iii ) the instrumental broadening @xmath82 ( @xmath83 : spectrum resolving power ) , as fwhm @xmath84 , and @xmath85 was assumed to be 2 according to our experience . the actual measurements of @xmath20(6708 ) were carried out on the `` smoothed '' spectrum after applying the 9-pixel boxcar function to the original spectrum in order to reduce the effect of noise . value ( 0.7 m@xmath16 ) of @xmath25 lyr at that time was actually measured in the `` original '' spectrum . in this study , however , the measurement was done on the `` smoothed '' spectrum , which is the reason for the appreciably larger value ( 1.2 m@xmath16 ) derived for this star in spite of the use of the same spectrum data of takeda et al . this fact implies the difficulty in evaluating @xmath20 of extremely weak line feature , which is subject to considerable uncertainties . ] figure 7 displays how the gaussian fitting for measuring @xmath20 was done in the relevant li i 6708 region of the spectra . ( 80mm,240mm)figure7.eps then , based on such evaluated @xmath20 ( or @xmath86 ) and the model atmosphere along with the non - lte departure coefficients ( appropriately interpolated from the precomputed ones on a model grid ; cf . takeda and kawanomoto 2005 for more details of the non - lte calculations ) , @xmath3(li ) ( non - lte abundance ) or its upper limit was determined for each star ( see table 2 for the adopted atomic parameters ) as given in table 1 . besides , in the similar manner as in subsection 3.1 , we also computed the corresponding non - lte corrections ( @xmath72 ) , and the abundance variations ( @xmath73 ) in response to perturbations of the atmospheric parameters . these [ @xmath20 , @xmath72 , @xmath3(li ) , @xmath73 ] results are plotted against @xmath4 in figure 8 , which are also presented in electronic table e. ( 50mm,160mm)figure8.eps regarding the abundance of k , we exclusively invoked the k i line at 7698.97 @xmath16 , since the alternative line of the doublet at 7664.91 @xmath16 tends to be severely influenced by strong telluric lines . its @xmath20 was measured on the spectrum ( in which telluric lines in the neighborhood had already been removed by dividing by that of a rapid rotator as in paper i ) in the same way as the case of the li i 6708 line ( cf . subsection 3.2 ) . if the line failed to be detected ( 4 stars among the 21 boao sample ) , we estimated the upper limit ( @xmath86 ) following equation ( 1 ) , where fwhm @xmath88 in this case . figure 9 displays how these measurements were done in the relevant k i 7699 region of the spectra . ( 80mm,240mm)figure9.eps based on such evaluated @xmath20 along with the non - lte departure coefficients ( appropriately interpolated from the precomputed ones on a model grid ; cf . takeda et al . 1996 for more details of the non - lte calculations ) , the non - lte k abundance ( @xmath3(k ) ) or its upper limit was determined for each star ( with the atomic parameter of the line in table 2 ) , as given in table 1 . these results of @xmath20 and @xmath3(k ) , along with the corresponding non - lte corrections ( @xmath72 ) and the abundance variations ( @xmath73 ) in response to perturbations of the atmospheric parameters , are plotted against @xmath4 in figure 10 , and are also presented in electronic table e. ( 50mm,160mm)figure10.eps we are now ready to answer the questions which motivated this study ( cf . section 1 ) : `` how do the abundances of na , li , and k behave in sharp - lined a - type stars ? how are they related with the am peculiarities frequently seen in slow rotators of this @xmath4 range ? '' in figure 11 are plotted the resulting [ @xmath3(na ) , @xmath3(li ) , and @xmath3(k ) ] against @xmath3(fe ) and @xmath23 ( determined in subsection 2.3 ) , both of which are the key quantities closely connected with the am phenomenon . based on this figure as well as figures 5c , 8c , and 10c ( showing the @xmath4-dependence ) , the abundance characteristics of these elements are discussed in the following subsections , where we also compare the observed trend with the theoretical prediction from the atomic diffusion theory by richer , michaud , and turcotte ( 2000 ) as well as talon , richard , and michaud ( 2006 ) . ( 80mm,200mm)figure11.eps regarding na , we notice in figure 11a a significant trend of remarkably tight correlation between @xmath3(na ) and @xmath3(fe ) . actually , a slight @xmath23-dependence in @xmath3(na ) ( similar to that seen for @xmath3(fe ) in figure 3b ) is observed in figure 11a@xmath89 . since @xmath3(fe ) is known to be one of the am indicators in the sense that larger overabundance of fe is accompanied by stronger am stars , this observational fact means that na behaves in accordance with fe ( i.e. , the excess of na is a measure of am anomaly just like that of fe ) . while na abundances of am(+fm ) and normal a(+f ) stars ( though mostly restricted to those with @xmath13 k ) in the field as well as in open clusters have been reported by several recent studies based on na i 5683/5688 or 6154/6161 lines being less sensitive to non - lte effects ( e.g. , lane & lester 1987 ; varenne & monier 1999 ; gebran et al . 2008 , 2010 ; gebran & monier 2008 ; fossati et al . 2007 , 2008 ) , nobody seems to have explicitly mentioned this trend . however , an inspection of the results of these papers suggests tendencies of ( i ) moderately supersolar [ na / h ] and ( ii ) [ na / h](am ) @xmath90 [ na / h](normal a ) . therefore , the existence of this positive correlation between [ na / h ] and [ fe / h ] appears almost certain . from a theoretical point of view , however , such a conformance behavior between na and fe has never been predicted by any element segregation simulation designed for explaining am peculiarities such as that of richer et al . ( 2000 ) or talon et al . ( 2006 ) , which suggests that [ na / h ] remains almost solar or even slightly subsolar in spite of an appreciable overabundance of [ fe / h ] ( cf . figure 14 of richer et al . or figure 16 of talon et al . ) . therefore , the current diffusion theory ( considered to be most promising for understanding the am phenomena ) is evidently imperfect at least in terms of na , which remains to be improved . here , the na fe correlation revealed in this study would serve as an important observational constraint . besides , it is interesting that the apparently metal - deficient stars with @xmath3(fe ) @xmath91 6.97.0 ( @xmath25 lyr and hd 218396=hr 8799 ) almost follow the same @xmath3(na)@xmath3(fe ) relation defined by am and normal stars with @xmath92 [ fe / h ] @xmath93 . since these two stars are known to be vega - like stars ( which may be related to so - called @xmath45 boo stars ; see , e.g. , kamp & paunzen 2002 , paunzen et al . 2002 ) , there might be some connection between am and @xmath45 boo anomalies . ( see also the last paragraph of subsection 2.3 . ) according to figures 11b and 11b@xmath89 , we do not see any clear @xmath3(fe)- as well as @xmath23-dependence in @xmath3(li ) which has values around or slightly lower than the solar - system abundance of @xmath94 . this may suggest that am and normal a - type stars do not show any manifest difference in terms of their surface li abundances . we recognize , however , a tendency of @xmath4-dependence for late a - type stars in the sense that @xmath3(li ) gradually decreases from @xmath94 ( @xmath95 k ) to @xmath96 ( @xmath97 k ) while such a systematic trend is not seen for early a - type stars ( @xmath14 k ) where @xmath3(li ) tends to be around the solar system abundance of @xmath94 ( figure 8c ) . this @xmath4-dependent tendency of @xmath3(li ) in late - a stars is also observed in figure 2 of north et al . ( 2005 ) , and is more or less consistent with the theoretical prediction by richer et al . ( 2000 ; cf . figure 14 therein ) . it should be noted , however , that several stars apparently deviate from this general trend and show considerable li - deficiencies ( i.e. , only the upper limit was derived since the li line could not be detected ) . here , especially important are the following 4 stars : hd 204188 ( 7622 k , @xmath98 ) , hd 141795 ( 8367 k , @xmath99 ) , @xmath21 peg ( 9453 k , @xmath100 ) , and @xmath25 cma ( 9938 k , @xmath101 ) . what makes such an exceptional depletion of li ? all these are classified as am stars ( @xmath21 peg is also an am star according to adelman 1988 , though the conventional type of a1 iv is given in table 1 ) , though am alone can not be the condition to trigger such a marked underabundance of li , because there are a number of other am stars that do show normal behavior of li . actually , most of the present sample of low @xmath23 stars more or less show some degree of chemical peculiarities . in this connection , burkhart and coupry ( 1991a ) stated ( based on their study of late a - type stars ) that `` occurrence of li - deficient stars is not exceptional without any clearcut division between normal and am stars . '' yet , as far as early a - type stars are concerned , the fact that li tends to remain almost unaffected in near - normal stars such as @xmath26 gem ( believed to have almost the solar abundances ; e.g. , adelman & philip 1996 , lehmann et al . 2002 ) and @xmath15 dra ( which were previously considered as normal even though later found to have only weak am anomaly ; cf . sadakane & okyudo 1990 ) might suggest the existence of some connection between the conspicuous li deficit of @xmath25 cma or @xmath21 peg and their am phenomenon . here , we have to remark that the conclusion of paper i ( the photospheric li abundance of @xmath21 peg almost coincides with that of the solar - system composition ; cf . appendix a therein ) , which was based on the @xmath20(li 6708 ) value of 1.3 m@xmath16 taken from table 1 of coupry and burkhart ( 1992 ) , was not correct . as can be seen from figure 7 , the li i 6708 line is not detectable even on the spectrum of s / n @xmath29 ; and we derived the upper limit of its equivalent width as @xmath20(6708 ) @xmath102 m@xmath16 ( cf . the electronic table e ) following equation ( 1 ) . we suspect , therefore , that coupry and burkhart ( 1992 ) misidentified / mismeasured this line , since the quality of their spectrum with typical s / n of @xmath103 ( cf . section 1 therein ) is evidently insufficient for measuring such an extremely weak line . our revised conclusion for @xmath21 peg is @xmath3(li ) @xmath104 , which means that its photospheric li abundance is at least by @xmath105 dex deficient compared to the solar - system value . regarding @xmath3(k ) , we can observe ( from figure 10c ) a trend in terms of @xmath4 just similar to the case of @xmath3(li ) : @xmath3(k ) of late a - type stars appears to systematically decrease from @xmath8 ( @xmath9 k ) to @xmath10 ( @xmath11 k ) , while such a systematic trend is not seen for early a - type stars ( @xmath14 k ) where @xmath3(k ) is almost solar around @xmath8 ( an exception is @xmath21 peg which shows a clearly subsolar abundance of @xmath106 ) . meanwhile , it also appears that @xmath3(k ) s of late a - type stars are weakly dependent upon @xmath3(fe ) as well as @xmath23 in the sense that k gets more deficient as the am anomaly becomes stronger ( cf . the red circles in figures 11c and 11c@xmath89 ) . accordingly , the deficiency of k ( particularly seen in late - a stars of @xmath13 k ) may be controlled by two factors ( @xmath4 and the degree of am peculiarity ) . richer et al s ( 2000 ) simulation based on the theory of atomic diffusion ( along with radiative acceleration ) predicts a small overabundance of k at higher @xmath4 ( @xmath107 k ) region , whereas a marginal underabundance is expected at the lower @xmath4 side ( @xmath108 k ) ( cf . figure 14 therein ) . such a predicted trend of positive @xmath109(k)/@xmath110 gradient is in qualitative agreement with what we found from figure 10c . as mentioned in section 1 , k - abundance determinations of a - type stars have so far been tried only by fossati et al . ( 2007 ) in their lte analysis of the k i 7699 line for late - a stars in the praesepe cluster . according to their results for 5 stars , we see @xmath111[k / h]@xmath112 @xmath113 ( cf . figure 10 and table 3 therein ) . considering that the typical ( negative ) non - lte corrections amount to @xmath12 0.30.4 dex at the relevant @xmath4 range ( cf . figure 10b ) , we may state that their results would turn out to be mildly subsolar when the non - lte effect is taken into consideration , and become more consistent with our conclusion . motivated by the fact that our current understanding on the abundances of alkali elements in a - type stars is considerably insufficient , we conducted a comprehensive non - lte analysis to establish the photospheric abundances of na , li , and k for 24 selected sharp - lined a - type stars ( @xmath0 km s@xmath1 , 7000 k @xmath2 10000 k ) , many of which are am stars showing different degree of chemical peculiarity . for this purpose , we primarily invoked the spectra of moderately high - dispersion ( @xmath24 ) and high s / n ratio ( typically a few hundreds ) obtained with boes at boao / korea , though spectra of much higher quality ( @xmath28 and s / n @xmath114 ) secured with hides at oao / japan were additionally used for 7 apparently bright stars ( mostly of early a - type ) . we first carried out spectrum fitting analyses applied to the 61466163 @xmath16 region and derived @xmath18 , @xmath23 , @xmath3(o ) , and @xmath3(fe ) , from which the degree of am characteristics for each star could be confirmed . the abundances of sodium , which were determined by the spectrum fitting method applied to the region comprising na i 5682/5688 lines , revealed a significant trend that @xmath3(na ) tightly scales with @xmath3(fe ) , which means that na becomes enriched similarly to fe in accordance with the degree of am phenomenon . this result should be regarded as important because it seriously contradicts the prediction from the atomic diffusion theory . regarding lithium , @xmath3(li ) derived from the weak li i line at 6708 @xmath16 showed a tendency of @xmath4-dependence for late a - type stars in the sense that @xmath3(li ) gradually decreases from @xmath94 ( @xmath95 k ) to @xmath96 ( @xmath97 k ) , while @xmath3(li ) for early a - type stars ( @xmath14 k ) tends to be around the solar system abundance of @xmath94 . however , several stars apparently deviating from this general trend and showing marked underabundances do exist , though what triggers such a li deficiency is not clear . the abundances of potassium derived from the k i 7699 line revealed a tendency somewhat similar to the case of lithium : an apparent @xmath4-dependence ( and possibly with a weak anti - correlation with am peculiarity ) exists for late - a stars where @xmath3(k ) tend to be mildly subsolar systematically decreasing from @xmath8 ( @xmath9 k ) to @xmath10 ( @xmath11 k ) , whereas those for most early - a stars remain near - solar around @xmath12 5.05.2 . when these observational facts are compared with the theoretical prediction of richer et al . ( 2000 ) or talon et al . ( 2006 ) based on their atomic diffusion calculations , a serious discrepancy is found for na ( almost no anomaly or slightly subsolar tendency is predicted , while an enrichment of na just like fe is suggested from our analysis ) , though the simulated results appear to qualitatively reproduce the general trends observed for li and k. further improvement and development on the theoretical side would be desirably awaited to explain the results concluded in this study , which may serve as important constraints for any theory aiming to account for the origin of chemical anomalies in am stars . this research has made use of the simbad database , operated by cds , strasbourg , france . i. han acknowledges the financial support for this study by kicos through korea ukraine joint research grant ( grant 07 - 179 ) . lee acknowledges the astrophysical research center for the structure and evolution of the cosmos ( arsec , sejong university ) of the korea science and engineering foundation ( kosef ) through the science research center ( src ) program . adelman , s. j. 1996 , mnras , 280 , 130 adelman , s. j. 1988 , mnras , 230 , 671 adelman , s. , & philip , a. g. d. 1996 , mnras , 282 , 1181 anders , e. , & grevesse , n. 1989 , geochim . acta , 53 , 197 burkhart , c. , & coupry , m. f. 1991a , a&a , 249 , 205 burkhart , c. , & coupry , m. f. 1991b , mem . soc . astron . it . , 62 , 91 burkhart , c. , & coupry , m. f. 1995 , mem . astron . it . , 66 , 357 burkhart , c. , & coupry , m. f. 2000 , a&a , 354 , 216 burkhart , c. , coupry , m. f. , faraggiana , r. , & gerbaldi , m. 2005 , a&a , 429 , 1043 caliskan , h. , & adelman , s. j. 1997 , mnras , 288 , 501 coupry , m. f. , & burkhart , c. 1992 , a&as , 95 , 41 esa 1997 , the hipparcos and tycho catalogues , esa sp-1200 , available from nasa - adc or cds in a machine - readable form ( file name : hip_main.dat ) flower , p. j. 1996 , apj , 469 , 355 fossati , l. , bagnulo , s. , landstreet , j. , wade , g. , kochukhov , o. , monier , r. , weiss , w. , & gebran , m. 2008 , a&a , 483 , 891 fossati , l. , bagnulo , s. , monier , r. , khan , s. a. , kochukhov , o. , landstreet , j. , wade , g. , & weiss , w. 2007 , a&a , 476 , 911 fossati , l. , ryabchikova , t. , bagnulo , s. , alecian , e. , grunhut , j. , kochukhov , o. , & wade , g. 2009 , a&a , 503 , 945 gebran , m. , & monier , r. 2007 , in convection in astrophysics , ed . f. kupka , i. w. roxburgh , & k. l. chan , proc . 239 ( cambridge : cambridge university press ) , 160 gebran , m. , & monier , r. 2008 , a&a , 483 , 567 gebran , m. , monier , r. , & richard , o. 2008 , a&a , 479 , 189 gebran , m. , vick , m. , monier , r. , & fossati , l. 2010 , a&a , 523 , 71 girardi , l. , bressan , a. , bertelli , g. , & chiosi , c. 2000 , a&as , 141 , 371 gray , d. f. 2005 , the observation and analysis of stellar photospheres , 3rd ed . ( cambridge : cambridge university press ) , ch.9 grevesse , n. , & noels , a. 1993 , in origin and evolution of the elements , ed . n. prantzos , e. vangioni - flam , & m. cass ( cambridge : cambridge university press ) , 15 hoffleit , d. 1982 , the bright star catalogue , 4th ed . ( new haven , connecticut : yale university observatory ) izumiura , h. 1999 , in proc . 4th east asian meeting on astronomy , observational astrophysics in asia and its future ed . p. s. chen ( kunming : yunnan observatory ) , 77 kamp , i. , & paunzen , e. 2002 , mnras , 335 , l45 kurucz , r. l. 1993 , kurucz cd - rom , no . 13 ( harvard - smithsonian center for astrophysics ) kurucz , r. l. , & bell , b. 1995 , kurucz cd - rom , no . 23 ( harvard - smithsonian center for astrophysics ) landstreet , j. d. , kupka , f. , ford , h. a. , officer , t. , sigut , t. a. a. , silaj , j. , strasser , s. , & townshend , a. 2009 , a&a , 503 , 973 lane , m. c. , & lester , j. b. 1987 , apjs , 65 , 137 lehmann , h. , andrievsky , s. m. , egorova , i. , hildebrandt , g. , korotin , s. a. , panov , k. p. , scholz , g. , & schnberner , d. 2002 , a&a , 383 , 558 monier , r. 2005 , a&a , 442 , 563 napiwotzki , r. , schnberner , d. , & wenske , v. 1993 , a&a , 268 , 653 north , p. , betrix , f. , & besson , c. 2005 , in element stratification in stars , 40 years of atomic diffusion , eds . g. alecian , o. richard , and s. vauclair , eas publication series , vol . 17 , p.333 ( eas ) paunzen , e. 2004 , in the a - star puzzle , proc . 224 , eds . j. zverko , j. @xmath115i@xmath116@xmath117ovsk@xmath118 , s.j . adelman , w.w . weiss ( cambridge : cambridge university press ) , p.443 paunzen , e. , iliev , i. kh . , kamp , i. , & barzova , i. s. 2002 , mnras , 336 , 1030 richer , j. , michaud , g. , & turcotte , s. 2000 , apj , 529 , 338 sadakane , k. 2006 , pasj , 58 , 1023 sadakane , k. , & okyudo , m. 1990 , pasj , 42 , 317 smith , v. v. , lambert , d. l. , & nissen , p. e. 1998 , apj , 506 , 405 takeda , y. 1995 , pasj , 47 , 287 takeda , y. 2008 , mnras , 388 , 913 ( paper i ) takeda , y. , han , i. , kang , d .- i . , lee , b .- , & kim , k .- 2008a , jkas , 41 , 83 takeda , y. , kambe , e. , sadakane , k. , & masuda , s. 2010 . pasj , 62 , 1239 takeda , y. , kang , d .- , han , i. , lee , b .- , & kim , k .- m . 2009 , pasj , 61 , 1165 ( paper ii ) takeda , y. , kato , k. , watanabe , y. , & sadakane , k. 1996 , pasj , 48 , 511 takeda , y. , & kawanomoto , s. 2005 , pasj , 57 , 45 takeda , y. , kawanomoto , s. , & ohishi , n. 2007 , pasj , 59 , 245 takeda , y. , ohkubo , m. , & sadakane , k. 2002 , pasj , 54 , 451 takeda , y. , & sadakane , k. 1997 , pasj , 49 , 367 takeda , y. , sato , b. , & murata , d. 2008b , pasj , 60 , 781 takeda , y. , & takada - hidai , m. 1994 , pasj , 46 , 395 talon , s. , richard , o. , & michaud , g. 2006 , apj , 645 , 634 van leeuwen , f. 2007 , hipparcos , the new reduction of the raw data ( berlin : springer ) varenne , o. , & monier , r. 1999 , a&a , 351 , 247 rrccccrrrrccclhd # & hr # & name & sp.type & @xmath4 & @xmath31 & @xmath18 & @xmath23 & @xmath119 & @xmath120 & @xmath121 & @xmath122 & @xmath123 & remark + + 195725 & 7850 & @xmath124 cep & a7iii & 7816 & 3.74 & 4.45 & 52.0 & 7.60 & 8.47 & 3.27 & 6.35 & 4.81 & + 95418 & 4295 & @xmath35 uma & a1v & 9489 & 3.85 & @xmath1252.50 & 46.8 & 7.73 & 8.45 & ( @xmath1264.34 ) & 6.45 & ( @xmath1265.47 ) & + 27819 & 1380 & @xmath127tau & a7v & 8047 & 3.95 & 3.93 & 46.5 & 7.45 & 8.78 & 3.35 & 6.23 & 4.88 & h + 43378 & 2238 & 2 lyn & a2vs & 9210 & 4.09 & @xmath1252.10 & 45.4 & 7.34 & 8.70 & ( @xmath1264.07 ) & 6.13 & ( @xmath1265.18 ) & + 218396 & 8799 & @xmath128 & a5v & 7091 & 4.06 & @xmath1253.00 & 41.4 & 6.90 & 8.88 & 2.82 & 5.83 & 5.01 & vega - like star + 84107 & 3861 & 15 leo & a2iv & 8665 & 4.31 & 3.57 & 38.1 & 7.50 & 8.62 & ( @xmath1263.44 ) & 6.21 & 4.93 & + 33204 & 1670 & @xmath128 & a5 m & 7530 & 4.06 & 4.14 & 36.4 & 7.76 & 8.46 & 3.11 & 6.47 & 4.72 & h + 204188 & 8210 & @xmath128 & a8 m & 7622 & 4.21 & @xmath1253.90 & 35.3 & 7.50 & 8.74 & ( @xmath1262.77 ) & 6.24 & 4.75 & + 141795 & 5892 & @xmath54 ser & a2vm & 8367 & 4.24 & 4.04 & 34.4 & 7.72 & 8.17 & ( @xmath1263.16 ) & 6.46 & 5.02 & + 173648 & 7056 & @xmath129lyr & a4 m & 8004 & 3.90 & 4.73 & 32.4 & 7.74 & 8.31 & 3.21 & 6.47 & 4.82 & + 27628 & 1368 & 60 tau & a3 m & 7218 & 4.05 & 4.74 & 32.3 & 7.74 & 8.43 & 3.03 & 6.46 & 4.75 & h + 28546 & 1428 & 81 tau & a5 m & 7640 & 4.17 & 3.63 & 28.1 & 7.74 & 8.58 & 3.11 & 6.47 & 4.86 & h + 172167 & 7001 & @xmath25 lyr & a0va & 9435 & 3.99 & 1.85 & 21.9 & 6.96 & 8.63 & ( @xmath1263.84 ) & 6.00 & ( @xmath1265.38 ) & vega - like star + 60179 & 2891 & @xmath25 gem & a1v & 9122 & 3.88 & 2.33 & 19.7 & 7.52 & 8.42 & ( @xmath1263.71 ) & 6.15 & 5.10 & + 95608 & 4300 & 60 leo & a1 m & 8972 & 4.20 & 2.93 & 17.2 & 7.84 & 8.19 & ( @xmath1263.43 ) & 6.46 & 5.32 & + 48915 & 2491 & @xmath25 cma & a1vm & 9938 & 4.31 & @xmath1252.53 & 16.5 & 7.89 & 8.41 & ( @xmath1264.01 ) & 6.52 & ( @xmath1265.19 ) & + 27749 & 1376 & 63 tau & a1 m & 7448 & 4.21 & 3.90 & 12.7 & 7.97 & 8.25 & 3.09 & 6.56 & 4.50 & h + 33254 & 1672 & 16 ori & a2 m & 7747 & 4.14 & 3.07 & 12.2 & 7.99 & 8.18 & ( @xmath1262.53 ) & 6.60 & 4.66 & h + 72037 & 3354 & 2 uma & a2 m & 7918 & 4.16 & 2.48 & 11.6 & 7.89 & 7.98 & ( @xmath1262.55 ) & 6.53 & 4.66 & + 47105 & 2421 & @xmath26 gem & a0iv & 9115 & 3.49 & 1.81 & 10.9 & 7.55 & 8.76 & ( @xmath1263.50 ) & 6.35 & 4.90 & + 27962 & 1389 & 68 tau & a2iv & 8923 & 3.94 & 3.44 & 10.4 & 7.73 & 8.54 & ( @xmath1263.23 ) & 6.46 & 5.22 & h + + 182564 & 7371 & @xmath15 dra & a2iiis & 9125 & 3.80 & 3.43 & 27.3 & 7.83 & 8.48 & 3.47 & 6.65 & 4.98 & + 172167 & 7001 & @xmath25 lyr & a0va & 9435 & 3.99 & 1.90 & 21.9 & 6.97 & 8.63 & 3.21 & 6.05 & 5.13 & vega - like star + 48915 & 2491 & @xmath25 cma & a1vm & 9938 & 4.31 & 2.53 & 16.6 & 7.89 & 8.42 & ( @xmath1263.18 ) & 6.59 & 5.15 & + 47105 & 2421 & @xmath26 gem & a0iv & 9115 & 3.49 & 2.49 & 11.1 & 7.48 & 8.74 & 3.17 & 6.33 & 5.19 & + 189849 & 7653 & 15 vul & a4iii & 7870 & 3.62 & 5.16 & 10.7 & 7.34 & 8.65 & 3.03 & 6.31 & 4.60 & + 27962 & 1389 & 68 tau & a2iv & 8923 & 3.94 & 3.99 & 10.5 & 7.69 & 8.54 & 3.17 & 6.46 & 5.10 & h + 214994 & 8641 & @xmath21 peg & a1iv & 9453 & 3.64 & 3.07 & 6.0 & 7.63 & 8.52 & ( @xmath1262.94 ) & 6.46 & 4.74 & + note . + in columns 1 through 6 are given the hd number , hr number , star name ( with constellation ) , spectral type ( taken from hoffleit 1982 ) , effective temperature ( in k ) , and logarithmic surface gravity ( in cm s@xmath130 ) . columns 7 through 10 show the results derived from 61466163 @xmath16 region fitting : the microturbulent velocity ( in km s@xmath1 ) , projected rotational velocity ( in km s@xmath1 ) , ( lte ) abundance of fe , and ( non - lte ) abundance of o. the non - lte abundances of li , na , and k finally resulting from this investigation are presented in columns 1113 . all abundance results are expressed in the usual normalization of @xmath3(h ) = 12.00 . in each of the boao ans oao samples , the objects are arranged in the descending order of @xmath131 . in the remark in column 14 , hyades cluster stars are denoted with `` h '' . + @xmath125regarding these @xmath18 values , literature values were adopted unlike the others , because the @xmath18-determination from the 61466163 @xmath16 region fitting did not work successfully . see subsection 2.3 for more details . cclrldesig . & species & @xmath132 & @xmath133 ( ev ) & @xmath134 + li 6708 & li i & 6707.756 & 0.00 & @xmath135 + & li i & 6707.768 & 0.00 & @xmath136 + & li i & 6707.907 & 0.00 & @xmath137 + & li i & 6707.908 & 0.00 & @xmath138 + & li i & 6707.919 & 0.00 & @xmath139 + & li i & 6707.920 & 0.00 & @xmath137 + o 6156 & o i & 6155.961 & 10.74 & @xmath140 + & o i & 6155.971 & 10.74 & @xmath141 + & o i & 6155.989 & 10.74 & @xmath138 + o 6157 & o i & 6156.737 & 10.74 & @xmath142 + & o i & 6156.755 & 10.74 & @xmath137 + & o i & 6156.778 & 10.74 & @xmath143 + o 6158 & o i & 6158.149 & 10.74 & @xmath144 + & o i & 6158.172 & 10.74 & @xmath145 + & o i & 6158.187 & 10.74 & @xmath146 + na 5682 & na i & 5682.633 & 2.10 & @xmath147 + na 5688 & na i & @xmath148 & 2.10 & @xmath149 + na 6154 & na i & 6154.226 & 2.10 & @xmath150 + na 6160 & na i & 6160.747 & 2.10 & @xmath151 + si 5684 & si i & 5684.484 & 4.95 & @xmath152 + si 5688 & si ii & 5688.817 & 14.19 & + 0.40 + si 6155 & si i & 6155.134 & 5.62 & @xmath153 + s 7696 & s i & 7696.758 & 7.87 & @xmath154 + k 7699 & k i & 7698.974 & 0.00 & @xmath155 + ca 6162 & ca i & 6162.173 & 1.90 & + 0.10 + sc 5684 & sc ii & 5684.202 & 1.51 & @xmath141 + fe 5686 & fe i & 5686.524 & 4.55 & @xmath156 + fe 6147 & fe ii & 6147.741 & 3.89 & @xmath157 + fe 6149 & fe ii & 6149.258 & 3.89 & @xmath157 + fe 6150 & fe ii & 6150.098 & 11.45 & @xmath158 + fe 6157 & fe i & 6157.725 & 4.08 & @xmath151 + fe 6705 & fe i & 6705.101 & 4.61 & @xmath159 + fe 6708 & fe ii & 6708.885 & 10.91 & @xmath160 + ni 5682 & ni i & 5682.198 & 4.11 & @xmath161 + note . + all data are were taken from kurucz & bell s ( 1995 ) compilation , except for those of li i which were adopted from smith et al . ( 1998 ) as in takeda and kawanomoto ( 2005 ) . + @xmath125since this line comprises two close components of stronger @xmath162 ( at 5688.205 @xmath16 ) and weaker @xmath163 ( at 5688.194 @xmath16 ) , we adopted @xmath164 as the sum of these two . regarding the li i lines , we considered only the lines of @xmath165li , since we neglected @xmath166li in our analysis .

we calculate the relative versions of embedded contact homology , contact homology and cylindrical contact homology of the sutured solid torus , where consists of parallel longitudinal sutures .

0911.0055

the embedded contact homology ( ech ) of a closed , oriented @xmath3-manifold with a contact form was introduced by hutchings in @xcite and is a variant of the symplectic field theory @xcite of eliashberg , givental and hofer . it is defined in terms of a contact form but is an invariant of the underlying @xmath3-manifold . this invariance has been established by taubes in @xcite via the identification with seiberg - witten floer ( co-)homology as defined by kronheimer and mrowka @xcite and in particular implies the weinstein conjecture in dimension three . ech is also conjecturally isomorphic to ozsvth szab heegaard floer homology defined in @xcite . we would like to mention that kutluhan , lee and taubes , and independently colin , ghiggini and honda have recently announced two different proofs of the isomorphism between hutchings s embedded contact homology and heegaard floer homology . a natural condition to impose on a compact , oriented contact @xmath4-manifold @xmath5 with boundary is to require that @xmath6 be _ convex _ , i.e. , there is a contact vector field @xmath7 transverse to @xmath8 . to a transverse contact vector field @xmath7 we can associate the _ dividing set _ @xmath9 , namely the set of points @xmath10 such that @xmath11 . by the contact condition , @xmath12 is a @xmath13-dimensional contact submanifold of @xmath14 ; the isotopy class of @xmath15 is independent of the choice of @xmath7 . we will denote by @xmath16 the contact manifold @xmath5 with convex boundary and dividing set @xmath17 with respect to some transverse contact vector field @xmath7 . note that the actual boundary condition we need is slightly different and is called a _ sutured boundary condition_. ( in the early 1980 s , gabai developed the theory of sutured manifolds @xcite , which became a powerful tool in studying @xmath3-manifolds with boundary . ) for the moment we write @xmath18 to indicate either the convex boundary condition or the sutured boundary condition . it turns out that there is a way to generalize embedded contact homology to sutured @xmath3-manifolds . this is possible by imposing a certain convexity condition on the contact form . this construction is completely described in the paper of colin , ghiggini , honda and hutchings @xcite . heegaard floer homology also admits a sutured version , namely the sutured floer homology ( sfh ) of juhsz @xcite , which is an invariant of sutured manifolds . finally , kronheimer and mrowka in @xcite introduced the sutured version of seiberg - witten floer homology . extending the conjectured equivalence of heegaard floer homology and embedded contact homology , the following conjecture was formulated in @xcite : [ conj_echsfh ] @xmath19 , where @xmath20 denotes the relative spin@xmath21-structure determined by @xmath22 and @xmath23 . in this paper , we construct sutured contact solid torus with @xmath2 parallel longitudinal sutures , where @xmath24 , using the gluing method of colin , ghiggini , honda and hutchings @xcite and calculate the sutured embedded contact homology of it . we apply the gluing method in such a way that the constructed sutured solid torus is equipped with a nondegenerate contact form satisfying the property that all closed embedded reeb orbits are noncontractible , define the same homology class and have the same symplectic action . it turns out that for the constructed sutured manifolds the sutured version of embedded contact homology coincides with sutured floer homology . the corresponding calculation in sutured floer homology has been done by juhsz in @xcite . so far , this is the first series of nontrivial examples where these two theories provide the same answer . [ suturedech ] let @xmath25 be a sutured manifold , where @xmath1 is a set of @xmath2 parallel longitudinal curves and @xmath24 . then there is a contact form @xmath26 which makes @xmath27 a sutured contact manifold and @xmath28 where @xmath29 and @xmath30 is an adapted almost complex structure . hence , @xmath31 there is a floer - type invariant of a closed , oriented contact odd - dimensional manifold , called contact homology . contact homology was introduced by eliashberg and hofer and is a special case of the symplectic field theory . in @xcite , colin , ghiggini , honda and hutchings generalized contact homology to sutured manifolds . for the sutured contact manifold from theorem [ suturedech ] , we calculated the sutured versions of cylindrical contact homology and contact homology . [ suturedcylconthom ] let @xmath32 be a sutured contact manifold from theorem [ suturedech ] . then @xmath33 is defined , is independent of the contact form @xmath26 for the given contact structure @xmath22 and almost complex structure @xmath30 , @xmath34 and hence @xmath35 where @xmath36 is the homological grading and @xmath37 is the conley - zehnder grading . [ suturedconthom ] let @xmath32 be a sutured contact manifold from theorem [ suturedech ] . then @xmath38 is defined , is independent of the contact form @xmath26 for the given contact structure @xmath22 and almost complex structure @xmath30 , @xmath39 and hence @xmath40 where @xmath36 is the homological grading , @xmath37 is the conley - zehnder grading and @xmath41 denotes the coefficient of @xmath42 in the generating function @xmath43 . this paper is organized as follows : in section [ background ] we review definitions of embedded contact homology , cylindrical contact homology and contact homology for sutured contact manifolds ; section [ construction ] describes the construction of sutured contact solid torus with @xmath2 longitudinal sutures , where @xmath24 ; finally in section [ calculations ] we calculate the relative versions of embedded contact homology , cylindrical contact homology and contact homology of the solid torus constructed in section [ construction ] . the author is deeply grateful to ko honda for his guidance , help and support far beyond the call of duty . he also thanks oliver fabert , jian he and michael hutchings for helpful suggestions and interest in his work . in addition , the author is grateful to the referee of an earlier version of this paper for many valuable comments and suggestions . the goal of this section is to review definitions of embedded contact homology , cylindrical contact homology and contact homology for contact sutured manifolds . this section is essentially a summary of @xcite . in this section we repeat some definitions from @xcite . a _ liouville manifold _ ( often also called a liouville domain ) is a pair @xmath44 consisting of a compact , oriented @xmath2-dimensional manifold @xmath45 with boundary and a @xmath46-form @xmath47 on @xmath45 , where @xmath48 is a positive symplectic form on @xmath45 and the _ liouville vector field _ @xmath49 given by @xmath50 is positively transverse to @xmath51 . it follows that the @xmath46-form @xmath52 ( this notation means @xmath47 pulled back to @xmath51 ) is a positive contact form with kernel @xmath53 . a compact oriented @xmath54-dimensional manifold @xmath55 with boundary and corners is a _ sutured manifold _ if it comes with an oriented , not necessarily connected submanifold @xmath56 of dimension @xmath57 ( called the _ suture _ ) , together with a neighborhood @xmath58\times[-1 , 1]\times \gamma$ ] of @xmath59 in @xmath55 , with coordinates @xmath60\times [ -1,1]$ ] , such that the following holds : * @xmath61\times \gamma)\cup([-1,0]\times \{-1\ } \times \gamma)\cup ( [ -1,0]\times \{1\}\times \gamma)$ ] ; * @xmath62 , where the orientation of @xmath8 agrees with that of @xmath63 and is opposite that of @xmath64 , and the orientation of @xmath1 agrees with the boundary orientation of @xmath63 ; * the corners of @xmath55 are precisely @xmath65 . the submanifold @xmath66 is called the _ horizontal boundary _ and @xmath67\times \gamma$ ] the _ vertical boundary _ of @xmath55 . let @xmath68 be a sutured manifold . if @xmath22 is a contact structure on @xmath55 ( this means that @xmath55 is now @xmath69-dimensional ) , we say that @xmath70 is a _ sutured contact manifold _ if @xmath22 is the kernel of a positive contact 1-form @xmath26 such that : * @xmath71 and @xmath72 are liouville manifolds ; * @xmath73 inside @xmath74 , where @xmath75 and @xmath47 is independent of @xmath76 and does not have a @xmath77-term ; * @xmath78 , where @xmath79 is a liouville vector field for @xmath80 . such a contact form is said to be _ adapted _ to @xmath68 . let @xmath81 be a sutured contact manifold with an adapted contact form @xmath26 . the form @xmath26 is then given by @xmath82 on the neighborhoods @xmath83\times r_{+}(\gamma)$ ] and @xmath84\times r_{-}(\gamma)$ ] of @xmath85 and @xmath86 , where @xmath87\cup [ 1-\varepsilon,1]$ ] extends the @xmath76-coordinate on @xmath88 . on @xmath88 , @xmath89 , @xmath90 , and @xmath91 is a liouville vector field @xmath49 for @xmath47 . following the procedure explained in @xcite we can complete " @xmath92 to a noncompact contact manifold @xmath93 . we first extend @xmath26 to @xmath94 and @xmath95\times r_{-}(\gamma)$ ] by taking @xmath96 as appropriate . the boundary of this new manifold is @xmath97 . notice that since @xmath98 , the form @xmath99\times \{t\}\times \gamma}$ ] is the symplectization of @xmath100 in the positive @xmath101-direction . we glue @xmath102 with the form @xmath103 , where @xmath104 is the pullback of @xmath47 to @xmath105 . let @xmath106 be the noncompact extension of @xmath55 described above and @xmath107 be the extension of @xmath26 to @xmath106 . for convenience , we extend the coordinates @xmath108 so far defined only on the ends of @xmath106 to functions on @xmath106 so that @xmath109 $ ] and @xmath110 $ ] . we then say that @xmath111 corresponds the top ( t ) , @xmath112 corresponds to the bottom ( b ) , and @xmath113 corresponds to the side ( s ) . let @xmath114 be the extension / completion of @xmath115 , obtained by extending to ( s ) . let @xmath116 be a sutured contact manifold with an adapted contact form @xmath26 and @xmath93 be its completion . the _ reeb vector field _ @xmath117 that is associated to a contact form @xmath107 is characterized by @xmath118 a _ reeb orbit _ is a closed orbit of the reeb flow , i.e. , a smooth map @xmath119 for some @xmath120 such that @xmath121 . [ reeborbincompl ] every periodic orbit of @xmath117 lies in @xmath55 . hence , the set of periodic reeb orbits of @xmath122 coincides with the set of periodic reeb orbits of @xmath123 . consider reeb orbit @xmath124 passing through a point @xmath125 . the linearization of the reeb flow on the contact planes along @xmath124 determines a linearized return map @xmath126 . this linear map is symplectic and it does not depend on @xmath127 ( up to conjugation ) . the reeb orbit @xmath124 is _ nondegenerate _ if @xmath128 . note that nondegeneracy can always be achieved by a small perturbation , i.e. , for any contact structure @xmath22 on @xmath55 , there exists a contact form @xmath26 for @xmath22 such that all closed orbits of @xmath123 are nondegenerate . for simplicity , we assume that all reeb orbits of @xmath123 , including multiply covered ones , are nondegenerate . a reeb orbit @xmath124 is called _ elliptic _ or _ positive _ ( respectively _ negative _ ) _ hyperbolic _ if the eigenvalues of @xmath129 are on the unit circle or the positive ( resp . negative ) real line respectively . if @xmath101 is a trivialization of @xmath22 over @xmath124 , we can then define the conley - zehnder index . in @xmath3-dimensional situation this is given explicitly as follows : if @xmath124 is elliptic , then there is an irrational number @xmath130 such that @xmath129 is conjugate in @xmath131 to a rotation by angle @xmath132 , and @xmath133 where @xmath132 is the total rotation angle with respect to @xmath101 of the linearized flow around the orbit . if @xmath124 is positive ( respectively negative ) hyperbolic , then there is an even ( respectively odd ) integer @xmath134 such that the linearized flow around the orbit rotates the eigenspaces of @xmath129 by angle @xmath135 with respect to @xmath101 , and @xmath136 in this section we repeat some definitions from section 3.1 in @xcite . let @xmath5 be a contact manifold with a contact form @xmath26 such that @xmath137 . an almost complex structure @xmath30 on the symplectization @xmath138 is _ @xmath26-adapted _ if @xmath30 is @xmath139-invariant ; @xmath140 with @xmath141 for nonzero @xmath142 ; and @xmath143 , where @xmath144 denotes the @xmath145-coordinate and @xmath123 is a reeb vector field associated to @xmath26 . let @xmath44 be a liouville manifold and @xmath53 be the contact structure given on @xmath51 by @xmath146 , where @xmath52 . in addition , let @xmath147 be the completion of @xmath44 , i.e. , @xmath148 and @xmath149 , where @xmath101 is the @xmath150-coordinate . an almost complex structure @xmath151 on @xmath152 is _ @xmath153- adapted _ if @xmath151 is @xmath104-adapted on @xmath154 ; and @xmath155 for all nonzero tangent vectors @xmath156 on @xmath45 . let @xmath116 be a sutured contact manifold , @xmath26 be an adapted contact form and @xmath93 be its completion . we say that an almost complex structure @xmath30 on @xmath157 is _ tailored to _ @xmath158 if the following hold : * @xmath30 is @xmath107-adapted ; * @xmath30 is @xmath159-invariant in a neighborhood of @xmath160 ; * the projection of @xmath30 to @xmath161 is a @xmath162-adapted almost complex structure @xmath151 on the completion @xmath163 of the liouville manifold @xmath164 . moreover , the flow of @xmath159 identifies @xmath165 and @xmath166 . first , let @xmath55 be a closed , oriented @xmath3-manifold , @xmath26 be a contact @xmath46-form on @xmath55 and let @xmath30 be an @xmath26-adapted almost complex structure on @xmath138 . for simplicity , we assume that all reeb orbits of @xmath123 , including multiply covered ones , are nondegenerate . an _ orbit set _ is a finite set of pairs @xmath167 , where the @xmath168 s are distinct embedded orbits of @xmath123 and the @xmath169 s are positive integers . the orbit set @xmath170 is _ admissible _ if @xmath171 whenever @xmath172 is hyperbolic . the homology class of @xmath170 is defined by @xmath173:=\sum\limits_{i } m_i[\alpha_i ] \in h_1(m ; \mathbb z).\end{aligned}\ ] ] if @xmath174 and @xmath175 are two orbit sets with @xmath176=[b]$ ] , let @xmath177 denote the set of relative homology classes of @xmath178-chains @xmath179 in @xmath55 with @xmath180 the _ ech chain complex _ @xmath181 is a free @xmath182-module with one generator for each admissible orbit set @xmath170 with @xmath176=h$ ] . if @xmath183 and @xmath184 are orbit sets with @xmath176 = [ b]$ ] , let @xmath185 denote the moduli space of @xmath30-holomorphic curves @xmath186 with positive ends at covers of @xmath172 with total multiplicity @xmath187 , negative ends at covers of @xmath188 with total multiplicity @xmath189 , and no other ends . note that the projection of each @xmath190 to @xmath55 has a well - defined relative homology class @xmath191\in h_{2}(m , a , b)$ ] . for @xmath192 we then define @xmath193 = z\}.\end{aligned}\ ] ] if @xmath174 is an orbit set , define the _ symplectic action _ @xmath194 [ le1_symplactfech ] for an adapted almost complex structure @xmath30 , if @xmath195 is non - empty , then : * @xmath196 . * if @xmath197 , then @xmath198 and every element of @xmath199 maps to a union of trivial cylinders . if @xmath200 , define the _ ech index _ @xmath201 here @xmath202 denotes the relative intersection pairing , which is defined in @xcite . any @xmath30-holomorphic curve @xmath203 can be uniquely written as @xmath204 , where @xmath205 and @xmath206 are unions of components of @xmath186 , each component of @xmath205 maps to an @xmath139-invariant cylinder , and no component of @xmath206 does . [ pr1_echind ] suppose that @xmath30 is generic and @xmath207 . then : * @xmath208 with equality if and only if @xmath209 . * if @xmath210 , then @xmath186 contains one embedded component @xmath206 with @xmath211 and @xmath205 does not intersect @xmath206 . to fix the signs in the differential , fix some ordering of all the embedded positive hyperbolic reeb orbits in @xmath55 . two curves @xmath186 and @xmath212 in @xmath213 are equivalent if their embedded components @xmath214 and @xmath215 are the same up to translation , and if their other components cover each embedded trivial cylinder @xmath216 with the same total multiplicity . the differential @xmath217 in @xmath218 counts @xmath219 curves in @xmath220 where @xmath170 and @xmath221 are admissible orbit sets . such curves may contain multiple covers of the @xmath139-invariant cylinder @xmath216 when @xmath124 is an elliptic embedded reeb orbit . the differential @xmath217 only keeps track of the total multiplicity of such coverings for each @xmath124 . finiteness of the count results from the ech compactness theorem ( * ? ? ? * lemma 9.8 ) . for the sign of the count we refer to @xcite . let @xmath116 be a sutured contact @xmath3-manifold with an adapted contact form @xmath26 , @xmath93 be its completion and @xmath30 be an almost complex structure on @xmath222 which is tailored to @xmath93 . the sutured embedded contact homology group @xmath223 is defined to be the embedded contact homology of @xmath224 . the following theorems have been proven by colin , ghiggini , honda and hutchings in @xcite : the ech compactness theorem ( * ? ? ? * lemma 9.8 ) holds for @xmath30-holomorphic curves in the symplectization of the completion of a sutured contact @xmath3-manifold , provided that we choose the almost complex structure @xmath30 on @xmath225 to be tailored to @xmath93 . let @xmath116 be a sutured contact @xmath3-manifold with an adapted contact form @xmath26 , @xmath93 be its completion and @xmath30 be an almost complex structure on @xmath222 which is tailored to @xmath93 . then the embedded contact homology group @xmath226 is defined . lemma [ le1_symplactfech ] and proposition [ pr1_echind ] hold for @xmath30-holomorphic curves in the symplectization of the completion of a sutured contact manifold , provided that we choose the almost complex structure @xmath30 on @xmath225 to be tailored to @xmath93 . recall that embedded contact homology is an invariant of the underlying closed , oriented @xmath3-manifold . hence , it is natural to expect the following : the embedded contact homology group @xmath227 does not depend on the choice of contact form @xmath26 , contact structure @xmath137 , and almost complex structure @xmath30 . let @xmath116 be a sutured contact manifold with an adapted contact form @xmath26 , @xmath93 be its completion and @xmath30 be an almost complex structure on @xmath222 which is tailored to @xmath93 . for simplicity , we assume that all reeb orbits of @xmath123 , including multiply covered ones , are nondegenerate . let @xmath124 be an embedded reeb orbit . we are also interested in the multiple covers @xmath228 of @xmath124 , @xmath229 . there are 2 ways the conley - zehnder index of @xmath230 can behave : * the parity of @xmath231 is the same for all @xmath232 . * the parity for the even multiples @xmath233 , @xmath234 , disagrees with the parity for the odd multiples @xmath235 , @xmath236 . in the second case , the even multiples @xmath237 , @xmath234 , are called _ bad orbits_. an orbit that is not bad is called _ good_. the _ sutured contact homology algebra _ @xmath238 is defined to be the contact homology of @xmath224 in the following sense : the contact homology chain complex @xmath239 is the free supercommutative @xmath240-algebra with unit generated by good reeb orbits , where the grading and the boundary map @xmath217 are defined in the usual way ( as in @xcite ) with respect to the @xmath107-adapted almost complex structure @xmath30 . the homology of @xmath241 is the sutured contact homology algebra @xmath242 . we define the _ sutured cylindrical contact homology group _ @xmath243 to be the cylindrical contact homology of @xmath244 . the cylindrical contact homology chain complex @xmath245 is the @xmath240-module freely generated by all good reeb orbits , where the grading and the boundary map @xmath217 are defined as in @xcite with respect to the @xmath107-adapted almost complex structure @xmath30 . the homology of @xmath245 is the sutured cylindrical contact homology group @xmath246 . for our calculations we will need the following fact which is a consequence of lemma 5.4 in @xcite : [ energyconthom ] let @xmath247 be a closed , oriented contact manifold with nondegenerate reeb orbits and @xmath248 be a @xmath30-holomorphic curve in @xmath249 , where @xmath124 and @xmath250 s are all good reeb orbits , @xmath30 is an @xmath26-adapted almost complex structure on @xmath138 and @xmath251 is a moduli space of @xmath30-holomorphic curves that we consider in contact homology . then @xmath252 with equality if and only if the image of @xmath253 is contained in a trajectory of @xmath123 , i.e. , @xmath186 maps to a trivial cylinder over @xmath254 , where @xmath254 is an embedded orbit of @xmath123 , and hence @xmath255 for some @xmath256 and @xmath257 with @xmath258 . in addition , we recall the following fact proven by eliashberg and hofer : [ welldefcylconthom ] let @xmath247 be a closed , oriented contact manifold with nondegenerate reeb orbits . let @xmath259 be the cylindrical contact homology complex , where @xmath36 is a homotopy class of reeb orbits and @xmath37 corresponds to the conley - zehnder grading . if there are no contractible reeb orbits , then for every free homotopy class @xmath36 * @xmath260 ; * @xmath261 is independent of the contact form @xmath26 for @xmath22 , the almost complex structure @xmath30 and the choice of perturbation for the moduli spaces . now we remind the following theorems which have been proven by colin , ghiggini , honda and hutchings in @xcite : the sft compactness theorem ( * ? ? ? * theorem 10.1 ) holds for @xmath30-holomorphic curves in the symplectization of the completion of a sutured contact manifold , provided that we choose the almost complex structure @xmath30 on @xmath225 to be tailored to @xmath93 . [ welldefconthom ] let @xmath116 be a sutured contact @xmath3-manifold with an adapted contact form @xmath26 , @xmath93 be its completion and @xmath30 be an almost complex structure on @xmath222 which is tailored to @xmath93 . then the contact homology algebra @xmath262 is defined and independent of the choice of contact @xmath46-form @xmath26 with @xmath263 , adapted almost complex structure @xmath30 , and abstract perturbation . [ actchomcchom ] fact [ energyconthom ] and fact [ welldefcylconthom ] hold for @xmath30-holomorphic curves in the symplectization of the completion of a sutured contact manifold , provided that we choose the almost complex structure @xmath30 on @xmath225 to be tailored to @xmath93 . note that theorem [ welldefconthom ] and remark [ actchomcchom ] rely on the assumption that the machinery , needed to prove the analogous properties for contact homology and cylindrical contact homology in the closed case , works . now we briefly describe the procedure of gluing sutured contact manifolds , together with compatible reeb vector fields which was first described by colin and honda in @xcite and generalized in @xcite . in @xcite , gabai defined the notion of a sutured manifold decomposition for sutured 3-manifolds , which is the inverse construction of the sutured gluing . let @xmath264 be a sutured contact @xmath3-manifold with an adapted contact form @xmath265 . we denote by @xmath266 the projection along @xmath267 defined on @xmath268 . if we think of @xmath269\times \gamma'$ ] as a subset of @xmath270 ( resp . @xmath271 ) , then we denote the projection by @xmath272 ( resp . @xmath273 ) . by definition , the horizontal components @xmath274 are liouville manifolds . we denote by @xmath275 their liouville vector field . the contact form @xmath265 is @xmath276 on the neighborhoods @xmath277 $ ] and @xmath278 $ ] of @xmath279 and @xmath280 . in addition , we may assume without loss of generality that the reeb vector field @xmath281 is given by @xmath267 on @xmath268 . take a @xmath178-dimensional submanifolds @xmath282 such that @xmath283 is the union of @xmath284 , @xmath285 and @xmath283 is positively transversal to the liouville vector field @xmath275 on @xmath286 . whenever we refer to @xmath287 and @xmath288 , we assume that closures are taken as appropriate . moreover we make the assumption that @xmath289 . let @xmath290 be a diffeomorphism which sends @xmath291 to @xmath292 and takes @xmath293 to @xmath294 and @xmath295 to @xmath296 . we will refer to the triple @xmath297 as the _ gluing data_. for the purposes of gluing , we only need @xmath298 and @xmath299 to match up on @xmath300 , since we can linearly interpolate between primitives of positive area forms on a surface . topologically , we construct the sutured manifold @xmath301 from @xmath302 and the gluing data @xmath297 as follows : let @xmath303 , where * @xmath304 for all @xmath305 ; * @xmath306 if @xmath307 and @xmath308 . then @xmath309 and @xmath310 in dimension @xmath3 , for the purposes of studying holomorphic curves , we want to stretch in @xmath76-direction . in higher dimensions , one needs to stretch in both @xmath101- and @xmath76-directions . the construction depends on the parameter @xmath311 , where @xmath311 is a stretching parameter in @xmath76-direction , and the resulting glued - up sutured contact manifold is written as @xmath312 . let @xmath313 ( we will suppress @xmath311 to avoid cluttering the notation ) be the manifold obtained from the completion @xmath314 by removing the side ( s ) , i.e. , @xmath315).\end{aligned}\ ] ] then construct @xmath316 from @xmath317)),\end{aligned}\ ] ] by taking closures and identifying : * @xmath318 with @xmath319 ; * @xmath320 with @xmath321 ; * @xmath322 $ ] with @xmath323 $ ] ; all via the identification @xmath324 . next we take @xmath325 and truncate the top and bottom of @xmath316 to obtain the ( compact ) sutured manifold @xmath326 so that @xmath327 contains @xmath328)\cup ( ( \overline{r_{-}(\gamma')\setminus p_{-}})\times [ -n',-1]),\end{aligned}\ ] ] the reeb vector field is transverse to the horizontal boundary , and the vertical boundary @xmath329 is foliated by interval reeb orbits with fixed action @xmath330 . attaching @xmath331\times e$ ] to @xmath327 for some specific @xmath332 gives us @xmath333 . the horizontal boundary which is positively ( resp . negatively ) transverse to @xmath334 will be called @xmath335 ( resp . @xmath336 ) . for more details we refer to @xcite . in this section we construct a sutured contact solid torus with @xmath2 longitudinal sutures , where @xmath24 . now we construct @xmath337 . the flow of the hamiltonian vector field associated to @xmath338 will play a role of gluing map when we will apply the gluing construction described in section [ section : glsutcontman ] to the sutured contact solid cylinder constructed in section [ section : gluing ] . we fix @xmath339 and consider @xmath340 given by @xmath341 in polar coordinates about @xmath127 , where @xmath342 and @xmath343 . note that @xmath344 is singular only at @xmath127 . we obtain @xmath345 from @xmath344 by perturbing @xmath344 on a small disk @xmath346 about @xmath127 in such a way that @xmath338 has @xmath347 nondegenerate saddle points and interpolates with no critical points with @xmath344 on @xmath346 . in other words , @xmath348 on @xmath349 . for the level sets of @xmath344 and @xmath338 in the case @xmath350 we refer to figure [ levelset3 ] . the construction of @xmath338 was initially described by cotton - clay as a construction of a hamiltonian function whose time-@xmath46 flow is a symplectic smoothing of the singular representative of pseudo - anosov map in a neighborhood of a singular point with @xmath351 prongs in @xcite . since some of the properties of @xmath338 described in @xcite will be important for further discussion , we will state them in the next remark . [ re1_closedlevsets ] we can choose @xmath338 in such a way that it satisfies the following properties : * @xmath338 can be written as @xmath352 in some coordinates @xmath353 in a connected neighborhood containing its critical points ; * there are no components of level sets of @xmath338 which are circles ; * there is an embedded curve which is a component of one of the level curves of @xmath338 and connects all the saddle points of @xmath338 . we call this embedded curve @xmath124 . for the detailed construction of @xmath338 we refer to lemmas 3.4 and 3.5 in @xcite . ( left ) and the level sets of @xmath338 ( right ) in the case @xmath350.,title="fig : " ] [ le1_prarfm ] let @xmath144 be a saddle point of @xmath338 . then there are coordinates @xmath353 about @xmath144 such that @xmath354 for @xmath355 . first observe that from remark [ re1_closedlevsets ] it follows that @xmath356 . by morse lemma , there are coordinates @xmath357 about @xmath144 such that @xmath358 . given @xmath356 , we can write @xmath359 now let @xmath360 clearly @xmath127 and @xmath361 satisfy the statement of the lemma . in addition , the orientation of the pair @xmath353 coincides with the orientation of @xmath357 . let @xmath362 be a neighborhood of the @xmath256-th saddle point @xmath363 of @xmath338 from lemma [ le1_prarfm ] . if @xmath364 is an exact symplectic manifold , i.e. , @xmath365 , then the flow @xmath366 of a hamiltonian vector field @xmath367 consists of exact symplectic maps , i.e. , @xmath368 since @xmath369 , @xmath370 since by definition @xmath371 , the integrand is equal to @xmath372\circ \varphi_{x_{h}}^{s}).\end{aligned}\ ] ] thus @xmath373 notice that our definition of @xmath367 is slightly different from the standard one ; usually @xmath367 is defined by @xmath374 . note that the condition that @xmath375 is equivalent to the condition that @xmath376 . [ re1_funcanalcon ] let @xmath377 . in addition , let @xmath378 be a region such that @xmath375 on @xmath379 and @xmath380 \}$ ] . then @xmath381 and @xmath382 on @xmath383 . in the next two lemmas we construct a @xmath46-form @xmath47 on @xmath384 with @xmath385 and show that @xmath47 is `` adapted '' to @xmath338 , i.e. , @xmath386 near the saddle points of @xmath338 and on the region far enough from @xmath346 , where @xmath367 is a hamiltonian vector field with respect to @xmath387 and @xmath388 is the time-@xmath46 map of the flow of @xmath367 . the condition that @xmath389 and remark [ re1_funcanalcon ] will play a crucial role when we will compare @xmath390 and @xmath47 . [ le1_formanalcon ] there exists a 1-form @xmath47 on @xmath391 satisfying the following : * @xmath392 ; * the singular foliation given by @xmath393 has isolated singularities and no closed orbits ; * the elliptic points of the singular foliation of @xmath47 are the saddle points of @xmath338 ; @xmath394 on @xmath362 with respect to the coordinates from lemma [ le1_prarfm ] , where @xmath395 and @xmath396 is a small positive real number ; * @xmath397 on @xmath398 with respect to the polar coordinates whose origin is at the center of @xmath346 ; * the hyperbolic points of the singular foliation of @xmath47 are located on @xmath124 , outside of @xmath362 s and distributed in such a way that between each two closest elliptic points there is exactly one hyperbolic point . consider a singular foliation @xmath399 on @xmath391 which satisfies the following : * @xmath399 is morse - smale and has no closed orbits . * the singular set of @xmath399 consists of elliptic points and hyperbolic points . the elliptic points are the saddle points of @xmath338 . the hyperbolic points are located on @xmath124 and distributed in such a way that between each two closest elliptic points there is exactly one hyperbolic point . in addition , the hyperbolic points are outside of @xmath362 s . * @xmath399 is oriented , and for one choice of orientation the flow is transverse to and exits from @xmath400 . ( left ) and the characteristic foliation of @xmath47 ( right ) in the case @xmath350,title="fig : " ] next , we modify @xmath399 near each of the singular points so that @xmath399 is given by @xmath401 on @xmath362 with respect to the coordinates from lemma [ le1_prarfm ] , and @xmath402 near a hyperbolic point . in addition , on @xmath349 , @xmath403 with respect to the polar coordinates whose origin is at the center of @xmath346 . finally , we get @xmath399 given by @xmath104 , which satisfies @xmath404 near the singular points and on @xmath349 . now let @xmath405 , where @xmath406 is a positive function with @xmath407 outside of @xmath408 , @xmath409 , @xmath410 and @xmath7 is an oriented vector field for @xmath399 ( nonzero away from the singular points ) . since @xmath411 , @xmath412 guarantees that @xmath413 . here @xmath396 is a small positive real number . [ re1_formandfunc ] from the previous lemma we get @xmath47 defined on @xmath391 with the following properties : * @xmath414 on @xmath391 . * @xmath394 and @xmath354 on @xmath362 for @xmath415 . in other words , the saddle points of @xmath338 are exactly the elliptic singularities of @xmath47 . * @xmath397 and @xmath416 on @xmath349 . for the comparison of the level sets of @xmath338 with the singular foliation of @xmath47 in the case @xmath350 we refer to figure [ charfol ] . [ le2_formanalcon ] let @xmath47 be a @xmath46-form from lemma [ le1_formanalcon ] . the hamiltonian vector field @xmath367 of @xmath338 with respect to the area form @xmath387 satisfies @xmath375 on @xmath417 . first we work on @xmath362 , where @xmath395 . from remark [ re1_formandfunc ] it follows that @xmath394 and @xmath354 on @xmath362 . let @xmath367 be a hamiltonian vector field defined by @xmath418 . we show that @xmath419 is a solution of the equation @xmath420 on @xmath362 . we calculate @xmath421 and @xmath422 next , by remark [ re1_formandfunc ] , @xmath397 and @xmath423 on @xmath349 . as in the previous case , we show that @xmath424 is a solution of equation ( [ e1_formanconstr ] ) on @xmath398 . we calculate @xmath425 and @xmath426 let @xmath427 be the time-@xmath144 flow of @xmath367 . consider @xmath428 \}\quad \mbox{and}\\ & v_{k}:=\ { x\in u_{k } : \varphi_{x_{h}}^{s}(x)\in u_{k}\ \forall s\in[0,1 ] \}.\end{aligned}\ ] ] since the saddle points of @xmath338 are the fixed points of @xmath427 for @xmath429 $ ] , @xmath430 contains an open neighborhood about the @xmath256-th saddle point of @xmath338 . in addition , note that @xmath379 contains the level sets of @xmath338 which do not intersect @xmath346 , and @xmath431 , where @xmath432 . for ease of notation , we write @xmath388 instead of @xmath433 . [ re1_analconstr ] lemma [ le2_formanalcon ] implies that @xmath434 on @xmath435 . [ re1_phiperpoints ] from remark [ re1_closedlevsets ] and the fact that the flow of @xmath367 preserves the level sets of @xmath338 it follows that @xmath436 is the set of periodic points of @xmath388 on @xmath391 . in the next lemma we construct a contact form @xmath26 on @xmath437 $ ] such that @xmath123 has vertical trajectories . [ le1_almvertical ] let @xmath438 and @xmath439 be two @xmath46-forms on @xmath440 such that @xmath441 in a neighborhood of @xmath442 and @xmath443 then there exists a contact @xmath46-form @xmath26 with reeb vector field @xmath123 on @xmath444 \times d^2 $ ] with coordinates @xmath445 , where t is a coordinate on @xmath444 $ ] and @xmath127 is a coordinate on @xmath440 , with the following properties : * @xmath446 in a neighborhood of @xmath447 ; * @xmath448 in a neighborhood of @xmath449 ; * @xmath123 is collinear to @xmath450 on @xmath444\times d^2 $ ] ; * @xmath451 in a neighborhood of @xmath444\times \partial d^2 $ ] . here @xmath396 is a small positive number . since @xmath440 is simply connected and @xmath452 , there exists a function @xmath453 such that @xmath454 . let @xmath455\to [ 0,1]$ ] be a smooth map for which @xmath456 for @xmath457 , @xmath458 for @xmath459 and @xmath460 for @xmath461 $ ] , where @xmath462 is a small positive number . in addition , we define @xmath463 . consider @xmath444 \times d^2 $ ] equipped with a @xmath46-form @xmath464 we then compute @xmath465 and hence @xmath466 if @xmath396 is sufficiently small , then @xmath26 satisfies the contact condition , i.e. , @xmath467 . now let us show that the reeb vector field @xmath123 is given by @xmath468 first we compute @xmath469 then we check the normalization condition , i.e. , @xmath470 : @xmath471 since @xmath472 in a neighborhood of @xmath473 , @xmath474 in a neighborhood of @xmath442 and hence @xmath451 in a neighborhood of @xmath444\times \partial d^2 $ ] . finally , we see that @xmath26 satisfies conditions @xmath475 . fix @xmath476 such that there is an annular neighborhood @xmath477 of @xmath478 in @xmath384 with @xmath479 . consider @xmath480 with two @xmath46-forms @xmath481 , where @xmath47 is a @xmath46-form from lemma [ le1_formanalcon ] , and @xmath482 . by remark [ re1_analconstr ] , @xmath483 in addition , we have @xmath484 from equations ( [ e1_cylinf ] ) and ( [ e2_cylinf ] ) it follows that @xmath104 and @xmath485 satisfy the conditions of lemma [ le1_almvertical ] . now take @xmath444\times d(r_{\ast})$ ] with the contact @xmath46-form @xmath26 from lemma [ le1_almvertical ] with @xmath104 and @xmath485 as in the previous paragraph . note that @xmath486 for @xmath487 . we can rewrite this equation as @xmath488 let us remind that @xmath489 on @xmath391 , where @xmath490 hence , @xmath491 satisfies equation ( [ e3_cylinf ] ) . from remark [ re1_funcanalcon ] it follows that @xmath492 on @xmath493 . thus , @xmath474 on @xmath493 . since @xmath474 on @xmath494 , by the construction of @xmath26 , @xmath495 on @xmath496\times v_{k})\cup([-1,1]\times ( d(r_{\ast})\cap s))$ ] . let @xmath497 and @xmath498 , where @xmath396 is a constant from lemma [ le1_almvertical ] which makes @xmath26 contact . in this section we will construct the sutured contact solid torus with @xmath2 parallel longitudinal sutures , where @xmath24 . first we construct surfaces with boundary @xmath499 with the following properties : * @xmath500 ; * @xmath501 and @xmath502 ; * @xmath388 maps @xmath503 to @xmath504 in such a way that @xmath505 and @xmath506 ; * @xmath507 . recall that @xmath508 on @xmath509 . note that @xmath367 is collinear to @xmath510 for @xmath511 , where @xmath512 . for simplicity , let us denote @xmath513 where @xmath514 . fix @xmath334 such that @xmath515 and there is an annular neighborhood @xmath516 of @xmath517 in @xmath480 satisfying @xmath518 . consider @xmath519 . let @xmath520 be a segment on @xmath517 which starts at @xmath521 and ends at @xmath522 , i.e. , @xmath523 . consider @xmath524 . it is easy to see that every level set of @xmath338 which intersects @xmath520 intersects it only once . hence , using that there are no closed level sets of @xmath338 and @xmath367 is @xmath525-symmetric on @xmath526 , we get @xmath527 let @xmath528 . by possibly making @xmath334 and @xmath529 big enough , we can make @xmath530 s to be in @xmath516 . consider the endpoints of @xmath530 s . since @xmath367 is @xmath525-symmetric outside of @xmath346 , it is easy to see that @xmath531 and @xmath532 , where @xmath533 , @xmath534 and @xmath535 . in addition , observe that @xmath536 is the same for all endpoints of @xmath520 s . let @xmath537 be a set of embedded curves on @xmath480 with the following properties : * @xmath538 starts at the terminal point of @xmath539 and ends at the initial point of @xmath520 , where @xmath256 is considered mod @xmath351 ; * @xmath540 and @xmath541 for @xmath542 ; * @xmath543 for @xmath542 ; * the region bounded by @xmath520 s and @xmath544 s has smooth boundary ; * each level set of @xmath338 which intersects @xmath544 intersects it only once . for simplicity , we take @xmath525-symmetric @xmath544 s , i.e. , @xmath545 can be obtained from @xmath544 , where @xmath256 is considered mod @xmath351 , by doing @xmath525-positive rotation about the center of @xmath480 . , @xmath504 and @xmath546 in the case @xmath350,title="fig : " ] note that properties @xmath547 and @xmath548 and the form of @xmath367 on @xmath526 imply that @xmath549 where @xmath512 . again , using that the level sets of @xmath338 which intersects @xmath544 intersects it only once , there are no closed level sets of @xmath338 and @xmath367 is @xmath525-symmetric , we obtain @xmath550 let @xmath551 . from formula ( [ e1_infgl ] ) and the construction of @xmath520 s it follows that @xmath552 then we connect the terminal point of @xmath520 with the initial point of @xmath553 by the line segment @xmath554 and the terminal point of @xmath553 with the initial point of @xmath555 by the line segment @xmath556 . from the construction above it follows that @xmath554 intersects @xmath557 only at the terminal point of @xmath520 , and @xmath556 intersects @xmath557 only at the initial point of @xmath555 . then we round the corners between @xmath554 and @xmath520 , @xmath554 and @xmath553 , @xmath556 and @xmath553 , @xmath556 and @xmath555 . finally , we get a surface whose boundary consists of @xmath520 s , @xmath553 s , @xmath556 s and @xmath554 s , which we call @xmath546 . see figure [ constofd ] . [ re1_boundparts ] by the construction , @xmath558 s , @xmath559 s and @xmath560 lie in @xmath561 . now we take @xmath444\times d$ ] with a contact form @xmath562\times d}$ ] and contact structure @xmath563 . let @xmath564 in @xmath444\times d$ ] and @xmath565\times [ -1,1]\times \gamma$ ] be a neighborhood of @xmath1 with coordinates @xmath566\times [ -1,1]$ ] , where @xmath76 is a usual @xmath76-coordinate on @xmath444\times d$ ] . by remark [ re1_boundparts ] , we can make @xmath74 such that @xmath567\times ( d\cap s).\end{aligned}\ ] ] observe that @xmath568\times d)=r_{+}\cup r_{-}$ ] , where @xmath569 and @xmath570 with respect to the coordinates on @xmath444\times d$ ] . in addition , @xmath571\times d)=[-1,1]\times \partial d$ ] with respect the coordinates on @xmath444\times d$ ] . let @xmath572 . [ le1_suturbefglug ] @xmath573\times d , \gamma , u(\gamma ) , \xi)$ ] is a sutured contact manifold and @xmath26 is an adapted contact form . first note that @xmath574 and @xmath575 . let us check that @xmath576 and @xmath577 are liouville manifolds . from the construction of @xmath80 it follows that @xmath578 . since @xmath579 on @xmath580 , by equation ( [ e3_infgl ] ) , we have @xmath581 on @xmath74 . recall that @xmath582 on @xmath583 . hence , @xmath584 . the calculation @xmath585 implies that the liouville vector fields @xmath586 are equal to @xmath587 . from the construction of @xmath546 it follows that @xmath79 is positively transverse to @xmath588 . therefore , @xmath589 and @xmath590 are liouville manifolds . as we already mentioned , @xmath581 on @xmath74 . finally , if we take @xmath101 such that @xmath591 , then @xmath573\times d , \gamma , u(\gamma ) , \xi)$ ] becomes a sutured contact manifold with an adapted contact form @xmath26 . now observe that from the construction of @xmath26 it follows that @xmath592\times d}=dt+\beta_{-}$ ] and @xmath593\times d}=dt+\beta_{+}$ ] . then we define @xmath594 . let @xmath503 be a region bounded by @xmath520 s and @xmath544 s in @xmath595 and let @xmath504 be a region bounded by @xmath530 s and @xmath553 s in @xmath596 . note that from remark [ re1_boundparts ] it follows that @xmath558 s and @xmath559 are in @xmath379 . hence , by lemma [ le1_suturbefglug ] , @xmath597 along @xmath283 . the construction of @xmath558 s and @xmath559 implies that @xmath79 is positively transverse to @xmath283 . from the construction we made it is easy to see that @xmath598 , @xmath599 , @xmath600 and @xmath601 . if @xmath602\times d \to d$ ] is a projection to @xmath546 along @xmath267 , then from equation ( [ e2_infgl ] ) it follows that @xmath603 . observe that @xmath604 and @xmath605 , @xmath606 for @xmath542 . hence , by definition of @xmath607 , @xmath388 sends @xmath503 to @xmath504 in such a way that @xmath293 maps to @xmath294 and @xmath295 maps to @xmath296 . next , we follow the gluing procedure overviewed in section [ section : glsutcontman ] and completely described in @xcite . we get a sutured contact manifold @xmath608 . for simplicity , we omit index @xmath311 . observe that the region enclosed by @xmath473 and @xmath530 in @xmath609 contains @xmath520 and the region enclosed by @xmath473 and @xmath544 in @xmath610 contains @xmath553 . then from the gluing construction and the form of @xmath388 near the boundary of @xmath503 it follows that @xmath611 has @xmath2 parallel longitudinal components . consider @xmath608 obtained in section [ section : gluing ] . recall that @xmath611 consists of @xmath2 parallel longitudinal curves let @xmath612 denote the contact structure defined by @xmath613 and @xmath614 denote the reeb vector field defined by @xmath613 . the main goal of this section is to understand the set of embedded , closed orbits of @xmath614 . let @xmath379 be a non - empty set with two non - empty subsets @xmath615 and @xmath616 such that @xmath617 , and let @xmath618 . a point @xmath619 is called a _ periodic point _ of @xmath253 of period @xmath351 if @xmath620 is well - defined , i.e. , @xmath621 for @xmath622 , and @xmath623 . [ le1_anconstnumboforb ] @xmath614 has @xmath347 embedded , closed orbits . first consider @xmath624 . recall that from the construction of @xmath504 and @xmath503 it follows that @xmath625 . hence , by remark [ re1_phiperpoints ] , @xmath626 is the set of periodic points of @xmath624 . from the construction of @xmath26 on @xmath444\times d$ ] and the gluing construction it follows that there is a one - to - one correspondence between the set of embedded reeb orbits and the set of periodic points of @xmath624 . thus , there are @xmath347 embedded closed orbits of @xmath614 . let @xmath627 be the embedded , closed orbit , which corresponds to the periodic point @xmath363 , i.e. , @xmath627 is obtained from @xmath444\times \{p_{k}\}\subset [ -1,1]\times d$ ] . [ le1_anconstsymplact ] @xmath628 is a nondegenerate orbit for @xmath629 and @xmath630 . moreover @xmath631 is a set of positive hyperbolic orbits and @xmath632 for @xmath633 . let @xmath634\times d)\cup ( r_{+}(\gamma)\times[1;\infty))\cup ( r_{+}(\gamma)\times ( -\infty;-1]))\end{aligned}\ ] ] and @xmath635 in addition , let @xmath636 denote the contact form on @xmath637 and let @xmath638 denote the contact structure defined by @xmath636 . consider @xmath444\times d\subset \tilde{m}$ ] . from the construction of @xmath26 it follows that @xmath639\times v_{k}}=dt+\beta_{-}$ ] . since the contact structure on @xmath640 is given by @xmath641 and the contact structure on @xmath642\times p_{-}$ ] is given by @xmath643 , @xmath644\times v_{k}}=dt+\beta_{-}$ ] on @xmath645\times v_{k}\subset \tilde{m}$ ] . therefore , we get @xmath646\times \{p_{k}\ } } \alpha_{\tilde{m}}=2n.\end{aligned}\ ] ] from the gluing construction and equation ( [ e10_anconstlength ] ) it follows that @xmath647 . note that @xmath648 does not depend on @xmath256 . hence , @xmath649 for @xmath633 . now observe that @xmath650 and hence @xmath651 where @xmath652 . let the symplectic trivialization of @xmath638 along @xmath645\times \{p_{k}\}$ ] be given by the framing @xmath653 . note that the symplectic trivialization of @xmath638 gives rise to the symplectic trivialization of @xmath612 along @xmath627 . it is easy to see that the linearized return map @xmath654 is given by @xmath655 since the eigenvalues of @xmath654 are positive real numbers different from @xmath46 , @xmath627 is a positive hyperbolic orbit . hence , @xmath631 is a set of positive hyperbolic orbits of @xmath614 . in addition , @xmath656 . therefore , the eigenvalues of @xmath657 are different from @xmath46 . hence , @xmath628 is a nondegenerate orbit for @xmath630 . in this section we will calculate the sutured embedded contact homology , the sutured cylindrical contact homology and the sutured contact homology of the sutured contact solid torus constructed in section [ section : gluing ] . consider the symplectization @xmath658 of @xmath659 , where @xmath144 is the coordinate on @xmath139 and @xmath660 is the completion of @xmath661 . let @xmath30 be an almost complex structure on @xmath662 tailored to @xmath663 . consider the set of embedded , closed orbits of @xmath614 . by lemma [ le1_anconstnumboforb ] , @xmath614 has @xmath347 embedded , closed orbits @xmath664 , which are positive hyperbolic by lemma [ le1_anconstsymplact ] . in addition , lemma [ le1_anconstsymplact ] implies that all reeb orbits are nondegenerate . from the gluing construction , i.e. , since @xmath665 is a set of fixed points of @xmath388 , it follows that @xmath666 $ ] is a generator of @xmath667 for @xmath668 and @xmath666=[\gamma_{j}]$ ] for @xmath669 . from now on we identify @xmath670 with @xmath182 in such a way that @xmath666\in h_{1}(s^1\times d^2 ; \mathbb z)$ ] is identified with @xmath671 for @xmath672 . recall that multiplicities of hyperbolic orbits in an admissible orbit set must be equal to @xmath46 . hence , from lemma [ le1_anconstsymplact ] it follows that the admissible orbit sets are of the form @xmath673 , where @xmath674 . note that @xmath675 is an admissible orbit set . for ease of notation , we write @xmath676 instead of @xmath677 and @xmath46 instead of @xmath675 , where @xmath678 . fix @xmath680 . let @xmath681 be a set of admissible orbit sets with homology class @xmath36 . it is easy to see that @xmath682 from lemma [ le1_anconstsymplact ] it follows that for every @xmath683 , @xmath684 . let @xmath685 be different admissible orbit sets . then , as we already mentioned , @xmath686 from equation ( [ e11_anconstdifforbss ] ) and the second part of lemma [ le1_symplactfech ] it follows that @xmath687 is empty . in addition , by the second part of lemma [ le1_symplactfech ] , every element in @xmath688 maps to a union of trivial cylinders . hence , by proposition [ pr1_echind ] and definition of @xmath217 , @xmath679 . note that trivial cylinders are regular and hence we can omit the genericity assumption in proposition [ pr1_echind ] . thus , for every admissible orbit set @xmath170 , @xmath679 . by counting the number of element in @xmath681 , we get @xmath689 by equation ( [ e9_acech ] ) and lemma [ le1_anconstechdiff ] , we get @xmath690 here @xmath691 is the exterior algebra over @xmath182 generated by @xmath664 . thus , we obtain @xmath692 this completes the proof of theorem [ suturedech ] . note that for the constructed sutured contact solid torus , the sutured floer homology coincides with the sutured embedded contact homology . in fact , they agree in each spin@xmath21-structure . this follows from proposition 9.2 in @xcite , where the sutured floer homology of every sutured manifold @xmath25 has been computed by juhsz . [ cch_indep ] note that there are no contractible reeb orbits . hence , from fact [ welldefcylconthom ] , remark [ actchomcchom ] and the fact that @xmath693 it follows that for all @xmath694 @xmath695 is defined , i.e. , @xmath260 , and is independent of contact form @xmath613 for the given contact structure @xmath612 and the almost complex structure @xmath30 . from lemma [ le1_anconstsymplact ] it follows that @xmath631 is a set of positive hyperbolic orbits . hence , the definition of the conley - zehnder index implies that @xmath698 is even for @xmath699 and @xmath630 . then , according to the definition of a good orbit , it follows that @xmath700 is a good orbit for @xmath701 and @xmath630 . hence , we get @xmath702 here @xmath703 is a @xmath240-module freely generated by @xmath704 . now recall that lemma [ le1_anconstsymplact ] says that @xmath705 , where @xmath706 . therefore , @xmath707 for @xmath706 and @xmath630 . remark [ actchomcchom ] , equation ( [ e21_anconsymact ] ) and definition of the cylindrical contact homology differential imply that @xmath708 for all @xmath694 . thus , using equation ( [ e15_accch ] ) , we obtain @xmath709 finally , equation ( [ e22_anconstcch ] ) implies that @xmath710 this completes the proof of theorem [ suturedcylconthom ] . recall that from lemma [ le1_anconstnumboforb ] it follows that all closed orbits of @xmath614 are nondegenerate . from the discussion in the previous section it follows that @xmath700 is a good orbit for @xmath699 and @xmath630 . hence , the supercommutative algebra @xmath711 is generated by @xmath700 for @xmath701 and @xmath630 . note that @xmath711 splits as @xmath712 where @xmath713 is generated , as a vector space over @xmath240 , by monomials of total homology class @xmath36 . hence , @xmath714 where @xmath715 denotes the coefficient of @xmath42 in the generating function @xmath43 . in ( * ? ? ? * corollary 4.2 ) , fabert proved that the differential in contact homology and rational symplectic field theory is strictly decreasing with respect to the symplectic action filtration . in other words , branched covers of trivial cylinders do not contribute to contact homology and rational symplectic field theory differentials . from lemma [ le1_anconstsymplact ] it follows that all generators of @xmath713 have the same symplectic action and hence @xmath716 for all @xmath680 . thus , @xmath717 and hence @xmath718 this completes the proof of theorem [ suturedconthom ] .

a multi - reference configuration mixing scheme is used to describe the ground state , characterized by well defined spin and space group symmetry quantum numbers as well as doping fractions , of one dimensional hubbard lattices with nearest - neighbor hopping and periodic boundary conditions . within this scheme , each ground state is expanded in a given number of nonorthogonal and variationally determined symmetry - projected configurations . the results obtained for the ground state and correlation energies of half - filled and doped lattices with 30 , 34 and 50 sites , compare well with the exact lieb - wu solutions as well as with the ones obtained with other state - of - the - art approximations . the structure of the intrinsic symmetry - broken determinants resulting from the variational procedure is interpreted in terms of solitons whose translational and breathing motions can be regarded as basic units of quantum fluctuations . it is also shown that in the case of doped 1d lattices , a part of such fluctuations can also be interpreted in terms of polarons . in addition to momentum distributions , both spin - spin and density - density correlation functions are studied as functions of doping . the spectral functions and density of states , computed with an ansatz whose quality can be well - controlled by the number of symmetry - projected configurations used to approximate the electron systems , display features beyond a simple quasiparticle distribution , as well as spin - charge separation trends .

1402.2968

disentangling the effects of electron - electron interactions in the ground and excited states of low - dimensional systems has become an exciting challenge in contemporary condensed matter physics . @xcite in particular , the discovery of high - tc superconductivity in the cuprates @xcite has acted as a driving force to develop theoretical models able to account for the most relevant correlations in many - electron systems in the simplest possible way . within this context , the repulsive hubbard model @xcite has been widely studied for several reasons including that it represents a prototype for the mott transition between a metal and an antiferromagnetic insulator @xcite and the suggestion @xcite that it contains the basic physics associated with high - tc superconductivity . the phenomenon of colossal magnetic resistance @xcite has also attracted considerable attention . on the other hand , the study of the high - tc iron - based superconductors @xcite has become a very active research area . @xcite here , calculations in terms of multi - orbital hubbard - like models have already provided valuable insight into the interplay between doping and the strength of the electronic correlations in these exotic superconductors . @xcite hubbard models represent valuable tools to study cold fermionic atoms in optical lattices @xcite as well as the properties of graphene . @xcite it has also become clear that their strong coupling limits , i.e. , the heisenberg models , @xcite can be quite useful to study low - dimensional magnets whose properties might be relevant for real materials found in nature and/or synthesized by means of crystal growing . @xcite the previous examples already show the central role of hubbard - like lattice models and their strong coupling versions to obtain insight into the properties associated with the emergent complexity in many - electron problems . precisely , it is this complexity that requires the use of different approximations to describe low - dimensional systems . among the available theoretical tools we have , for example , exact diagonalizations for small lattices @xcite while for the larger ones , we can resort to quantum monte carlo , @xcite variational monte carlo , @xcite coupled cluster , @xcite variational reduced - density - matrix @xcite and density matrix renormalization group @xcite methods , as well as approximations based on matrix product and tensor network states . @xcite both frequency - dependent and frequency - independent embedding approaches @xcite are also actively pursued . [ table1 ] cccccccccccccccccccc + @xmath2 & & & & & @xmath3=30 & @xmath4 & & @xmath3=26 & @xmath4 & & & @xmath3=22 & @xmath4 & & @xmath3=18 & @xmath4 & & @xmath3=14 & @xmath4 + + + 2@xmath5 & & & rhf & & -23.2671 & & & -26.1642 & & & & -26.8921 & & & -25.5587 & & & -22.3390 & + & & & hf & & -23.4792 & 10.02 & & -26.1642 & 0 & & & -26.8921 & 0 & & -25.5587 & 0 & & -22.3390 & 0 + & & & fed & & -25.3800 & 99.83 & & -28.0201 & 99.72 & & & -28.4391 & 99.68 & & -26.7816 & 99.69 & & -23.2365 & 99.74 + & & & fed@xmath6 & & -25.3730 & 99.50 & & - & - & & & - & - & & - & - & & - & + & & & reshf & & -25.3436 & 98.11 & & -27.9979 & 98.52 & & & -28.4268 & 98.88 & & - & - & & - & - + & & & exact & & -25.3835 & & & -28.0253 & & & & -28.4441 & & & -26.7854 & & & -23.2388 & + + + 4@xmath5 & & & rhf & & -8.2671 & & & -14.8975 & & & & -18.8254 & & & -20.1587 & & & -19.0723 & + & & & hf & & -14.0732 & 64.75 & & -17.3756 & 34.96 & & & -20.1328 & 23.42 & & -21.1011 & 22.45 & & -19.8855 & 27.84 + & & & fed & & -17.2081 & 99.71 & & -21.9193 & 99.04 & & & -24.3497 & 99.03 & & -24.3222 & 99.19 & & -21.9824 & 99.63 + & & & fed@xmath6 & & -17.1789 & 99.39 & & - & - & & & - & - & & - & - & & - & + & & & reshf & & -17.0542 & 98.00 & & -21.5720 & 94.15 & & & -24.1582 & 95.56 & & - & - & & - & - + & & & exact & & -17.2335 & & & -21.9868 & & & & -24.4057 & & & -24.3561 & & & -21.9932 & + + + 8@xmath5 & & & rhf & & 21.7329 & & & 7.6358 & & & & -2.6921 & & & -9.3587 & & & -12.5390 & + & & & hf & & -7.8329 & 93.65 & & -11.2049 & 79.45 & & & -14.9299 & 69.32 & & -18.1922 & 70.12 & & -19.0005 & 77.96 + & & & fed & & -9.8260 & 99.95 & & -15.8927 & 99.22 & & & -20.1711 & 99.01 & & -21.8777 & 99.38 & & -20.8062 & 99.75 + & & & fed@xmath6 & & -9.7612 & 99.75 & & - & - & & & - & - & & - & - & & - & + & & & reshf & & -9.5378 & 98.46 & & -15.4059 & 97.17 & & & -19.5552 & 95.52 & & - & - & & - & - + & & & exact & & -9.8387 & & & -16.0761 & & & & -20.3462 & & & -21.9555 & & & -20.8271 & + + the exact bethe ansatz solution of the one - dimensional ( 1d ) hubbard hamiltonian is well known , @xcite which is not the case for the two - dimensional ( 2d ) model . on the other hand , in spite of the considerable progress already made , the exact 1d wave functions still remain difficult to handle in practice when computing several physical properties . @xcite it has also remained difficult to obtain an intuitive physical picture of the basic units of quantum fluctuations @xcite using the lieb - wu solutions @xcite and/or within the theoretical frameworks already mentioned above . this task is further complicated by the fact that quantum fluctuations can exhibit unconventional features in low - dimensional systems . a typical example , is the spin - charge separation in the strong coupling regime of the 1d hubbard model . @xcite angle - resolved photoemission spectroscopy results also reveal a complex pattern of spin - charge coupling / decoupling both in 1d and 2d systems @xcite in the weak and intermediate - to - strong interaction regimes . therefore , it is highly desirable to explore the performance of alternative wave function based approaches that , on the one hand , could complement existing state - of - the - art theoretical tools and , on the other hand , lead us to compact states whose ( intrinsic ) structures are simple enough to be interpreted in terms of basic units of quantum fluctuations in half - filled and doped lattices . within this context , the 1d hubbard model represents a challenging testing ground since both its exact solution and highly accurate density matrix renormalization group results are available . the last years have seen some progress in this direction within the framework of the single - reference ( sr ) and multi - reference ( mr ) symmetry - projected approximations @xcite which are routinely used in nuclear structure physics . @xcite note that in quantum chemistry the names single - component and multi - component have been adopted @xcite instead of single - reference and multi - reference , respectively . recently , a hierarchy of symmetry - projected variational approaches @xcite has been applied to describe both the ground and excited states of the 1d and 2d hubbard models . @xcite in its simplest ( i.e. , sr ) form , the symmetry - projected variation - after - projection ( vap ) method @xcite resorts to a hartree - fock - type @xcite ( hf ) trial state @xmath7 that deliberately breaks several symmetries of the considered hamiltonian . the method then superposes , with the help of projection operators , @xcite a degenerate manifold of goldstone states @xmath8 , with @xmath9 being a symmetry operation . in this way one recovers a set of quantum numbers associated with the original symmetries of the hamiltonian . already at this sr level genuine defects are induced in the intrinsic determinant @xmath7 resulting from the vap procedure . therefore , its structure differs from the one obtained within the standard hf approximation . @xcite this kind of sr symmetry - projected framework has already enjoyed considerable success in quantum chemistry . @xcite an extension of the sr method @xcite has been previously used @xcite to describe half - filled and doped 2d hubbard lattices . with the help of chains of vap calculations , it provides a ( truncated ) basis consisting of a few ( orthonormalized ) symmetry - projected configurations which can then be used to further diagonalize the considered hamiltonian . in this way , one can account on an equal footing for additional correlations in both ground and excited states keeping well defined symmetry quantum numbers . the method also provides a well - controlled ansatz to compute both spectral functions ( sfs ) and density of states ( dos ) . @xcite in quantum chemistry , the first benchmark calculations on the @xmath10 dimer have shown that , with a modest basis set , this methodology provides a high quality description of the low - lying spectrum for the entire dissociation profile . in addition , the same methodology has been applied to obtain the full low - lying spectrum of formaldehyde as well as to a challenging model @xmath11 insertion pathway for beh@xmath12 . @xcite obtained with the fed approximation is plotted as a function of the inverse of the number of ghf transformations for a lattice with @xmath13=30 sites and @xmath3=22 electrons . results are shown for on - site repulsions of @xmath2=2@xmath5 , 4@xmath5 , and 8@xmath5 . for details , see the main text . , scaledwidth=50.0% ] however , being already more sophisticated than the sr framework , @xcite the extension @xcite mentioned above still essentially describes a given ground and/or excited state in terms of a single symmetry - projected configuration . this certainly limits the amount of correlations that can be accessed for those states . a more correlated description is encoded in a mr scheme . @xcite here , one resorts to a set of symmetry - broken hf states @xmath14 and superposes their goldstone manifolds @xmath15 . in this way , a given state with a well defined set of quantum numbers is expanded in terms of @xmath16 nonorthogonal symmetry - projected configurations @xcite that are optimized with the help of the ritz variational principle @xcite applied to the projected energy . there are several ways to perform the self - consistent optimization of the intrinsic hf states @xmath14 within a mr approach . one possible vap strategy is represented by the resonating hf @xcite ( reshf ) scheme within which all the determinants @xmath14 are optimized at the same time . another vap strategy is represented by the few determinant @xcite ( fed ) approach where , the hf transformations @xmath17 are optimized one - at - a - time . in both the reshf and fed schemes , the corresponding configuration mixing coefficients are determined through resonon - like equations . @xcite we note that there is no need for the fed expansion to be short , as its name would imply , although this is a desirable feature . in the present study , we keep the acronym to remain consistent with the literature . @xcite hybrid mr approximations are also possible . for example , one could optimize @xmath18 states using the reshf scheme and @xmath19 states using the fed one . the fed methodology has already been used @xcite to compute ground state energies , spin - spin correlation functions ( sscfs ) in real space , magnetic structure factors ( msfs ) as well as spin - charge separation tendencies in the sfs of half - filled 1d hubbard lattices of different sizes in the weak , intermediate - to - strong , and strong interaction regimes . we have shown @xcite that short reshf and fed expansions can provide an accurate description of chemical systems such as the nitrogen and water molecules along the entire dissociation profile , as well as an accurate interconversion profile among the peroxo and bis(@xmath20-oxo ) forms of [ cu@xmath12o@xmath12]@xmath22 comparable to other state - of - the - art quantum chemical methods . recent calculations , @xcite have also considered the complex binding pattern in the mo@xmath12 molecule . in addition to ground state properties , the excited few determinant @xcite ( excited fed ) scheme has also been used @xcite to treat excited states , with well defined quantum numbers , as expansions in terms of nonorthogonal symmetry - projected configurations . as a byproduct of vap calculations , the excited fed provides a ( truncated ) basis consisting of a few gram - schmidt orthonormalized states , each of them expanded in a given number of nonorthogonal symmetry - projected configurations , which may be used to perform a final diagonalization of the hamiltonian to account for a more correlated description of both ground and excited states . let us stress , that each of the mr approaches already mentioned has its own advantages and drawbacks . a reshf wave function is stationary with respect to arbitrary changes in the hf transformations @xcite @xmath23 ( i=1 , @xmath24 , @xmath16 ) while a fed one displays stationarity only with respect to the last added transformation . therefore , the reshf wave functions become easier to work with in the evaluation of those properties depending on derivatives of the wave function . however , in a reshf optimization , @xmath25 hamiltonian and norm kernels have to be recomputed at every iteration while only @xmath26 kernels are required in an efficient implementation of the fed method . @xcite regardless of the fed and/or reshf vap strategy adopted , the mr approximations are not restricted by the dimensionality ( i.e. , they can be equally well applied to 1d and 2d systems ) and/or the topology of the considered lattices . on the other hand , one of the most attractive features of the mr approximations is that they offer compact wave functions , with well defined quantum numbers , whose quality can be systematically improved by increasing the number of symmetry - projected configurations included in the corresponding ansatz . @xcite obviously , one is always limited in practical applications to a finite number @xmath16 of symmetry - projected terms in the fed and/or reshf expansions . however , it should also be kept in mind that both the reshf and fed wave functions are nothing else than a discretized form of the exact coherent state representation of a fermion state @xcite and , therefore , become exact in the limit @xmath27 . all in all , we believe that symmetry - projected approximations , already quite successful in nuclear physics , @xcite lead to a rich conceptual landscape and deserve further attention in quantum chemistry @xcite and condensed matter physics . @xcite they also provide @xcite high quality trial states that can be used within the constrained - path monte carlo @xcite scheme , increasing the energy accuracy and decreasing the statistical variance as more symmetries are broken and restored . in this paper we apply , for the first time , the fed approach to doped systems . therefore , our main goal is to test its performance using benchmark calculations . to this end , we have selected the 1d hubbard model for which both exact and highly accurate density matrix renormalization group ( dmrg ) results can be obtained . for the sake of completeness and comparison we will also discuss half - filling results . in sec . [ theory ] , we briefly describe the key ingredients of our mr approach . for a more detailed account , the reader is referred to our previous works . @xcite in sec . [ results ] , we discuss the results of our calculations . we have first paid attention to lattices with @xmath13=30 sites and @xmath3=14 , 18 , 22 , 26 , 30 electrons as illustrative examples . calculations have been performed for the on - site repulsions @xmath2=2@xmath5 , 4@xmath5 , and 8@xmath5 representing the weak , intermediate - to - strong ( i.e. , noninteracting band width ) , and strong interaction regimes , respectively . in sec . [ gsandcoenergies ] , we compare our ground state and correlation energies with the exact ones as well as with those obtained using other theoretical methods . the dependence of the correlation energies predicted for doped lattices with the number of transformations included in our mr ansatz is also discussed in the same section . the basic units of quantum fluctuations in the case of doped lattices are discussed in sec . [ unitsqf ] , where we consider the structure of the symmetry - broken determinants resulting from the fed vap procedure . next , in secs . [ modistribution ] and [ correfunctions ] , we benchmark our results for momentum distributions , sscfs , and density - density ( ddcfs ) correlation functions with dmrg ones obtained with the open - source alps software . @xcite a typical outcome of our calculations for sfs and dos @xmath28 is presented in sec . [ spectral ] , where we consider a lattice with @xmath13=30 sites and @xmath3=26 electrons at @xmath2=4@xmath5 . next , in sec . [ larger - sites ] , we illustrate the performance of the fed method in the case of larger lattice sizes . finally , sec . [ conclusions ] is devoted to the concluding remarks and work perspectives . we consider the 1d hubbard hamiltonian @xcite @xmath29 where the first term represents the nearest - neighbor hopping ( t @xmath30 0 ) and the second is the repulsive on - site interaction ( u @xmath30 0 ) . the fermionic @xcite spin-1/2 operators @xmath31 and @xmath32 create and destroy an electron with spin - projection @xmath33 on a lattice site j=1 , @xmath24 , @xmath13 . the operators @xmath34 = @xmath35 are the local number operators . we assume periodic boundary conditions and a lattice spacing @xmath36=1 . the starting point @xcite of our fed approach is a set of ghf determinants@xcite @xmath14 ( @xmath37 , n ) , which deliberately break several symmetries of the hamiltonian like rotational ( in spin space ) and spatial ones . to restore these broken symmetries , we explicitly use the spin @xmath38 and space group @xmath39 projection operators . @xcite in eq.([proj - s ] ) , @xmath40 is the rotation operator in spin space , the label @xmath41 stands for the set of euler angles , and @xmath42 are wigner matrices . @xcite in eq.([proj - space - group ] ) , @xmath43 and @xmath44 represent the dimension of the irreducible representation and the number of space group operations for a given lattice . on the other hand , @xmath45 is an irreducible representation @xcite while @xmath46 represents the corresponding point group symmetry operations parametrized in terms of the label g. the linear momentum @xmath47 is given in terms of the quantum number @xmath48 which takes the values allowed inside the brillouin zone ( bz ) . @xcite the high symmetry momenta @xmath19=0,@xmath49 are also labelled by the parity of the corresponding irreducible representation . @xcite in what follows we will not explictly write this label . the total projection operator can then be written in the following shorthand form @xmath50 where @xmath51 represents the set of ( spin and linear momentum ) symmetry quantum numbers and @xmath52 . the key idea of the fed approach is to superpose the set of degenerate goldstone states @xcite @xmath53 through the following ansatz @xmath54 , with well defined symmetry quantum numbers @xmath55 , in terms of @xmath16 nonorthogonal symmetry - projected configurations @xmath56 . the sum over @xmath57 in eq.([fed - state - general ] ) is necessary in order to remove an unphysical dependence of @xmath58 on the orientation of the ghf states @xmath14 . @xcite the fed wave function is formally similar to the one adopted within the reshf method . @xcite it is determined by applying the ritz variational principle @xcite to the energy ( independent of k ) @xmath59 written in terms of hamiltonian and norm kernels @xmath60 which require the knowledge of the symmetry - projected matrix elements between all the ghf determinants used in the expansion eq.([fed - state - general ] ) . in the case of the mixing coefficients , we obtain a resonon - like @xcite eigenvalue equation @xmath61 with the constraint @xmath62 ensuring the normalization of the solution . within the fed approach the energy eq.([ojo - ojo ] ) is varied only with respect to the last added ghf determinant @xmath63 keeping all the other transformations @xmath23 ( i=1 , @xmath24 , @xmath64 ) , obtained in previous chains of vap calculations , fixed . @xcite note that at variance with the reshf approximation , @xcite where all the transformations @xmath17 are optimized at the same time , the fed vap strategy optimizes them one - at - a - time . this is particularly relevant for alleviating our numerical effort if one keeps in mind that we use the most general symmetry - broken ghf states and therefore a full 3d spin projection eq.([proj - s ] ) is required . regardless of the adopted fed and/or reshf strategy , the variation with respect to the transformations @xmath23 can be efficiently parametrized with the help of the thouless theorem . @xcite all the fed calculations discussed in this paper have been carried out with an in - house parallel implementation @xcite of our vap procedure . we have used a limited - memory quasi - newton method @xcite to handle the optimization . note that the fed expansion of a given ground state @xmath58 by @xmath16 nonorthogonal symmetry - projected ghf configurations enlarges the flexibility in our wave functions , with respect to a sr description , to a total number @xmath65 of variational parameters . in sec . [ spectral ] , we will also discuss both the sfs and dos @xmath28 . the key point , is to superpose the goldstone hole @xmath66 and particle @xmath67 ( i= 1 , @xmath24 , @xmath68 ) manifolds in the wave functions of the @xmath69 and @xmath70 electron systems , respectively . the amplitudes of these superpositions are then determined through the corresponding generalized eigenvalue equations similar to eq.([hw-1 ] ) . with these ingredients at hand , as well as the fed solution @xmath58 , the sfs and dos can be computed according to eqs.(25 ) and ( 26 ) in our previous work.@xcite ) for @xmath13=30 lattices with 14 ( orange triangles ) , 18 ( magenta triangles ) , 22 ( cyan triangles ) , 26 ( green triangles ) , and 30 ( red triangles ) electrons are shown for @xmath2=2@xmath5 ( a ) , 4@xmath5 ( b ) , and 8@xmath5 ( c ) . dmrg results ( open black circles ) are also included in panels ( a ) to ( c ) for comparison . the fed values ( filled triangles ) are compared in panel ( d ) with a power - law [ eq.([power - law - kdist ] ) ] fitting ( dashed lines ) of the momentum distributions . , scaledwidth=42.0% ] finally , for the convenience of the reader we summarize the acronyms used in the present study for different types of slater determinants : * rhf is used for those symmetry - adapted states preserving all the symmetries of the hamiltonian eq.([ham - hubbard1d ] ) . * uhf is used for those states preserving the @xmath71-symmetry while possibly breaking all others . * ghf states are those that break all the symmetries of the considered hamiltonian . in this section , we discuss the results of our fed calculations . first , we pay attention to lattices with @xmath13=30 sites and @xmath3=14 , 18 , 22 , 26 , 30 electrons . results are presented for on - site repulsions @xmath2=2@xmath5 , 4@xmath5 , and 8@xmath5 , respectively . in sec . [ gsandcoenergies ] , we compare the ground state and correlation energies with the exact ones , as well as with results obtained using other theoretical approaches . we also discuss , for the case of doped lattices , the dependence of the correlation energies on the number @xmath16 of nonorthogonal symmetry - projected ghf configurations . next , in sec . [ unitsqf ] , we consider the structure of the intrinsic ghf determinants resulting from our vap procedure in the case of doped lattices . the momentum distributions are presented in sec . [ modistribution ] while the fourier transforms of the sscfs and ddcfs in real space are shown in sec . [ correfunctions ] . they are compared with those obtained within the dmrg framework retaining 1024 states in the renormalization procedure . on the other hand , in sec . [ spectral ] , we discuss spin - charge separation tendencies in the sfs and dos @xmath28 of a lattice with @xmath13=30 sites and @xmath3=26 electrons . finally , in sec . [ larger - sites ] , we illustrate the performance of the fed method in the case of larger lattices . in table [ table1 ] , we compare the exact @xcite and the predicted fed ground state energies for half - filled and doped lattices with @xmath13=30 sites and @xmath3=14 , 18 , 22 , 26 , 30 electrons . the corresponding @xmath55 = ( 0,0 ) ground states have a@xmath72 symmetry , i.e. , they are symmetric under the reflection @xmath73 . for @xmath2=2@xmath5 and 4@xmath5 , the fed energies shown in the table have been obtained by including @xmath16=60 nonorthogonal symmetry - projected ghf configurations in the ansatz eq.([fed - state - general ] ) . on the other hand , @xmath16=150 ghf transformations have been used for @xmath2=8@xmath5 . besides the rhf energies , we have included in table [ table1 ] the lowest possible hf solution for completeness . at @xmath2=2@xmath5 , the lowest - energy hf solution coincides with the rhf one in the case of doped systems while an uhf solution is obtained at half - filling . at @xmath2=4@xmath5 the hf state corresponds to an uhf wave function while for @xmath2=8@xmath5 we have found a ghf solution with predominant ferromagnetic character . in the same table , we also show our previous results @xcite ( fed@xmath6 ) , based on @xmath16=25 ghf transformations , and the reshf ones @xcite obtained with @xmath16=30 uhf transformations . we have computed the ratio @xmath74 in order to check how well the fed correlation energies reproduce the exact ones . for the other approximations , such a ratio is obtained from a similar expression . = 30 lattices with 14 ( orange diamonds ) , 18 ( magenta diamonds ) , 22 ( cyan diamonds ) , 26 ( green diamonds ) , and 30 ( red triangles ) electrons are shown for @xmath2=2@xmath5 ( a ) , 4@xmath5 ( b ) , and 8@xmath5 ( c ) . dmrg results ( black triangles ) are also included for comparison . starting with 26 electrons , all the curves have been successively shifted by 0.3 to accomodate them in a single plot . , scaledwidth=40.0% ] the first noticeable feature from table [ table1 ] is that , at most , the standard hf solutions account for 93.65 @xmath75 of the exact correlation energy . regardless of the filling and/or the interaction strength , the mr fed expansion clearly recovers a very large portion of correlation energy ( i.e. , @xmath76 @xmath77 99 @xmath75 ) in all cases studied . for the half - filled case as well as for the lattices with 26 and 22 electrons , the ground state and correlation energies improve the reshf and fed@xmath6 ones obtained in previous studies . @xcite on the other hand , the dmrg energies ( not shown in the table ) are exact to all the quoted figures . we have further used the dmrg results in secs . [ modistribution ] and [ correfunctions ] to benchmark our calculations for momentum distributions and correlation functions . from these results and the ones obtained in our previous work , @xcite we conclude that the fed scheme provides a reasonable starting point to obtain correlated ground state wave functions , with well defined symmetry quantum numbers , in both half - filled and doped 1d hubbard lattices . in addition , the method offers a systematic way to improve , through chains of vap calculations , the quality of such wave functions by increasing the number @xmath16 of nonorthogonal symmetry - projected ghf configurations included in the fed ansatz . this is illustrated in fig . [ en - vs - tra ] where we have plotted , as a function of the inverse @xmath78 of the number of transformations , the ratio @xmath76 for a doped lattice with @xmath13=30 sites and @xmath3=22 electrons . one sees that @xmath76 increases smoothly and approaches the exact result as the number of symmetry - projected configurations is increased . for example , a single symmetry - projected configuration provides @xmath76=93.14 @xmath75 , 85.67 @xmath75 , 87.08 @xmath75 while increasing the number of ghf transformations up to @xmath16=10 we obtain @xmath76=98.37 @xmath75 , 96.23 @xmath75 and 94.35 @xmath75 for @xmath2=2@xmath5 , 4@xmath5 and 8@xmath5 , respectively . some comments are in order here . first , since the nature of the quantum correlations varies for different doping fractions @xmath79 and on - site repulsions , one can expect that the number of ghf transformations required to obtain a given @xmath76 ratio depends on both of them . as already mentioned the fed wave functions become exact in the limit @xmath27 . in practice we are always limited to a finite number of nonorthogonal symmetry - projected configurations in the fed expansion and it is difficult to assert beforehand how many of them are required . therefore , their number @xmath16 should be tailored , through chains of vap calculations , so as to reach a reasonable accuracy not only in the ground state energy but also in other physical quantities like , for example , the spin - spin correlators . in the present study we have used a fixed number @xmath16=60 for both the weak and intermediate - to - strong interaction regimes while a larger number @xmath16=150 is required to obtain the energies reported in table [ table1 ] at @xmath2=8@xmath5 . as we will see later on in sec . [ correfunctions ] , a larger number of transformations is also required , especially close to half - filling , to improve the quality of the predicted correlation functions . this can be qualitatively understood from the crossover in the sscfs @xcite that explains how the antiferromagnetic spin correlation at half - filling grows near half - filling . one may then expect strong quantum fluctuations near half - filling , whose basic units ( see , sec . [ unitsqf ] ) can only be captured with larger fed expansions . the performance of the fed method for larger lattices will be discussed in sec . [ larger - sites ] . in our previous study @xcite of the half - filled 1d hubbard model we have considered two orders parameters , i.e. , the spin density ( sd ) @xmath80 and the charge density ( cd ) @xmath81 associated with an arbitrary symmetry - broken determinant @xmath7 , with j=1 , @xmath24 , @xmath13 being the lattice index . the comparison of the sd and cd computed with the standard uhf solution and the ones obtained using the ghf determinants @xmath14 resulting from the fed vap procedure , reveals that @xmath82 displays neutral [ i.e. , @xmath83=0 ] solitons @xcite whose translational and breathing motions can be regarded @xcite as the basic units of quantum fluctuations in the fed wave functions eq.([fed - state - general ] ) . the question naturally arises , as to what are the basic units of the quantum fluctuations captured within the fed vap optimization in the case of doped 1d lattices . among all the ghf transformations @xmath17 used to describe the lattice with @xmath13=30 sites and @xmath3=18 electrons ( see , sec . [ gsandcoenergies ] ) , we have selected some typical examples to plot in figs . [ solitons - n30-ne18-u2 ] , [ solitons - n30-ne18-u4 ] and [ solitons - n30-ne18-u8 ] the corresponding sd @xmath82 [ panels ( a ) to ( d ) ] and cd @xmath83 [ panels ( e ) to ( h ) ] as functions of lattice site . results obtained with the lowest - energy standard hf solutions are also included in the plots ( red ) for comparison . in the case of the rhf solution ( fig.[solitons - n30-ne18-u2 ] ) at @xmath2=2@xmath5 , the corresponding sd vanishes while the cd takes the constant value 0.4 . due to the symmetry - broken nature of the uhf solution ( fig.[solitons - n30-ne18-u4 ] ) , the corresponding sd and cd exhibit oscillating patterns around the expected values ( i.e. , 0 and 0.4 , respectively ) at @xmath2=4@xmath5 . in the case of the intrinsic ghf solution ( fig.[solitons - n30-ne18-u8 ] ) at @xmath2=8@xmath5 the sd displays a very fast oscillating pattern , which is a direct consequence of its predominant ferromagnetic character [ note the presence of the factor @xmath84 in the definition eq.([sdw1 ] ) ] . in fact , the energy of this intrinsic ghf solution ( i.e , -18.1922 t ) is only slightly lower than the one ( i.e. , -18.0974 t ) corresponding to a fully ferromagnetic uhf solution with all the spins aligned along the z - direction . on the other hand , the cd takes the constant value 0.4 . regardless of the considered interaction regime , the ghf determinants @xmath85 associated with the fed solution eq.([fed - state - general ] ) exhibit pairs of solitons ( black squares ) where the sd @xmath82 changes its sign . the space group projection operator provides a translational motion for such soliton pairs . when different determinants show soliton pairs with different widths , this can be interpreted as a breathing mechanism . isolated points ( blue circles ) where @xmath82 becomes zero are also apparent from figs . [ solitons - n30-ne18-u2 ] , [ solitons - n30-ne18-u4 ] and [ solitons - n30-ne18-u8 ] . in the case of doped lattices these new defects represent polarons . @xcite in addition , the cd @xmath83 displays local variations around the constant value 0.4 for all the considered on - site repulsions . other ghf determinants @xmath85 ( not shown in the figures ) display the same qualitative features . similar results also hold for other u values and lattices . therefore , in the case of doped 1d lattices , the ansatz eq.([fed - state - general ] ) superposes manifolds @xmath86 containing both solitons and polarons . one is then left with an intuitive physical picture @xcite in which the basic units of quantum fluctuations in 1d lattices can be mainly associated with the translational and breathing motions of neutral and charged solitons . however , in the case of doped 1d systems , a part of such fluctuations can also be described by polarons . within both the fed @xcite and the reshf @xcite schemes , the interference between the defects belonging to different symmetry - broken determinants @xmath14 is accounted for through eq.([hw-1 ] ) . ) ] plotted as functions of ln @xmath87 , with @xmath87 being the corresponding doping parameter . results are shown for on - site repulsions @xmath2=2@xmath5 ( red diamonds ) , 4@xmath5 ( blue diamonds ) , and 8@xmath5 ( green diamonds ) . dmrg values are plotted with open circles . a straight line has been fitted to guide the eye . for more details , see the main text . , scaledwidth=45.0% ] for a given set of quantum numbers @xmath55 , the momentum distribution @xmath88 can be computed as @xmath89 where @xmath90 is the @xmath91-occupation operator at wave vector q. note that , due to the particular form of the operator @xmath90 , the momentum distribution eq.([nkdist ] ) does not depend explicitly on @xmath92 . the ground state momentum distributions for @xmath13=30 lattices with 14 ( orange triangles ) , 18 ( magenta triangles ) , 22 ( cyan triangles ) , 26 ( green triangles ) , and 30 ( red triangles ) electrons are plotted in panels ( a ) , ( b ) and ( c ) of fig . [ momentum - distribution ] for on - site repulsions @xmath2=2@xmath5 , 4@xmath5 , and 8@xmath5 , respectively . regardless of the interaction strength , the fed and dmrg ( open black circles ) momentum distributions agree well . in all cases , we have obtained a jump at @xmath93 , with @xmath94 being the fermi momentum , that becomes less pronounced , especially at half - filling , for larger u values . such a jump is also found in calculations based on the exact solution at @xmath2=@xmath95 @xcite as well as in previous studies . @xcite as can seen from panel ( c ) , the momentum distribution presents a slight nonmonotonic behavior close to half - filling ( i.e. , @xmath3=26 ) due to a small feature near @xmath96 . previous works @xcite have shown that , contrary to an ordinary fermi liquid , the momentum distribution of the 1d hubbard model exhibits a power - law behavior around @xmath93 given by @xmath97 at half - filling or u @xmath98 @xmath95 . we have used the momentum distributions obtained with the fed approach to fit the functional dependence eq.([power - law - kdist ] ) . the results are shown in panel ( d ) of fig . [ momentum - distribution ] for @xmath2=8@xmath5 . despite the fact that the functional form eq.([power - law - kdist ] ) has been obtained using the exact @xmath2=@xmath95 solution , we observe that it nicely reproduces the trends in the fed ( and also dmrg ) results . let us now turn our attention to the predicted fed sscfs and ddcfs . we compare them with the corresponding dmrg values in the case of @xmath13=30 lattices with 14 ( orange triangles ) , 18 ( magenta triangles ) , 22 ( cyan triangles ) , 26 ( green triangles ) , and 30 ( red triangles ) electrons . this comparison will allow us to reveal to which extent the fed scheme can capture the main short , medium and long range features in these correlators , especially in the case of doped lattices . we have resorted to the momentum space representation ( i.e. , the fourier transforms ) of the sscfs and ddcfs in real space . for a discussion of the predicted fed sscfs in the case of half - filled lattices , the reader is also referred to our previous work . @xcite but for the fourier transforms of the ground state density - density correlation functions in real space . starting with 26 electrons , all the curves have been successively shifted by 0.2 to accomodate them in a single plot . , scaledwidth=40.0% ] the sscfs in real space are given by @xmath99 where the subindex @xmath100 accounts for the dependence with respect to the particular row of the space group irreducible representation used in the projection . let us stress that the wave functions @xmath101 eq.([fed - state - general ] ) are pure spin states where orbital relaxation is allowed . both conditions have already been shown to be important ingredients to improve the description of the long - range behavior of the sscfs . @xcite the fourier transforms ( ft - sscfs ) @xmath102 of the sscfs eq.([ss - cf ] ) are depicted in panels ( a ) , ( b ) and ( c ) of fig . [ spinspin - momentum - space ] for the ground states of the lattices considered in the present study at @xmath2=2@xmath5 , 4@xmath5 and 8@xmath5 , respectively . the first feature apparent from fig . [ spinspin - momentum - space ] , is the prominent antiferromagnetic peak at the wave vector q=@xmath49 in the case of the half - filled system . the peaks of the ft - sscfs always occur at @xmath96 . such peaks have also been found @xcite with the exact @xmath2=@xmath95 solution of the 1d hubbard model as well as in previous calculations . @xcite they are shifted towards smaller linear momenta as we move away from half - filling . in the same plot , we have also included the results of our dmrg calculations ( black triangles ) for comparison . it is satisfying to observe that the predicted fed values closely follow the trend obtained within the dmrg approach . the largest differences between the fed and dmrg ft - sscfs arise in the values of the corresponding peaks near half - filling for large u. for the lattice with 26 electrons the fed peak , obtained with @xmath16=150 ghf transformations , at @xmath2=8@xmath5 overestimates the dmrg one by 5 @xmath75 . on the other hand , using a smaller number @xmath16=60 of transformations , we obtain a poorer ( 10 @xmath75 overestimation ) description of the sscfs and ft - sscfs in the strong interaction regime . of the fed fourier - transformed density - density correlation functions in real space [ eq.([denden - cf ] ) ] are plotted as functions of ln @xmath87 for @xmath2=2@xmath5 ( red diamonds ) , 4@xmath5 ( blue diamonds ) , and 8@xmath5 ( green diamonds ) . dmrg values are plotted with open circles . for more details , see the main text . , scaledwidth=45.0% ] a previous study @xcite has shown the universal character of the crossover in the sscfs as we approach half - filling . as a result of this crossover , the peaks observed in the ft - sscfs at @xmath96 display a linear logarithmic dependence with the doping parameter @xmath103 . the results of our fed calculations are compared in fig . [ spinspin - maxima ] with the dmrg ones . for each u value , we have fitted a straight line to guide the eye . for @xmath2=2@xmath5 and 4@xmath5 , the fed and dmrg values agree well ( we have therefore included only the fitting of the former in the plot ) and exhibit an almost linear behavior as a function of ln @xmath87 . the same linear trend is observed at @xmath2=8@xmath5 though in this case the discrepancy between the fed and dmrg values , arising from a poorer description of the ft - sscfs in the former ( see , fig . [ spinspin - momentum - space ] ) , is larger . the ddcfs in real space can be computed as @xmath104 where @xmath105 and @xmath106 . their ft - ddcfs @xmath107 are shown in panels ( a ) , ( b ) , and ( c ) of fig . [ denden - momentum - space ] for @xmath2=2@xmath5 , 4@xmath5 , and 8@xmath5 , respectively . dmrg values ( black triangles ) are also plotted for comparison . similar to the momentum distributions and ft - sscfs already discussed above , the fed ft - ddcfs closely follow the trends observed in the dmrg ones , with the largest differences arising for @xmath108 in the case of the lattice with 26 electrons at @xmath2=8@xmath5 . from panels ( a ) , ( b ) , and ( c ) of fig . [ denden - momentum - space ] , we also observe the appearence of inflection points around @xmath96 which become less pronounced as u increases . they are plotted in fig . [ denden - maxima ] as functions of ln @xmath87 . we observe a shift down of the curves as u increases reflecting that the charge fluctuations decrease for larger u values . in addition , we note that the curves bend down more for larger on - site repulsions . a typical outcome of our calculations is shown in panel ( a ) of fig . [ spectral - dos ] , where we have plotted the hole ( black ) and particle ( blue ) sfs , as functions of the excitation energy @xmath109 , for a lattice with @xmath13=30 sites and @xmath3=26 electrons at @xmath2=4@xmath5 . calculations have been performed along the lines described in our previous work . @xcite the fed ground state of the system with @xmath3=26 electrons has been approximated by @xmath16=60 ghf transformations while for the systems with @xmath1 electrons we have superposed @xmath68=25 hole and particle manifolds ( see , sec . [ theory ] ) . smaller values @xmath68=5 and @xmath68=15 have also been investigated . however , increasing the number of hole and particle manifolds to @xmath68=25 leads to a shift of the main peaks and a redistribution of the strength of some of the peaks found for @xmath68=5 and @xmath68=15 as a result of the small number of configurations used in the calculations . in all cases , a lorentzian folding of width @xmath110=0.05 t has been used . we observe prominent hole peaks belonging to the spinon band . such a band resembles the one found in the half - filled case @xcite though the spectral weight of some of the hole peaks found in the latter is redistributed to the particle sector due to the presence of doping . another prominent feature of the sfs shown in panel ( a ) is the very extended distribution of the spectral weight for linear momenta @xmath111 . the splitting of the strength in the corresponding particle sfs reveals that the present finite size results at the intermediate - to - strong interaction regime already display spin - charge separation tendencies beyond a simple quasiparticle distribution as well as shadow features . this agrees well with results obtained using other theoretical approximations . @xcite in panel ( b ) of fig . [ spectral - dos ] we have plotted the dos @xmath28 corresponding to the lattice with @xmath13=30 sites and @xmath3=26 electrons ( black ) . for the sake of comparison , we have also included in the same panel the dos in the half - filled case ( red ) . the last one , has also been computed using @xmath16=60 and @xmath68=25 . it exhibits the characteristic hubbard gap @xcite and particle - hole symmetry . @xcite as can be seen from the figure , this particle - hole symmetry is lost in the doped case . as a result of states intruding the original gap , a smaller pseudogap is developed at @xmath2=4@xmath5 . however , our calculations indicate that such a pseudogap progressively disappears for increasing doping fractions @xmath112 . similar results also hold for both @xmath2=2@xmath5 and @xmath2=8@xmath5 though the effect is less pronounced in the latter due to the larger value of the gap at half - filling . [ table2 ] ccccccccccccc + @xmath13 & & @xmath3 & & u & & fed & & @xmath16 & & exact & & @xmath113 ( @xmath75 ) + + 34 & & 14 & & 4@xmath5 & & -23.0991 & & 60 & & -23.1137 & & 99.46 + + 34 & & 14 & & 4@xmath5 & & -23.1048 & & 100 & & -23.1137 & & 99.67 + + 34 & & 18 & & 4@xmath5 & & -26.4440 & & 60 & & -26.4842 & & 98.99 + + 34 & & 18 & & 4@xmath5 & & -26.4587 & & 100 & & -26.4842 & & 99.35 + + 34 & & 22 & & 4@xmath5 & & -27.8402 & & 60 & & -27.9207 & & 98.48 + + 34 & & 22 & & 4@xmath5 & & -27.8673 & & 100 & & -27.9207 & & 98.99 + + 34 & & 26 & & 4@xmath5 & & -27.1248 & & 60 & & -27.2553 & & 98.05 + + 34 & & 26 & & 4@xmath5 & & -27.1634 & & 100 & & -27.2553 & & 98.63 + + 34 & & 30 & & 4@xmath5 & & -24.2555 & & 60 & & -24.3967 & & 98.28 + + 34 & & 30 & & 4@xmath5 & & -24.2966 & & 100 & & -24.3967 & & 98.78 + + 34 & & 34 & & 4@xmath5 & & -19.4646 & & 60 & & -19.5258 & & 99.40 + + 34 & & 34 & & 4@xmath5 & & -19.4876 & & 100 & & -19.5258 & & 99.62 + + 34 & & 34 & & 8@xmath5 & & -11.0562 & & 60 & & -11.1473 & & 99.74 + + 34 & & 34 & & 8@xmath5 & & -11.0883 & & 100 & & -11.1473 & & 99.83 + + 50 & & 50 & & 2@xmath5 & & -42.1748 & & 60 & & -42.2443 & & 98.46 + + 50 & & 50 & & 2@xmath5 & & -42.1957 & & 100 & & -42.2443 & & 98.62 + + 50 & & 50 & & 4@xmath5 & & -28.1924 & & 60 & & -28.6993 & & 96.62 + + 50 & & 50 & & 4@xmath5 & & -28.2999 & & 100 & & -28.6993 & & 97.33 + + 50 & & 50 & & 8@xmath5 & & -15.7739 & & 60 & & -16.3842 & & 98.84 + + 50 & & 50 & & 8@xmath5 & & -15.9770 & & 100 & & -16.3842 & & 99.22 + now , we turn our attention to larger lattices . let us stress that our aim in this section is not to be exhaustive but to illustrate the fed results in such lattices . to this end the predicted ground state energies are compared with the exact ones in table ii . results are presented for @xmath13=34 lattices with @xmath3=14 , 18 , 22 , 26 , 30 , 34 electrons as well as for the half - filled lattice with @xmath13=50 sites . for the considered on - site repulsions we have performed two different sets of fed calculations based on @xmath16=60 and @xmath16=100 ghf transformations . the ratio of correlation energies @xmath113 , has been computed according to eq.([formulace ] ) . as can be seen from the table , the fed approach , based on @xmath16=60 symmetry - projected configurations , already provides @xmath114 @xmath30 98 @xmath75 . these values significantly improve the ones obtained with the standard hf approximation . for example , for the half - filled lattice with @xmath13=50 , the uhf approximation accounts for @xmath115=12.02 @xmath75 , 65.02 @xmath75 , 92.26 @xmath75 while @xmath114= 98.46 @xmath75 , 96.62 @xmath75 , 98.84 @xmath75 at @xmath2=2@xmath5 , 4@xmath5 , and 8@xmath5 , respectively . note that , for the same on - site repulsions , the variational monte carlo method provides @xmath113 values of around 87 @xmath75 , 92 @xmath75 , and 96 @xmath75 , respectively . @xcite for the same half - filled lattice , the ground state energies obtained with @xmath16=60 also improve the values reported in our previous work @xcite using a smaller number ( i.e. , @xmath16=25 ) of nonorthogonal symmetry - projected ghf configurations in the fed expansion eq.([fed - state - general ] ) as well as the ones provided by the reshf approximation @xcite based on @xmath16=30 uhf transformations . moreover , further increasing up to @xmath16=100 leads to @xmath114 @xmath77 99.35 @xmath75 in the case of lattices with @xmath13=34 sites . in the case @xmath116=50 , we have obtained @xmath114= 98.62 @xmath75 , 97.33 @xmath75 , 99.22 @xmath75 for @xmath2=2@xmath5 , 4@xmath5 , and 8@xmath5 , respectively . the previous results show that the fed scheme also provides a resonable starting point to obtain correlated wave functions in lattices larger than the ones considered in sec . [ gsandcoenergies ] . in particular , one sees that for increasing lattice sizes the quality of the fed wave functions can also be systematically improved , for different doping fractions , by increasing the number of symmetry - projected configurations included in the mr ansatz eq.([fed - state - general ] ) . we are unable at the moment to anticipate the number of symmetry - projected configurations necessary to achieve a given quality in the fed wave functions for arbitrary lattice sizes and doping fractions . nevertheless , we stress once more that the exact answer can always be approached in a systematic constructive way . finally , as a typical example of the results obtained for the lattices considered in this section , we have plotted in fig . [ momentum - distribution - n34 ] the momentum distributions corresponding to @xmath13=34 sites with 18 ( magenta triangles ) , 26 ( green triangles ) , and 34 ( red triangles ) electrons at @xmath2=4@xmath5 . we have resorted to @xmath16=100 nonorthogonal symmetry - projected states in the calculations . as can be seen from the figure , the momentum distributions still display the main feature already discussed in sec . [ modistribution ] for the case of half - filled and doped lattices with @xmath13=30 sites , i.e. , a jump at @xmath93 that becomes less pronounced at half - filling . ) for @xmath13=34 lattices with 18 ( magenta triangles ) , 26 ( green triangles ) , and 34 ( red triangles ) electrons are shown for @xmath2=4@xmath5 . , scaledwidth=42.0% ] in the present study we have applied , for the first time , the fed approach to doped hubbard systems . our main goal has been to test its performance using benchmark calculations in 1d hubbard lattices . half - filled systems have also been discussed . we have compared the results of our calculations for ground state and correlation energies with those obtained using other theoretical approximations . from the results of our previous study @xcite and those obtained in the present work based on a larger number of nonorthogonal symmetry - projected ghf configurations in the mr expansion , we conclude that the fed scheme provides a reasonable starting point to obtain ( compact ) correlated wave functions in both half - filled and doped 1d hubbard lattices . we have shown that the quality of such wave functions can be systematically improved , through chains of vap calculations , in a constructive manner by increasing the number of transformations in the corresponding fed ansatz . the analysis of the structure of the ( intrinsic ) symmetry - broken slater determinants resulting from our vap procedure reveals that they differ from that provided by the standard hf approximation . in particular , in the case of doped lattices they contain defects ( i.e. , solitons and polarons ) . the translational and breathing motions of such solitons can be regarded as the basic units of quantum fluctuations for the considered lattices . in addition , in the case of doped 1d systems , a part of the quantum fluctuations can also be described by polarons . on the other hand , though the fed results are not as accurate as the dmrg ones for the considered 1d lattices , our benchmark calculations for momentum distributions and correlations functions show that the former captures the main physics trends found in the latter . we have also shown that the fed scheme can be used to access dynamical properties of doped 1d hubbard lattices such as sfs and the dos . to this end , in addition to the corresponding fed ground state based on @xmath16 ghf transformations , we have considered anstze for the @xmath70 and @xmath69 electron systems that superpose @xmath68 particle and hole manifolds , respectively . for the case of a doped lattice with @xmath13=30 sites and @xmath3=26 electrons our scheme provides hole and particle sfs that qualitatively agree with results obtained using other theoretical frameworks . they point point to a distribution of the spectral strength beyond the one expected for a simple quasiparticle distribution and display spin - charge separation tendencies in all the considered interaction regimes . we believe that the finite size fed calculations already show that vap mr expansions , based on nonorthogonal symmetry - projected slater determinants , represent a useful theoretical tool to study low - dimensional correlated electronic systems with different doping contents that complement other existing approaches and could even be combined with them . within this context , we have recently used @xcite sr symmetry - projected wave functions as trial states within the constrained - path monte carlo framework . it has been shown that the use of such sr symmetry - projected states increases the energy accuracy while decreasing the statistical variance in calculations for large lattices . given the fact that short fed - like expansions encode a more correlated description of the considered systems , they might be seen as plausible candidates for further improving the previous results . the mr expansions used in the present study still offer a rich conceptual landscape for further development . in particular , small vibrations around symmetry - projected mean fields ( i.e. , symmetry - projected tamm dancoff and random phase approximations ) can be consistently formulated . @xcite such approximations can then be used to access a large number of excited states as required in studies of the optical conductivity in lattice models . @xcite such calculations are in progress and will be reported elsewhere . let us stress that symmetry - projected approximations are not restricted by the dimensionality of the considered lattices . in this respect , our studies have paved the way for applying the mr methodology to the systematic description of both ground and excited states of 2d square , honeycomb , kagome , and triangular lattices , as well as more involved multi - orbital hubbard models relevant to iron - based superconductors . @xcite in the realm of quantum chemistry , we plan to further enlarge our current developments @xcite for the molecular hamiltonian . this work was supported by the department of energy , office of basic energy sciences , grant no . de - fg02- 09er16053 . g.e.s . is a welch foundation chair ( c-0036 ) . some of the calculations in this work have been performed at the titan computational facility , oak ridge national laboratory , national center for computational sciences , under project chm048 . the authors also acknowledge a computational grant received from the national energy research scientific computing center ( nersc ) under the project projected quasiparticle theory . one of us ( r.r - g . ) would like to thank prof . k. w. schmid , institut fr theoretische physik der universitt tbingen , for valuable discussions . r. jrdens , n. strohmaier , k. gnter , h. moritz and t. esslinger , nature * 455 * , 204 ( 2008 ) ; u. schneider , l. hackermller , s. will , th . best , i. bloch , t. a. costi , r. w. helmes , d. rasch and a. rosch , science * 322 * , 1520 ( 2008 ) ; i. bloch , j. dalibard and w. zwerger , rev . * 80 * , 885 ( 2008 ) . k. m. shen , f. ronning , d. h. lu , w. s. lee , n. j. c. ingle , w. meevasana , f. baumberger , a. d. damascelli , n. p. armitage , l. l. miller , y. kohsaka , m. azuma , m. takano , h. takagi and z. x. shen , phys . lett . * 93 * , 267002 ( 2004 ) . j. l. stuber and j.paldus , _ symmetry breaking in the independent particle model_. fundamental world of quantum chemistry : a tribute volume to the memory of per - olov lwdin ; edited by e. j. brandas and e. s kryachko ( kluwer academic publishers : dordrecht , the netherlands , 2003 ) . k. penc , k. hallberg , f. mila and h. shiba , phys . lett . * 77 * , 1390 ( 1996 ) ; j. favand , s. haas , k. penc , f. mila and e. dagotto , phys . b * 55 * , r4859 ( 1997 ) ; a. parola and s. sorella , phys . rev . b * 45 * , r13156 ( 1992 ) ; m. ogata , t. sugiyama and h. shiba , phys . b * 43 * , 8401 ( 1991 ) ; m. ogata and h. shiba , phys . rev . b * 41 * , 2326 ( 1990 ) .

this is a survey of several exciting recent results in which techniques originating in the area known as additive combinatorics have been applied to give results in other areas , such as group theory , number theory and theoretical computer science . we begin with a discussion of the notion of an approximate group and also that of an approximate field , describing key results of freman - ruzsa , bourgain - katz - tao , helfgott and others in which the structure of such objects is elucidated . we then move on to the applications . in particular we will look at the work of bourgain and gamburd on expansion properties of cayley graphs on and at its application in the work of bourgain , gamburd and sarnak on nonlinear sieving problems .

0911.3354

the subject of _ additive combinatorics _ has grown enormously over the last ten years and now comprises a large collection of tools with many applications in number theory and elsewhere , for example in group theory and theoretical computer science . it has often been thought a little difficult to specify to an outsider exactly what the subject _ _ is__. however the following point of view seems to be gradually crystallising : additive combinatorics is the study of _ approximate mathematical structures _ such as approximate groups , rings , fields , polynomials and homomorphisms . it is interested in what the right definitions of these approximate structures are , what can be said about them , and what applications this has to other parts of mathematics . this article has three main aims . firstly , we wish to introduce the above point of view to a general audience , focussing in particular on the basic theory of approximate groups and approximate fields . secondly , we wish to sketch some beautiful applications of these ideas . one of them has to do with the beautiful picture on the cover ( for which we thank cliff reiter ) of an apollonian circle packing . it is classical that the radii of these circles are all reciprocals of integers . we will describe work of bourgain , gamburd and sarnak giving upper bounds for the number of circles at `` depth @xmath1 '' which have radius the reciprocal of a prime . thirdly , we wish to hint at the extraordinary variety of different areas of mathematics which have started to interact with additive combinatorics : geometric group theory , analytic number theory , model theory and point - set topology are just the ones we shall mention here . what we offer here is merely a taste of this viewpoint of additive combinatorics as the theory of approximate structure and of its applications . we do not touch on the theory of approximate polynomials ( a.k.a . the theory of gowers norms ) or say much at all about approximate homomorphisms , or anything about the many applications of these two notions . these topics will be covered in detail in forthcoming lecture notes of the author @xcite . before we can define an approximate group , we need to recall what an exact one is . we shall be concerned with finite groups , and we shall be working inside some ambient group @xmath2 , so that it makes sense to talk about multiplication of elements and taking inverses . if @xmath3 is a finite set then we shall write @xmath4 and @xmath5 . later on we shall see more general notations such as @xmath6 and @xmath7 whose meaning , we hope , will be evident . the following proposition , whose proof is an exercise in undergraduate group theory , gives various criteria for @xmath8 to be a subgroup or something very closely related . [ subgroup - char ] let @xmath8 be a finite subset of some ambient group @xmath2 . then we have the following statements we mean simply the proportion of all pairs @xmath9 for which @xmath10 also lies in @xmath8 . ] : 1 . @xmath11 , with equality if and only if @xmath12 for some subgroup @xmath13 and some element @xmath14 ; 2 . @xmath15 , with equality if and only if @xmath12 for some subgroup @xmath13 and some element @xmath16 in the normaliser @xmath17 ; 3 . the number of quadruples @xmath18 with @xmath19 is at most @xmath20 , with equality if and only if @xmath12 for some subgroup @xmath13 and some element @xmath14 ; 4 . the number of quadruples @xmath18 with @xmath21 is at most @xmath20 , with equality if and only if @xmath12 for some subgroup @xmath13 and some element @xmath22 ; 5 . @xmath23 , with equality if and only if @xmath24 for some subgroup @xmath13 ; 6 . @xmath25 , with equality if and only if @xmath24 for some subgroup @xmath13 . this would be a rather odd proposition to see formulated in an algebra text . however each of the statements ( 1 ) ( 6 ) has been constructed as an inequality in such a way that one may ask when equality approximately holds . before we can talk about such approximate variants , however , we need to know how approximate they will be . for this purpose we introduce a parameter @xmath26 ; larger values of @xmath27 will indicate more approximate , and thus less structured , objects is very different from the theory when , for example , @xmath28 . in the former setting , these approximate notions of subgroup constitute very small perturbations of the exact characterisations of proposition [ subgroup - char ] , and it turns out ( though is not always trivial to prove ) that the approximate objects so defined are small perturbations of the exact objects characterised by proposition [ subgroup - char ] . in conversation tao and i tend to refer to this regime as `` the 99% world '' , an expression i would not be averse to popularising . in this paper @xmath27 will be much larger , causing the theory to become much richer . tao and i call this the `` 1% world '' although the parameter @xmath27 could be anything between @xmath29 ( say ) and some small power of @xmath30 . ] . consider , then , the following list of properties that a finite set @xmath3 might enjoy . 1 . @xmath31 ; 2 . @xmath32 ; 3 . the number of quadruples @xmath18 with @xmath19 is at least @xmath33 ; 4 . the number of quadruples @xmath18 with @xmath21 is at least @xmath33 ; 5 . @xmath34 ; 6 . . now these are by no means as closely equivalent as the properties ( 1 ) ( 6 ) in proposition [ subgroup - char ] . let us give an example in which the ambient group is @xmath36 , and where we use additive rather than multiplicative notation . take @xmath37 . then it is easy to check that ( 3 ) and ( 4 ) are both satisfied with any @xmath38 , as @xmath39 , this being because there are @xmath40 solutions to @xmath41 with @xmath42 . on the other hand the sumset @xmath43 contains the numbers @xmath44 for each pair @xmath45 with @xmath46 . since these numbers are all distinct , we have @xmath47 , which means that if @xmath1 is sufficiently large depending on @xmath27 then ( 2 ) is not satisfied at all . rather remarkably , however , there is a sense in which the concepts ( 1 ) ( 6 ) are all _ roughly _ the same . to say what we mean by that , we introduce the following notion of rough equivalence rather than @xmath48 is a little arbitrary . the notion of rough equivalence will , in this survey , be applied to classes of sets ( such as ( 1 ) ( 6 ) here ) which are invariant under conjugation , in which case whether we multiply on the left or on the right in the definition makes little difference . ] . suppose that @xmath8 and @xmath49 are two finite sets in some ambient group and that @xmath26 is a parameter . then we write @xmath50 to mean that there is some @xmath16 in the ambient group such that @xmath51 . we say that @xmath8 and @xmath49 are roughly equivalent with parameter @xmath27 . the remarkable fact alluded to above is the following . for every choice of @xmath52 , suppose that some set @xmath8 satisfies condition ( @xmath53 ) in the list above with parameter @xmath27 . then there is a set @xmath49 satisfying condition ( @xmath54 ) with parameter @xmath55 ( some polynomial in @xmath27 ) such that @xmath56 . of particular note is the fact that the weak `` statistical '' properties ( 3 ) ( 6 ) imply the apparently more structured properties ( 1 ) and ( 2 ) . the proof of this is not at all trivial and the main content of it is the so - called balog - szemerdi - gowers theorem @xcite , generalised to the nonabelian setting in the fundamental paper of tao @xcite , as well as a collection of `` sumset estimates '' which , in the abelian case , i refer to collectively as _ ruzsa calculus _ @xcite . these estimates of ruzsa have such a classical role in the theory that we record two of them , in the abelian setting , here : we will mention these two again later on . [ ruzsa - sumsets ] suppose that @xmath57 and @xmath58 are finite sets in some ambient abelian group . then @xmath59 and @xmath60 . the original paper @xcite , the book @xcite or the notes @xcite may be consulted for more details . the first estimate is true in general groups but adapting the second requires care : see @xcite . it might be remarked that for many pairs ( @xmath53 ) and ( @xmath54 ) the correspondence between the relevant properties is a little tighter than mere rough equivalence , and often this can be useful . we shall not dwell on this point here . in the paper of tao just mentioned one finds what has become the `` standard '' notion of an approximate group . [ approx - gp - def ] suppose that @xmath8 is a finite subset of some ambient group and that @xmath26 is a parameter . then we say that @xmath8 is a @xmath27-approximate group if it is symmetric ( that is , if @xmath61 then @xmath62 , and the identity lies in @xmath8 ) and if there is a set @xmath63 in the ambient group with @xmath64 and such that @xmath65 . this notion , it turns out , is roughly equivalent to ( 1 ) - ( 6 ) above . it has certain advantages over ( 1 ) - ( 6 ) , for example as regards its behaviour under homomorphisms . it is also clear that an approximate group in this sense enjoys good control of iterated sumsets . thus , for example , @xmath66 , which means that @xmath67 , and similarly @xmath68 where @xmath69 denotes the set of all products @xmath70 with @xmath71 . from now on , when we speak of an approximate group , we will be referring primarily to definition [ approx - gp - def ] . with this discussion in mind , we can introduce what might be termed the rough classification problem of approximate group theory . [ rough - classification - problem ] consider the collection @xmath72 of all @xmath27-approximate groups @xmath8 in some ambient group @xmath2 . is there some `` highly structured '' subcollection @xmath73 such that every @xmath74 is roughly equivalent to some set @xmath75 with parameter @xmath76 , where @xmath76 depends only on @xmath27 ? this question has been addressed in a great many different contexts , starting with the freman - ruzsa theorem @xcite , which gives an answer for subsets of @xmath36 . here , it is possible to take @xmath73 to consist of the so - called _ generalised arithmetic progressions _ , that is to say sets @xmath49 of the form @xmath77 where @xmath78 , the quantities @xmath79 are `` lengths '' and @xmath80 . note in particular that , even in the highly abelian setting of the integers @xmath36 , approximate groups are a more general kind of object than genuine subgroups . that is , the theory of approximate groups , even up to rough equivalence , is a little richer than the theory of finite subgroups of @xmath36 ( which is in fact a rather trivial theory ) . the remarkable feature of the freman - ruzsa theorem is that the theory is not _ much _ richer , in the sense that generalised progressions remain highly `` algebraic '' objects . here is a list of other contexts in which the question has been at least partially answered : * abelian groups @xcite ; * nilpotent and solvable groups @xcite ; * free groups @xcite ; * linear groups @xmath81 @xcite , + @xmath82 @xcite , + @xmath83 @xcite , + @xmath84 ( sketched in @xcite ) , + `` bounded '' subsets of @xmath85 including @xmath86 @xcite , + @xmath0 @xcite , + @xmath87 @xcite + and @xmath88 for various other @xmath89 ( cf . @xcite ) . it is generally felt that approximate groups in quite general contexts can be controlled by objects built up from genuine subgroups and nilpotent objects ; this has been found in all of the examples just mentioned and is suggested by the famous theorem of gromov on groups with polynomial growth @xcite and the recent quantitative formulation of it due to shalom and tao @xcite . quite precise suggestions along these lines have been made by helfgott , lindenstrauss and others : more information on this can be found on tao s blog @xcite . before leaving this subject , we remark that even ( perhaps _ especially _ ) in the abelian case the issue of the dependence of @xmath76 on @xmath27 is far from being resolved . no examples are known to rule out the possibility that , with the right definition of the highly - structured class @xmath73 , @xmath76 can be taken to be polynomial in @xmath27 . in particular this is conjectured when the ambient group @xmath2 is @xmath90 , the countable infinite vector space over the field of two elements , and @xmath73 consists of ( finite ) subgroups of @xmath2 . this assertion ; see @xcite . ] is known as the _ polynomial freman - ruzsa conjecture _ @xcite , see also @xcite . it is equivalent to the following question which , for many years , i have tried to advertise to those for whom the word cohomology holds no fear . suppose that @xmath91 is a map such that @xmath92 takes on at most @xmath27 different values as @xmath93 range over @xmath94 . is it true that @xmath95 , where @xmath96 is linear and @xmath97 takes on at most @xmath55 different values ? it is a very easy exercise to obtain such a statement with @xmath98 but , so far as i know , no serious improvement of this bound has ever been obtained . ] . fortified by the experiences of the last section , one might attempt to come up with a sensible notion of an approximate _ ring_. a natural one , based perhaps on ( 2 ) in the previous section , is as follows : if @xmath8 is a finite subset of some ambient ring @xmath99 , we say that it is a @xmath27-approximate ring if @xmath100 and @xmath101 . here , of course , @xmath102 and @xmath103 as before . if @xmath104 is actually a field ( or an integral domain , which embeds into its field of fractions ) then we refer to @xmath8 as an approximate field , noting that approximate closure under division is essentially automatic in view of the rough equivalence of the notions ( 1 ) and ( 2 ) of approximate group . the study of approximate rings and fields was initiated in a paper of erds and szemerdi @xcite who proved ( though not in this language ! ) that a @xmath27-approximate subfield of @xmath36 must have size @xmath105 . they in fact conjectured that the right bound is @xmath106 for any @xmath107 , but this is so far unresolved ; the best exponent so far obtained is @xmath108 , a result of solymosi @xcite . note that this is equivalent to , and more usually stated as , the lower bound @xmath109 for all finite sets @xmath110 . in a different paper @xcite , solymosi generalised the erds - szemerdi result to show that every @xmath27-approximate subfield of @xmath111 has size at most @xmath112 . the general theory of approximate fields can be said to have started with the papers of bourgain - katz - tao @xcite and bourgain - glibichuk - konyagin @xcite , where ) approximate subrings of @xmath113 ; this restriction was removed in @xcite . very often the approximate fields under consideration in a given setting will have size at least @xmath114 , and for this reason one often refers to the bourgain - katz - tao theorem . ] the following result is established . [ bkt - theorem ] let @xmath115 be a prime and let @xmath116 . then every @xmath27-approximate subfield of @xmath113 has size at most @xmath117 or at least @xmath118 , for some absolute constant @xmath119 . the arguments on page 384 of @xcite , though they are phrased in a more limited context , essentially prove that every approximate subfield ( in an arbitrary ambient field ) must be roughly equivalent to a genuine finite subfield . this unifies the results of erds - szemerdi and solymosi with theorem [ bkt - theorem ] . in fact something similar is true for approximate rings , at least provided the ambient ring @xmath99 does not have `` too many '' zero divisors . these issues are comprehensively explored in an interesting paper @xcite of tao , which also has a very comprehensive collection of references . suppose that @xmath8 is a @xmath27-approximate field in some ambient field @xmath120 , that is to say both @xmath121 and @xmath122 are bounded by @xmath123 . we are going to sketch a proof that @xmath120 must contain a genuine subfield @xmath49 which is `` close '' to @xmath8 . the first step is to prove the _ katz - tao lemma _ , which asserts that @xmath8 ( or , more precisely , a large subset @xmath124 ) behaves in a manner which more strongly resembles that of a field : that is to say , @xmath8 is almost closed under both addition / subtraction and multiplication / division _ simultaneously_. to give a ( relevant ) example , the set @xmath125 has size @xmath126 , where @xmath127 . a slick proof of the katz - tao lemma is given in ( * ? * section 2.5 ) and we shall say little more about it here other than to remark that it involves a combination of ruzsa s sumset calculus and clever elementary arguments . personally , i regard it as part of the `` basic '' theory of approximate fields as opposed to the `` structural theory '' , to be regarded on the same level as the arguments used to show that definitions ( 1 ) ( 6 ) of an approximate group are roughly equivalent ( namely , ruzsa s sumset calculus and the balog - szemerdi - gowers theorem ) . in other words one might argue that the smallness of @xmath128 , or of similar objects , might be taken as an alternative _ definition _ of approximate field . suppose that @xmath8 is known to have this property , that is to say @xmath129 . then it is possible to establish an intriguing dichotomy : if @xmath130 then either @xmath131 or @xmath132 here , @xmath133 refers to the set of all @xmath134 with @xmath9 . to see why this is so , note that @xmath135 and that equality occurs if and only if the elements @xmath134 are all distinct . if equality does not occur then we may find a nontrivial solution to @xmath136 , which means that @xmath137 . but then every element of @xmath133 has the form @xmath138 and thus lies in @xmath128 . on the other hand , it is not hard to see using ruzsa calculus one requires an inequality controlling @xmath139 in terms of the @xmath140 . ] that if @xmath141 satisfy then for @xmath142 we have @xmath143 for absolute constants @xmath144 . if @xmath27 is a sufficiently small power of @xmath30 then this means that can not hold , forcing us to conclude that holds for @xmath145 . in this way we identify the set . ] of all @xmath145 satisfying as a genuine subfield of @xmath120 . straightforward additional arguments allow one to show that this subfield and @xmath120 are roughly equivalent . the original argument of @xcite is different and specific to @xmath113 but rather fun and , given the preceding discussion , it is not hard to say a few meaningful words about it . suppose for the sake of illustration that @xmath146 is a @xmath27-approximate subfield of size @xmath147 ; our task is to derive a contradiction if ( say ) @xmath148 . suppose that the katz - tao lemma has already been applied , so that @xmath128 , as defined above , is known to be small . the sets @xmath149 arising from ( boundedly many ) successive applications of this operation may also be shown to be small . now simple averaging arguments ( using nothing more than the fact that @xmath150 ) show that @xmath113 has dimension at most 100 ( say ) as a `` vector space '' over @xmath8 ; that is , there exist @xmath151 such that @xmath152 now @xmath153 can not be a `` basis '' for @xmath113 over @xmath8 since otherwise we would have @xmath154 , contrary to the assumption that @xmath115 is prime . thus there must exist some @xmath155 which is representable in two different ways as @xmath156 with @xmath157 . suppose , without loss of generality , that @xmath158 . then @xmath159 by substituting this expression for @xmath160 into , we see that @xmath161 repeating the argument gives ( without loss of generality ) @xmath162 and we may continue in this fashion to get , eventually , @xmath163 this is contrary to the fact that none of the sets @xmath164 has size much larger than that of @xmath8 itself , namely about @xmath165 , and a contradiction ensues . remarkably , the main `` dimension reduction '' idea here comes from a paper in point - set topology , namely edgar and miller s solution of the erds - volkmann ring problem @xcite ( that is , the statement that all borel subrings of @xmath166 have dimension @xmath167 or @xmath168 ) . see in particular lemma 1.3 of that paper . in this section we discuss the results of helfgott @xcite concerning approximate subgroups of @xmath169 helfgott proves the following . [ helfgott - theorem ] suppose that @xmath170 is a @xmath27-approximate group . then @xmath8 is roughly @xmath117-equivalent to an upper - triangular @xmath117-approximate subgroup of @xmath171 that is , an approximate subgroup conjugate to a set of upper - triangular matrices . rather than discuss helfgott s result itself , we discuss the analogous question for @xmath172 . here the answer is rather simpler and is given in @xcite , based on helfgott s work . [ sl2-c - theorem ] suppose that @xmath173 is a @xmath27-approximate group . then @xmath8 is roughly @xmath117-equivalent to an _ abelian _ @xmath117-approximate subgroup of @xmath172 . if desired the abelian approximate group could itself be controlled by a generalised progression using the freman - ruzsa theorem . we will only sketch a proof of the weaker result that @xmath8 is @xmath117-equivalent to an upper - triangular @xmath117-approximate subgroup , that is to say the direct analogue of helfgott s result . in @xmath172 , additional arguments may then be applied to prove theorem [ sl2-c - theorem ] ; there are no such arguments in @xmath171 , since the upper - triangular `` borel subgroup '' @xmath174 is not close to abelian . the proof of this weak form of theorem [ sl2-c - theorem ] is simpler than that of theorem [ helfgott - theorem ] in two major ways . firstly since @xmath111 is algebraically closed we may talk about eigenvalues , eigenvectors and diagonalization without the need to pass to an extension field , whereas over @xmath113 we would have to involve the quadratic extension @xmath175 . secondly , the structure of @xmath27-approximate subfields of @xmath111 is easy to describe : by the theorem of solymosi @xcite they are all sets of size at most @xmath112 . theorem [ bkt - theorem ] , by contrast , has to allow for those approximate fields which are almost all of @xmath113 . worse still , to handle @xmath171 helfgott must in fact classify approximate subfields of @xmath175 , and this involves the additional possibility of sets which are close to the subfield @xmath113 . for the sake of exposition , we will assume in the first instance that @xmath8 is a genuine finite subgroup of @xmath172 ; our task is to show that @xmath8 contains a large upper - triangular subgroup . when we have sketched how helfgott s argument looks in this case we will remark on the additional technicalities required to make the argument `` robust '' enough to apply to @xmath27-approximate groups . the key idea in helfgott s argument , referred to by subsequent authors as _ trace amplification _ , involves examining the set of traces @xmath176 we will sketch a proof that a large subset of this set of traces is a @xmath177-approximate subfield of @xmath111 of size greater than @xmath178 . this contradicts solymosi s theorem @xcite and so we must be in one of those degenerate situations . careful analysis of each of them leads to the conclusion that @xmath8 is roughly upper - triangular . the first degenerate situation to analyse is that in which @xmath179 is small , an appropriate notion of _ small _ being @xmath180 . now a linear algebra computation ( lemma 4.2 of @xcite ) confirms that if @xmath181 are elements without a common eigenvector in @xmath182 then the map @xmath183 is at most two - to - one . this , or rather the fact that something like this holds , is not at all surprising : indeed knowledge of @xmath184 together with the fact that @xmath185 provides four pieces of information which , generically , ought to more - or - less determine the four entries of the matrix @xmath16 . if @xmath8 contains two such elements @xmath186 then it follows that we have @xmath187 and so @xmath30 is also small as `` small '' . `` bounded by an absolute constant '' might be more appropriate . ] . if , by contrast , @xmath8 does not contain two such elements , and if @xmath188 , then it is easy to see that there is some @xmath189 which is an eigenvector for all of @xmath8 simultaneously . with respect to a basis containing @xmath190 , every matrix in @xmath8 is upper - triangular . suppose , then , that @xmath191 . in particular ( ! ) there is some element @xmath192 which is non - parabolic , or in other words @xmath193 ; such elements have distinct eigenvalues and so are diagonalisable . write @xmath124 for the set of non - parabolic elements ; then @xmath194 . now in @xmath172 the trace of a non - parabolic element @xmath195 completely determines the conjugacy class of @xmath195 . it follows that there is some non - parabolic @xmath192 such that the conjugacy class of @xmath8 containing @xmath195 has size at most @xmath196 . by the orbit - stabiliser theorem , the centraliser is for _ torus _ , the word used for such a subgroup in lie theory . ] @xmath197 has size at least @xmath198 . but by changing basis so that @xmath195 is in diagonal form ( with distinct diagonal entries ) it is not hard to check that @xmath199 consists entirely of diagonal matrices . no single trace can arise from more than two of these elements , and so @xmath200 . we shall show that the set @xmath201 is a @xmath177-approximate subfield of @xmath111 . noting that @xmath202this is contrary to solymosi s theorem . in order to do this we play around a little with traces . such playing around is most productive if , in the basis just selected , there is an element @xmath203 with @xmath204 . the absence of such an element is another degenerate situation to analyse , and once again one can check that @xmath8 must be either upper - triangular or else equal to one of the dihedral groups , each of which has an index two abelian subgroup . now let us note that @xmath205 this being a consequence of the fact that @xmath206 let us also note that @xmath207 where @xmath208 and @xmath209 , which means that @xmath210 in particular @xmath211 which , by ruzsa s inequalities ( theorem [ ruzsa - sumsets ] , applied with @xmath212 and @xmath213 ) implies that @xmath214 . this , together with , implies that @xmath99 is a @xmath177-approximate subring of @xmath111 . by solymosi s theorem this implies that @xmath215 , contrary to .@xmath216 in the above sketch we assumed , of course , that @xmath8 was actually a finite subgroup . however the argument was of a type that can be made to work for @xmath27-approximate groups also . to explain what we mean by this let us remark , rather vaguely , on how one or two of the steps adapt and then offer some general remarks . _ orbit - stabiliser theorem . _ if @xmath8 is a group and if @xmath217 then we used the fact that the size of the conjugacy class @xmath218 containing @xmath16 and that of the centraliser @xmath219 are related by @xmath220 . in fact we only used the inequality @xmath221 , giving us an element with large centraliser , and here is a simple way of seeing why this holds : all of the conjugates @xmath222 , @xmath61 , lie in @xmath218 , and so by the pigeonhole principle there must be distinct elements @xmath223 , @xmath224 , with @xmath225 . but then the elements @xmath226 , @xmath227 , centralise @xmath16 . now if @xmath8 is only a @xmath27-approximate group then this argument does not quite work , as there is no well - defined notion of conjugacy class . however a similar pigeonhole argument nonetheless gives us an element with large centraliser , since the conjugates @xmath228 are all constrained to lie in @xmath229 , a set of size at most @xmath230 . _ escape from subvarieties . _ a more interesting point concerns the location of an element of @xmath8 which , in a given basis , has no zero entries . whilst this might not be _ a priori _ possible if @xmath8 is only an approximate group , it _ is _ possible to find such an element in @xmath69 for some bounded @xmath1 ( independent of the approximation parameter @xmath27 ) , and this is good enough for helfgott s purposes . this is a special case of a nice lemma of eskin , mozes and oh @xcite called `` escape from subvarieties '' . the point is that the group @xmath231 generated by @xmath8 , if it is not almost upper - triangular , contains an element with no zero entries indeed this fact was used in the above sketch . in other words , @xmath231 is not contained in the subvariety of @xmath172 defined by @xmath232 the eskin - mozes - oh result states that in such a situation we can find `` evidence '' for the non - containment of @xmath231 inside @xmath233 by taking just a bounded number , depending only on @xmath233 , of products of @xmath8 . it seems , then , that certain types of argument in some sense those involving `` bounded length '' computations in the ambient group adapt very well from the traditional group theory setting to approximate groups . at the moment we do not have anything approaching a precise formulation of this principle and indeed at present the passage from the `` exact '' to the approximate is as much an art as a science . nonetheless , there seems to be merit in looking for `` bounded length '' proofs in traditional group theory which might be adapted to the approximate setting . perhaps this is as good a place as any to mention the remarkable recent paper of hrushovski @xcite in which tools from model theory have been applied to the study of approximate groups . the ramifications of that paper are not yet completely clear , but it looks as though theorem 1.3 of that paper together with some structure theory of algebraic groups ought to lead , without too much difficulty , to a proof of the following statement . suppose that @xmath234 is a @xmath27-approximate group . then there is a @xmath76-approximate group @xmath49 which is _ nilpotent _ and @xmath76-controls @xmath8 , where @xmath76 depends only on @xmath27 . it seems reasonable to conjecture that @xmath76 can be taken to depend polynomially on @xmath27 , although in their present form hrushovski s techniques will not give this . we move on now to applications of the theory of approximate groups . in this section we discuss the paper @xcite of bourgain and gamburd . this paper concerns _ expander graphs_. for the purposes of this discussion these are @xmath235-regular graphs @xmath236 on @xmath1 vertices for which there is a constant @xmath237 such that for any set @xmath63 of at most @xmath238 vertices of @xmath236 , the number of vertices outside @xmath63 which are adjacent to @xmath63 is at least @xmath239 . expander graphs share many of the properties of random regular graphs , and this is an important reason why they are of great interest in theoretical computer science . there are many excellent articles on expander graphs ranging from the very concise @xcite to the seriously comprehensive @xcite . a key issue is that of constructing explicit expander graphs , and in particular that of constructing _ families _ of expanders in which @xmath240 and @xmath241 are fixed but the number @xmath1 of vertices tends to infinity . many constructions have been given , and several of them arise from cayley graphs . let @xmath2 be a finite group and suppose that @xmath242 is a symmetric set of generators for @xmath2 . the cayley graph @xmath243 is the @xmath235-regular graph on vertex set @xmath2 in which vertices @xmath16 and @xmath244 are joined if and only if @xmath245 . such graphs provided some of the earliest examples of expanders @xcite . a natural way to obtain a family of such graphs is to take some large `` mother '' group @xmath246 admitting many homomorphisms @xmath247 from @xmath246 to finite groups , a set @xmath248 , and then to consider the family of cayley graphs @xmath249 as @xmath247 ranges over a family of homomorphisms . the work under discussion concerns the case @xmath250 , which of course admits homomorphisms @xmath251 for each prime @xmath115 . for certain sets @xmath248 , for example @xmath252 or @xmath253 spectral methods from the theory of automorphic forms may be used to show that @xmath254 is a family of expanders . see @xcite and the references therein . these methods depend on the fact that the group @xmath255 has finite index in @xmath250 and they fail when this is not the case , for example when @xmath256 in @xcite lubotzky asked whether the corresponding cayley graphs in this and other cases might nonetheless form a family of expanders , the particular case of being known as his `` 1 - 2 - 3 question '' . the paper of bourgain and gamburd under discussion answers this quite comprehensively , showing that all that is required is that the group generated by @xmath257 is not _ virtually abelian _ ( contains a finite index abelian subgroup ) . we will sketch the proof in the case that @xmath257 generates a nonabelian free subgroup of @xmath258 . this is essentially the most general case , since the kernel of the natural homomorpism from @xmath259 to @xmath260 is free and has index at most @xmath261 in @xmath259 . [ bg - theorem ] let @xmath250 as above and suppose that @xmath257 is a finite symmetric set generating a free subgroup of @xmath258 . then @xmath262 is a family of expanders . the notation we have introduced here is rather cumbersome , so let us write @xmath263 . for concreteness we will focus on the special case @xmath264 , where @xmath265 and @xmath266 are the matrices relevant to lubotzky s 1 - 2 - 3 question . the argument is almost identical in any other case . in this case , then , @xmath267 is the graph on vertex set @xmath171 in which @xmath16 is joined to @xmath244 if and only if @xmath268 is one of the elements @xmath269 or @xmath270 considered modulo @xmath115 . supposing that @xmath271 , each of these graphs is @xmath272-regular . the number of verices in @xmath267 is @xmath273 . the reader may be interested to see a proof , using the `` ping - pong '' technique of felix klein , that that the subgroup of @xmath258 generated by these @xmath8 and @xmath49 is indeed free . consider the natural action of @xmath8 and @xmath49 on the projective plane @xmath274 . write @xmath275 and @xmath276 and observe that @xmath63 and @xmath277 are disjoint and _ jouent au ping pong _ , that is to say @xmath278 and @xmath279 ( the origin of the name should be clear the `` players '' @xmath8 and @xmath49 hit the domains @xmath63 and @xmath277 back and forth as should the preference for the french term rather than the cumbersome `` play table tennis with one another '' . ) if the group generated by @xmath8 and @xmath49 is not free , then some nontrivial reduced word in @xmath8 and @xmath49 is equal to the identity , where `` reduced word '' means a finite word of the form @xmath280 with @xmath281 . the conjugate of such a word by an appropriate power of @xmath8 will still be the identity and will now have the form @xmath282 with @xmath283 . however by repeated application of the ping - pong properties we see that @xmath284 , certainly an impossibility since @xmath63 and @xmath277 are disjoint and @xmath285 is supposed to be the identity . following that slight digression let us focus once again on the cayley graphs @xmath267 , our aim being to prove that they form a family of expanders as @xmath115 ranges over the primes . to do this we begin by giving a spectral interpretation of the expansion property which we defined combinatorially above . for each @xmath115 we may consider the _ laplacian _ of the corresponding cayley graph , that is to say the operator @xmath286 defined by @xmath287 the eigenvalues of the laplacian lie in the interval @xmath288 $ ] . zero is certainly an eigenvalue , since @xmath289 . write the eigenvalues in ascending order as @xmath290 . it turns out the expansion properties of the graph @xmath267 ( in fact of _ any _ regular graph ) are intimately connected with the size of the second - smallest eigenvalue @xmath291 . the precise relation between the combinatorial property of expansion and this spectral property is discussed in section 2 of @xcite , but for our purposes we need only remark that it suffices to show that the second - smallest eigenvalue @xmath292 is bounded away from zero independently of @xmath1 ( in fact , this is also a necessary condition for expansion ) . the term _ spectral gap _ is used to describe this property : there is a gap at the bottom of the spectrum in which there are no eigenvalues apart from zero . to try to show that there is a spectral gap , consider the operator @xmath293 given by @xmath294 , that is to say @xmath295 the matrix consisting of the functions @xmath296 defined by @xmath297 if @xmath298 and @xmath167 otherwise . ] of @xmath199 is same thing as the adjacency matrix of the graph @xmath267 , that is to say the matrix whose @xmath299 entry is @xmath168 if @xmath300 and zero otherwise . the eigenvalues of @xmath199 are of course @xmath301 , @xmath302 , and it is a very well - known and easy to establish fact that the @xmath303th moment @xmath304 is equal to @xmath1 times @xmath305 , the number of closed walks of length @xmath303 from the identity to itself . it follows that we have @xmath306 note in particular that @xmath307 , since all the terms are non - negative . at first glance it looks as though the only way to use to bound @xmath308 away from zero would be to get rather precise estimates on @xmath305 , and in particular one would at the very least want to show that @xmath309 . however a remarkable observation , used earlier in related contexts by sarnak and xue @xcite and gamburd @xcite , comes into play . this is that any eigenspace of the laplacian is @xmath171-invariant , where the action of @xmath171 on @xmath310 is the right - regular one given by @xmath311 . in other words , any such eigenspace has the structure of a representation of @xmath171 and thus , by basic representation theory , decomposes into irreducible representation of @xmath171 . but by a classical theorem of frobenius all such representations have dimension at least @xmath312 . this means that @xmath313 for some @xmath314 , and hence from we in fact have the bound @xmath315 this enables a meaningful spectral gap ( lower bound on @xmath308 ) to be obtained from somewhat weaker upper bounds on @xmath305 . the main new content of @xcite , then , is to obtain those upper bounds on @xmath305 , the number of walks of length @xmath303 starting and finishing at the identity , for appropriate @xmath316 . a nice way of thinking about these walks is in terms of _ convolution powers _ of the probability measure @xmath317 on @xmath171 , where @xmath318 if @xmath319 and @xmath167 otherwise . this measure @xmath320 is a very singular or `` spiky '' probability measure , supported on just the four points @xmath269 and @xmath270 . now the convolution @xmath321 is supported on words of length at most two in @xmath269 and @xmath270 , or alternatively those @xmath16 in the graph @xmath267 which can be reached from the identity by a path of length two , the value of @xmath322 being @xmath323 times the number of paths of length two from the identity to @xmath16 . similarly higher convolution powers @xmath324 give @xmath325 times the number of paths of length @xmath316 from the identity to @xmath16 . the idea of the proof is to examine these convolution powers , showing that they become progressively more `` spread out '' until , for suitable @xmath316 , @xmath326 vaguely resembles the uniform measure @xmath327 which assigns weight one to each point of @xmath171 . then , in particular , @xmath328 , meaning that @xmath329 . combined with , this is enough to establish the desired spectral gap . the notion of a probability measure @xmath330 on @xmath171 being `` spread out '' may be quantified using the @xmath331-norm @xmath332 the @xmath331-norm of a delta measure @xmath333 is @xmath334 , which is huge , whilst that of the uniform measure @xmath327 is equal to one , the smallest value possible by the cauchy - schwarz inequality . it is not hard to show that convolution can not increase the @xmath331-norm , and so we have the chain of inequalities @xmath335 the aim is to show that this sequence is , in fact , rather rapidly decreasing . roughly speaking one shows that @xmath336 for some @xmath337 ; this @xmath338 turns out to be an appropriate choice to substitute into in order to reach the desired conclusions . it turns out that this sequence gets off to a rather good start . this is a consequence of an observation of margulis @xcite , namely that the freeness of the subgroup of @xmath258 generated by @xmath8 and @xmath49 persists to some extent even when reduced modulo @xmath115 . indeed let us take a reduced word @xmath339 with @xmath340 and suppose that this equals the identity when reduced modulo @xmath115 , that is to say in @xmath171 . lifting back up to @xmath258 we have @xmath341 but the freeness of the lifted group means that @xmath342 , and thus in order to be congruent to the identity mod @xmath115 the matrix @xmath343 must have at least one entry of size at least @xmath344 . but by some simple matrix inequalities this is impossible provided that @xmath345 for some absolute constant @xmath237 . it follows that the subgroup of @xmath171 generated by @xmath8 and @xmath49 is `` free up to words of length @xmath346 '' . in terms of the cayley graphs @xmath267 this means that up to retracing steps there is a unique walk of length @xmath316 between the identity and @xmath16 for any @xmath347 and for any @xmath348 . this implies that the measures @xmath349 , @xmath350 are already rather spread out . to quantify this ( and in particular to deal with the issue of `` retracing steps '' ) a result of kesten concerning random walks in the free group may be applied . the conclusion is that@xmath351 for some @xmath352 . this is good progress on the way to and represents a significant improvement on the initial bound @xmath353 . it is convenient to imagine , for the rest of the argument , that all probability measures @xmath330 on @xmath2 have the form @xmath354 for some set @xmath3 , the `` support '' of @xmath330 . whilst this is clearly not true , various ( somewhat technical ) decompositions into level sets may be used to reduce to this case . for such a measure we have @xmath355 and so the bound corresponds to @xmath356 , certainly a reasonable level of spreadoutness . the rest of the argument , which constitutes the heart of the paper , involves examining the convolution powers between @xmath357 and @xmath358 for a suitable @xmath359 , the aim being to establish . an application of the `` dyadic pigeonholing argument '' , used to great effect by bourgain in many papers , is employed : if @xmath360 is much larger than 1 , this means that the sequence can not decay too rapidly between @xmath357 and @xmath358 and so there must be two convolution powers @xmath349 and @xmath326 , @xmath361 , such that @xmath362 . let us be deliberately vague about the meaning of @xmath363 here . suppose that @xmath364 for some set @xmath3 . noting that @xmath365 , it is not hard to compute that the ratio @xmath366 is actually equal to @xmath367 times the number of quadruples @xmath368 with @xmath21 . this may be compared with condition ( 4 ) in the list of properties which are known to roughly characterise approximate groups . thus , being even rougher at this point , @xmath369 for some approximate group @xmath370 . note that the rough equivalence of ( 4 ) and other , more flexible definitions such as definition [ approx - gp - def ] is one of the deeper equivalences mentioned in [ approx - groups - sec ] , being reliant on the nonabelian balog - szemerdi - gowers theorem of tao @xcite . if @xmath371 is already all of @xmath171 then is telling us that @xmath349 is close to the uniform distribution , in which case so is @xmath358 , hence is established and we are done . if not then we apply helfgott s result , theorem [ helfgott - theorem ] , to conclude that @xmath371 is essentially upper - triangular , and hence that @xmath357 has significant mass on an upper - triangular subgroup of @xmath171 . the support of @xmath357 , however , consists of words of length at most @xmath372 in the generators @xmath269 and @xmath270 and , as we stated , these elements behave freely up to words of this length . this is highly incompatible with upper - triangularity , which in particular implies that we always have the commutator relation are 2-step solvable . ] @xmath373,[g_3,g_4 ] ] = { \operatorname{id}}.\ ] ] a pleasant group - theoretic argument formalises this incompatibility and allows one to show that any set of words of length at most @xmath372 in the generators @xmath269 and @xmath270 satisfying has size at most @xmath374 . this represents a tiny proportion of the set of all such words , which ( counted with multiplicity at least ) has cardinality @xmath375 . this contradiction finishes the sketch proof of theorem [ bg - theorem].@xmath216 before moving on , we wish to record , for use in the next section , a further observation concerning the measures @xmath349 . we sketched a proof that @xmath376 for some @xmath377 , that is to say @xmath358 vaguely resembles the uniform distribution on @xmath171 . by taking further convolutions and using the fact that irreducible representations have large degree once more , this may be bootstrapped to show that @xmath349 becomes exponentially well uniformly - distributed : @xmath378 for some absolute @xmath237 and for all @xmath316 . alternatively , such a statement can be deduced directly from the spectral gap property , as is done for example in @xcite . it is interesting to ask whether the arguments might adapt to deal with cayley graphs on @xmath379 with @xmath380 . a recent paper of bourgain and gamburd @xcite shows that this is the case when @xmath381 . the argument is , in large part , quite similar to the above , except of course that helfgott s theorem on approximate subgroups of @xmath171 must be replaced by his more difficult result @xcite on approximate subgroups of @xmath382 . there is one significant extra difficulty , however , which is that there are proper subgroups of @xmath382 which are not close to upper - triangular , an obvious example being a copy of @xmath171 . to deal with this a deep algebro - geometric result of nori @xcite is brought into play , which states that any proper subgroup of @xmath382 , @xmath115 sufficiently large , must satisfy a non - trivial polynomial equation . to obtain a contradiction , it must be shown that the set of words of length @xmath372 in the generators @xmath8 and @xmath49 ( say ) does not concentrate on the corresponding subvariety of @xmath84 , and here techniques from the theory of random matrix products and a certain amount of `` quantitative algebraic geometry '' are brought into play . in this section we discuss work of bourgain , gamburd and sarnak @xcite . the goal of _ sieve theory _ , traditionally viewed as a part of analytic number theory , is to find prime numbers or at least to say something about them . historically , the sieve arose through work of brun and merlin on the twin prime problem , that is to say the problem of finding infinitely may primes @xmath115 such that @xmath383 is also prime . whilst this remains a famous open problem , approximations to it have been found . for example , brun established the following result . there are infinitely many integers @xmath1 such that @xmath384 has at most @xmath385 prime factors . much later , chen @xcite replaced @xmath385 by @xmath386 . one way of stating this type of result is as follows : there are infinitely many @xmath1 for which both @xmath384 is a 3-_almost prime _ , that is to say a positive integer with at most 3 prime factors . the aim of @xcite is to discover almost primes in more exotic locales , and specifically in _ orbits _ of linear groups . we will sketch a proof of the following result . [ bgs - special - case ] let @xmath8 and @xmath49 be two matrices in @xmath258 generating a free subgroup . then there is some @xmath387 such that this group contains infinitely many @xmath387-almost prime matrices matrices , the product of whose entries is @xmath387-almost prime . henceforth we shall say `` almost prime '' instead of `` @xmath387-almost prime for some @xmath387 '' . we remark that in the specific case we focussed on in the last section , when @xmath265 and @xmath266 , the theorem as stated follows from classical sieve theory of the type used to prove brun s theorem . indeed ( for example ) @xmath388 , and the product of the entries here is @xmath389 , which will be almost prime for infinitely many @xmath1 by a simple variant of brun s analysis . the issue here is that the subgroup generated by @xmath8 and @xmath49 contains unipotent elements ( in this case both @xmath8 and @xmath49 are themselves unipotent ) . we start with a ( very ) elementary discussion of what a sieve is . suppose one has a finite set @xmath63 of integers and that one wishes to find primes or almost primes in @xmath63 . the most nave way to do this would be to try to adapt the sieve of eratosthenes , using the inclusion - exclusion principle to compute @xmath390 where @xmath391 is the set of elements of @xmath63 which are divisible by @xmath89 . unfortunately it is well - known that , even when @xmath63 is an extremely simple set such as @xmath392 , it is not generally possible to evaluate @xmath393 sufficiently accurately to avoid the error terms in this long sum blowing up . in this simple case just mentioned , for example , we have @xmath394 . however the floor function is rather unpleasant and it is tempting to write instead @xmath395 , but then one finds that there are so many @xmath396 errors that the sieve of eratosthenes becomes useless . by and large , _ sieve theory _ is concerned with what it is possible to say about primes or almost primes in @xmath63 given `` reasonably nice '' information about the size of the sets @xmath391 . although the sieve of eratosthenes is bad , other sieves fare rather better . these other sieves are generally cleverly weighted versions of the sieve of eratosthenes , but we will not dwell upon their construction here . a typical example of `` reasonably nice '' information about @xmath393 would be @xmath397 for all squarefree @xmath398 , where @xmath399 is some pleasant multiplicative function and the error @xmath400 is small in the sense that @xmath401 for some @xmath352 . for example , if @xmath402 then this is true with @xmath403 and for any @xmath404 . the fundamental theorem of the combinatorial sieve states , roughly speaking , that such information is enough to find almost primes in @xmath63 ; in fact , one can even estimate the number of almost primes . what is meant by `` almost prime '' that is , how many prime factors these numbers will have depends on how large we can take @xmath405 as well as on the so - called dimension of the sieve , which has to do with the average size of the quantities @xmath406 . we will not delay ourselves by expanding upon the details here . let us instead refer the reader to @xcite for the precise formulation convenient to the application there and to the book @xcite or the unpublished notes @xcite for a more wide - ranging discussion of sieves in general with full proofs . all we shall take from the preceding discussion is the notion that , given a finite set @xmath63 to be sieved in order to locate almost primes , we should be looking for good asymptotics for the size of the sets @xmath393 , @xmath89 squarefree . returning to theorem [ bgs - special - case ] , the first obvious question to answer is that of what the set @xmath63 to be sieved should be . the set in which we wish to find almost primes is @xmath407 now @xmath408 is of course an infinite set of integers . rather than truncate in the usual way and take @xmath409 , it is much more natural to truncate in a manner that respects the group structure more . this we do by taking @xmath410 where @xmath411 is the set of words of length @xmath316 in @xmath269 and @xmath270 and @xmath63 is counted with multiplicity so that @xmath412 . suppose that @xmath115 is a prime . then @xmath413 is equal to the number of words @xmath414 , counted with multiplicity , which , when reduced modulo @xmath115 , give rise to a matrix in @xmath171 with at least one zero entry . writing @xmath415 for the set of such matrices , it is easy to compute that @xmath416 . now the number of words @xmath414 which reduce modulo @xmath115 to some @xmath347 is , in the notation of the last section , precisely @xmath417 , and so @xmath418 however at the end of the last section we saw . ] that @xmath419 becomes very close to @xmath168 . in fact , in we noted the bound @xmath420 . using this we obtain @xmath421 where @xmath422 and @xmath423 for some @xmath352 . thus the expansion property of the cayley graphs @xmath424 gives exactly the kind of information that can be input into the combinatorial sieve ! there is , however , a very major caveat . what we have just said applies only to @xmath425 when @xmath115 is a prime , and for the sieve one must understand @xmath391 when @xmath89 is a general squarefree number . to do this requires the establishment of theorem [ bg - theorem ] for the family @xmath426 , where now @xmath89 ranges over all squarefrees and not just over primes . the broad scheme of the proof is the same , but every single ingredient must be generalised to the more general setting , starting from the classification of approximate subrings of @xmath427 . the situation here is more complicated because this ring will in general have many approximate subrings , namely @xmath428 with @xmath429 . one of the main technical results of @xcite ( occupying some 20 pages ) is the statement that , very roughly speaking , these are the only approximate subrings of @xmath427 . although this is a deeply technical argument of a type that this author would struggle to summarise meaningfully even to an expert audience , it might be compared with the 92-page proof @xcite of the corresponding assertion without the squarefree assumption on @xmath89 . thankfully this is not required for the present application . once the classification of approximate subrings of @xmath427 for @xmath89 squarefree is in place a suitable analogue of helfgott s argument is applied to roughly classify approximate subgroups of @xmath88 . even the statement of this result ( proposition 4.3 in the paper ) is rather technical . finally , the majority of the argument outlined in the last section in the case @xmath89 prime goes over without substantial change . this concludes our discussion of the proof of theorem [ bgs - special - case ] . to conclude this survey , we wish to mention a beautiful application , mentioned in the original paper @xcite and in other articles such as @xcite , of these nonlinear sieving ideas . this has to do with apollonian packings such as the one in the attractive image above . for a very pleasant and gentle introduction to apollonian packings , see @xcite . referring to figure [ figure-1 ] , inside each circle is an integer which represents the curvature of that circle , or in other words the reciprocal of the radius . some of the number theory associated with the integers that arise in this way is discussed in the letter @xcite where , for example , it is shown that infinitely many of these curvatures are prime and in fact that there are infinitely many touching pairs of circles with prime curvature . now a pleasant exercise in euclidean geometry gives a theorem of descartes , namely that the relation between the four integers @xmath430 inside four mutually touching circles is given by @xmath431 examples of quadruples @xmath432 which are related in this way and easily visible in the picture are @xmath433 and @xmath434 . take a quadruple @xmath435 of touching circles with curvatures @xmath436 there is another circle @xmath437 tangent to @xmath438 and @xmath439 , and it has curvature @xmath440 . to find a general relation between @xmath441 and @xmath442 we may note that @xmath443 are roots of regarded as a quadratic in @xmath441 and thereby obtain the relation @xmath444 this may of course be written as @xmath445 that is , if one starts with some fixed vector such as @xmath446 then one may obtain another quadruple of curvatures of circles in the apollonian packing by applying the matrix @xmath447 by playing the same game with @xmath448 and @xmath449 we can make the same assertion with the matrices @xmath450 @xmath451 and @xmath452 this leads naturally to consideration of the orbit @xmath453 , where @xmath454 every vector in which consists entirely of curvatures of circles in the apollonian packing . this puts us in a situation very similar to that studied in theorem [ bgs - special - case ] , except now we appear to be dealing with a subgroup of @xmath455 rather than of @xmath258 . it turns out , however , that this situation is essentially a two - dimensional one in disguise , and for this we need to add to the list of areas of mathematics we touch upon by hinting at lie theory and special relativity ! the matrices @xmath456 belong to @xmath457 , the subgroup of @xmath455 consisting of @xmath458 matrices with determinant one which preserve the quadratic form @xmath459 ( cf . ) . by the standard theory of quadratic forms ( over @xmath166 ) this is equivalent to the lorentz form @xmath460 , and so we may identify @xmath461 with the orthogonal group @xmath462 preserving this latter form . but it is very well - known that this group admits @xmath463 as a double cover : this is because the set @xmath464 may be identified with the set of @xmath465 hermitian matrices @xmath466 with determinant 1 via @xmath467 and so any @xmath468 gives rise to an element of @xmath462 via the transformation @xmath469 . by lifting to this double cover the group @xmath259 can be lifted to a subgroup of @xmath470)$ ] . the proof of theorem [ bgs - special - case ] goes through with relatively minimal changes , although once again the group generated by @xmath471 and @xmath472 contains unipotents and so , if the aim is simply to find infinitely many circles or pairs / quadruples of touching circles with almost - prime curvatures , more elementary approaches work just as well . those elementary approaches do not , however , give sharp quantitative results , whereas the techniques we have sketched do . to explain one such result , imagine figure [ figure-1 ] being generated as follows . start with the outer circle ( which has curvature @xmath473 ) and the three largest inner circles , with curvatures 13 , 21 and 24 . this is the _ first generation_. the second generation consists of those circles touching three from the first generation : they have curvatures 28,37,61 and 124 . the third generation contains those new circles touching three circles from either the first or the second generations : these have curvatures 45 , 60 , 69 , 93 , 124 , 132 , 133 , 156 , 220 , 292 , 301 and 325 . carry on in this vein : the @xmath1th generation will contain @xmath474 circles . the number of circles at generation @xmath1 which have prime curvature is bounded by @xmath475 , for some absolute constant @xmath119 . we conclude by remarking that there are some very interesting unsolved questions connected with apollonian packings @xcite . in that paper the very interesting question is raised of whether , in figure [ figure-1 ] , a positive proportion of all positive integers appear as curvatures . j. bourgain has recently indicated to me that he and elena fuchs have obtained new information on this question . see also @xcite for an asymptotic formula for the number of circles in the packing of curvature at most @xmath63 . it seems that the question of describing this set of integers more precisely remains open : are they given , from some point on , by finitely many congruence conditions ? i am very grateful to cliff reiter for his permission to use the image of an apollonian packing . in addition to the cited references , the notes from the fourth marker lecture of terence tao were very useful in preparing what is written here . i thank jean bourgain , emmanuel breuillard , alex gamburd , harald helfgott , alex kontorovich , peter sarnak and terry tao for their comments on an earlier draft of this survey . finally , i am very grateful to lilian matthiesen and vicky neale for correcting a number of typographical errors . j. bourgain , a. a. glibichuk and s. v. konyagin , _ estimates for the number of sums and products and for exponential sums in fields of prime order , _ j. london math . soc . ( 2 ) * 73 * ( 2006 ) , no . 2 , 380398 . a. lubotzky , _ cayley graphs : eigenvalues , expanders and random walks , _ surveys in combinatorics , 1995 ( stirling ) , 155189 , london math . lecture note ser . , * 218 * , cambridge univ . press , cambridge , 1995 . i. z. ruzsa , _ on the cardinality of @xmath484 and @xmath485 , _ combinatorics ( proc . fifth hungarian colloq . , keszthely , 1976 ) , vol . ii , pp . 933938 , colloq . math . jnos bolyai , 18 , north - holland , amsterdam - new york , 1978 .

this paper is concerned with dependence of discrete sturm - liouville eigenvalues on problems . topologies and geometric structures on various spaces of such problems are firstly introduced . then , relationships between the analytic and geometric multiplicities of an eigenvalue are discussed . it is shown that all problems sufficiently close to a given problem have eigenvalues near each eigenvalue of the given problem . so , all the simple eigenvalues live in so - called continuous simple eigenvalue branches over the space of problems , and all the eigenvalues live in continuous eigenvalue branches over the space of self - adjoint problems . the analyticity , differentiability and monotonicity of continuous eigenvalue branches are further studied .

1505.07531

a discrete sturm - liouville problem ( briefly , slp ) considered in the present paper consists of a discrete sturm - liouville equation ( briefly , sle ) @xmath0,\ ] ] and the boundary condition ( briefly , bc ) @xmath1 where @xmath2 is an integer , @xmath3 and @xmath4 are the forward and backward difference operators , respectively , i.e. , @xmath5 and @xmath6 ; @xmath7 , @xmath8 and @xmath9 are complex - valued sequences such that @xmath10 , \;\;w_n\neq0\;{\rm for}\ ; n\in [ 1,n];\ ] ] @xmath11 is the spectral parameter ; the interval @xmath12 $ ] denotes the set of integers @xmath13 ; and @xmath14 and @xmath15 are @xmath16 complex matrices such that @xmath17 throughout this paper , by @xmath18 , @xmath19 , and @xmath20 denote the sets of the complex numbers , real numbers , and integer numbers , respectively ; and by @xmath21 denotes the complex conjugate of @xmath22 . moreover , when a capital latin letter stands for a matrix , the entries of the matrix are denoted by the corresponding lower case letter with two indices . for example , the entries of a matrix @xmath23 are @xmath24 s . the dependence of the continuous sturm - liouville eigenvalues on the problems and its applications have been extensively studied ( cf . , @xcite ) . in @xcite , kong and zettl proved that the eigenvalues of continuous slps depend not only continuously but also smoothly on problems and then gave an expression for the derivative of the @xmath25-th eigenvalue with respect to a given parameter in the continuous slp . later , they , together with wu , gave a natural geometric structure on the space of bcs of continuous slps in @xcite . this structure is the base for studying the dependence of sturm - liouville eigenvalues on the bcs . in addition , they investigated the differentiability of continuous eigenvalue branches based on this structure , and discussed the relationships between the algebraic and geometric multiplicities of an eigenvalue . along another line , research on discrete spectral problems and their inverse problems has been of growing interest in recent years ( cf . , e.g.@xcite ) . atkinson @xcite and jirari @xcite studied spectral problems of second - order discrete scalar self - adjoint slps with separate bcs . in @xcite , the third author of the present paper with her coauthor chen investigated the following vector difference equation @xmath26,\;\;\;n\geq2,\ ] ] with the general boundary condition @xmath27 where @xmath28)$ ] , @xmath29 , and @xmath30(@xmath31 $ ] ) are hermitian @xmath32 matrices , @xmath33 and @xmath34 are nonsingular , @xmath35 for @xmath31 $ ] ; r and s are @xmath36 matrices with @xmath37 . it is evident that the bc is included in the bc . the spectral results obtained in @xcite will be used in the study of the multiplicity of eigenvalues in the present paper . further , the third author of the present paper with her coauthor lv studied error estimate of eigenvalues of perturbed problems , sufficiently close to a given sturm - liouville problem and , by some variational properties of the eigenvalues under a certain non - singularity condition in @xcite . so we obtained the continuous dependence of eigenvalues on problems under the nonsingularity condition . in chapter 2 of @xcite , kato investigated perturbation problems for linear operators in finite - dimensional spaces . he studied how the eigenvalues change with the operator , in particular when the operator analytically depends on a parameter . his method is based on function - theoretic study of the corresponding resolvent . obviously , the eigenvalue problem of the self - adjoint discrete slps consisting of - corresponds to that of an operator in a finite - dimensional space . note that the operator defined by - may be multi - valued since @xmath38 and @xmath39 may not be uniquely determined by the bc , and the problem discussed in the present paper is dependent on multi - parameters . however , the operators are all single - valued and their perturbations are only referred to one single parameter in @xcite . so the results in @xcite can not be directly available in our study . in the present paper , we shall investigate dependence of eigenvalues on the slp consisting of and . there are two main motivations for our study . firstly , it is helpful to clarify the common features and differences between the class of continuous slps and that of discrete slps . secondly , it is hoped that findings of such work will form a theoretical foundation for numerical works on discrete slps and their inverse problems , and such numerical works will shed light on numerical works on continuous slps and their inverse problems . many results in the continuous case may be obtained from the corresponding results in the discrete case , via certain limit procedures , but not vice verse ; while some results in the discrete case have relatively direct proofs . in this way , shorter proofs of results in the continuous case may be found . this paper is organized as follows . in section 2 , we give topologies and geometric structures on various spaces of discrete slps , which are fundamental for further developments . in section 3 , we first discuss properties of the analytic and geometric multiplicities of eigenvalues of the discrete slps and their relationships , and then study continuous dependence of eigenvalues on the problems . in section 4 , we investigate some fundamental properties of continuous eigenvalue branches including their analyticity , differentiability and monotonicity . finally , several examples illustrating results of these sections are presented in section 5 . [ r11 ] we shall apply the results obtained in the present paper to study some other topics about discrete sturm - liouville problems , including dependence of the @xmath25-th eigenvalue on problems , inequalities among eigenvalues for different problems , and index problems for eigenvalues in our forthcoming papers . in this section , we shall first introduce the topologies and geometric structures on the spaces of discrete sles , bcs , and self - adjoint bcs , separately , and then give the geometric structures of the spaces of discrete slps and self - adjoint discrete slps . on the one hand , unlike in the continuous case , the space of discrete sles in this paper has an easy and obvious structure . on the other hand , the space of bcs and the space of self - adjoint bcs have the same geometric structures as those in the continuous case . let the discrete sle be abbreviated as @xmath40 . then the space of discrete sles can be written as @xmath41 and is equipped with the topology deduced from the complex space @xmath42 . bold faced lower case greek letters , such as @xmath43 , are used to denote elements of @xmath44 . the subspace @xmath45 of @xmath44 has its obvious meaning . for convenience , the maximum norm on @xmath42 will be used : @xmath46}\left\{|1/f_n|,|q_n|,|w_n|\right\}\right\}.\ ] ] note that @xmath44 is a connected open subset of @xmath42 . similarly , @xmath45 is an open subset of @xmath47 and has @xmath48 connected components , two of which are @xmath49,w_n>0\;{\rm for}\ ; n\in [ 1,n]\},\\[1.0ex ] \omega_n^{+,\mathbb{r},+}:=\{(1/f , q , w)\in\omega_n^{\mathbb{r}}:f_n>0\;{\rm for}\ ; n\in [ 0,n],w_n>0\;{\rm for}\ ; n\in [ 1,n]\}.\end{array}\ ] ] we also set @xmath50\},\end{array}\ ] ] which has @xmath51 connected components . since equivalent linear algebraic systems of the form define the same bc following @xcite , we will take the quotient space @xmath52 equipped with the quotient topology , as the space of bcs , where @xmath53 gl(2,\mathbb{c}):=\{2\times2\ ; { \rm comlplex \;matrix}\ ; t:{\rm det}\ ; t \neq0\}.\end{array}\vspace{-0.05cm}\ ] ] note that @xmath54 is an open subspace of @xmath55 implies that @xmath56 if @xmath57 , @xmath58 . each bc is an equivalence class of coefficient matrices of systems of the form ; that is , an element of @xmath59 we use @xmath60 $ ] to denote the bc represented by the system . bold faced capital latin letters , such as @xmath61 , are also used for bcs . note that the space of bcs is independent of the equation either in the continuous or in the discrete case , and so it has the same topology and geometric structure in the discrete case as that in the continuous case . for convenience , we present them as follows . we refer to theorems 3.1 and 3.3 in @xcite for details . [ th21 ] the space @xmath62 of bcs is a connected and compact complex manifold of complex dimension 4 , while the space @xmath63 of real bcs is a connected and compact real - analytic manifold of dimension 4 . in addition , @xmath62 has the following canonical atlas of local coordinate systems : @xmath64:\;b_{ij}\in\mathbb{c},\ ; i , j= 1 , 2\right \},\vspace{3mm}\\ \mathcal{n}_{1,3}^{\mathbb{c}}=\left \{\left [ \begin{array } { cccc}1&a_{12}&0&b_{12}\\ 0&a_{22}&-1&b_{22}\end{array } \right ] : \;a_{i2},b_{i2}\in\mathbb{c},\ ; i=1,2\right \},\vspace{3mm}\\ \mathcal{n}_{1,4}^{\mathbb{c}}=\left \{\left [ \begin{array } { cccc}1&a_{12}&b_{11}&0\\ 0&a_{22}&b_{21}&1\end{array } \right ] : \ ; a_{i2},b_{i1}\in\mathbb{c},\ ; i=1,2\right \},\vspace{3mm}\\ \mathcal{n}_{2,3}^{\mathbb{c}}=\left \{\left [ \begin{array } { cccc}a_{11}&-1&0&b_{12}\\ a_{21}&0&-1&b_{22}\end{array } \right ] : \ ; a_{i1},b_{i2}\in\mathbb{c},\;i=1,2\right \},\vspace{3mm}\\ \mathcal{n}_{2,4}^{\mathbb{c}}=\left \{\left [ \begin{array } { cccc}a_{11}&-1&b_{11}&0\\ a_{21}&0&b_{21}&1\end{array } \right ] : \ ; a_{i1},b_{i1}\in\mathbb{c},\;i=1,2\right \},\vspace{3mm}\\ \mathcal{n}_{3,4}^{\mathbb{c}}=\left \{\left [ \begin{array } { cccc}a_{11}&a_{12}&-1&0\\ a_{21}&a_{22}&0&-1\end{array } \right ] : \ ; a_{ij}\in\mathbb{c},\;i , j=1,2\right \},\end{array}\ ] ] which are the so - called canonical coordinate systems on @xmath62 . the map @xmath65 @xmath66&\mapsto&(b_{11 } , b_{12 } , b_{21 } , b_{22})\end{array},\ ] ] is homeomorphic , and the coefficient matrix of the bc @xmath61 can be written as the form @xmath67 which is called the corresponding normalized form . there are similar statements about @xmath68 , @xmath69 , @xmath70 , @xmath71 , @xmath72 . one of @xmath73 , @xmath74 , @xmath75 , @xmath76 , is called a coordinate chart on @xmath77 . the above discussion gives a differentiable structure on @xmath77 . in addition , the space @xmath63 has a similar atlas of canonical coordinate systems , given by with @xmath18 replaced by @xmath19 everywhere . using the canonical coordinate systems on @xmath62 and @xmath63 , it is easy to determine how close to each other any two given bcs are . for a point @xmath78 in a differential manifold @xmath79 , we denote by @xmath80 the tangent space of @xmath79 at @xmath78 . now we give descriptions of the tangent spaces of @xmath62 and @xmath63 . if @xmath81 , then @xmath82 if @xmath83 , then @xmath84 etc . the tangent spaces of @xmath63 have similar descriptions . [ de21 ] * a bc @xmath60 $ ] is said to be self - adjoint if @xmath85 where@xmath86 and @xmath87 denotes the complex conjugate transpose of @xmath14 . we use @xmath88 to denote the space of self - adjoint bcs . * a bc is said to be degenerated if it can be written as the form @xmath89 & { \rm or } & \left [ \begin{array } { llll}0&0&1&0\\ 0&0&0&1\end{array } \right ] .\end{array}\ ] ] * a bc is said to be separated if it can be written as the form @xmath90.\ ] ] we use @xmath91 to denote the space of separated self - adjoint bcs . * a bc is said to be coupled if it is neither separated nor degenerated . [ r21 ] note that @xmath92 satisfies if and only if @xmath93 does , where @xmath94 . therefore , the self - adjointness is well - defined . moreover , the definition of self - adjointness is equivalent to definition 2.1 in @xcite . the following result gives the canonical forms of separated and coupled self - adjoint bcs , respectively . we refer to theorem 10.4.3 in @xcite for details . [ l21 ] the separated self - adjoint bcs can be written as @xmath95,\ ] ] where @xmath96;\ ] ] and the coupled self - adjoint bcs can be written as @xmath97,\ ] ] where @xmath98 [ r22 ] the coupled self - adjoint bcs together form an open subset of @xmath88 and @xmath99:\gamma\in[0,\pi),k\in sl(2,\mathbb{r})\right \}=\mathcal{n}_{3,4}^\mathbb{c}\cap\mathcal{b}^\mathbb{c}.\ ] ] the following result gives the topology and the geometric structure of @xmath88 . [ th22 ] the space @xmath88 equals the union of the following relative open subsets : @xmath100:\ ; a_{12},b_{22}\in\mathbb{r},z\in \mathbb{c}\right \},\vspace{3mm}\\ \mathcal{o}_{1,4}^{\mathbb{c}}=\left \{\left [ \begin{array } { cccc}1&a_{12}&\bar{z}&0\\ 0&z&b_{21}&1\end{array } \right ] : \ ; a_{12},b_{21}\in\mathbb{r},z\in \mathbb{c}\right \},\vspace{3mm}\\ \mathcal{o}_{2,3}^{\mathbb{c}}=\left \{\left [ \begin{array } { cccc}a_{11}&-1&0&\bar{z}\\ z&0&-1&b_{22}\end{array } \right ] : \ ; a_{11},b_{22}\in\mathbb{r},z\in \mathbb{c}\right \},\vspace{3mm}\\ \mathcal{o}_{2,4}^{\mathbb{c}}=\left \{\left [ \begin{array } { cccc}a_{11}&-1&\bar{z}&0\\ z&0&b_{21}&1\end{array } \right ] : \ ; a_{11},b_{21}\in\mathbb{r},z\in \mathbb{c}\right \}.\end{array}\ ] ] moreover , @xmath88 is a connected and compact real - analytic manifold of dimension 4 . direct calculations yield that all bcs in @xmath101 , @xmath102 , @xmath103 and @xmath104 are self - adjoint . evidently , all separated self - adjoint bcs are in these subsets by lemma [ l21 ] . moreover , by lemma 3.18 in @xcite , every coupled self - adjoint bc also lies in these subsets . thus , @xmath88 is the union of these subsets . as a canonical coordinate system on @xmath105 , @xmath68 is open in @xmath105 . it can be easily verified by a direct calculation that @xmath106 and hence @xmath101 is a relatively open set in @xmath88 . similarly , it can be shown that @xmath102 , @xmath103 and @xmath104 are relatively open sets in @xmath88 . since each of @xmath101 , @xmath102 , @xmath103 and @xmath104 is connected and intersects the other three , @xmath88 is connected . the proof of the rest part is the same as that of theorem 3.11 in @xcite . [ r23 ] * there are similar statements ( except the dimension ) about the space @xmath107 of real self - adjoint bcs to those about @xmath88 in theorem [ th22 ] . note that @xmath107 has dimension 3 and is a submanifold of @xmath108 ( see theorem 3.9 in @xcite for detailed discussion ) . * theorem [ th22 ] says that @xmath101 , @xmath102 , @xmath103 , and @xmath104 together form an atlas of local coordinate systems on @xmath88 . if @xmath109 , then the corresponding coordinate chart is given by @xmath110\to(a_{12},a , b , b_{22}),\ ] ] where @xmath111 , @xmath112 , @xmath113 . the normalized form of the coefficient matrix of a bc in @xmath114 is given naturally . others are given similarly . the above discussion gives a differentiable structure on @xmath115 . there is a similar statement about @xmath107 . * this result has been first mentioned in theorem 3.11 in @xcite , and a proof can be deduced from it . * the tangent spaces of @xmath88 and @xmath107 can be similarly described as in , , etc . here we omit the details , but they are mentioned in theorem [ th45 ] in section 4 about @xmath88 . [ de22 ] the discrete slp consisting of a discrete sle @xmath40 and a bc @xmath61 is said to be self - adjoint if @xmath40 is in @xmath116 and @xmath61 is self - adjoint . from the above discussions , we immediately deduce the following conclusions , which give the geometric structures of the spaces of discrete slps and self - adjoint discrete slps , respectively . [ th23 ] the space @xmath117 of discrete slps is a connected complex manifold of complex dimension @xmath118 , while the space @xmath119 of self - adjoint discrete slps is a real - analytic manifold of dimension @xmath118 and has @xmath51 connected components . [ r24]note that the differentiable structure of the product space @xmath117 can be given by that of @xmath44 and @xmath105 naturally . there is a similar statement about @xmath119 . in this section , we shall first discuss properties and relationships of analytic and geometric multiplicities of eigenvalues and then study continuous dependence of eigenvalues on problems . we shall point out that these relationships of the multiplicities of eigenvalues are very important in the following investigations because continuous eigenvalue branches are defined according to the analytic multiplicity of eigenvalues ( see theorem [ th35 ] and remark [ r32 ] ) , while the study on their properties , such as differentiability and monotonicity , is related to the geometric multiplicity of eigenvalues ( see section 4 ) . these relationships allow us to simplify the discussion about the above properties . in this subsection , we shall first study properties of geometric and analytic multiplicities of eigenvalues of discrete slps , separately , and then establish their relationships . especially , we shall show that they are equal by a direct method if the problem is self - adjoint . let @xmath120 denote the following natural difference operator corresponding to equation : @xmath121,\ ] ] and let @xmath122 = \left\{y = \{y_n \}_{n=0}^{n+1}:\ ; y_{n}\in \mathbb{c},0\leq n \leq n+1 \right\}.\ ] ] [ de31 ] * a complex number @xmath11 is called an eigenvalue of the discrete slp - if there exists @xmath123 $ ] which is non - trivial and solves the problem . the non - trivial solution @xmath124 is called an eigenfunction corresponding to @xmath11 . * the complex vector space spanned by the eigenfunctions for an eigenvalue is called the eigenspace corresponding to the eigenvalue , while the dimension of the eigenspace is called the geometric multiplicity of the eigenvalue . * an eigenfunction @xmath123 $ ] corresponding to an eigenvalue of the self - adjoint discrete slp is said to be normalized provided that @xmath125 [ r31 ] * a solution @xmath124 of is said to be non - trivial if there exists @xmath126 $ ] such that @xmath127 . * since has exactly 2 linearly independent solutions , the geometric multiplicity of each eigenvalue is either 1 or 2 . the following uniqueness of solutions of initial value problems of can be easily verified . [ l31 ] let @xmath128 $ ] , and @xmath129 , @xmath130 } \in \mathbb{c}$ ] . then , for each @xmath131 , the discrete sle has a unique solution @xmath132 $ ] satisfying @xmath133}.\ ] ] moreover , for each @xmath134 $ ] , @xmath135 and @xmath136 are polynomials of @xmath11 . for each @xmath137 , let @xmath138 and @xmath139 be the solutions of satisfying the initial conditions , respectively , @xmath140 then , by lemma [ l31 ] any solution of is a linear combination of @xmath138 and @xmath139 . set @xmath141 , \;\ ; \lambda\in\mathbb{c}.\ ] ] equation can be rewritten as @xmath142 f_{n-1 } \delta y_{n-1}+(q_n-\lambda w_n ) y_{n-1 } , \;\;\ ; n\in[1,n].\ ] ] so we have@xmath143.\ ] ] @xmath144 is called the transfer matrix of . by induction from , the leading terms of @xmath145 , @xmath146 , @xmath147 , and @xmath148 as polynomials of @xmath11 are @xmath149(-1)^{n}\left(w_n\prod\limits_{i=1}^{n-1}({w_i}/{f_i})\right ) \lambda^{n},&(-1)^{n}\left(({w_n}/{f_0})\prod\limits_{i=1}^{n-1}({w_i}/{f_i})\right ) \lambda^{n},\end{array}\ ] ] respectively . it follows from and that @xmath150.\ ] ] the following result says that the transfer matrix @xmath151 determines the eigenvalues of the problem for every bc . [ l32 ] a number @xmath152 is an eigenvalue of the discrete slp - if and only if @xmath11 is a zero of the polynomial @xmath153 therefore , either all the complex numbers are eigenvalues of the problem or the problem has only finitely many eigenvalues . fix a @xmath154 . let @xmath155 , where @xmath156 . inserting @xmath157 into the bc yields @xmath158 then @xmath157 is a non - trivial solution of and , and hence @xmath11 is an eigenvalue of the slp if and only if the determinant of the coefficient matrix in vanishes ; that is , @xmath159 . moreover , by lemma [ l31 ] @xmath160 is a polynomial of @xmath11 . this completes the proof . [ de32 ] * the polynomial function @xmath161 , unique up to a non - zero constant multiple , is called the characteristic function of the discrete slp , for its importance . * the order of an eigenvalue as a zero of @xmath162 is called the analytic multiplicity ( or simply just multiplicity ) of the eigenvalue . an eigenvalue is said to be simple if it has multiplicity 1 , while an eigenvalue of multiplicity 2 is called a double eigenvalue . the following result can be easily deduced from and via direct calculations . it is useful in some situations . [ l33 ] the characteristic function of the discrete slp - can be written as @xmath163 where @xmath164 @xmath165 [ th31 ] for each @xmath166 , among all boundary conditions , @xmath167 $ ] is the unique one that has @xmath11 as an eigenvalue of geometric multiplicity @xmath168 . from , a complex bc @xmath60 $ ] has @xmath11 as an eigenvalue of geometric multiplicity 2 if and only if @xmath169 . it follows that @xmath170 since rank@xmath171 from , @xmath15 is nonsingular by . thus , the only bc that has @xmath11 as an eigenvalue of geometric multiplicity 2 is the one @xmath167 $ ] . the proof is complete . now , we discuss relationships between the analytic and geometric multiplicities of an eigenvalue of an slp . [ th32 ] the analytic multiplicity of an eigenvalue is greater than or equal to its geometric multiplicity . it suffices to prove that the analytic multiplicity of any eigenvalue @xmath172 of geometric multiplicity 2 is at least 2 by ( ii ) of remark [ r31 ] . by theorem [ th31 ] , we only need to show that as an eigenvalue for the bc @xmath173 $ ] , @xmath174 has multiplicity @xmath175 . now , in this case it follows from lemma [ l33 ] that the characteristic function is given by @xmath176 by we obtain that @xmath177 -\psi'_n(\lambda)(f_n\delta\phi_n(\lambda ) ) -\psi_n(\lambda)(f_n\delta\phi'_n(\lambda ) ) \equiv0,\;\;\lambda\in \mathbb{c}.\end{array}\ ] ] then , and together yield that @xmath178 ; that is , the analytic multiplicity of @xmath174 is at least 2 . the proof is complete . we shall remark that the analytic and geometric multiplicities of an eigenvalue are not necessarily equal for an slp in general , see examples [ e51 ] and [ e52 ] . however , we shall show that they are equal in the case that the slp is self - adjoint . next , we consider the self - adjoint case . the self - adjoint slp - can be written as - by setting @xmath179 @xmath180 then @xmath181 obviously , @xmath182 . further , we have @xmath183 the following result is a direct consequence of theorem 4.1 in @xcite . [ l34 ] the sum of geometric multiplicities of all the eigenvalues of a self - adjoint slp ( 1.1)-(1.2 ) is @xmath184 , and moreover , all its eigenvalues are real . the following result can be deduced from theorem 4.3 in @xcite . we shall give an alternative and direct proof as follows . [ th33 ] the analytic and geometric multiplicities of each eigenvalue of a self - adjoint slp ( 1.1)-(1.2 ) are the same . for convenience , by @xmath185 and @xmath186 denote the sum of the analytic multiplicities and that of the geometric multiplicities of all the eigenvalues of the self - adjoint slp , respectively , by @xmath187 denote the distinct eigenvalues of the slp and by @xmath188 and @xmath189 denote the corresponding analytic and geometric multiplicities , respectively , where @xmath190 , @xmath191 . the rest proof is divided into two steps . * step 1 . * @xmath185=@xmath186 . we divide the discussion into three cases . * case 1 . * @xmath192 . by lemma [ l34 ] , @xmath193 . from , we get that @xmath194 by lemma [ l33 ] , , and one can get that the leading term of @xmath160 as a polynomial of @xmath11 is @xmath195 and then @xmath196 . hence , @xmath197 . * case 2 . * @xmath198 . by lemma [ l34 ] , @xmath199 . it follows from that @xmath200 , and then @xmath201 thus , @xmath202 by definition [ de32 ] . further , we have that @xmath203 by theorem [ th32 ] . hence , @xmath204 . * @xmath205 . by lemma [ l34 ] , @xmath206 . from , we get that @xmath207 by inserting into and by , the bc can be written as the form @xmath208 this implies that there exists a @xmath209 such that @xmath210 where @xmath211 by lemma [ l33 ] we get that @xmath212 then we have that @xmath213 which , together with , implies that the coefficients of the terms @xmath214 and @xmath215 of @xmath160 are equal to zero . thus , @xmath216 by definition [ de32 ] . again by theorem [ th32 ] we have that @xmath203 . hence , @xmath217 . * step 2 . * @xmath218 for @xmath219 . by theorem [ th32 ] , @xmath220 for @xmath221 . suppose that there exists a @xmath222 , @xmath223 , such that @xmath224 . then @xmath225 which contradicts to @xmath185=@xmath186 . therefore , the assertion holds and the entire proof is complete . the following result is a direct consequence of theorems [ th31 ] and [ th33 ] . it has been firstly given in theorem 4.3.1 in @xcite for a special class of separated self - adjoint boundary conditions and then in theorem 2.2.6 in @xcite for more general case . [ c31 ] assume that is in @xmath226 . then all the eigenvalues for each separated self - adjoint boundary condition are simple . in this subsection , using the topologies and geometric structures on the space of discrete slps introduced in section 2 , we shall show that sufficiently close discrete slps have near - by eigenvalues in a given bounded region of @xmath18 , and explain how such eigenvalues form the so - called continuous eigenvalue branches . in a general case , all the simple eigenvalues live in so - called continuous simple eigenvalue branches over the space of problems . however , we can get a better result in the self - adjoint case that all the eigenvalues , which may be simple or not simple , live in continuous eigenvalue branches over the space of the problems . to indicate the dependence of @xmath144 on the discrete sle , we write @xmath227 with @xmath228 . the following result can be deduced from and . [ l35 ] let @xmath229 . for each @xmath230 , there is @xmath231 such that if @xmath232 satisfies @xmath233 , then @xmath234 , \;\;|\lambda| \leq 1/\varepsilon,\ ] ] where @xmath235 is the maximum norm for the @xmath16 matrix . now , we are ready to prove the locally continuous dependence of eigenvalues on the corresponding discrete slp . [ th34 ] let @xmath236 be an eigenvalue of an slp @xmath237 with multiplicity @xmath238 , @xmath239 a bounded open subset of @xmath18 such that @xmath240 , and @xmath174 the only eigenvalue of @xmath241 in the closure @xmath242 of @xmath239 . then , there is a neighborhood @xmath243 of @xmath244 in @xmath245 such that each problem in @xmath246 has exactly @xmath238 eigenvalues in @xmath242 , counting multiplicity , and they all lie in @xmath239 . to indicate the dependence of @xmath160 on the slp @xmath247 , we write @xmath248 . let @xmath249 be a coordinate system in containing @xmath61 . for all bcs in @xmath249 , we compute the characteristic function using the corresponding normalized form of the coefficient matrices of the bcs . by lemma [ l35 ] , when @xmath250 is sufficiently close to @xmath247 , @xmath251 is also in @xmath249 , and @xmath252 is close to @xmath248 on @xmath242 . since @xmath248 ( or @xmath253 ) is a polynomial of @xmath11 and the degree of @xmath248 ( or @xmath252 ) in @xmath11 is less than or equal to @xmath254 by lemma [ l33 ] and , we can set @xmath255 @xmath256 where @xmath257 , @xmath258 is sufficiently close to @xmath247 , and the value of @xmath259 is dependent on @xmath260 , @xmath261 . since the boundary set @xmath262 is a compact subset of @xmath18 and @xmath263 for all @xmath264 , there exists @xmath265 such that @xmath266 one can choose @xmath230 satisfying that @xmath267 since @xmath268 uniformly for @xmath269 as @xmath270 , there exists a neighborhood @xmath271 of @xmath272 in @xmath245 such that @xmath273 which , together with , yields that @xmath274 therefore , we have by that @xmath275 by rouche s theorem in complex analysis , @xmath276 and @xmath248 have the same number of zeros in @xmath239 , counting order . therefore , the slp @xmath277 has exactly @xmath238 eigenvalues in @xmath242 , counting multiplicity , and they all lie in @xmath239 . this proof is complete . the following result is a direct consequence of theorem [ th34 ] . [ c32 ] for each @xmath278 , the set of discrete slps having at least @xmath238 eigenvalues , counting multiplicity , is open in @xmath279 . combining the reality of the eigenvalues for a self - adjoint discrete slp and theorem [ th34 ] yields the following result : [ c33 ] let @xmath280 be an eigenvalue of a discrete slp @xmath281 with multiplicity @xmath238 , @xmath282 a bounded open interval of @xmath19 such that @xmath283 , and @xmath174 the only eigenvalue of @xmath241 in the close interval @xmath284 $ ] . then , there is a neighborhood @xmath243 of @xmath244 in @xmath285 such that each problem in @xmath246 has exactly @xmath238 eigenvalues in @xmath284 $ ] , counting multiplicity , and they all lie in @xmath282 . based on the above discussion , we now form the continuous eigenvalue branches over @xmath286 or @xmath287 through a fixed eigenvalue . [ th35 ] * let @xmath174 be a simple eigenvalue for a discrete slp @xmath288 . then there is a continuous function @xmath289 defined on a connected neighborhood @xmath290 of @xmath291 in @xmath292 such that * * @xmath293 ; * * for any @xmath294 , @xmath295 is a simple eigenvalue of @xmath296 . * let @xmath174 be an eigenvalue of a self - adjoint discrete slp @xmath291 with multiplicity @xmath168 . fix a small @xmath297 such that @xmath174 is the only eigenvalue of @xmath291 in the interval @xmath298 $ ] . then , there are continuous functions @xmath299 , @xmath300 defined on a connected neighborhood @xmath301 of @xmath291 in @xmath287 such that * * @xmath302 ; * * @xmath303 for each @xmath304 ; * * for every @xmath304 , @xmath305 and @xmath306 are eigenvalues of @xmath296 . assertion ( 1 ) can be straightforward shown by theorem [ th34 ] . now , we show assertion ( 2 ) . by corollary [ c33 ] there is a neighborhood @xmath307 of @xmath244 in @xmath285 such that each problem @xmath296 in @xmath308 has exactly @xmath168 eigenvalues , which are denoted by @xmath309 and @xmath310 with @xmath311 , respectively , in @xmath312 $ ] and they all lie in @xmath313 . note that @xmath307 can be chosen such that it belongs to a connected component of @xmath285 by theorem [ th23 ] . then one can define two functions @xmath299 , @xmath300 such that @xmath314 and @xmath315 . it can be easily verified that @xmath299 and @xmath316 satisfy ( i)-(iii ) . next , we prove that @xmath299 , @xmath300 are continuous functions . fix a @xmath317 . if @xmath318 , there exists a neighborhood @xmath319 of @xmath296 in @xmath287 such that @xmath320 for each open interval @xmath282 satisfying @xmath321 $ ] and each @xmath322 by corollary [ c33 ] . now , we assume that @xmath323 . then , there exists a positive number @xmath324 such that @xmath325 and @xmath326 by corollary [ c33 ] there exists a neighborhood @xmath327 of @xmath296 in @xmath287 such that @xmath328 , @xmath329 for each @xmath330 satisfying @xmath331 and each @xmath332 . therefore , @xmath299 and @xmath316 are continuous in @xmath307 . this completes the proof . [ r32 ] * in the case of general discrete slp , that is , @xmath333 , any two such functions as @xmath334 in ( 1 ) of theorem [ th35 ] agree on the common part of their domains , which is still a neighborhood of @xmath291 in @xmath292 . so , by the continuous simple eigenvalue branch over @xmath286 through @xmath174 , we mean any such function . in the case of self - adjoint discrete slp , that is , @xmath335 , and @xmath174 is an eigenvalue with multiplicity @xmath168 , there are actually different functions on @xmath308 . locally they are the only such functions , to be called the continuous eigenvalue branches over @xmath336 through @xmath174 . * statement ( 1 ) of theorem [ th35 ] holds if we replace @xmath337 and @xmath338 by @xmath287 and @xmath339 , respectively . this gives the continuous simple eigenvalue branch over @xmath287 through @xmath174 . * there are similar results for subspaces of @xmath279 , such as @xmath340 and @xmath341 to ( 1 ) of theorem [ th35 ] . there are similar results for subspaces of @xmath336 , such as @xmath342 to ( 2 ) of theorem [ th35 ] and ( 2 ) of remark [ r32 ] . * the third author of the present paper , together with lv , obtained the continuous dependence of eigenvalues on problems under a non - singularity condition ( see theorem3.1 and corollary 3.1 in @xcite ) . the following result can be directly obtained by theorem [ th32 ] and ( 1 ) of theorem [ th35 ] . [ c34 ] let @xmath174 be a simple eigenvalue of a discrete slp @xmath343 , and @xmath334 the continuous simple eigenvalue branch over @xmath292 through @xmath174 . then , there exists a connected neighborhood @xmath290 of @xmath291 such that for each @xmath344 , @xmath295 has geometric multiplicity 1 . in this section , we shall investigate analyticity and differentiability of continuous eigenvalue branches under some assumptions on their multiplicities , and then study monotonicity of continuous eigenvalue branches of self - adjoint discrete slps on boundary conditions and equations , separately . in this subsection , we shall study the analyticity and differentiability of continuous simple eigenvalue branches . to do this , we need the following two lemmas ( see theorem 2.1.2 in @xcite and chapter v in @xcite , separately ) : [ l41 ] let @xmath345 , @xmath346 , be analytic functions of @xmath347 @xmath348 in a neighborhood of a point @xmath349 in @xmath350 , and assume that @xmath351 , @xmath346 , and@xmath352 then the equations @xmath353 , @xmath346 , have a uniquely determined analytic solution @xmath354 in a neighborhood of @xmath355 such that @xmath356 . moreover , the derivative formula in the neighborhood of @xmath355 is determined by @xmath357 [ l42 ] assume that @xmath358 is an open set and @xmath359 is a @xmath360 function for some @xmath361 . for @xmath362 , we write @xmath363 with @xmath364 and @xmath365 . assume that @xmath366 and @xmath367 let @xmath368 . then , there are open sets @xmath369 containing @xmath370 and @xmath371 containing @xmath372 with @xmath373 , and a @xmath360 function @xmath374 such that @xmath375 and @xmath376 further , for each @xmath377 , @xmath378 is the unique @xmath379 such that @xmath380 . [ th41 ] let @xmath381 be a simple eigenvalue of a discrete slp @xmath382 . then , the continuous simple eigenvalue branch @xmath334 defined on a neighborhood @xmath307 of @xmath383 in @xmath384 through @xmath174 is analytic . for a fixed discrete sle , the derivative of @xmath334 at @xmath385 $ ] is given by @xmath386 where the coefficient matrices @xmath387 and @xmath388 are defined by @xmath389 @xmath390 for the fixed problem @xmath383 , we assume that @xmath391\in \mathcal n_{1,2}^{\mathbb{c}}.\ ] ] for all bcs in @xmath392 , we compute @xmath162 using the corresponding normalized form of the coefficient matrices of the bcs . define @xmath393 @xmath394&\mapsto&(1/f',q',w',b'_{11 } , b'_{12 } , b'_{21 } , b'_{22}).\end{array}\ ] ] then @xmath395 is a coordinate chart on @xmath396 . for convenience , we set @xmath397 @xmath398 now , consider @xmath162 restricted to the region @xmath399 . by and lemma [ l32 ] , @xmath162 is a polynomial and hence an analytic function of all variables in @xmath399 . since @xmath400 is simple , we have that@xmath401 by lemma [ l41 ] @xmath402 is exactly the uniquely determined analytic solution to the equation @xmath403 on @xmath11 in a neighborhood of @xmath404 @xmath405 such that @xmath406 hence , @xmath334 is analytic at @xmath383 . if we replace @xmath383 by @xmath407 , a similar argument above yields that @xmath334 is analytic at @xmath408 . therefore , @xmath334 is analytic in the neighborhood @xmath307 of @xmath383 . fix a discrete sle . for any @xmath409 , applying to @xmath162 in a neighborhood of @xmath410 , one can deduce @xmath411 where@xmath412 which , together with , implies . if @xmath413 , where @xmath414 , @xmath74 , the proof can be completed by a method analogous to that used above . with the help of lemma [ l42 ] , one can deduce the following result using the same method above : [ th42 ] let @xmath415 be a simple eigenvalue of a self - adjoint discrete slp @xmath416 . then , the continuous simple eigenvalue branch @xmath334 defined on a neighborhood of @xmath383 in @xmath417 through @xmath174 is a @xmath418 function . [ r41 ] if @xmath415 is an eigenvalue of @xmath416 with multiplicity 2 , the continuous eigenvalue branch @xmath334 defined on a neighborhood of @xmath383 in @xmath417 through @xmath174 is not necessarily differentiable . please see examples [ e54]-[e57 ] for illustration . in this subsection , we shall investigate monotonicity of continuous simple eigenvalue branches on boundary conditions in several subsets of @xmath115 for self - adjoint discrete slps using the derivative formulas of continuous simple eigenvalue branches with respect to the corresponding bc . [ l43 ] let @xmath174 be a simple eigenvalue of a discrete slp @xmath419 , and @xmath334 the continuous simple eigenvalue branch defined on a neighborhood @xmath243 of @xmath420 in @xmath292 through @xmath174 . then , there is a continuous choice @xmath421 $ ] of eigenfunction corresponding to @xmath295 for all @xmath422 sufficiently close to @xmath291 . here , the continuity of @xmath423 means that for each @xmath424 sufficiently close to @xmath291 , @xmath425 as @xmath426 in @xmath292 . every eigenfunction of the slp @xmath291 corresponding to @xmath293 can be written as @xmath427 where @xmath428 , @xmath429 are dependent on @xmath291 . inserting into , we get @xmath430 set @xmath431 since @xmath174 is simple , @xmath295 is continuous in @xmath243 and has geometric multiplicity 1 for each @xmath432 by theorem [ th35 ] and corollary [ c34 ] , and one has that@xmath433 which implies @xmath434 for some @xmath435 . without loss of generality , we assume that @xmath436 . by replacing @xmath291 with @xmath432 in - , the similar equations denoted by @xmath437 still hold . obviously , there exists a neighborhood @xmath438 of @xmath291 with @xmath439 such that @xmath440 for each @xmath441 . it is evident that @xmath442 is a solution of @xmath443 for each @xmath441 by @xmath444 . hence , @xmath423 defined by @xmath445 is an eigenfunction corresponding to @xmath295 and continuous in @xmath446 by the fact that @xmath447 is continuous in @xmath243 . this completes the proof . [ r42 ] if @xmath292 is replaced by @xmath336 in lemma [ l43 ] , then it still holds for the self - adjoint discrete slps . by corollary [ c31 ] , all eigenvalues of every self - adjoint slp with separated bc are simple . with the help of the preceding lemma we can now give related derivative formulas of continuous simple eigenvalue branch @xmath334 with respect to the parameters of the separated self - adjoint bcs . to indicate the dependence of @xmath448 on the parameters @xmath449 and @xmath450 , we sometimes write @xmath451 for @xmath452 . [ th43 ] assume that @xmath174 is an eigenvalue of a self - adjoint discrete slp @xmath453 with @xmath454 . let @xmath123 $ ] be a normalized eigenfunction for @xmath174 , and @xmath334 the continuous eigenvalue branch over @xmath455 through @xmath174 . then , its derivatives are given by@xmath456 we first show that the first relation in holds . fix all the components of @xmath453 except @xmath449 . let @xmath457 . by corollary [ c31 ] , @xmath174 is a simple eigenvalue of @xmath453 . by remark [ r42 ] , we can choose an eigenfunction @xmath458 with respect to @xmath459 for @xmath460 sufficiently small such that @xmath461 as @xmath462 . from we get that @xmath463,\ ] ] where @xmath464:=y_n(f_n\delta\bar{z}_n)-(f_n\delta y_n)\bar{z}_n$ ] . hence,@xmath465-[y_n , z_n].\ ] ] the bc @xmath466 with respect to @xmath450 implies that @xmath467=0.\ ] ] in the case that @xmath468 , by the bc @xmath466 with respect to @xmath449 , together with and , we get that @xmath469 dividing both sides of by @xmath470 and taking the limit as @xmath462 , we obtain that @xmath471 in the other case that @xmath472 , by the bc @xmath473 with respect to @xmath449 , together with and , we get that @xmath474 and @xmath475 dividing both sides of by @xmath470 and taking the limit as @xmath462 , we obtain that @xmath476 hence , the first relation in follows . with a similar argument to that used in the above discussion , one can show that the second relation in holds . this completes the proof . the following result is directly derived from corollary [ c31 ] and theorem [ th43 ] . [ th44 ] assume that is in @xmath226 . then each continuous eigenvalue branch over @xmath455 is always strictly decreasing in the @xmath449-direction and always strictly increasing in the @xmath450-direction . the following results give the derivative formulas of continuous simple eigenvalue branch @xmath334 over @xmath287 with respect to @xmath115 . [ th45 ] fix @xmath477 . let @xmath174 be a simple eigenvalue of @xmath247 for a self - adjoint boundary condition @xmath478 , @xmath479 $ ] be a normalized eigenfunction for @xmath174 , and @xmath334 be the continuous simple eigenvalue branch of @xmath480 for @xmath481 through @xmath174 . then , we have the following derivative formulas : * when @xmath482 , @xmath483 for all @xmath484 in@xmath485 * when @xmath486 , @xmath487 for all @xmath484 in @xmath488 * when @xmath489,@xmath490 for all @xmath484 in @xmath491 * when @xmath492,@xmath493 for all @xmath484 in @xmath494 we first show that holds . for @xmath385\in\mathcal o_{1,3}^{\mathbb c}$ ] given by , we define @xmath495=\left [ \begin{array } { cccc}1&a_{12}+h_{12}&0&\bar{z}+\bar{h}_{22}\\ 0&z+h_{22}&-1&b_{22}+l_{22}\end{array } \right ] , \ ] ] where @xmath496 , @xmath497 , @xmath498 . obviously , @xmath499 as @xmath500 . by remark [ r42 ] , we can choose an eigenfunction @xmath501 of @xmath502 such that @xmath503 as @xmath499 . from , we get @xmath504-[y_n , z_n],\ ] ] which , together with the boundary conditions , @xmath505 and @xmath506 implies that @xmath507 further , using the following equalities @xmath508 where @xmath509 with @xmath510 , one can easily conclude that holds . with similar arguments , one can show that , , and hold . this proof is complete . next , we give an important application of theorem [ th45 ] . [ th46 ] assume that is in @xmath226 . then , in each of the coordinate systems @xmath511 , @xmath512 , @xmath513 , and @xmath514 in @xmath115 , every continuous eigenvalue branch is always increasing in the two real axis directions . for example , in @xmath511 , every continuous eigenvalue branch is always increasing in the @xmath515-direction and in the @xmath516-direction . note that the monotonicity in theorem [ th46 ] is not necessarily strict ( see example [ e54 ] ) . let @xmath517 and @xmath518 . by lemma 3.7 in @xcite we know that@xmath519:\;s\in\mathbb r\right\}\bigcup\left\{\left[\begin{array}{cccc } 0&1&0&0\\ 0&0&-1&b_{22}\end{array}\right]\right\}\ ] ] is a real - analytic loop . let @xmath334 be a continuous eigenvalue branch on a subset of @xmath520 . note that both @xmath521 and the curve @xmath522,\;\;\;\;\lambda\in \mathbb r,\ ] ] are real - analytic . so , either their intersection is discrete in @xmath520 or they agree completely . in the former case , @xmath334 is simple on a dense subset of its domain . fix an @xmath523 and @xmath524 such that @xmath525,\;\;s\in(s_0-\delta , s_0+\delta),\ ] ] lies in the domain of @xmath334 when @xmath526 is simple . assume that @xmath124 is a normalized eigenfunction for @xmath526 . then , by we get that @xmath527 since @xmath334 is simple in a dense open subset of @xmath528 , a similar argument implies that @xmath334 has a non - negative derivative at each point in the dense open subset . thus , @xmath334 is increasing in the dense open subset . assume that @xmath529 is an eigenvalue of multiplicity 2 with @xmath530 . by the continuity of @xmath334 one has that @xmath531 which , together with the monotonicity of @xmath334 in the dense open subset , implies that @xmath334 is increasing in a neighborhood of @xmath532 . thus , @xmath334 is increasing in @xmath528 . the latter case may happen for at most one pair @xmath533 and @xmath534 . the monotonicity of @xmath334 can be deduced from the former case by perturbing @xmath535 or @xmath516 . for example , by perturbing @xmath516 , namely by @xmath536 , @xmath334 is increasing on the domain according to discussion in the former case . so , setting that @xmath537 , we have that @xmath538 one can show the rest of the claims similarly . the proof is complete . [ l44 ] if @xmath539 and @xmath540 are eigenfunctions for eigenvalues of two self - adjoint discrete slps @xmath541 and @xmath542 , respectively , with the same bc @xmath478 , then@xmath543 first , consider the coupled self - adjoint bc @xmath61 . it follows from lemma [ l21 ] that @xmath544 $ ] for some @xmath545 and @xmath546 . so , we have that @xmath547 @xmath548 thus , @xmath549 hence , holds in the coupled case . the separated case can be treated similarly . the proof is complete . [ th47 ] fix @xmath550 . let @xmath174 be a simple eigenvalue of a self - adjoint discrete slp @xmath551 , @xmath123 $ ] a normalized eigenfunction for @xmath174 , and @xmath334 the continuous simple eigenvalue branch over @xmath226 through @xmath174 . then , @xmath552 for all @xmath553 denote @xmath554 and @xmath555 . by remark [ r42 ] we can choose an eigenfunction @xmath556 with respect to @xmath557 for @xmath558 sufficiently small such that @xmath461 as @xmath559 . for convenience , we set @xmath560 with @xmath561 , @xmath562 , @xmath563 . using and lemma [ l44 ] , we get that @xmath564 @xmath565 @xmath566 which yields that holds . this completes the proof . [ th48 ] fix a self - adjoint bc . then , each continuous eigenvalue branch @xmath334 over @xmath226 is decreasing in every @xmath567-direction with @xmath568 , independent of @xmath569 , and increasing in every @xmath570-direction ; while the positive parts of @xmath334 are decreasing in every @xmath571-direction , and the negative parts of @xmath334 are increasing in every @xmath571-direction . the proof is similar to that of theorem [ th46 ] and hence omitted . [ r43 ] the monotonicity in theorem [ th48 ] is not necessarily strict . please see examples [ e55]-[e57 ] for illustration . in this section , we shall give some examples to illustrate some results obtained in sections 3 and 4 . [ e51 ] consider the modified discrete fourier equation with @xmath576 , and let @xmath577 . then , by using lemma [ l33 ] and , the characteristic function for the separated bc @xmath578\ ] ] is @xmath579 thus , 0 is an eigenvalue for @xmath580 with geometric multiplicity 1 and analytic multiplicity 2 . so , the analytic and geometric multiplicities of an eigenvalue are not equal in general . note that the bcs @xmath580 are not self - adjoint . [ e52 ] take the discrete slp consisting of the modified discrete fourier equation and the separated bc @xmath581,\ ] ] where @xmath582 . by using lemma [ l33 ] and , direct calculations yield that the characteristic function of the problem is @xmath583 thus , if @xmath584 then 0 is an eigenvalue with geometric multiplicity 1 and analytic multiplicity 2 . in this case , the weight function @xmath585 is indefinite since @xmath586 by . [ e53 ] consider the discrete fourier equation with @xmath576 , and let @xmath587 . then , from lemma [ l33 ] and , we see that the characteristic function for the bc @xmath588 ( see ) is @xmath589 thus , the self - adjoint discrete slp consisting of the discrete fourier equation and @xmath590 has no eigenvalues ; and when @xmath591 , the only eigenvalue for @xmath588 is @xmath592 so , in this case , there is only one continuous eigenvalue branch over @xmath593 , i.e. , the function given by for @xmath591 . see figure 5.1 . similarly , the characteristic function for the bc @xmath594 is @xmath595 thus , the only eigenvalue for @xmath596 is @xmath597 ; and when @xmath591 , the two eigenvalues for @xmath598 are @xmath599 @xmath600 so , in this case , each continuous eigenvalue branch over @xmath601 is locally a part of one of the following functions : @xmath602 see figure 5.2 . note that for each @xmath598 , there are one or two continuous eigenvalue branches defined on a neighborhood of @xmath598 in @xmath603 . this example demonstrates that the index of the eigenvalue in a continuous eigenvalue branch over @xmath336 can change as the problem varies . [ e54 ] consider the discrete fourier equation with @xmath576 . let @xmath604 and @xmath605 . then , the characteristic function for the bc @xmath606 \in \mathcal o_{1,4}^{\mathbb c}\ ] ] is @xmath607 \lambda -(a_{12}-1 ) \lambda^2.\ ] ] thus , the two eigenvalues for @xmath608 are @xmath609 @xmath610 where @xmath611 let @xmath612 and @xmath613 . these are the two continuous eigenvalue branches over @xmath614 let @xmath615 . since @xmath616 for each @xmath617 , then @xmath618 and @xmath619 are two different and simple eigenvalues in this case . by lemma [ l42 ] and theorem [ th42 ] , @xmath299 and @xmath316 are the only two different @xmath418 eigenvalue branches on @xmath620 . on the other hand , setting @xmath621 , from and we deduce that @xmath622 see figure 5.3 . therefore , the two continuous eigenvalue branches @xmath623 and @xmath624 are not differentiable at @xmath625 . note that @xmath626 is the eigenvalue of multiplicity 2 of the problem when @xmath627 and @xmath621 . this demonstrates that the multiplicity assumptions in theorems [ th41 ] and [ th42 ] can not be omitted in general ( see remark [ r41 ] ) . [ e55 ] let @xmath630 . consider the 1-parameter family of self - adjoint discrete slps consisting of the discrete sles with @xmath631 and the bc @xmath632 \in \mathcal o_{1,3}^{\mathbb c}.\ ] ] then , by lemma [ l33 ] , direct calculations deduce that the characteristic function is @xmath633 thus , the two continuous eigenvalue branches are @xmath634 see figure 5.4 . therefore , in general , continuous eigenvalue branches are not differentiable with respect to @xmath635 , and their monotonicity with respect to @xmath635 is not strict . [ e56 ] let @xmath637 . take the 1-parameter family of self - adjoint discrete slps consisting of the discrete sles with @xmath638 and the bc given in . then , the characteristic function is @xmath639 thus , the two continuous eigenvalue branches are @xmath640 see figure 5.5 . therefore , in general , continuous eigenvalue branches are not differentiable with respect to @xmath570 , and their monotonicity with respect to @xmath570 is not strict . [ e57 ] let @xmath642 . consider the 1-parameter family of self - adjoint discrete slps consisting of the discrete sles with @xmath643 and the bc given in . then , the characteristic function is @xmath644 thus , the two continuous eigenvalue branches are @xmath645 see figure 5.6 . therefore , in general , continuous eigenvalue branches are not differentiable with respect to @xmath571 , and the monotonicity of their positive parts with respect to @xmath571 is not strict . w. peng , m. racovitan , h. wu , geometric aspects of sturm - liouville problems , v. natural loops of boundary conditions for monotonicity of eigenvalues and their applications , pac . j. appl . math . * 4 * ( 2006 ) 253273 .

we show that all the time - dependent statistical properties of the rightmost points of a branching brownian motion can be extracted from the traveling wave solutions of the fisher - kpp equation . we show that the distribution of all the distances between the rightmost points has a long time limit which can be understood as the delay of the fisher - kpp traveling waves when the initial condition is modified . the limiting measure exhibits the surprising property of superposability : the statistical properties of the distances between the rightmost points of the union of two realizations of the branching brownian motion shifted by arbitrary amounts are the same as those of a single realization . we discuss the extension of our results to more general branching random walks .

1011.4864

a branching random walk is a collection of points which , starting from a single point , diffuse and branch independently of the time , of their positions or of the other points , as in figure [ fig : brw ] . branching random walks appear in many contexts ranging from mathematics @xcite to biology @xcite . they can for example be used to describe how a growing population invades a new environment . in the one dimensional case , see figure [ fig : brw ] , there is , at a given time @xmath0 , a rightmost individual at position @xmath1 , a second rightmost at @xmath2 and so on . ( note that the rightmost @xmath3 at a time @xmath4 is not necessarily a descendant of the rightmost @xmath1 at time @xmath0 . ) the expected position @xmath5 of the rightmost individual as well as the probability distribution of its position @xmath1 around @xmath6 are well understood @xcite ; the goal of the present paper is to describe the statistical properties of the positions of all the rightmost points in the system , in particular the distribution of the distances between the two rightmost points , the average density of points at some fixed distance from the rightmost @xmath1 , etc . one motivation for studying these distances is that the problem belongs to the broader context of extreme value statistics @xcite : trying to understand the statistical properties of the rightmost points in a random set of points on the line is a problem common to the studies of the largest eigenvalues of random matrices @xcite , of the extrema of random signals @xcite , or of the low lying states of some disordered systems such as spin glasses @xcite . in fact , the points generated after some time @xmath0 by a branching random walk can be viewed as the energies of the configurations of a directed polymer in a random medium @xcite , and the distances between the rightmost points as the gaps between the low lying energy states . the most studied example of branching random walk is the branching brownian motion : one starts with a single point at the origin which performs a brownian motion and branches at a given fixed rate ( right part of figure [ fig : brw ] ) . whenever a branching event occurs , the point is replaced by two new points which evolve themselves as two independent branching brownian motions . while the number of points generated after some time @xmath0 grows exponentially with time , the expected position @xmath6 of the rightmost point increases only linearly with time @xcite . in one dimension , mc kean @xcite and bramson @xcite have shown that the probability distribution of the rightmost point is given by the traveling wave solution of the fisher - kpp equation , with a step initial condition . here we will see that all the statistical properties of the rightmost points can be understood in terms of solutions to the fisher - kpp equation with appropriate initial conditions @xcite . we will also show that the distribution of the distances between these rightmost points has a long time limit which exhibits the striking property of superposability : the distances between the rightmost points of the union of two realizations of the branching brownian motion have the same statistics as those of a single realization . this paper is organized as follows : in section [ statistics ] we introduce some generating functions useful to study random sets of points on the line and show how one can use them to obtain all the properties of these random sets . in section [ sec : bbmfkpp ] we show that , for the branching brownian motion , all these generating functions are solutions of the fisher - kpp equation . we also show that the distribution of all the rightmost points as seen from @xmath6 or , alternatively , as seen from @xmath1 , has a long time limit which can be computed as the delay of fisher - kpp traveling waves . this distribution has the property of superposability . in section [ quantitative ] , we present results , mostly numerical , on some specific aspects of the limiting distribution of points in the branching brownian motion , namely the distribution of the distance between the two rightmost points and the average density seen from the rightmost point . in section [ sec : disc ] we explain how the results on the branching brownian motion can be extended to more general branching random walks . finally , we study in section [ statmeas ] the distribution of all the rightmost points in a specific frame which depends on the realization and which was introduced by lalley and sellke @xcite . in this section , we introduce some useful quantities ( generating functions ) to characterize random sets of points on the line such that the number @xmath7 defined as @xmath8 is finite and vanishes for @xmath9 large enough . the first generating function one can define is @xmath10 from the knowledge of this function , one can extract the probability distribution function @xmath11 of the position @xmath9 of the @xmath12-th rightmost point . indeed , by definition of @xmath13 , @xmath14 where @xmath15 is the probability that there are exactly @xmath12 points on the right of @xmath9 . one can notice that @xmath16 is the probability to have less than @xmath12 points on the right of @xmath9 . the generating function of these sums is , from , @xmath17\lambda^2+\big[q_0(x)+q_1(x)+q_2(x)\big]\lambda^3+\cdots . \label{sumterms}\ ] ] but @xmath16 is also the probability that the @xmath12-th rightmost point , if it exists , is on the left of @xmath9 . therefore , @xmath18 where @xmath19 is the probability that the @xmath12-th rightmost point exists and is in the interval @xmath20 $ ] . ( note that @xmath21 is the probability that there are at least @xmath12 points on the line . ) the knowledge of @xmath22 gives in particular the average distances between the points : from ( [ pix ] ) , one can see that @xmath23,\notag\\ & = \langle x_1 \rangle-\lambda[\langle x_1 \rangle-\langle x_2 \rangle ] -\lambda^2[\langle x_2 \rangle-\langle x_3 \rangle]-\cdots,\end{aligned}\ ] ] where @xmath24 is the average position of the @xmath12-th point ( with the convention that @xmath25 if there are less than @xmath12 points in the system ) . therefore @xmath26 = \sum_{i\ge1 } \langle d_{i , i+1}\rangle \lambda^i , \label{psid}\ ] ] where @xmath27 is the average distance between the @xmath12-th and the @xmath28-th point . to obtain the correlations between the positions of pairs of points , one can start , for @xmath29 , from the generating function @xmath30 the coefficient in front of @xmath31 in the expansion of @xmath32 in powers of @xmath33 and @xmath34 is the probability that there are exactly @xmath12 points on the right of @xmath9 and @xmath35 points on the right of @xmath36 . as in , the coefficient of @xmath31 in the expansion of @xmath37 is the probability that there are less than @xmath12 points on the right of @xmath9 and less than @xmath35 points on the right of @xmath36 , which is also the probability that the @xmath12-th rightmost point ( if it exists ) is on the left of @xmath9 and the @xmath35-th rightmost point ( if it exists ) is on the left of @xmath36 . thus , for @xmath38 , @xmath39 where @xmath40 is the probability that both the @xmath12-th and @xmath35-th rightmost points exist and lie respectively in the intervals @xmath20 $ ] and @xmath41 $ ] . one can generalize ( [ defpsi],[defpsi2 ] ) by defining , for @xmath42 , the generating functions @xmath43 of the numbers @xmath44 of points on the right of positions @xmath45 , and get as in ( [ pix],[twopoint ] ) all the higher correlation functions . in that way , all the statistical properties of the measure can be derived from the knowledge of the generating functions ( [ defpsi3 ] ) . in the following we will often try to characterize the random set of points as seen from the rightmost point ( i.e. in the frame where the rightmost point is at the origin ) . to do so , let us define the generating functions of the numbers @xmath46 of points at the right of @xmath47 in the frame of the rightmost point . ( note that if @xmath48 is the position of the rightmost , then @xmath49 and one has @xmath50 for @xmath51 and @xmath52 for @xmath53 . ) @xmath54 ( as in , we assume @xmath55 . ) these generating functions , as in section [ generating - functions ] , allow one to calculate all the statistical properties of the measure in the frame of the rightmost point ( in particular the distribution of the relative distances between the points ) . they can be determined from the knowledge of the generating functions @xmath56 defined in ( [ defpsi3 ] ) by @xmath57 in section [ quantitative ] we will calculate the density of probability @xmath58 that the two rightmost points are separated by a distance @xmath59 ( and that there are at least two points on the line ) and the average density @xmath60 at a distance @xmath59 from the rightmost point . from ( [ defchi3 ] ) one can see that @xmath61 then using that @xmath62 , one gets @xmath63 by a similar calculation one can show that the average density @xmath60 of points at distance @xmath59 from the rightmost point is @xmath64 we now describe a few examples of such measures . our first example is a poisson process on the line with a density @xmath65 . we assume that @xmath65 decays fast enough to the right so that a rightmost point exists , and that @xmath66 so that there are infinitely many points on the line . by definition of a poisson process , each infinitesimal interval @xmath20 $ ] is occupied by a point with probability @xmath67 and empty with probability @xmath68 , and the occupation numbers of disjoint intervals are uncorrelated . the probability @xmath15 that there are exactly @xmath12 points on the right of @xmath9 is given by @xmath69 from this , we obtain @xmath22 from ( [ defpsi],[exppsi ] ) and @xmath70 from ( [ defpsi2 ] ) in the poisson process : @xmath71 using , the generating function of the average @xmath72 between the @xmath12-th and @xmath28-th points is @xmath73.\ ] ] the probability distribution function @xmath58 that the distance @xmath74 is equal to @xmath59 and the average density @xmath60 seen at a distance @xmath59 from the rightmost point are given by @xmath75 these expressions can be understood directly from the definition of the poisson process or , with a little more algebra , from ( [ p12],[rho ] ) . one can notice that @xmath58 and @xmath60 are given by the same expression with @xmath59 replaced by @xmath76 and are therefore analytic continuations of each other whenever @xmath65 is analytic . in the special case where the density of the poisson process is an exponential @xmath78 , one can simply replace @xmath79 in the previous expressions by @xmath80 . this gives @xmath81 = \exp \left [ -e^{-\alpha \big(x - \frac{\ln(1- \lambda)}\alpha\big)}\right ] , \\ \psi_{\lambda\mu}(x , y ) & = \exp \left [ - \mu ( 1-\lambda ) \frac{e^{-\alpha x}}\alpha - ( 1-\mu ) \frac{e^{-\alpha y}}\alpha \right ] , \end{aligned } \label{a1}\ ] ] so that from ( [ psid ] ) @xmath82 and thus @xcite @xmath83 one also has from @xmath84 start with a collection of points @xmath85 , distributed according to some measure @xmath86 and , independently for each point @xmath85 , replace it by a realization of another measure @xmath87 shifted by @xmath85 . we say that the points @xmath85 are _ decorated _ by the measure @xmath87 and call the resulting measure as @xmath86 decorated by @xmath87 . we assume that @xmath86 and @xmath87 are such that the decorated measure has a rightmost point . if the functions @xmath22 , @xmath70 , for the measure @xmath87 are known , the decorated measure is characterized by functions @xmath88 , @xmath89 , given by @xmath90 where the average is over all realizations @xmath91 of the measure @xmath86 . for instance , if @xmath86 is a poisson process of density @xmath92 , then @xmath93 = \exp \left [ \int \big [ \psi_{\lambda}(x - u ) -1\big ] r(u ) \ , \diffd u \right ] , \\ \psi_{\lambda\mu}(x , y)= \exp \left [ \int \big [ \psi_{\lambda\mu}(x - u , y - u ) -1\big ] r(u ) \ , \diffd u \right ] . \end{gathered } \label{psi}\ ] ] for a decorated measure where the decoration @xmath87 is a poisson process of density @xmath94 , the average over the @xmath85 s in leads in general to complicated expressions for @xmath88 or @xmath95 . the expressions for @xmath58 and @xmath60 are however the same as in for the pure poisson process of density @xmath94 . in fact , all the statistical properties of the distances between the rightmost points are the same as those in the exponential poisson process . this can be understood from the following reason : decorating the points @xmath96 by independent realizations of a poisson process of density @xmath94 is equivalent to drawing a single realization of a poisson process of density @xmath97 , which is just the same as one realization of a poisson process of density @xmath94 shifted by the random variable @xmath98 . the same argument applies to ruelle cascades , which can be defined as follows @xcite : take an increasing sequence of positive numbers @xmath99 and start with a poisson process of density @xmath100 . at each step @xmath101 , each point in the system is decorated by a poisson process of density @xmath102 . at step @xmath103 , the measure of points in the system is simply , from the previous argument , a poisson process of density @xmath102 globally shifted by a random variable which depends on the positions of the points at step @xmath104 . therefore , the statistics of the distances of the rightmost points is the same as for the poisson process of density @xmath102 . we are now going to see how the generating functions ( [ defpsi],[defpsi2],[defpsi3 ] ) can be determined when the random set of points on the line are the points generated at time @xmath0 by a branching brownian motion . to define the branching brownian motion we start at time @xmath105 with a single point at the origin . this point diffuses and branches , and its offspring do the same . after some time @xmath0 , a realization of the process consists of a finite number of points located at positions @xmath106 for @xmath107 then , during the next time interval @xmath108 , each point , independently of what the others do , moves a random distance @xmath109 with @xmath110 and @xmath111 , and , with probability @xmath112 , is replaced by two new points located at the same position @xmath106 . for any function @xmath113 one can define the generating function @xmath114 by @xmath115\right\rangle , \label{dual}\ ] ] where the @xmath106 for @xmath116 are the positions of the @xmath117 points of the branching brownian motion at time @xmath0 and @xmath118 denotes an average over all the possible realizations . by analyzing what happens during the very first time interval @xmath112 , one can see that the evolution of @xmath114 satisfies @xmath119 the first term in the right hand side represents the motion of the initial point during the first time interval @xmath112 and the second term represents the branching event which occurs with probability @xmath112 during this first time interval . taking @xmath112 to zero , one gets @xmath120 which is the fisher - kpp equation @xcite . ( the fisher - kpp equation is often written as @xmath121 , but this is identical to by the change of variable @xmath122 . ) because there is a single point at the origin at time @xmath105 , the initial condition is simply , from , @xmath123 the generating function ( [ defpsi3 ] ) at time @xmath0 @xmath124 can be written , for @xmath125 , as @xmath126\right\rangle = h_\phi(x , t ) , \label{psi - h}\ ] ] where the function @xmath127 is given by @xmath128 and where @xmath129 is the heaviside step function defined by @xmath130 see figure [ figphi ] for the general shape of . . ] with the choice ( [ choicef ] ) of @xmath113 , the generating function ( [ defpsi3 ] ) and , therefore , all the properties of the point measure in the branching brownian motion at time @xmath0 can be obtained as solutions of the fisher - kpp equation with the initial condition ( [ initcond ] ) . in the special case @xmath131 and @xmath132 of , i.e. for the initial condition @xmath133 , one gets @xmath134 and one recovers the well - known fact @xcite that the solution @xmath135 of the fisher - kpp equation with a step initial condition is the cumulative distribution function of the position of the rightmost point . in section [ quantitative ] we will choose @xmath136 and @xmath137 , other special cases of , given by @xmath138 to calculate the generating functions ( [ defpsi],[defpsi2 ] ) at time @xmath0 @xmath139 needed to determine the distribution @xmath58 and the density @xmath60 defined at the end of section [ statistics ] . the fisher - kpp equation has two homogeneous solutions : @xmath140 , which is unstable , and @xmath141 , which is stable . when the initial condition @xmath127 is given by the step function @xmath129 , see , the solution @xmath135 of ( [ fkpph ] ) becomes a traveling wave with the phase @xmath142 invading the phase @xmath143 @xcite . as the front is an extended object , one can define its position @xmath6 in several ways ; for example one could define @xmath6 as the solution of @xmath144 for some @xmath145 . here it will be convenient to use the following definition @xmath146 one can see using ( [ hprob ] ) that @xmath6 defined by ( [ defmt ] ) is the average position of the rightmost point . if the initial condition ( [ initcond ] ) is not a step function but is such that @xmath147 for all large enough @xmath9 and @xmath127 is a constant smaller than 1 for all large negative @xmath9 , as in ( [ choicef],[choicefsimple ] ) , the solution @xmath114 of ( [ fkpph ] ) becomes also a traveling wave . its position @xmath148 can be defined as in ( [ defmt ] ) by @xmath149 we are now going to show that the whole measure seen from the rightmost point can be written in terms of this position @xmath150 : one can rewrite ( [ the - frame ] ) as @xmath151 then from and ( [ defmtbis ] ) one gets @xmath152 where @xmath113 is the function with @xmath132 . therefore , _ with the definition ( [ defmtbis ] ) of the position of the front _ , the whole information about the measure in the frame of the rightmost point , at any time @xmath0 , can be extracted from the @xmath113 dependence of @xmath153 in the long time limit , it is known @xcite that the traveling wave solution @xmath135 of ( [ fkpph ] ) , with the initial condition ( [ teta ] ) , takes an asymptotic shape , @xmath154 . this means that @xmath155 { } f(x ) , \label{hf2}\ ] ] where @xmath154 satisfies @xmath156 it is also known , since the work of bramson @xcite , that , in the long time limit , the traveling wave moves at a velocity 2 and that its position ( [ defmt ] ) is given by @xmath157 if the function @xmath127 is not the step function but is of the form ( [ choicef],[choicefsimple ] ) , the solution @xmath114 of ( [ fkpph ] ) becomes also a traveling wave with the same shape @xmath154 . this wave is centered around the position @xmath148 , defined in ( [ defmtbis ] ) , and one has @xmath158 { } f(x ) . \label{hmphif}\ ] ] for large times @xmath148 is still given by ( [ bramson ] ) , but with a different constant @xcite . this means that @xmath159 { } f[\phi ] , \label{defdelay}\ ] ] where @xmath160 $ ] is the long time delay in the position of the front due to the modified initial condition , as compared to a front starting with a step function . taken together , and give @xmath161 { } f\big(x+f[\phi]\big ) . \label{delay}\ ] ] using , this becomes @xmath162 { } f\big(x+f[\phi]\big ) , \label{stat1}\ ] ] which shows that the measure of @xmath163 ( the rightmost points in the branching brownian motion seen from the @xmath6 frame ) does converge when @xmath164 to a limiting point measure characterized by the functions @xmath154 and @xmath160 $ ] . the measure of @xmath165 ( the rightmost points in the branching brownian motion seen from the @xmath1 frame ) also has a well - defined limit when @xmath164 . indeed , using ( [ the - frame-2 ] ) and , one gets @xmath166 { } 1 - \left ( \partial_{z_1 } + \cdots + \partial_{z_{k } } \right ) f[\phi ] . \label{the - frame-3}\ ] ] therefore , in the long time limit , all the information on the distribution of the rightmost points seen from @xmath1 is contained in the @xmath113 dependence of the delay @xmath160 $ ] . note that , in contrast to ( [ the - frame-2 ] ) which requires the position to be defined by ( [ defmtbis ] ) , the delay @xmath160 $ ] in or ( [ the - frame-3 ] ) depends only on @xmath113 : it would not change if we had chosen another definition of the front position . let us now consider @xmath167 independent branching brownian motions starting at @xmath105 at positions @xmath168 . following the same argument as in section [ sec : fkpp ] , the generating function ( [ defpsi3 ] ) of the union of the points at time @xmath0 of these @xmath167 branching brownian motions is given by the following generalization of ( [ psi - h ] ) @xmath169 where @xmath114 is the same solution of ( [ fkpph ] ) with the initial condition ( [ choicef ] ) as in the case of a single branching brownian motion starting at the origin . in the long time limit , using , @xmath170 { } \prod_{\alpha=1}^m f(x+ f[\phi]-u_\alpha ) . \label{stat2}\ ] ] this means that here again , there is a limiting measure when @xmath164 for the rightmost points in the @xmath6 frame . this measure is not the same as before ( when one starts with a single point at the origin ) , as can be seen by comparing and . in particular , the distribution of the rightmost point is different . in the frame of the rightmost point , however , one can see using and ( [ the - frame-2-bis ] ) that @xmath171 { } 1 - \left ( \partial_{z_1 } + \cdots + \partial_{z_{k } } \right ) f[\phi],\ ] ] as in . it is remarkable that the generating function @xmath172 does depend neither on the number @xmath167 of starting points nor on their positions @xmath173 . the picture which emerges is that if we superpose the rightmost points of several branching brownian motions , starting at arbitrary positions , the limiting measure in the frame of the rightmost point is , when @xmath174 , the same as for a single branching brownian motion . we will say that , in the long time limit , the measure of the distances between the rightmost points in a branching brownian motion becomes _ superposable _ : the union of two ( or more ) realizations of the process ( even moved by arbitrary translations @xmath173 ) leads to the same measure in the frame of the rightmost point as for a single branching brownian motion . as a remark , it is easy to check , that the poisson process with an exponential density @xmath175 , see section [ sec : pppexp ] , is an example of a superposable measure : the superposition of @xmath167 such poisson processes translated by arbitrary amounts @xmath168 is identical to a single poisson process with an exponential distribution translated by @xmath176 . one can also check that , for the same reason , all the decorated measures of section [ sec : deco ] are superposable when @xmath86 is a poisson process with an exponential density . in section [ statmeas ] , we will state a stronger version of the superposability property of the branching brownian motion . in this section we obtain , by integrating numerically the equation ( [ fkpph ] ) with the appropriate initial condition , some statistical properties of the limiting measure seen from the rightmost point . the analytic calculation of the delay @xmath160 $ ] is in general not easy . for @xmath136 given by ( [ choicefsimple ] ) , however , it was possible to show @xcite that when @xmath177 , the delay is given by @xmath178 \simeq - \ln(1-\lambda ) - \ln [ - \ln(1-\lambda ) ] + { \cal o}(1 ) , \label{fphi1asympt}\ ] ] and , from this , one could deduce that , in the long time limit , the average of the distance @xmath179 between the @xmath12-th and the @xmath28-th rightmost points is given for large @xmath12 by @xmath180 in @xcite , the numerical values of the distances between the rightmost points were also obtained by integrating the fisher - kpp equation with the initial condition @xmath181 in ( [ choicefsimple ] ) and by using ( [ psid ] ) ( in practice we integrated numerically the equations satisfied by the coefficients of the expansion of @xmath22 in powers of @xmath182 . it was found that @xmath183 the results ( [ asympdi],[valuesdn ] ) gave evidence that the distances between the rightmost points of the branching brownian motion were different from those of a poisson process with an exponential density ( [ diexpo ] ) . according to ( [ p12 ] ) , to obtain the distribution @xmath58 of the distance between the two rightmost points , one needs to calculate @xmath184 to first order in @xmath34 . we first remark that at time @xmath0 for @xmath185 , @xmath186= \operatorname{proba}[n(x - a)=0]=h_\theta(x - a , t)$ ] where @xmath135 is the standard fisher - kpp front with the step initial condition ( it is also easy to see from the definition of @xmath187 . ) then writing at time @xmath0 that @xmath188 is solution of the fisher - kpp equation , and using the initial condition @xmath187 in , one gets @xmath189 then , from one gets @xmath190 figure [ fig : distrib_d ] shows our numerical result for the distribution @xmath58 of the distance between the two rightmost points in the long time limit . more details on our numerical procedure is given in appendix [ sec : numeric ] . of observing a distance @xmath59 between the two rightmost points in the @xmath164 limit , as a function of @xmath59 . for @xmath59 small ( left part ) , the distribution is very close to @xmath191 . for larger values of @xmath59 , one observes a faster exponential decay of order @xmath192.,title="fig : " ] of observing a distance @xmath59 between the two rightmost points in the @xmath164 limit , as a function of @xmath59 . for @xmath59 small ( left part ) , the distribution is very close to @xmath191 . for larger values of @xmath59 , one observes a faster exponential decay of order @xmath192.,title="fig : " ] we see that @xmath58 is very close to @xmath193 for the values of @xmath59 which have a significant probability of occurring . this is of course consistent with an average distance ( [ valuesdn ] ) close to @xmath194 . for large @xmath59 ( events with a small probability ) , however , the exponential decay is faster . we now present a simple argument leading to the following prediction , which is consistent with our numerical data , @xmath195 in the long time limit , the right frontier of the branching brownian motion moves at velocity @xmath196 . let us assume that a large distance @xmath59 between the two rightmost points is produced by the following scenario : by a rare fluctuation , the rightmost point escapes and , without branching , goes significantly ahead while the rest of the points go on as usual , forming a frontier moving at velocity @xmath196 . such an event leads to the distance @xmath59 between the two rightmost points if , during a time @xmath197 , the rightmost point moves ( by diffusion alone ) by a distance @xmath198 without branching . the probability of such a scenario is @xmath199\times e^{-\tau}. \label{prob1}\ ] ] the first term is the probability of diffusing over a distance @xmath200 during time @xmath197 , and the second term is the probability of not branching . the probability to observe a large distance @xmath201 is then dominated by the events with @xmath197 chosen to maximize , that is @xmath202 and this leads to ( [ decayratedist2 ] ) in good agreement with the numerical data of figure [ fig : distrib_d ] . there is a remarkable relation between the decay rate in ( [ decayratedist2 ] ) and the shape of the traveling wave solution of ( [ deff2 ] ) . around the _ stable _ region @xmath203 , the equation can be linearized and one has @xmath204 we emphasize that this is a linear analysis of the _ stable _ region , which is usually uninteresting ( in contrast to the _ unstable _ region which determines the velocity ) . the solutions for @xmath205 are @xmath206 @xmath207 is the correct root as it is the only positive solution and @xmath208 has to vanish . the other solution @xmath209 ( the wrong root ) coincides ( up to the sign ) with the decay rate of the distribution @xmath58 for the distance @xmath59 between the two rightmost points ( [ decayratedist2 ] ) . as explained in appendix [ sec : largedeviation ] , this coincidence exists in a broad class of branching processes : each variant of the branching brownian motion is linked to a variant of the fisher - kpp equation , and the wrong root in the linear analysis of the stable region always gives the asymptotic decay rate of @xmath58 . to obtain the average density of points at a distance @xmath59 on the left of the rightmost point , one needs , according to , to calculate @xmath184 for @xmath34 close to 1 . as in section [ sec : distances ] , we first remark , from the definition , that @xmath210=h_\theta(x , t)$ ] is the standard fisher - kpp front with the step initial condition . then , writing at time @xmath0 that @xmath211 \label{expnpsirtilde}\ ] ] is solution of the fisher - kpp equation , and using the initial condition @xmath187 in , one gets @xmath212 it is the same equation as for @xmath213 in , but with a different initial condition . then , from one gets @xmath214 one can notice the great similarity between the expressions for the average density @xmath60 of points at a distance @xmath59 from the rightmost ( [ eqratilde],[rarho ] ) and the probability distribution @xmath58 for the distance between the two rightmost points ( [ eqra],[rap12 ] ) : one goes from one to the other by simple changes of signs , as in the example of a poisson process . figure [ fig : density ] presents our numerical results for @xmath60 in the long time limit . we see that @xmath60 increases as @xmath215 note that a poisson process with such a density would lead to asymptotic distances between points given by . the branching brownian motion is however not a poisson process as the points are correlated , at least near the tip . of points at a distance @xmath59 of the rightmost in the long time limit grows like @xmath216 . when the data is multiplied by @xmath217 , as shown in the figure , the linear prefactor is clearly visible . ] all the results of sections [ sec : bbmfkpp ] and [ quantitative ] can be generalized to other branching processes on the line where points move and branch independently of the positions and of the motions of the other points . in such systems , the function @xmath114 defined in is also solution of an equation similar to the fisher - kpp equation . here are four examples : * the points perform brownian motions and branch as before , but at each branching event there is a probability @xmath218 to branch into three points and @xmath219 to branch into two . then @xmath114 evolves according to @xmath220 * time is discrete with steps of duration @xmath221 ; at each time step , a point at position @xmath9 branches into two points at positions @xmath222 and @xmath223 , where the @xmath224 take independent random values distributed according to some given @xmath225 . the evolution of @xmath114 is then given by @xmath226 ^ 2 . \label{ex3}\ ] ] in this example , the positions of the points can be thought of as the possible energies of a directed polymer on a caley tree with independent random energies @xmath227 on the edges of the tree @xcite . * time is continuous but space is discrete with steps 1 ; during @xmath112 , each point at position @xmath9 has a probability @xmath112 of being removed and replaced by two points at position @xmath228 . the equation satisfied by @xmath114 is @xmath229 this example is relevant to the theory of binary search trees @xcite . * time and space are discrete with steps @xmath230 for space and @xmath221 for time ; in a given time step , a point at position @xmath9 has a probability @xmath221 of branching into two points at position @xmath9 , a probability @xmath231 of jumping to the left , @xmath231 of jumping to the right , and @xmath232 of doing nothing . then : @xmath233 , \label{ex5}\ ] ] which is of course a discretized version of the original fisher - kpp equation . is actually the equation we used in our numerical simulations , see appendix [ sec : numeric ] . in all cases , these equations have @xmath140 as an unstable fixed point , and @xmath141 as a stable fixed point . for initial conditions @xmath234 of the type ( [ choicef],[teta],[choicefsimple ] ) , the function @xmath114 develops into a traveling wave moving at a specific velocity @xmath235 . we recall briefly the procedure to determine the asymptotic velocity @xmath235 of the front ( which is also , through , the velocity of the rightmost point in the branching process ) . one looks for traveling wave solutions moving at velocity @xmath236 of the form @xmath237 and solve the linearized equation around the unstable fixed point by writing @xmath238 . this leads to a relation between @xmath239 and @xmath236 , and the minimal value @xmath235 of @xmath236 reached at some @xmath240 is the velocity selected by the front @xcite . ( we only consider here cases where the function @xmath241 has a minimum . ) for our four examples @xmath242 , \\\text{c)}\quad & v= \frac{2e^\gamma-1}\gamma;\ v^*\simeq 4.311 , & \text{d)}\quad & v=\frac1{\gamma\delta}\ln\left[1 + 2\delta\frac{\cosh(\gamma s)-1}{s^2}+\delta\right ] . \end{aligned } \label{vgammas}\ ] ] once the equation for @xmath243 of a particular branching process is written , one has access to all the generating functions @xmath244 , @xmath245 , etc . , see ( [ defpsi],[defpsi2],[defpsi3 ] ) , by choosing the appropriate initial conditions ( [ choicef],[choicefsimple ] ) for the front equation . the whole measure in the frame of the rightmost point is then obtained from at any finite time @xmath0 . note that to prove the existence of a long time limit to the point measure in this frame for a specific branching process , one would need a version of bramson s result for this process which is , to our knowledge , not known in the general case . it is natural to ask which properties of the branching brownian motion can be extended to other branching processes . if the measure for the distances between the rightmost points has a long time limit , then the arguments of section [ superpos1 ] can be easily generalized and one can show that it is superposable . we have checked that the analytical argument @xcite leading to the asymptotic expression for the average distances @xmath246 at large times can be extended in case b ) for a large class of densities @xmath225 and yields @xmath247 we have also checked numerically on examples c ) and d ) that the density at a distance @xmath59 of the rightmost point is , as in , @xmath248 for the tail of the distribution @xmath58 of the distance @xmath59 between the two rightmost points , we discussed a scenario , at the end of section [ sec : distances ] , which can be generalized ( see appendix [ sec : largedeviation ] ) to calculate the exponential decay @xmath58 for more general branching processes . this scenario , however , can only hold if points can move without branching , as in our examples a ) and d ) ; for instance , in example a ) , it predicts an exponential with a decay rate equal to @xmath249 . in examples b ) and c ) , the points branch whenever they move and the tail of @xmath58 is in general not an exponential . note that special care should be taken if the points are located on a discrete lattice , as in cases c ) , d ) and possibly b ) : quantities such as @xmath58 , see ( [ p12 ] ) , become probabilities rather than densities of probability and quantities such as @xmath60 , see ( [ rho ] ) , become average numbers rather than average densities , and all the formulas in the previous sections need to be adapted : integrals become discrete sums , derivatives become finite differences , etc . if one interprets @xmath7 as the number of points _ strictly _ on the right of @xmath9 , then the generating functions @xmath13 , @xmath32 , are still related to @xmath243 as in ( [ psi - h],[phi1phi2 ] ) with the choices ( [ choicef],[choicefsimple ] ) for the initial condition @xmath113 . then , for instance , one can show easily that becomes @xmath250 = \sum_{i\ge1 } \langle d_{i , i+1}\rangle \lambda^i,\end{gathered}\ ] ] where @xmath230 is the lattice spacing . for these systems on the lattice , there are new properties that can be investigated . as an example , if @xmath251 is the number of points on the rightmost occupied site , then it is easy to check that @xmath252.\ ] ] the whole distribution of @xmath251 can then be determined by numerical integration . in the case of our example c ) , the number @xmath251 corresponds to the number of leaves at the deepest level in a binary search tree @xcite and we found numerically that , at large times , @xmath253 , @xmath254 , @xmath255 , @xmath256 , etc . we recall from the results of section [ sec : fkpp ] , see and , that the distribution of the rightmost point is given in the long time limit by @xmath257=f(x ) , \label{nerg}\ ] ] where @xmath6 is the average position of the rightmost point and is asymptotically given by and where @xmath258 is defined in . the property is however non - ergodic , as shown by lalley and sellke @xcite in the sense that for a single realization of the branching brownian motion @xmath259\ne f(x ) . \label{nonergo}\ ] ] in fact , the left hand side of is not self - averaging and depends on the realization . this is illustrated in figure [ fig2bbm ] : for the realization on the left , there were few branching events at early times and the first points wandered to the left , leading at larger times to an asymmetric picture . for the realization on the right of figure [ fig2bbm ] , there were many branching events early . for the right realization , the rightmost point is almost always on the right of @xmath6 while it is almost always on the left of @xmath6 in the left realization . . the horizontal direction represents space , and time increases downwards . the dotted gray lines are @xmath6 and @xmath260 , the average positions of the rightmost and leftmost points , as measured from equation .,title="fig : " ] . the horizontal direction represents space , and time increases downwards . the dotted gray lines are @xmath6 and @xmath260 , the average positions of the rightmost and leftmost points , as measured from equation .,title="fig : " ] visually , these strong memory effects of the early stages of the branching brownian motion do not seem to decay with time , and it looks like the fluctuating right frontier of the system settles at some random fixed distance @xmath261 from @xmath6 : @xmath262 where @xmath261 would depend on the realization but not the time , and where @xmath263 would be a time - dependent random number centered around zero . a natural question is whether it is possible to define @xmath261 for each realization in such a way that the distribution of @xmath263 becomes in the long time limit independent of @xmath261 and @xmath0 , the idea being that the branching brownian motion at long times seen from @xmath264 would `` look the same '' for any realization , whatever is the value of @xmath261 . a related question was addressed by lalley and sellke @xcite in the following way : for each realization of the branching brownian motion , define @xmath265 as @xmath266e^{x_i(t)-2t}. \label{defz}\ ] ] ( the sum is over all the points @xmath106 present at time @xmath0 . ) as shown in @xcite , @xmath267 has a limit @xmath265 for almost every realization ; that limit is finite and positive . lalley and sellke prove then a limit theorem for the frontier of the branching brownian motion which we interpret as follows : @xmath268}\big ) , \label{lsint}\ ] ] where @xmath269 is a constant related to the large @xmath9 behavior of @xmath154 , see . in words , if one considers only the realizations of the branching brownian motion with a given value of @xmath265 , then the large time distribution of the rightmost point is given by a gumble located around @xmath270 . to make the link with , @xmath271 in appendix [ apxls ] , we present the precise theorem stated by lalley and sellke and we argue that should be equivalent to their result . a natural extension to lalley s and sellke s result is to write for all the points @xmath12 @xmath272 as in , and ask whether the joint distribution of @xmath263 , @xmath273 , @xmath274 , reaches a long time limit which is independent of the value of @xmath265 . we show in appendix [ apxls ] that lalley s and sellke s result can be extended to all the generating functions @xmath243 . our interpretation of this extension is @xmath275}}\big ) = \exp\big(- e^{-[x-\ln(az)+f[\phi]]}\big ) , \label{ls2int}\ ] ] where the delay function @xmath160 $ ] is the same as in . by choosing @xmath276 ( the step function ) , reduces to . by choosing @xmath113 as in , one sees from that the distribution of points at the right of the branching brownian motion conditioned by @xmath265 reaches a long time limit where @xmath265 only appears through the global shift @xmath277 . this means that at large times , the distribution of the rightmost points in a branching brownian motion has a well defined measure _ independent of @xmath265 _ located around @xmath278 . as an example , if one chooses the function @xmath181 defined by , one can easily show from and that , in the @xmath278 frame , the average density of points at any position is infinite in the long time limit . if one considers two branching brownian motions @xmath59 and @xmath279 starting at arbitrary positions , then the points in @xmath59 at large time will be characterized by a random value @xmath280 and a realization of the point measure described by ; idem for the points in @xmath279 . if one considers the union of these two branching brownian motions , one gets from @xmath281}}\big ) = \exp\big(- e^{-[x-\ln(az)+f[\phi]]}\big),\ ] ] with @xmath282 this means that the point measure reached in the long time limit in the @xmath278 frame is the same whether one started initially with one , two or , by extension , any finite number of initial points at arbitrary positions on the line . what does depend on the initial number of points is only the law of the random number @xmath265 , not the positions around @xmath278 . this is to be related to the discussion in section [ superpos1 ] , where we showed that , in the long time limit , the measure seen from @xmath6 depends on the initial number of points while the measure seen from @xmath1 does not . furthermore , the large time measure of the points in the @xmath278 frame has the following property : @xmath283 ( think of @xmath284 as the offspring of @xmath59 in the @xmath285 $ ] frame and of @xmath286 as @xmath287 ; idem for @xmath279 . the shifts @xmath286 and @xmath288 are arbitrary because @xmath280 and @xmath289 are unbounded independent random numbers . ) the property of the point measure in the @xmath278 frame is a stronger version of the superposability property discussed in section [ superpos1 ] : clearly , it implies that the distribution of distances between the rightmost points is invariant by superposition , but it gives more information on the measure as it encompasses the position of the rightmost point . in particular , one can check that , in any such measure , the rightmost point is gumble distributed . the simplest point measure with the property is the poisson process with an exponential density @xmath290 , for an arbitrary @xmath291 . furthermore , all the decorated measures of section [ sec : deco ] when @xmath86 is a poisson process with an exponential density @xmath290 are also superposable measures . a natural question is then : can any superposable point measure be constructed as a decorated exponential poisson process for a well chosen decoration measure ? a more specific question would be : is the limiting point measure of the branching brownian motion in the @xmath278 frame can be constructed as a decorated exponential poisson process ? we conjecture that the answer to the last question is yes . indeed , it is known @xcite that if one considers two points chosen at random among the rightmost points in a branching brownian motion at a large time @xmath0 , then the time one needs to go back to find the most recent common ancestor of those two points is either very short ( of order 1 ) or very long ( of the order of the age @xmath0 of the system ) . this means that one can group the rightmost points into families where two points belong to the same family if the branching event that generated them occurred recently . the branching brownian motion could then be a decorated exponential poisson process , where the rightmost points of all the families are poisson distributed with an exponential density @xcite and where the members of a given family form the decoration . this interpretation helps to understand a question raised by our previous work @xcite : the distances between points in the branching brownian motion are given by ( [ asympdi],[valuesdn ] ) , but if one considers only the rightmost point of each family , then the distances are given by with @xmath292 , as in the grem or in the ruelle cascade , see section [ ruellecascade ] . in the present work , we have shown that all the statistical properties of the rightmost points in a branching brownian motion can be obtained by solving a front equation with a well - chosen initial condition . the distribution of the positions of the rightmost points seen in the frame @xmath6 ( the average position of the rightmost ) has a long time limit . the properties of the limiting distribution can be expressed as the long time delays of the traveling wave solution of the fisher - kpp equation when one varies the initial condition . this limiting distribution is however modified if one considers the union of several branching brownian motions . if one considers , however , only the distances between the points , for example if one looks at the distribution of all the positions of the rightmost points seen in the frame of the rightmost one , one obtains at large times another limiting distribution which does not depend on the initial positions of the branching brownian motions ( as long as there are finitely many of them ) . we called this property superposability . in section [ quantitative ] we have measured a few properties of this limiting distribution , and in section [ sec : disc ] we explained how our results can be extended to more general branching random walks . lastly , in section [ statmeas ] , we argued that in the lalley and sellke frame , the branching brownian motion satisfies a stronger version of the superposability property , see . in the future , it would be interesting to characterize more precisely the limiting measure of the branching brownian motion and of the branching random walks to see whether some universal properties emerge . for example , we believe that the average density seen from the rightmost point should always grow as in . it would also be interesting to check whether , in lalley s and sellke s frame , the limiting distribution is indeed a decorated exponential , and to determine the properties of the decorating measure . the question of ergodicity , raised at the beginning of section [ statmeas ] , is also an interesting open question . we would like to thank julien berestycki and simon harris for interesting discussions . we performed the numerical simulations by discretizing the branching brownian motion in space ( with a grid length @xmath221 ) and time ( with steps @xmath230 ) as in the example d ) of section [ sec : disc ] . the corresponding front equation is given by , which is of course the most straightforward discretization of the fisher - kpp equation . the solutions to the discrete equation converge to the solutions of the fisher - kpp equation , if @xmath293 with @xmath231 held constant and small enough . we used three sets of values for @xmath230 and @xmath221 and computed the exact asymptotic velocity @xmath235 of the front and the decay rate @xmath240 of the asymptotic shape @xmath154 by minimizing numerically the function @xmath241 given in ( [ vgammas]d ) . the decay rate @xmath288 for the probability of observing a distance @xmath59 between the two rightmost points , see equation , was computed using the recipe given at the end of section [ sec : distances ] and explained in appendix [ sec : largedeviation ] . all these values are presented in table [ tabval ] .values of @xmath235 , @xmath240 and @xmath288 for our discretized branching processes , compared to the values in the fisher - kpp case . [ cols= " < , < , < , < , < " , ] the simulations were made on a finite but large domain centered around the position of the front ; typically it extended to about a distance 1000 ahead and behind the center of the front ( respectively 4000 , 10000 or 20000 lattice sites depending on @xmath230 ) . the values at @xmath294 were exactly computed and used for the boundaries of the domain . whenever the front moved by more than one unit space lattice , the whole data set was recentered . the simulations were performed up to large times of order @xmath295 ( 0.8 to 24 millions of time steps ) and the data was extrapolated to obtain a value at @xmath296 . to do this extrapolation , we used a more precise asymptotic expansion of the position of the front than : according to @xcite , @xmath297 where the number @xmath298 does not depend on the initial condition . as we measure the delay @xmath299 , many terms cancel and one gets @xmath300+\frac{\delta a_1}t+\frac{\delta a_{3/2}}{t^{3/2}}+\cdots . \label{deltaasympt}\ ] ] all the quantities we measure are derivatives of @xmath160 $ ] , see , and have therefore the same large time expansion as . thus , we extrapolated our numerical data to the large time limit by fitting it with the function @xmath301 for times larger than ( typically ) 5000 , see figure [ fit ] , and by using @xmath269 as the end result . by the function @xmath301 ( line ) . the inset shows the quality of the fit by displaying the ratio between the data points and the fitting function . ] on figure [ fig : distrib_d ] , the three data points were presented together ; on figure [ fig : density ] , we have drawn together for each data set the function @xmath302 using in each case the value of @xmath240 of table [ tabval ] . in both cases , the superposition was nearly perfect , and so we expect that on the scales of the figure , the curves would not change noticeably for smaller values of @xmath230 and @xmath221 . in this appendix we generalize , to any branching random walk , the argument leading to the asymptotic decay of the distribution of the distance between the two rightmost points in the branching brownian motion . we consider a generic branching random walk in discrete space ( with spacing @xmath230 ) and time ( with intervals @xmath221 ) defined by the following family of functions @xmath303 we assume that @xmath304 , so that there is no extinction . then @xmath305 can be thought as the probability that the point does not branch but moves by a distance @xmath205 . the continuous time and/or space cases can be obtained as suitable @xmath293 and/or @xmath306 limits . let @xmath307 $ ] be the generating function of the displacement during time @xmath0 of one point _ conditioned on the fact that this point does not branch _ : @xmath308 as the time steps are independent , the function @xmath309 can be computed during the time interval @xmath221 which gives @xmath310 note that @xmath311 as soon as the branching probability is non - zero . we want now to evaluate the probability that a point moves a distance @xmath205 , without branching , during time @xmath0 . for large @xmath0 , it takes the form @xmath312,\ ] ] where @xmath313 is a large deviation function . using , one finds that @xmath313 and @xmath309 are related by a legendre transform @xmath314 now , assuming as in section [ sec : distances ] that the events which contribute most to a large distance @xmath59 between the two rightmost points are those where the rightmost point moves , without branching , a distance @xmath59 ahead of the frontier of the branching brownian motion , one gets @xmath315 \right\ } , \label{prob2}\ ] ] where @xmath235 is the velocity of the front . for large @xmath59 , the optimal @xmath197 is also large and it satisfies , by derivation , @xmath316 let @xmath317 . using , equation becomes @xmath318 remarkably , this equation does not depend on @xmath59 . replacing into gives @xmath319 the asymptotic decay rate of the probability distribution function of the distance between the two rightmost points is therefore simply the positive solution @xmath288 of with @xmath309 given by . in the branching brownian motion , @xmath320 , @xmath321 , so that gives indeed @xmath322 . as can be checked easily from ( [ giveg ] ) , the function @xmath309 is convex . therefore , as @xmath311 , equation has at most one positive solution and at most one negative solution . the positive solution is the relevant one here . we are now going to show that the negative solution of ( [ whichbeta ] ) , if it exists , gives the asymptotic shape of the associated traveling wave . we write the front equation associated to the branching point process . using the same method as in section [ sec : bbmfkpp ] , we find @xmath323 we look at the shape @xmath258 of the traveling wave solution for a step initial condition , which moves asymptotically at the velocity @xmath235 : @xmath324 using , we see that in the _ stable _ region @xmath325 one has @xmath326 we look for an exponential solution to this linearized equation : @xmath327 with @xmath328 , as @xmath329 . ( note that a periodic modulation of this exponential could occur as @xmath205 takes only discrete values . ) inserting into and using , one finds that the equation for @xmath33 is @xmath330 which is the same equation as for @xmath331 . to summarize , a _ positive _ solution to gives the exponential decay rate of the probability distribution of the distance between the two rightmost points , see , while a _ negative _ solution gives the coefficient @xmath332 governing the shape of the front @xmath154 in the stable region @xmath325 . lalley s and sellke s theorem @xcite is @xmath333 where @xmath269 is the constant appearing @xcite in the large @xmath9 expansion of the function @xmath154 defined in @xmath334 and @xmath265 is defined in . in words , given the positions @xmath335 at time @xmath230 , there is a @xmath164 limit to the probability that the rightmost is on the left of @xmath336 which depends , obviously , on the @xmath335 and is as such a random variable . as @xmath230 goes to infinity , this random variable converges almost surely to the gumble distribution around @xmath277 . this result can be extended into the following : for any suitable function @xmath113 ( see section [ bbm - frame ] ) , one has @xmath337 \big| \{x_i(s)\}\big\rangle = \exp\big({-a z e^{-x - f[\phi]}}\big ) , \label{ls2}\ ] ] where @xmath160 $ ] is the delay function . for @xmath276 , reduces to . we first give an outline of lalley s and sellke s proof applied to the case . given the positions @xmath338 of the points at time @xmath230 , the system as time @xmath339 can be seen as a collection of independent branching brownian motions at time @xmath340 starting from the @xmath338 . therefore @xmath341 \big| \{x_i(s)\}\big\rangle = \prod_i h_\phi\big(x - x_i(s),t - s\big),\ ] ] where the product in the right hand side is made on all the points present at time @xmath230 . we replace @xmath9 by @xmath342 , to center around the position of the front , and suppose @xmath0 large . it is easy to see from bramson s formula that @xmath343 as @xmath0 becomes large , so that @xmath344 \big| \{x_i(s)\}\big\rangle = \prod_i h_\phi\big(m_{t - s}+2s + x - x_i(s)+{o}(1),t - s\big),\ ] ] and , using , @xmath345 \big| \{x_i(s)\}\big\rangle = \prod_i f\big(2s+x - x_i(s)+f[\phi]\big).\ ] ] we now take @xmath230 large . of all the points present at time @xmath230 , the rightmost is around @xmath346 , see . therefore , @xmath347 diverges for all @xmath12 . using , @xmath345 \big| \{x_i(s)\}\big\rangle \simeq\exp\big(-\sum_i \big[a\big(2s+x - x_i(s)+f[\phi]\big)+b\big]e^{-2s - x+x_i(s)-f[\phi ] } \big).\ ] ] following lalley and sellke , we introduce the quantities @xmath348e^{-2s+x_i(s)},\ ] ] see , so that @xmath345 \big| \{x_i(s)\}\big\rangle \simeq\exp\left(- \left[a z_s + \big(ax+af[\phi]+b\big)y_s\right]e^{-x - f[\phi]}\right ) . \label{lstls}\ ] ] finally , the most technical part of lalley s and sellke s proof is that @xmath349 and @xmath350 are martingales converging when @xmath351 to @xmath352 and @xmath353 respectively , which leads to . we do not reproduce this part of the proof here as it does not concern our extension with the function @xmath113 and it works in exactly as in . in ( [ ls2 ] ) , the average is made on all the realizations with a given set @xmath335 of points at a large time @xmath230 but the only relevant quantity appearing in the generating function is @xmath350 . one would obviously have reached the same result if one had conditioned by @xmath350 instead of by the @xmath335 . furthermore , as @xmath350 converges quickly to @xmath265 , as illustrated on figure [ fig2bbm ] , we argue that conditioning by @xmath350 at a large time @xmath230 or directly conditioning by @xmath265 should be equivalent , hence ( [ lsint],[ls2int ] ) . kolmogorov , a. , petrovsky , i. , piscounov , n. : tude de lquation de la diffusion avec croissance de la quantit de matire et son application un problme biologique . tat moscou , a * 1*(6 ) , 125 ( 1937 )

planets in highly eccentric orbits form a class of objects not seen within our solar system . the most extreme case known amongst these objects is the planet orbiting hd 20782 , with an orbital period of 597 days and an eccentricity of 0.96 . here we present new data and analysis for this system as part of the transit ephemeris refinement and monitoring survey ( terms ) . we obtained chiron spectra to perform an independent estimation of the fundamental stellar parameters . new radial velocities from aat and paras observations during periastron passage greatly improve our knowledge of the eccentric nature of the orbit . the combined analysis of our keplerian orbital and hipparcos astrometry show that the inclination of the planetary orbit is , ruling out stellar masses for the companion . our long - term robotic photometry show that the star is extremely stable over long timescales . photometric monitoring of the star during predicted transit and periastron times using most rule out a transit of the planet and reveal evidence of phase variations during periastron . these possible photometric phase variations may be caused by reflected light from the planet s atmosphere and the dramatic change in star planet separation surrounding the periastron passage .

1511.08679

exoplanetary system architectures have revealed numerous surprises since the first exoplanets were discovered . one of the earliest surprises was the discovery of exoplanets in highly eccentric orbits , for which there is no analog in the solar system . these eccentric orbits were discovered for giant planets , such as hd 114762b @xcite and 70 vir b @xcite with eccentricities of 0.33 and 0.40 respectively . since those early discoveries , eccentric planets have presented a significant challenge for formation theories to account for the components of planet - planet scattering @xcite and tidal circularization @xcite . such planets tend to be discovered with the radial velocity ( rv ) technique since the observations are able to sample the entire keplerian planetary orbit . subsequent investigations of the eccentricity distribution of planetary orbits that take _ kepler _ transiting exoplanet discoveries into account show that small planets in multi - planet systems are more likely to have low eccentricities @xcite . the discovery and characterization of eccentric orbits is an on - going effort to understand the evolutionary history of these fascinating systems . a particularly eccentric exoplanet was discovered by @xcite orbiting the star hd 20782 . with a minimum mass twice that of jupiter and an orbital period of 597 days , the planet is typical of high - eccentricity planets . the orbit was further revised by @xcite and shown to have an eccentricity as high as 0.97 , making it the highest eccentricity exoplanet yet discovered . however , data during periastron passage is difficult to obtain for such systems since the rv variation predominantly occurs during a very small fraction of the orbital phase . the star continued to be monitored by the transit ephemeris refinement and monitoring survey ( terms ) to improve the orbital parameters of the system @xcite . such orbital refinement may be used to predict and observe events that occur during particular periods of the orbit , such as planetary transits @xcite or phase variations @xcite . here we present new results for the hd 20782 system , including rvs that sample several periastron passages and establish the planet as the most eccentric known exoplanet . follow - up photometry from both ground - based and space - based telescopes rule out a transit of the planet and show evidence of phase variations due to reflected light from the planet close to periastron passage . section [ science ] provides background information and discusses the science motivation for studying the system . section [ stellarprop ] presents analysis of new chiron spectra and the resulting fundamental parameters of the host star as well as stellar abundances . new rv data are combined with those published in section [ orbit ] and a new keplerian orbit for the planet is produced . section [ astrometry ] describes the use of _ hipparcos _ astrometry to constrain the orbital inclination of the planet . section [ transit ] discusses the transit prospects for the system and the effects of both orbital eccentricity and inclination . section [ photometry ] presents the ground - based photometry and an estimate of the stellar rotation period . data from most are used during the transit / periastron window to rule out a transit and also reveal the potential presence of a reflected light signature of the planet as it passes through periastron passage . we discuss future observing opportunities and make concluding remarks in section [ conclusions ] . the eccentricity distribution of exoplanets has a well - defined shape whereby the orbits diverge from circular beyond a semi - major axis of @xmath1 au @xcite , inside of which tidal circularization tends to force low eccentricity @xcite . the observed eccentricity distribution is a clear indicator of formation processes that are dependent upon initial system architectures , in particular planet - planet scattering . wide binaries may inadvertantly create a more suitable environment for the formation of highly - eccentric planetary orbits through gravitational perturbations from the companion star and the triggering of planetary ejections @xcite . . the orbits of the solar system planets ( dashed lines ) are shown for comparison.,width=321 ] hd 20782 is part of a wide binary with hd 20781 having a projected separation of 9,000 au , recently described by @xcite . the known planet orbiting hd 20782 lies at the very top of the exoplanet eccentricity distribution , though rv measurements during the crucial periastron passage were relatively rare . the extreme nature of the planet s orbital eccentricity may be seen in figure [ orbitfig ] , where the orbit is described using our expanded dataset ( see section [ orbit ] ) . our further investigations of this system are primarily motivated by a better characterization of the planetary orbit and performing follow - up observations at key orbital phases that can help to understand the nature of the planet . it is also important to establish that the secondary object is indeed a planet since a face - on orbital orientation would make it consistent with the eccentricity distribution of spectroscopic binaries @xcite . the orbital orientation depicted in figure [ orbitfig ] shows that the star planet separation along the line of sight to the observer is quite small , despite the @xmath218 month orbital period . this yields a relatively high transit probability equivalent to that of a hot jupiter ( see section [ transit ] ) . thus a primary motivation for follow - up observations is the possible detection of a planetary transit for a long - period eccentric planet @xcite . a previous example of such a system can be seen in the case of hd 80606b @xcite , where the secondary eclipse of the 0.93 eccentricity planet was detected by @xcite and later confirmed to also exhibit a primary transit @xcite . an additional motivation for obtaining high - precision photometry during the transit window and periastron passage for hd 20782b is the possibility of detecting reflected light from the planet since the small star planet separation will greatly increase the amplitude of the phase signature @xcite . such a detection would allow an estimate of the geometric albedo of the planet and place constraints upon the atmospheric properties and the atmosphere s radiative and advective time scales @xcite . note that since the orbital period is 18 months , an observing opportunity for a particular point in the orbit will only arise every 3 years since the star will be largely inaccessible to ground - based observers for each alternate orbit . a critical step in quantifying the properties of the planet lies in understanding the host star . here we provide new fundamental parameters and abundances for hd 20782 . we acquired a high s / n ( 300 second integration ) spectrum of hd 20782 on the night of july 6th , 2014 . the data were acquired using chiron , a fiber - fed echelle spectrometer @xcite , installed at the 1.5 m telescope at cerro tololo inter - american observatory ( ctio ) . chiron operates at a fixed wavelength range of 415880 nm and a resolution of @xmath3 . the spectrum was modeled using the spectroscopy made easy ( sme ) package , described in more detail by @xcite . sme uses an iterative technique that combines model atmosphere analysis with yonsei - yale model isochrones @xcite that utilize _ hipparcos _ photometry and distances @xcite . this approach produces a self - consistent convergence with the measured surface gravity @xcite . the results of this analysis are shown in table [ stellar ] , including values for the surface gravity @xmath4 , rotational velocity @xmath5 , atmospheric abundance [ fe / h ] , effective temperature @xmath6 and stellar isochrone solution ( mass @xmath7 , radius @xmath8 , and age ) . these parameters are consistent with previous estimates of the stellar properties , such as those calculated by @xcite . the revised parameters demonstrate that hd 20782 is quite similar to the sun , with the mass and radius being crucial properties for the subsequent analysis of the planetary companion in this paper . lc @xmath9 & 7.4 + @xmath10 & 0.63 + distance ( pc ) & @xmath11 + @xmath12 ( k ) & @xmath13 + @xmath4 & @xmath14 + @xmath5 ( kms@xmath15 ) & @xmath16 + @xmath17fe / h@xmath18 $ ] ( dex ) & @xmath19 + @xmath7 ( @xmath20 ) & @xmath21 + @xmath8 ( @xmath22 ) & @xmath23 + age ( gyrs ) & @xmath24 both components of the wide binary system , namely hd 20781 and hd 20782 , have had their elemental abundances measured by a number of different authors . however , due to the difference in size and spectral type , the abundances within hd 20782 are easier to determine . while half as many groups have measured hd 20781 than hd 20782 , there does remain some overlap by some , such as @xcite , @xcite , and @xcite who did a more in - depth comparison of the two stars . per the analysis in the hypatia catalog @xcite , the individual abundances within both stars were renormalized to the @xcite solar scale . the largest measurement discrepancy between datasets , known as the _ spread _ , was used to better quantify the uniformity , or lack thereof , between measurements . this technique was implemented in the hypatia catalog to better understand the variation in abundances seen when employing different reduction techniques , due to instances where the _ spread _ between groups was larger than associated error . for the cases where variations between groups were small , the median value was used as the ultimate abundance measurement . the overall median [ fe / h ] content in hd 20781 was 0.1 dex , as compared to 0.15 dex within hd 20782 , where the spread was 0.03 dex and 0.17 dex , respectively . in other words , the groups that measured hd 20781 , while fewer in number , were in closer agreement regarding the iron abundance than those that measured hd 20782 . the [ fe / h ] determinations for both stars are disparate compared to the abundances determined by @xcite , which are not part of the hypatia catalog , who measured 0.04@xmath250.03 and -0.02@xmath250.02 , respectively . these are consistent with our new [ fe / h ] determination shown in table [ stellar ] . a wide variety of @xmath26-elements ( carbon , magnesium , silicon , and titanium ) , odd - z elements ( sodium , aluminum , and scandium ) , and iron - peak elements ( vanadium , chromium , cobalt , and nickel ) have been measured for both stars . for all elements except for [ na / fe ] , the abundance measurements for hd 20781 and hd 20782 were found to be consistent to within error and markedly sub - solar , or @xmath2 -0.1 dex . the [ na / fe ] content in hd 20782 was found to be @xmath27 more than in the companion hd 20781 , where [ na / fe ] = -0.09@xmath250.06 dex and -0.22@xmath250.04 dex , respectively . the highly eccentric planet orbiting hd 20782 was first reported in @xcite and updated in @xcite . we now present a further six years of radial velocity data from the anglo - australian planet search ( aaps ) . the aaps is one of the world s longest - running planet searches , with more than 40 planet discoveries in its 16 years of operation ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? hd 20782 has been observed on 52 epochs from 1998 aug 9 to 2013 sep 19 ( table [ rvs_aat ] ) . precision doppler measurements are obtained with the ucles echelle spectrograph @xcite at the 3.9 m anglo - australian telescope ( aat ) . a 1-arcsecond slit delivers a resolving power of @xmath2845,000 . calibration of the spectrograph point - spread function is achieved using an iodine absorption cell temperature - controlled at 60.0@xmath250.1@xmath29c . the iodine cell superimposes a forest of narrow absorption lines from 5000 to 6200 , allowing simultaneous calibration of instrumental drifts as well as a precise wavelength reference @xcite . the result is a precise radial velocity shift measured relative to the epoch of the iodine - free `` template '' spectrum . aat velocities for hd 20782 span a total of 15 years and have a mean internal uncertainty of 2.4 ms@xmath15 . orbital fits to the aat data allowed predictions of the next periastron passage of the planet , estimated to be 15 january 2015 . we were able to observe hd 20782 during that passage using the physical research laboratory ( prl ) optical fiber - fed high - resolution cross - dispersed echelle spectrograph ( paras ) with the mount abu 1.2 m telescope in india . the paras spectrograph is temperature - controlled at 25.55@xmath250.01@xmath29c in an enclosure that is pressure - controlled at 0.10@xmath250.03 mbar . paras has a resolution of @xmath2867,000 and obtains rv data at a spectral range of 3800 to 6900with simultaneous wavelength calibration with a thorium - argon ( thar ) hollow cathode lamp . the uncertainties for the paras measurements were derived based on the photon noise estimation procedure explained by @xcite . further details of the paras instrument and the data reduction are described by @xcite . paras observations were made under high air mass conditions ( 1.71.9 ) with no atmospheric dispersion corrector ( adc ) . the five paras observations ( see table [ rvs_paras ] ) complete our rv dataset and bring the total number of observations to 57 . ccc 11035.31946 & 21.90 & 2.33 + 11236.93065 & -6.51 & 3.27 + 11527.01731 & 7.32 & 3.39 + 11630.88241 & 29.70 & 2.72 + 11768.30885 & -6.64 & 2.62 + 11828.11066 & -7.64 & 3.00 + 11829.27449 & -6.64 & 3.82 + 11856.13530 & -10.37 & 3.55 + 11919.00660 & -3.62 & 2.92 + 11919.99630 & -1.67 & 2.85 + 11983.89009 & 4.16 & 3.32 + 12092.30437 & 17.84 & 2.35 + 12127.26814 & 17.70 & 2.79 + 12152.16308 & 23.15 & 2.50 + 12187.15965 & 22.78 & 2.53 + 12511.20636 & -1.26 & 2.29 + 12592.04809 & 17.40 & 2.30 + 12654.96031 & 15.38 & 2.34 + 12859.30551 & -202.48 & 1.90 + 12946.13833 & -18.15 & 2.08 + 12947.12246 & -14.27 & 1.77 + 13004.00143 & -0.29 & 1.85 + 13044.02367 & 0.76 & 2.25 + 13045.96088 & -0.40 & 1.93 + 13217.28800 & 9.01 & 1.71 + 13282.22023 & 20.57 & 1.87 + 13398.96924 & 22.14 & 1.39 + 13403.96059 & 30.40 & 2.56 + 13576.30688 & -9.14 & 1.60 + 13632.28114 & -7.62 & 1.59 + 13665.18659 & 6.38 & 1.72 + 14013.21622 & 31.23 & 1.55 + 14040.13171 & 22.12 & 1.96 + 14153.97010 & -11.56 & 2.10 + 14375.24693 & 13.32 & 1.70 + 14544.89158 & 10.26 & 2.15 + 14776.10092 & -7.55 & 1.85 + 14843.02077 & 0.09 & 1.56 + 14899.92440 & -0.65 & 2.07 + 15107.24701 & 16.54 & 2.78 + 15170.05453 & 17.31 & 2.37 + 15204.97966 & 29.22 & 1.88 + 15253.91188 & -78.17 & 2.35 + 15399.32249 & -8.19 & 1.88 + 15426.31459 & -6.89 & 1.71 + 15461.23900 & -14.81 & 2.99 + 15519.13309 & 8.36 & 2.00 + 15844.13584 & -145.90 & 6.54 + 15845.17956 & -185.60 & 2.28 + 15846.13671 & -156.28 & 2.32 + 15964.93095 & 7.77 & 2.87 + 16499.33740 & -11.25 & 3.03 ccc 57036.16183 & 272.25 & 4.12 + 57038.14436 & 127.25 & 4.09 + 57039.13336 & 61.53 & 3.65 + 57040.15494 & 149.14 & 3.86 + 57042.12356 & 183.06 & 2.98 the rv data shown in tables [ rvs_aat ] and [ rvs_paras ] were used to produce a revised keplerian orbital solution . this was performed using the rvlin package ; a partially linearized , least - squares fitting procedure @xcite . the uncertainties for the orbital and associated physical parameters were estimated using the boottran bootstrapping routines described in @xcite . to be sure that known instabilities of the levenberg - marquardt - based rvlin algorithm did not prevent convergence at these high eccentricities , we reduced the number of nonlinear parameters in the problem by fixing the eccentricity at 100 values evenly spaced between 0.93 and 0.995 and selecting the value that produced the minimum @xmath30 fit . a fit to the aat and paras data with their instrumental uncertainties was unsatisfactory . the rms residuals to the paras data are 17 ms@xmath15 , with two excursions at and after rv minimum of over 20 ms@xmath15 , inconsistent with typical instrumental uncertainties of under 6 ms@xmath15 . further , the scatter about the fit to the aat data is 6.1 ms@xmath15 , including three excursions larger than 15 ms@xmath15 ( up to 17 ms@xmath15 ) , both significantly larger than the quoted errors of 2.3 ms@xmath15 . given that there are only 52 aat points , we do not expect to see 3 points ( @xmath25% ) with deviations of 15 ms@xmath15 from gaussian noise unless the errors are more like 6 ms@xmath15 . some component of the scatter about the best fit is due to intrinsic stellar variability , and some is due to the precision of the measurements ( due to both instrumental / algorithmic imprecision and photon noise ) . the stellar noise should be the same for both instruments , meaning that the large excursions seen in the paras data indicate a problem with either the fit or the paras data . close examination of points near periastron reveal that the problem must lie with the instrumental uncertainties , not the fit . paras and aat have two measurements ( each ) at very similar phases ( the expected change in rv between the points in each pair is @xmath31 ms@xmath15 ) . however , in both cases the difference in velocities is over 20 ms@xmath15 , and in different directions . the combined measurement uncertainties of the two instruments therefore must be of order 20 ms@xmath15 . we attempted a second fit , but inflated both instrumental uncertainties by adding , in quadrature , 5 ms@xmath15 and 19 ms@xmath15 to the aat and paras velocities , respectively . these inflations reflect a common stellar jitter component ( likely to be around 5 ms@xmath15 ) and an additional , instrument - dependent component added in quadrature . this resulted in a much more satisfactory fit : the residuals to the best fit for the two telescopes have standard deviations of 5.75 ms@xmath15 and 19.85 ms@xmath15 , respectively , and @xmath30 values of 1.03 and 1.01 , respectively . there is still a significant outlier to the aat fit ( at 15 ms@xmath15 ) , but at 2.5@xmath32 ( using the inflated measurement uncertainties ) this is not unexpected from 52 data points . lcc @xmath33 ( days ) & @xmath34 & @xmath35 + @xmath36 ( bjd 2,440,000 ) & @xmath37 & @xmath38 + @xmath39 ( bjd 2,440,000 ) & @xmath40 & @xmath41 + @xmath42 & @xmath43 & @xmath44 + @xmath45 ( deg ) & @xmath46 & @xmath47 + @xmath48 ( ms@xmath15 ) & @xmath49 & @xmath50 + @xmath51sin@xmath52 ( @xmath53 ) & @xmath54 & @xmath55 + @xmath56 ( au ) & @xmath57 & @xmath57 + @xmath58 ( ms@xmath15 ) & @xmath59 & @xmath60 + @xmath61 & 52 & 57 + rms ( ms@xmath15 ) & 5.91 & 8.06 + @xmath62 & 1.0 & 1.14 and [ rvs_paras ] ) along with the best - fit orbital solution ( table [ planet ] ) . rv offsets between datasets have been accounted for in this figure . middle : the rv data phased on the orbital the solution from table [ planet ] , where phase zero corresponds to superior conjunction . bottom : a zoomed version of the phased middle plot which shows the coverage during periastron passage.,title="fig:",width=309 ] + and [ rvs_paras ] ) along with the best - fit orbital solution ( table [ planet ] ) . rv offsets between datasets have been accounted for in this figure . middle : the rv data phased on the orbital the solution from table [ planet ] , where phase zero corresponds to superior conjunction . bottom : a zoomed version of the phased middle plot which shows the coverage during periastron passage.,title="fig:",width=313 ] + and [ rvs_paras ] ) along with the best - fit orbital solution ( table [ planet ] ) . rv offsets between datasets have been accounted for in this figure . middle : the rv data phased on the orbital the solution from table [ planet ] , where phase zero corresponds to superior conjunction . bottom : a zoomed version of the phased middle plot which shows the coverage during periastron passage.,title="fig:",width=317 ] we conclude that there is significant instrumental / observational systematic noise in the paras data due to air mass , of order 20 ms@xmath15 . we also examined the inclusion of a linear rv trend in our model but found that this does not improve the quality of the fit . the final orbital solution from the data is shown in table [ planet ] , where we have included the solution that uses the aat data only for comparison . the aat+paras orbital solution includes an offset between the aat and paras datasets as a free parameter , found to be @xmath63 ms@xmath15 . the @xmath58 parameter shown in table [ planet ] is the systemic velocity of the system with respect to the zero point of the extracted rvs ( relative to the template spectrum ) . thus , there is an offset between the @xmath58 value reported in table [ planet ] and the true systemic velocity , reported by @xcite to be 40.7 kms@xmath15 . our final aat+paras orbital solution is depicted in figure [ rv ] . the bottom panel of figure [ rv ] shows the quality of the combined data coverage during periastron passage for this highly eccentric planet , particularly the additional coverage provided by the paras data . we note that the transit time we calculate is sensitive to the weights assigned to the paras and aat data . the paras data favor a transit time that is 0.20.3 days later than the aat data . because we do not fully understand the source of the very large scatter in the paras data , we should not assume that our errors are gaussian . fortunately , boottran uses bootstrapping to determine its parameter uncertainties , which is appropriate for non - normally distributed residuals ( although the underlying fitter minimizes @xmath30 , and so does assume gaussian errors ) . to constrain the inclination of the system and possibly refine the estimate of the companion mass , we combine hipparcos astrometry of hd 20782 with the orbital parameters obtained from the radial velocity observations . we use the new reduction of the hipparcos data @xcite , which presents a significant improvement in the overall reliability of astrometric information @xcite and includes the intermediate astrometric data ( iad ) product in a format that facilitates the quest for signatures of orbital motion . following the method prescribed by @xcite , we use the spectroscopic elements derived from our rv solution ( table [ planet ] ) to search for an orbital signature . the 5 standard astrometric parameters for the hipparcos solution of a hd20782 can be obtained from the vizier catalogue @xcite ; these are right ascension ( ra , @xmath26=50.01@xmath64 ) , declination ( dec , @xmath65=-28.85@xmath64 ) , proper motion in ra ( @xmath66=349.33 mas / yr ) and dec ( @xmath67=-65.92 mas / yr ) , and parallax ( @xmath68=28.15 mas ) . the 5 spectroscopic parameters required from the radial velocity analysis are period ( @xmath33 ) , eccentricity ( @xmath42 ) , semi - amplitude ( @xmath48 ) , time of periastron ( @xmath69 ) , and argument of periastron ( @xmath45 ) . each hipparcos observation is reconstructed from the iad and fit with a comprehensive model based on 12 parameters : the 5 standard astrometric parameters , the 5 spectroscopically derived parameters , the inclination ( @xmath52 ) , and the longitude of the ascending node ( @xmath70 ) . in practice , the spectroscopic parameters are treated as constants since they are considered reliable , and we work with 7 free parameters . the details of the procedure are carefully described by @xcite , and we follow their methods to calculate inclination , a new orbit , and the significance of the orbit via the permutation test . we begin by constructing a two - dimensional @xmath71 grid , where we solve for the remaining five parameters of the 7-parameter model , and the corresponding @xmath30 . the parameter values identified by the minimum @xmath30 value are used as the starting point for an @xmath72 minimization , using the downhill simplex method , to supersede the limitations imposed by the resolution of the @xmath71 grid . we then perform 100,000 monte carlo simulations , where we generate 1,000 sets of hipparcos measurements from the existing data . for each set of hipparcos abscissae , we execute 100 random draws from the spectroscopic parameters , in order to inculcate their uncertainties into our result . each spectroscopic parameter is assumed to be a gaussian distribution , with the rv solution and its errors serving as the mean and standard deviation respectively . the monte carlo models are then solved as described above , to produce 100,000 sets of solution parameters . the final parameters are defined as the median of the associated distribution , while the errors are the interval between the 15.85 percentile and the 84.15 percentile , encompassing 68.3% of the values around the median . for completeness , we report here our full set of final parameters , as offsets to the hipparcos values . the changes in right ascension , declination , parallax , proper motion in ra and declination , and the argument of periastron are : @xmath73 mas , @xmath74 mas , @xmath75 mas , @xmath76 mas / yr , @xmath77 mas / yr . we find an inclination of @xmath78 and an argument of periastron @xmath79 , but the solution is very poorly constrained . the astrometric data covers approximately two orbits for this system , so phase coverage should not inhibit the recovery of any significant orbital signatures . unfortunately , the projected minimum semi - major axis of our new solution is very small ( @xmath80 mas ) compared to the median hipparcos single - measurement precision for this target ( 3.5 mas ) , as shown in figures [ fig : hist ] and [ fig : orbit ] . we perform a permutation test to verify the significance of our result by comparing the semimajor axis of the new solution orbit with 1,000 pseudo orbits generated from random permutations of the astrometric data , similar to @xcite . figure [ fig : hist ] illustrates the calculation of a low significance orbit ( 68.2% , which is almost exactly at the 1@xmath32 level of detection ) , confirming that the hipparcos data contains little or no orbital signature . the new hipparcos reduction has this target flagged as having a good fit with just the 5 astrometric parameters , which is consistent with the fact that adding the 7 rv parameters does not seem to change the solution . @xcite emphasize that orbital solutions at this significance level are prone to very large biases , and the calculated values and their errors should be considered highly suspect . we present our full set of orbital parameters here only to facilitate future comparison of analytical methods , and not for direct application . and length given by the residual of the orbital solution connect the measurements with the predicted location from our model . this illustrates the difficulty of detecting an orbit with such a small projected semi - major axis , given the median hipparcos single - measurement precision on this target.,width=309 ] on the other hand , simulations by @xcite show that orbits are always detected at the 3@xmath32 level when the semi - major axis is at least 70% of the hipparcos precision on a target . any orbital signature above 2.45 mas would have been detectable in this hipparcos dataset , and this helps to set an upper limit on the companion mass . using this assertion , we get a lower limit on inclination ( 1.22@xmath64 ) and an upper limit on the companion mass ( 66 @xmath53 ) . although the consideration of astrometric data does not allow us to put tight constraints on the inclination of the system , the non - detection of an orbit allows us to rule out a stellar binary system . verification of this could be achieved through high - contrast adaptive - optics imaging of the system at predicted apastron passage . figure [ sep ] shows the projected and angular separation of the planet and star for one complete face - on orbit , where phase zero corresponds to superior conjuction as described by @xcite . an additional consequence of our astrometric constraint is that the transit probability is increased by a small amount since inclinations below 1.22 degrees are ruled out . ) separation of hd 20782b from the host star as a function of orbital phase , where phase zero corresponds to superior conjunction.,width=309 ] as described in section [ science ] , one of the most interesting aspects of hd 20782b is the relatively large transit probability compared with the orbital period . the transit probability is a function of the stellar and planetary radii and the star planet separation along the line of sight @xcite . for hd 20782 , we use the stellar radius shown in table [ stellar ] and adopt a planetary radius of @xmath81 @xmath82 given the minimum mass of 1.41 @xmath53 ( see table [ planet ] ) and using the mass - radius relationship described by @xcite . if the planet were in a circular orbit with the same semi - major axis , the transit probability would be 0.4% . the extreme eccentricity of the orbit results in star planet separation of 0.061 au at periastron and 0.076 au at inferior conjunction where a transit is possible . such a separation is similar to that of a hot jupiter in a circular orbit . inferior conjunction occurs when @xmath83 ; in this case , the true anomaly is @xmath84 at the time of mid - transit . this orbital orientation results in an enhanced transit probability of 7.1% . a further influence of the high eccentricity on the transit parameters is the expected transit duration . since the separation at inferior conjunction is comparable to a hot jupiter , the duration is likewise reduced and has an amplitude of 0.13 days ( 3.1 hours ) for a central transit . the epoch of mid - transit shown in table [ planet ] was calculated using the same monte - carlo bootstrap method used to calculate the orbital parameter uncertainties . the time of mid - transit corresponds to a calendar date of 15 january 2015 and a ut of 7:02 . the uncertainty on this time is 0.1 days which results in a total 1@xmath32 transit window of 0.33 days ( 7.6 hours ) . the estimated depth of the transit is 0.96% and so should be readily observable in typical millimag photometry . however , the infrequent occurrence of such events ( see section [ science ] ) motivated observations from both ground and space . the derived physical and orbital properties of hd 20782b described in previous sections motivated photometric monitoring of the host star for stellar variability and planetary transit / phase signatures . here we describe our photometric observations and results in detail . we collected a total of 191 nightly photometric observations of hd 20782 during its 201314 , 201415 , and 201516 observing seasons to search for stellar variability . the observations were acquired with the t8 0.80 m automatic photoelectric telescope ( apt ) , one of several automated telescopes operated by tennessee state university ( tsu ) located at fairborn observatory in southern arizona . the t8 apt is equipped with a two - channel precision photometer that uses a dichroic filter and two emi 9124qb bi - alkali photomultiplier tubes to measure the strmgren @xmath85 and @xmath86 pass bands simultaneously . we computed the differential magnitudes of hd 20782 with respect to the mean brightness of its three constant comparison stars . to improve the precision further , we combined the differential @xmath85 and @xmath86 observations into a single @xmath87 `` passband . '' the tsu apts , their precision photometers , observing strategy , and data reduction techniques are described in detail by @xcite . a summary of the photometric data for hd 20782 is given in table [ aptsum ] . ccccc 201314 & 43 & 5662256701 & 1.01241 & 0.00183 + 201415 & 89 & 5694357045 & 1.01206 & 0.00228 + 201516 & 59 & 5729357390 & 1.01128 & 0.00171 the nightly observations of hd 20782 are plotted in the top panel of figure [ aptphot ] in our @xmath87 passband . the observing seasons are quite short from fairborn observatory , only three months in length , because of the star s southerly declination of @xmath88 . the observations scatter about their grand mean ( indicated by the horizontal dotted line ) with a standard deviation of 0.00205 mag , as given in the upper right corner of the top panel . this is essentially the limit of precision for the hd 20782 observations because the star s southerly declination results in all measurements being made at air mass values between 2.0 and 2.4 ( see @xcite , figure 8) . the middle and bottom panels of figure [ aptphot ] show the frequency spectrum of the data set and the phase curve computed with the best frequency , respectively . our frequency analysis is based on least - squares sine fits with trial frequencies between 0.01 and 0.95 c / d , corresponding to periods between one and 100 days . the goodness of fit at each frequency is measured as the reduction factor in the variance of the original data , whose value lies between the extremes of 0.0 and 1.0 . a reduction factor of 0.0 corresponds to the case where the variance in the residuals from a least - squares sine fit to the observational data at some trial frequency have the same value as the variance in the original data , i.e. , no reduction in the variance takes place at that particular frequency . a reduction factor of 1.0 corresponds to the extreme case where the variance in the residuals of the sine fit is 0.0 , i.e. , the sine curve fits the data perfectly with no residuals . the frequency spectrum in the middle panel shows several peaks with reduction factors near 0.1 , but no peak stands out above the others to suggest a stellar rotation period . we ran simulations adding computed sine curves to our data sets and found that coherent variations with peak - to - peak amplitudes of @xmath89 mag or larger would be detectable in our light curves . this places an upper limit to any periodic modulation for hd 20782 , such as rotational modulation caused by starspots . this is consistent with the low level of magnetic activity ( logr@xmath90 = -4.91 ) given in the discovery paper of @xcite and demonstrates that our best - fit period of 1.2619 days in the bottom panel is spurious . in addition to the absense of rotational modulation , we find no evidence for longer - term variability ; the three seasonal means in table [ aptsum ] scatter about their grand mean with a standard deviation of only 0.00058 mag . given the size of the transit windows and the relatively infrequent opportunities to observe them ( see sections [ science ] and [ transit ] ) , we elected to make use of the microvariability and oscillations of stars ( most ) satellite to observe hd 20782 during the next scheduled transit window . most has an aperture of 15 cm and a filter passband covering the range 375675 nm , making it well suited to obtain precision optical photometry of very bright stars @xcite . observations of hd 20782 commenced at hjd 2457035.3 ( 2015 january 12 19:11 ut ) and concluded 7 days later at hjd 2457042.3 ( 2015 january 19 19:11 ut ) . the predicted time of mid - transit ( see table [ planet ] ) was bjd 2457037.794 ( 2015 january 15 07:02 ut ) . the star is outside of most s continuous viewing zone and so required observations outside of normal operational parameters . for each 101 min orbit , most was able to acquire the target field for 20 mins . exposure times were 0.6 secs to allow for both the brightness of the target and scattered light due to the roll angle of the spacecraft with respect to the sun . this resulted in photometry with a 1@xmath32 rms precision of 0.07% . during the week of most observations , a total of 257 measurements of hd 20782 were acquired . the resulting relative photometry of the data are shown in figure [ transitfig ] , along with a solid line that indicates the predicted location and depth of a possible transit . the 1@xmath32 transit window ( 0.33 days ) was described in section [ transit ] . we use the 3@xmath32 transit window ( 0.43 days ) to draw vertical dashed lines in figure [ transitfig ] . a central transit of the planet ( impact parameter of @xmath91 ) is ruled out for most locations within the transit window . the cadence of the observations is such that a transit duration half that of a central transit could have been missed within the 1@xmath32 transit window . such a duration corresponds to an impact parameter of @xmath92 , above which transits can not be ruled out by our photometry . a further consideration is the detection of the rossiter - mclaughlin ( r - m ) effect during a possible transit . the amplitude of the r - m effect is shown by @xcite to be @xmath93 using our stellar parameters from table [ stellar ] and the transit parameters described in section [ transit ] , the amplitude of the r - m effect for a transit of hd 20782b is predicted to be @xmath215.5 ms@xmath15 . two of our rv measurements ( one each from aat and paras ) are within the transit window , shown close to 0.5 phase in the bottom panel of figure [ rv ] . neither of these measurements show evidence of any significant deviation from our keplerian model . thus the rv data are consistent with the most photometry leading to the conclusion that the planet does not transit the host star . the phase variations of a planet as it orbits the host star has become a detectable signature in the era of high - precision photometry . numerous examples of phase signatures have been detected from the planets detected with the _ kepler _ mission @xcite . exoplanets that are close to their host stars have generally been found to have low geometric albedos , such as the low geometric albedo of hat - p-7b @xcite and the null detection of phase variations from hd 209458b @xcite . there are exceptions to the rule , however , such as the case of kepler-7b @xcite , and it is likely that a greater understanding of atmospheric processes is needed to explain this diversity @xcite . @xcite developed a geometric albedo model that scales the geometric albedo with star planet separation . the implication of the model for planets in eccentric orbits is that the geometric albedo is time dependent , with an assumption that reflective / scattering condensates in the upper atmosphere are removed during periastron passage by the increase in radiative flux from the host star . the generalized expression for the planet to host flux ratio is given by @xmath94 where @xmath26 is the phase angle , @xmath95 is the geometric albedo , @xmath96 is the phase function , @xmath97 is the planetary radius , and @xmath98 is the star planet separation . this separation is given by @xmath99 where @xmath100 is the true anomaly . the phase angle , @xmath26 , is defined to be zero at superior conjunction . a model of geometric albedo time - dependence assumes that the planetary atmosphere responds to the change in incident flux on timescales comparable to the duration of the periastron encounter . this effect has been modeled for the eccentric planet hat - p-2b at infrared wavelengths by @xcite . thus @xcite predicted that , although the largest phase variations of eccentric planets occur during a relatively short fraction of their orbital phase , the amplitude of the signature would be lowered by the subsequent darkening of their atmospheres during periastron . + + + for hd 20782b , we calculated the predicted phase variations of the planet with respect to the inferior conjunction ( transit ) and periastron times , shown as a dashed line in the top panel of figure [ phasefig ] . these orbital locations are very close to each other ( see figure [ orbitfig ] ) , separated by only 0.66 days . the location of superior conjunction where @xmath101 occurs 5.63 days after the periastron passage . all three of these orbital locations are covered by the most observations described in section [ most ] . we include the additional effects of ellipsoidal variations @xcite and doppler boosting @xcite in the top panel of figure [ phasefig ] , shown as dotted and dot - dashed lines respectively . the combined effect of all three ( phase variations , ellipsoidal variations , and doppler boosting ) is shown as a solid line . for the ellipsoidal component , we have assumed a gravity darkening exponent of @xmath102 @xcite . for the doppler boosting coefficient , we calculate a value of @xmath103 using the stellar temperature from table [ stellar ] and the methodology of @xcite . using the model of a distance - dependent geometric albedo and hilton phase function @xcite , we determined that the amplitude of the phase variations is comparable to the doppler boosting , whereas the ellipsoidal component is a minor contribution to the total flux variations . another point worth noting is that this model assumes an orbit that is close to edge - on . the effect of orbital inclination on the relative amplitudes of the three contributing components is minor except for orbits close to face - on @xcite . as described in section [ most ] , the original intent of acquiring the most data was for the purpose of observing a potential transit event . evidence of phase variations was unexpected due to the low predicted amplitude shown in the top panel of figure [ phasefig ] . to determine the overall trend in the most data , we calculated a running mean of the data using 20 data points either side of each measurement to calculate the running mean at that location . the results of this calculation are shown as a solid line along with the individual measurements ( including error bars ) in the middle panel of figure [ phasefig ] , where we have adjusted the vertical scale of the plot to the range of the running mean values , using the average of the running mean values as the zero - point . the apparent brightening of the host star between the truncated dates of 38 and 39 on the plot is where the peak of the phase variations are predicted to occur . this was diagnosed from an instrumentation point of view , and it was determined that the change in the brightness was not caused by any aspect of the most instrumentation or data reduction issues . we tested the likelihood whether this could be caused by an alignment between intrinsic stellar variability and the expected periastron passage , by conducting a monte - carlo simulation in which we treat the observed data as representative of possible stellar variability and randomly rearrange the data to see how often a similar chance alignment can occur . each random permutation of the observed flux values to the times of observation resulted in a new dataset for which the running mean was calculated and then analyzed for significant peaks in the flux . the percentage of simulations for which a specific criteria was met was taken as the probability that the criteria would have been satisfied by chance . based on 10,000 realizations of this simulation , the probability of a peak occurring in the 3839 date range is @xmath217% , and the probability of that peak being of equal or greater amplitude than the observed peak is @xmath24% . if indeed the observed peak is related to the close passage of the planet to the star , the flux variations may indicate that the assumption by @xcite that the presence of reflective condensates in a planetary atmosphere changes on timescales comparable with the periastron passage is likely incorrect for highly eccentric orbits . in fact , the larger the eccentricity , the more inconsistent the assumption becomes with the radiative and advective timescales of the atmosphere . furthermore , the possible presence of phase variations indicates the companion is not self - luminous , further supporting the claim that the companion is planetary rather than stellar in nature @xcite . to investigate this further , we binned the most photometry into 15 evenly spaced time intervals . the best - fit model to the binned data is shown in the bottom panel of figure [ phasefig ] where the model includes ellipsoidal variations and doppler boosting as well as phase variations . the best - fit inclination of the planetary orbit is @xmath104 . a fit of the data to both the described model and a constant model resulted in @xmath105 which shows that the phase model is quantitatively favored . however , the model uses a companion radius of @xmath25 jupiter radii and a geometric albedo of unity , which is a physically unlikely scenario . thus there is either an additional component missing in our model of the data , or the data may be insufficient to fully characterize the flux variations , or some combination of the two . as noted above , the most compelling aspects of the variations described here are the timing of the variations with those predicted by the phase model , combined with the extreme eccentricity of the planet . this system is clearly highly unusual amongst the known exoplanets , and we can not exclude the possibility of unaccounted for physics occuring during the extreme conditions of periastron . 1400 k. the passband boundaries of most are indicated by the vertical dashed lines . the blackbody calculation assumes zero bond albedo and zero heat redistribution ( hot dayside model ) and thus represents a maximum flux scenario . of the integrated flux , 0.02% falls within the most passband.,width=309 ] a possible missing factor is that of thermal emission . this has been shown to be a significant component at the _ kepler _ passband @xcite . the _ kepler _ passband however is significantly broader than that used by most ( see section [ most ] ) . we calculated this component for our observations by estimating the equilibrium temperature of the planet . to do so , we assumed the most extreme case of zero heat redistribution ( hot dayside ) and zero bond albedo @xcite . this produces a peak equilibrium temperature at periastron of @xmath21400 k. the resulting blackbody spectrum is shown in figure [ blackbody ] along with the passband of most , depicted as vertical dashed lines . of the integrated flux from the thermal emission , only 0.02% of the total flux falls within the passband of our observations . this corresponds to a flux ratio of planet to star thermal emission in the most passband of @xmath106 . we conclude that any phase variations due to the planet are dominated by the optical component . further data with higher precision are needed to confirm the presence of the variations and constrain the reflective properties of this fascinating planet as it passes through periastron . exoplanets in eccentric orbits remain some of the most intriguing discoveries of recent decades . although the semi - amplitude of their rv variations is systematically higher , the orbits of highly eccentric planets are difficult to characterize due the rapid variation at periastron passage . we have refit the orbit of hd 20782b , consistent with it being the most extreme of these eccentricity cases and have provided new stellar and planetary parameters . our rv measurements acquired during the brief duration of periastron passage allow a detailed orbital ephemeris to be constructed , despite the relatively long period of @xmath218 months . our analysis of the _ hipparcos _ astrometry for hd 20782 constrains the inclination sufficiently such that the companion is likely to be planetary rather than stellar . the uncertainties associated with our astrometric analysis leave open the possibility that the companion lies within the brown dwarf mass regime . it is expected that further astrometric data from the gaia mission will significantly improve these constraints @xcite . even with a relatively high transit probability of @xmath27% , we have shown that the planet does not transit the host star . the possible phase variations soon after periastron might be induced in part by stellar light reflected off the planet s atmosphere . although our modeling is incomplete , if this hypothesis is true then it raises interesting questions regarding the conditions to which such an extreme orbit exposes the planet . in particular , the effects of rotation rate and radiative / advective timescales on atmospheric dynamics may be overwhelmed by the short yet intense conditions that occur at the closest approach of the planet to the star . it has been further noted by several studies that the brightest region of the planet is shifted westward of the substellar point , caused by a relatively cloudy western hemisphere @xcite . additionally , it seems likely that , although planets in short - period orbits tend to have relatively low geometric albedos , long - period planets in eccentric orbits retain a high geometric albedo during the periastron passage since the atmosphere does not have time to respond to the change in incident flux . the result of this is a higher than expected flux ratio of the planet to the star at optical wavelengths . thus , eccentric planets present a particularly lucrative observing opportunity for the study of planetary atmospheres , provided one is able to accurately predict when the peak flux variations are expected to occur . further observations of this system at times close to inferior conjunction are highly encouraged . the next two times of inferior conjunction predicted from our ephemeris are bjd @xmath107 ( 2016 september 3 8:36 ut ) and bjd @xmath108 ( 2018 april 23 10:10 ut ) . in each case , the subsequent superior conjunction occurs @xmath26.29 days after the inferior conjunction . matching these times to those when the target is most visible is not trivial and the timescale of the periastron passage is best suited to continuous space - based observations . possibilities for these upcoming windows would be a perfect use for upcoming missions that are optimized for bright star observations , such as the characterizing exoplanet satellite ( cheops ) . a deeper understanding of the orbits and atmospheres of eccentric planets are key milestones towards unlocking the origin and nature of these mysterious objects . the authors would like to thank the anonymous referee , whose comments greatly improved the quality of the paper . s.r.k and n.r.h . acknowledge financial support from the national science foundation through grant ast-1109662 . g.w.h . acknowledges long - term support from tennessee state university and the state of tennessee through its centers of excellence program . h.r.a.j acknowledges support from stfc via grants st / m001008/1 and leverhulme trust rpg-2014 - 281 . the authors thank the gurushikar , mt . abu observatory staff and the paras technical staff for the observations with paras . the paras program is supported completely by the physical research laboratory , dept . of space , govt . of india . the results reported herein benefited from collaborations and/or information exchange within nasa s nexus for exoplanet system science ( nexss ) research coordination network sponsored by nasa s science mission directorate . brewer , j.m . , giguere , m. , fischer , d.a . 2014 , pasp , 126 , 48 bouchy , f. , pepe , f. , queloz , d. 2001 , a&a , 374 , 733 butler , r.p . , marcy , g.w . , williams , e. , mccarthy , c. , dosanjh , p. , vogt , s.s . 1996 , pasp , 108 , 500 butler , r.p . , tinney , c.g . , marcy , g.w . , et al . 2001 , apj , 555 , 410 butler , r.p . , wright , j.t . , marcy , g.w . , et al . 2006 , apj , 646 , 505 chakraborty , a. , mahadevan , s. , roy , a. , et al . 2014 , pasp , 126 , 133 chatterjee , s. , ford , e.b . , matsumura , s. , rasio , f.a . 2008 , apj , 686 , 580 delgado mena , e. , israelian , g. , gonzlez hernndez , j.i . , bond , j.c . , santos , n.c . , udry , s. , mayor , m. 2010 , apj , 725 , 2349 demarque , p. , woo , j .- h . , kim , y .- c . , yi , s.k . 2004 , apjs , 155 , 667 demory , b .- o . , seager , s. , madhusudhan , n. , et al . 2011 , apj , 735 , l12 demory , b .- o . , 2014 , apj , 789 , l20 diego , f. , charalambous , a. , fish , a.c . , walker , d.d . 1990 , proc . photo - opt . eng . , 1235 , 562 drake , a.j . 2003 , apj , 589 , 1020 esteves , l.j . , de mooij , e.j.w . , jayawardhana , r. 2013 , apj , 772 , 51 esteves , l.j . , de mooij , e.j.w . , jayawardhana , r. 2015 , apj , 804 , 150 faigler , s. , mazeh , t. 2011 , mnras , 415 , 3921 fortney , j.j . , lodders , k. , marley , m.s . , freedman , r.s . 2008 , apj , 678 , 1419 fossey , s.j . , waldman , i.p . , kipping , d.m . 2009 , mnras , 396 , l16 garcia - melendo , e. , mccullough , p.r . 2009 , apj , 698 , 558 gaudi , b.s . , winn , j.n . 2007 , apj , 655 , 550 goldreich , p. , soter , s. 1966 , icarus , 5 , 375 henry , g.w . 1999 , pasp , 111 , 845 hinkel , n.r . , timmes , f.x . , young , p.a . , pagano , m.d . , turnbull , m.c . 2014 , aj , 148 , 54 hu , r. , demory , b .- o . , seager , s. , lewis , n. , showman , a.p . 2015 , apj , 802 , 51 jones , h.r.a . , butler , r.p . , tinney , c.g . , 2006 , mnras , 369 , 249 jones , h.r.a . , bulter , r.p . , tinney , c.g . , et al . 2010 , mnras , 403 , 1703 kaib , n.a . , raymond , s.n . , duncan , m. 2013 , nature , 493 , 381 kane , s.r . , von braun , k. 2008 , apj , 689 , 492 kane , s.r . , mahadevan , s. , von braun , k. , laughlin , g. , ciardi , d.r . 2009 , pasp , 121 , 1386 kane , s.r . , gelino , d.m . 2010 , apj , 724 , 818 kane , s.r . , gelino , d.m . 2011a , apj , 729 , 74 kane , s.r . , gelino , d.m . 2011b , apj , 741 , 52 kane , s.r . , henry , g.w . , dragomir , d. , et al . 2011 , apj , 735 , l41 kane , s.r . , gelino , d.m . 2012a , pasp , 124 , 323 kane , s.r . , gelino , d.m . 2012b , mnras , 424 , 779 kane , s.r . , ciardi , d.r . , gelino , d.m . , von braun , k. 2012 , mnras , 425 , 757 kane , s.r . 2013 , apj , 766 , 10 kane , s.r . , boyajian , t.s . , henry , g.w . , et al . 2015 , apj , 806 , 60 latham , d.w . , mazeh , t. , stefanik , r.p . , mayor , m. , burki , g. 1989 , nature , 339 , 38 laughlin , g. , deming , d. , langton , j. , et al . 2009 , nature , 457 , 562 lewis , n.k . , knutson , h.a . , showman , a.p . , et al . 2013 , apj , 766 , 95 lodders , k. , plame , h. , gail , h .- p . 2009 , landolt - brnstein - group vi astronomy and astrophysics numerical data and functional relationships in science and technology volume 4b : solar system . edited by j.e . trmper , 4b , 44 loeb , a. , gaudi , b.s . 2003 , apj , 588 , l117 lucy , l.b . 1967 , za , 65 , 89 mack , c.e . , schuler , s.c . , stassun , k.g . , norris , j. 2014 , apj , 787 , 98 marcy , g.w . , butler , r.p . 1996 , apj , 464 , l147 matthews , j.m . , kuschnig , r. , guenther , d.b . , 2004 , nature , 430 , 51 mazeh , t. 2008 , eas , 29 , 1 meibom , s. , mathieau , r.d . 2005 , apj , 620 , 970 moutou c. , hbrard , g. , bouchy , f. , et al . 2009 , a&a , 498 , l5 naef , d. , latham , d.w . , mayor , m. , et al . 2001 , a&a , 375 , l27 neves , v. , santos , n.c . , sousa , s.g . , correia , a.c.m . , israelian , g. 2009 , a&a , 497 , 563 otoole , s.j . , tinney , c.g . , jones , h.r.a . , butler , r.p . , marcy , g.w . , carter , b. , bailey , j. 2009 , mnras , 392 , 641 perryman , m. , hartman , j. , bakos , g.a . , lindegren , l. 2014 , apj , 797 , 14 petrovich , c. , tremaine , s. , rafikov , r. 2014 , apj , 786 , 101 pont , f. , husnoo , n. , mazeh , t. , fabrycky , d. 2011 , mnras , 414 , 1278 rowe , j.f . , matthews , j.m . , seager , s. , et al . 2008 , apj , 689 , 1345 sahlmann , j. , sgransan , d. , queloz , d. , et al . 2011 , a&a , 525 , 95 seager , s. , richardson , l.j . , hansen , b.m.s . , et al . 2005 , apj , 632 , 1122 shporer , a. , hu , r. 2015 , aj , 150 , 112 takeda , g. , ford , e.b . , sills , a. , rasio , f.a . , fischer , d.a . , valenti , j.a . 2007 , apjs , 168 , 297 tinney , c.g . , wittenmyer , r.a . , butler , r.p . . 2011 , apj , 732 , 31 tokovinin , a. , fischer , d.a . , bonati , m. , et al . 2013 , pasp , 125 , 1336 valenti , j.a . , butler , r.p . , marcy , g.w . 1995 , pasp , 107 , 966 valenti , j.a . , piskunov , n. 1996 , a&as , 118 , 595 valenti , j.a . , fischer , d.a . 2005 , apjs , 159 , 141 valenti , j.a . , fischer , d. , marcy , g.w . , et al . 2009 , apj , 702 , 989 van eylen , v. , albrecht , s. 2015 , apj , 808 , 126 van leeuwen , f. 2007a , hipparcos , the new reduction of the raw data , astrophys . space sci . lib . , 350 van leeuwen , f. 2007b , a&a , 474 , 653 vogt , s.s . , wittenmyer , r.a . , butler , r.p . 2010 , apj , 708 , 1366 wang , s.x . , wright , j.t . , cochran , w. , et al . 2012 , apj , 761 , 46 walker , g. , matthews , j. , kuschnig , r. , et al . 2003 , pasp , 115 , 1023 welsh , w.f . , orosz , j.a . , seager , s. , et al . 2010 , apj , 713 , l145 wittenmyer , r.a . , horner , j. , tuomi , m. , et al . 2012 , apj , 753 , 169 wittenmyer , r.a . , horner , j. , tinney , c.g . , et al . 2014 , apj , 783 , 103 wright , j.t . , howard , a.w . 2009 , apjs , 182 , 205 zucker , s. , mazeh , t. , alexander , t. 2007 , apj , 670 , 1326

we investigate different mean - field - like approximations for stochastic dynamics on graphs , within the framework of a cluster - variational approach . in analogy with its equilibrium counterpart , this approach allows one to give a unified view of various ( previously known ) approximation schemes , and suggests quite a systematic way to improve the level of accuracy . we compare the different approximations with monte carlo simulations on a reversible ( susceptible - infected - susceptible ) discrete - time epidemic - spreading model on random graphs .

1702.06822

in statistical mechanics , the _ cluster variational method _ ( cvm ) , originally proposed by kikuchi in 1951 @xcite , is a well - established technique , which provides a systematic way to improve the accuracy of the simplest mean - field approximation . the classical scope of application is that of equilibrium thermodynamics of lattice models with discrete degrees of freedom , such as ising or potts @xcite . the basic idea is to cut off correlations among site variables beyond the range of certain _ maximal clusters _ ( for instance , square `` plaquettes '' on a square lattice ) , so that the level of accuracy can be somehow `` tailored '' by means of a suitable choice of such clusters . the output of the calculation is the set of marginals of the boltzmann distribution over the maximal clusters . in particular , choosing single sites or nearest - neighbor pairs as maximal clusters , the cvm recovers , respectively , the ordinary mean - field ( bragg - williams ) approximation and the bethe - peierls approximation @xcite . more generally , one can show that a number of generalized mean - field approximations , originally developed by heuristic arguments , can be incorporated into the cvm framework @xcite . let us also mention the fact that the method has gained a renewed interest in recent years , because of the formal analogy between equilibrium statistical mechanics and a number of inference problems of high technological relevance ( error - correction decoding , pattern recognition , etc . ) , which ultimately require computing marginals of the boltzmann distribution for ( heterogeneous ) statistical - mechanical models defined on ( very large ) graphs @xcite . in the complementary field of non - equilibrium statistical mechanics , a rigorous foundation from first principles is still lacking , so that a considerable part of research work is devoted to model systems in which the ( stochastic ) dynamics is assumed to be known , and properly defined to describe some kind of phenomena . this class of models includes , for instance , epidemic - spreading models @xcite , kinetic ising or ising - like models @xcite ( including models of neural - network dynamics ) @xcite , models of opinion formation and social dynamics ( voter model and its numerous variants ) @xcite , transport models ( exclusion processes ) @xcite and so on . increasing research efforts have also been oriented to investigating the aforementioned dynamics on complex networks , with special attention to the role played by the latter @xcite . most of the stochastic processes involved in these models are of the markovian ( i.e. , memoryless ) type , fully characterized by a matrix of transition probabilities ( stochastic matrix ) . the problem can be generally rephrased in a variational form through a _ thermodynamic formalism _ , fully analogous to that of equilibrium statistical mechanics , where the transition probabilities play the role of interactions , and , within this formalism , one can develop mean - field - like approximation strategies along the lines of the cluster variational method . such a `` dynamical analog '' of the cvm was indeed put forward in the 1960s by kikuchi himself @xcite , who named it _ path probability method _ ( ppm ) . the latter method has not become so popular as its equilibrium counterpart , so that in fact most mean - field - like approaches to stochastic dynamics are still recently conceived on the basis of traditional heuristic reasoning , without resorting to the cluster - variational machinery @xcite . at odds with the cited works , a path - probability approach has been recently followed by one of the authors @xcite , to develop a new type of approximation which , due to the peculiar `` shape '' of the maximal clusters , has been called the _ diamond approximation_. the latter has been applied to kinetic ising models ( with glauber dynamics , and either symmetric or asymmetric couplings ) , but it can be straightforwardly used for other markov - chain models as well , provided the transition matrix has a suitable factorized form , namely that ( the probability of ) the next - time configuration of each site variable depends only on the current - time configuration of its neighbors . the method has been the subject of a detailed testing work by an independent group @xcite , still on various types of kinetic ising - like models , demonstrating a remarkable accuracy for both the transient and the steady state , with just a slight breakdown in the notoriously difficult case of spin glasses at low temperature . let us stress the fact that the derivation of the diamond approximation in reference @xcite introduces a novelty item with respect to a `` standard '' ppm approach @xcite , in that it removes an unnecessary constraint in the choice of maximal clusters , potentially opening the way to new approximation schemes . in this paper we analyze different alternative approximations that can be worked out in the framework of a path - probability approach along the lines of reference @xcite . this analysis leads in particular to generalizing the diamond approximation in such a way that the transition probability of a given site variable may also depend on the current configuration of the site itself . this step forward is necessary for the method to be applicable to models with this feature , such as , for instance , epidemic models or the like . we also trace a number of connections with different approximation schemes , previously derived either by the conventional `` heuristic '' approach or by the `` standard '' ppm approach . in the second part of the paper , we evaluate the performance of these different approximations in terms of both accuracy and computational complexity , with respect to monte carlo simulations . as a test model , we consider a susceptible - infected - susceptible ( sis ) epidemic model on large random graphs , possibly with heterogeneous infection parameters . it turns out that the generalized diamond approximation outperforms simpler , previously known approximation schemes , with only a slight increase in the computational complexity . let us note that our investigation is carried out under the assumption of a discrete - time dynamics , but all the approximations can be applied as well to describe continuous - time processes . concerning this issue , we prove in particular that the ( generalized ) diamond approximation actually degenerates , in the continuous - time limit , into a simpler ( pairwise ) approximation . from the technical point of view , the latter result entails that , even though it is still possible to get a more accurate approximation for continuous - time models , one necessarily has to resort to even larger maximal clusters , which of course requires a larger computational effort . the paper is organized as follows . section [ sec : cluster - variational_approach ] contains the main theoretical issues . sub [ subsec : the_reference_model ] defines the most general model we can deal with , which we denote as the _ reference model_. in sub [ subsec : thermodynamic_formalism ] we introduce the thermodynamic formalism and the basic concepts of the cluster variational method . in sub [ subsec : specific_approximations ] we describe the specific approximations which are the subject of the present paper , in particular the one which generalizes the diamond approximation . most analytical details , along with a scheme of the resulting computation procedures , are reported in appendices [ app : details ] ( generalized diamond approximation ) and [ app : other_approximations ] ( other approximations ) . appendix [ app : continuous_time ] deals with the special case of continuous - time processes , mentioned above . section [ sec : test_models ] describes the numerical tests , whereas section [ sec : conclusions ] contains a summary of the results and some concluding remarks . as mentioned above , we consider quite a generic model , made up of discrete random variables @xmath0 , associated with the vertices ( or _ sites _ ) @xmath1 of a given undirected graph , and depending on a discrete time index @xmath2 , where @xmath3 may possibly tend to infinity . this collection of random variables may be regarded as a multivariate stochastic process @xmath4 $ ] , whereas a particular realization of such a process @xmath5 may be denoted as a _ path_. accordingly , we define the _ path probability _ as the probability of each given path , namely @xmath6 in the assumption of a markovian dynamics , the path probability takes the general form @xmath7 where @xmath8 is the initial condition ( i.e. , the probability of the initial configuration @xmath9 ) and @xmath10 are the transition probabilities from @xmath11 to @xmath12 ( i.e. , the conditional probability of the latter configuration at time @xmath13 , given the former at time @xmath14 ) . the graph structure defines which variables in the `` current - time '' configuration are actually relevant to determine the `` next - time '' configuration , and therefore the structure of the transition probabilities . in many physical models , such as those mentioned in the introduction , the next - time configuration of each given site depends only on the current - time configuration of its neighbors and also possibly of the same site . in formulae , this statement means that the transition probabilities can be written in the following factorized form @xmath15 where the product runs over all sites , @xmath16 and @xmath17 ( respectively , @xmath18 and @xmath19 ) denote two generic configurations of the whole graph ( respectively , of the @xmath20 site ) and a next - time configuration by a letter @xmath17 . ] , and @xmath21 is a shorthand for a list of configuration variables @xmath22 ( i.e. , those associated with the @xmath20 site and its neighbors ) . we shall also assume that the initial configurations of different sites are statistically independent , namely that @xmath23 the latter assumption could also be partially relaxed , but usually there is no specific interest in doing so . let us now consider the following variational functional @xmath24 = \sum_{x^{(0)}\!,\dots , x^{(\tau ) } } { \wp}\,(x^{(0)}\!,\dots , x^{(\tau ) } ) \ln \frac { { \wp}\,(x^{(0)}\!,\dots , x^{(\tau ) } ) } { { p}^{(0)}(x^{(0 ) } ) \prod_{t=0}^{\tau-1 } { w}^{(t)}(x^{(t+1)}|\,x^{(t ) } ) } \ , , \ ] ] where the sum runs over all possible paths , the path probability @xmath25 is regarded as a variational parameter , and the initial distribution @xmath26 and the transition probabilities @xmath27 are assumed to be `` input parameters '' , incorporating our knowledge of the stochastic dynamics . from the mathematical point of view , this functional is a so - called _ kullback - leibler divergence _ between the two sides of equation , and is known to have an absolute minimum ( zero ) where the two distributions are equal , that is when equation is verified . introducing the shannon information entropy associated with the path probability , namely @xmath28 = - \sum_{x^{(0)}\!,\dots , x^{(\tau ) } } { \wp}\,(x^{(0)}\!,\dots , x^{(\tau ) } ) \ln { \wp}\,(x^{(0)}\!,\dots , x^{(\tau ) } ) \ , , \ ] ] we can rewrite the variational functional as @xmath24 = \sum_{x^{(0)}\!,\dots , x^{(\tau ) } } \wp\,(x^{(0)}\!,\dots , x^{(\tau ) } ) \ , \varepsilon\,(x^{(0)}\!,\dots , x^{(\tau ) } ) - \mathcal{s}[{\wp } ] \ , , \label{eq : free_energy}\ ] ] where we have also defined @xmath29 equation points out that the variational functional @xmath30 $ ] is analogous to a free energy for a system in which time plays the role of an extra dimension , and whose energy function @xmath31 is defined by . more precisely , such a system can be viewed as a `` stack '' of @xmath32 `` copies '' of the original graph , each one labeled by a time index @xmath33 , so that we shall denote it as a _ space - time thermodynamic system_. the path probability minimizing @xmath30 $ ] is analogous to a boltzmann distribution for the space - time thermodynamic system . up to this point , we have got just a reformulation of the problem , which is of no practical use in a fully general case . nevertheless , our assumptions and about , respectively , transition probabilities and initial conditions allow us to rewrite the energy function as @xmath34 the latter expression points out that ( the logarithm of ) the elementary factors of the transition probabilities and of the initial conditions play respectively the role of interaction energies and external fields , and that both kinds of term are _ local _ in both space and time . as a consequence , since the energetic term of the free energy is local , we expect that the ( inherently nonlocal ) entropic term could be effectively approximated by a truncation of its _ cumulant expansion _ , according to kikuchi s cluster - variational method @xcite . each cumulant is associated to a given _ cluster _ ( i.e. , a given subset of lattice sites ) , so that different levels of approximation may be defined by the set of _ maximal clusters _ retained in the expansion . the approximate entropy can be finally recast into a linear combination of _ cluster entropies _ , i.e. , information entropies of _ cluster distributions _ ( the latter being obviously defined as joint probability distributions for the random variables associated to all sites in the given cluster ) . thus , a generic cvm entropy takes the following form @xcite @xmath35 = \sum_{\alpha \in \mathsf{r } } a_{\alpha } \mathcal{s}[{{\wp}_{\alpha } } ] \ , , \label{eq : cvm - entropy}\ ] ] where the sum runs over the set @xmath36 of all maximal clusters and their subclusters , @xmath37 denotes the cluster distribution for cluster @xmath38 ( i.e. , a particular marginal of @xmath25 ) , and @xmath39 $ ] is the corresponding information entropy . the coefficients @xmath40 of the combination , usually denoted as _ mbius numbers _ @xcite , or _ ( over)counting numbers _ @xcite , depend only on the graph geometry and the choice of maximal clusters . they can be determined by a simple sum rule , which can be stated as follows @xcite @xmath41 where @xmath38 may be any maximal cluster or subcluster , and the sum runs over all ( maximal clusters or subclusters ) @xmath42 containing @xmath38 . the rule shows in particular that the coefficient of any maximal cluster is equal to @xmath43 , because in that case the sum contains only one term . let us finally note that the choice of maximal clusters is generally driven by physical intuition , since it defines the range of correlations that a given approximation will be able to take into account . therefore , in the framework of a space - time system like the one we are interested in , maximal clusters should necessarily extend over different time steps , in order to take into account time correlations . kikuchi s _ path probability method _ ( ppm ) can be regarded , in this framework , as a particular way of choosing the maximal clusters , namely , one first chooses maximal clusters for a one - time system , then every maximal cluster of the space - time system is the union of a maximal cluster at time @xmath44 with its translation at time @xmath13 . as already noticed in reference @xcite , different criteria for this choice are possible , and may be more convenient , depending on the structure of the transition probabilities . in the following we shall see on test models that indeed one can obtain a considerably better tradeoff between accuracy and computational complexity , with respect to the conventional ppm choice . in this section we introduce the specific approximations ( i.e. , the different choices of maximal clusters ) , whose performance is analyzed in the following . first of all , let us define precisely the types of clusters ( either maximal or not ) that appear in at least one of the cvm entropies considered . looking at table [ tab : clusters ] we see that each cluster extends over either a single timestep or two consecutive ones . of course , since we assume that the graph does not change with time , clusters at different timesteps are simply translations of one another . we introduce specific symbols ( letters ) to denote the different cluster types and the related probability distributions ; such symbols are defined in table [ tab : clusters ] as well . .the 1st column displays cluster type identifiers . the 2nd and 3rd columns report lists of sites that belong to each cluster at timesteps @xmath44 and @xmath13 , respectively ( as usual @xmath45 denotes all the nearest neighbors of @xmath20 , whereas @xmath46 denote a nearest neighbor pair ) . the 4th column displays the definition of the probability distribution function ( pdf ) of each cluster . of course @xmath47 is a shorthand for @xmath48 , whereas @xmath49 is a shorthand for @xmath50 ( and the same for @xmath13 ) . analogous shorthands are used for the `` dummy '' variables @xmath16 and @xmath17 . [ cols="^,<,<,<,^,<",options="header " , ] making use of the sum rule , one can determine the mbius numbers , and therefore the approximate entropy functional , for each specific approximation . this is a rather standard calculation , whose results are summarized in table [ tab : superposition_coefficients ] . note that , as previously mentioned , all ( though not only ) the mbius numbers associated to the maximal clusters are equal to unity . note also that , in the m approximation , p , q , and r clusters are not maximal , so their mbius numbers do not need to be ( and in fact they are not ) equal to unity . in the remainder of this section we report the entropy functionals in formulae . as far as the p approximation is concerned , we have @xmath51 = \sum_{t=0}^{\tau-1 } \left\ { \sum_{i } \mathcal{s}[{p_{i}}^{(t ) } ] - \sum_{ij } \mathcal{s}[{z_{ij}}^{(t ) } ] \right\ } - \sum_{t=1}^{\tau-1 } \sum_{i } \mathcal{s}[{a_{i}}^{(t ) } ] \ , , \ ] ] where @xmath52 and @xmath53 denote sums over all sites and over all nearest neighbor pairs , respectively , and @xmath54}$ ] denotes the shannon entropy of each given probability distribution . the arguments @xmath55 mean that the functional @xmath56 depend on all p- , z- , and a - type cluster distributions , i.e. , @xmath57 is a shorthand for @xmath58 , and so on . with analogous notations , the pq entropy approximation reads @xmath59 & = \sum_{t=0}^{\tau-1 } \left\ { \sum_{i } \mathcal{s}[{p_{i}}^{(t ) } ] + \sum_{ij } \mathcal{s}[{q_{ij}}^{(t ) } ] - \sum_{i , j \in \partial i } \mathcal{s}[{t_{i , ij}}^{(t ) } ] \right\ } + \nonumber \\ & - \sum_{t=1}^{\tau-1 } \left\ { \sum_{ij } \mathcal{s}[{z_{ij}}^{(t ) } ] - \sum_{i } ( d_{i}-1 ) \mathcal{s}[{a_{i}}^{(t ) } ] \right\ } \ , , \end{aligned}\ ] ] where @xmath60 denotes a double sum over all sites and over all neighbors of each given site . furthermore , the pqr entropy reads @xmath61 = \sum_{t=0}^{\tau-1 } \left\ { \sum_{i } \left [ \mathcal{s}[{p_{i}}^{(t ) } ] + \mathcal{s}[{r_{i}}^{(t ) } ] + ( d_{i}-1 ) \mathcal{s}[{v_{i}}^{(t ) } ] \right ] + \right . \nonumber \\ & \left . + \sum_{ij } \mathcal{s}[{q_{ij}}^{(t ) } ] - \sum_{i , j \in \partial i } \left [ \mathcal{s}[{t_{i , ij}}^{(t ) } ] + \mathcal{s}[{u_{ij , i}}^{(t ) } ] \right ] \right\ } - \sum_{t=1}^{\tau-1 } \left\ { \sum_{i } \mathcal{s}[{s_{i}}^{(t ) } ] - \sum_{ij } \mathcal{s}[{z_{ij}}^{(t ) } ] \right\ } \ , , \label{eq : entropia - pqr}\end{aligned}\ ] ] whereas the m entropy reads @xmath62 = \sum_{t=0}^{\tau-1 } \left\ { \sum_{i } \mathcal{s}[{m_{i}}^{(t ) } ] - \sum_{ij } \mathcal{s}[{q_{ij}}^{(t ) } ] \right\ } - \sum_{t=1}^{\tau-1 } \left\ { \sum_{i } \mathcal{s}[{s_{i}}^{(t ) } ] - \sum_{ij } \mathcal{s}[{z_{ij}}^{(t ) } ] \right\ } \ , .\ ] ] for each approximation , the free - energy functional is obtained from equation by replacing the entropic term @xmath63 with one of the approximate entropies . as previously mentioned , the energetic term does not require any further approximation , because the maximal clusters are properly chosen to `` contain '' the elementary coupling terms of the energy function . the minimization of the resulting functionals is far from being a straightforward problem . this is ultimately due to cluster overlapping , which entails that compatibility constraints between cluster distributions must be satisfied . in general , the constrained minimization problem for a generic cvm free energy can be solved iteratively by a message - passing procedure denoted as _ generalized belief propagation _ @xcite ( which is not guaranteed to converge ) , or by more complex ( provably convergent ) methods @xcite . nevertheless , the peculiar structure of the couplings , arising in a space - time system describing a markovian stochastic dynamics like that of our reference model , allows one to devise a much simpler procedure , which does not require iterative refinement , but only a `` natural '' sequence of iterations , starting from the initial conditions and following the ( probabilistic ) time evolution of the system . we report the details of such a procedure in appendix [ app : details ] , only for the pqr approximation . the differences occurring in the other approximations ( which are indeed very much analogous ) are discussed in appendix [ app : other_approximations ] . furthermore , analogous procedures can be used to investigate continuous - time processes ( in a discretized form ) , and , as mentioned in the introduction , in such a case one can argue that the pqr and pq approximations become equivalent . the latter issues are discussed in appendix [ app : continuous_time ] . before switching to a performance evaluation of the different approximations described above , we find it worth summarizing a number of connections with previous works , which emerge from our analysis . as mentioned in the introduction , in most cases equivalent approximations have been put forward without resorting to a variational scheme , but just on the basis of heuristic reasoning , and/or for specific model dynamics . in such cases , our approach retains some interest in that it offers a more general and unified view of the matter . first of all , it is possible to recognize that the p and pq schemes turn out to be respectively equivalent to the classical ( non - variational ) _ mean - field _ and _ pairwise _ approximations , which have been used , for instance , to investigate epidemic spreading phenomena on networks @xcite . in such a context , the p approximation is customarily denoted as _ quenched mean field _ , in order to distinguish it from a naive mean - field approach , in which the infection probability is assumed to be a node - independent quantity @xcite , and from the popular _ heterogeneous mean - field approach _ by pastor - satorras and vespignani @xcite , in which nodes are distinguished only on the basis of their degree . for epidemics on networks that are expected to be inherently homogeneous , such as random - regular graphs or regular lattices , the same approximations have been considered also by petermann and de los rios @xcite , and other authors @xcite . we point out these equivalences in appendix [ subsec : pqandp ] along with some details about the respective iterative procedures . secondly , the m approximation turns out to be an instance of kikuchi s ( variational ) path - probability method , as we have mentioned in the previous subsection , but it has also been derived independently ( for the case of homogeneous epidemics ) again by petermann and de los rios @xcite , who denoted it as _ star approximation_. the latter equivalence is detailed in appendix [ subsec : m ] . furthermore , let us note that , in the special case of a spatially one - dimensional system ( i.e. , when the graph is a linear chain ) , a recent analysis of nonlinear voter models carried out by schweitzer and behera @xcite makes use of two different approximation schemes ( denoted as _ triplet _ and _ quintuplet _ approximations ) , which turn out to be special cases of the pq and m approximations , respectively . another special case of our reference model , which nonetheless includes several models of physical relevance ( such as the kinetic ising model , or the ordinary voter model ) , is that in which the elementary transition probability for a given site @xmath20 ( namely , @xmath64 ) does not depend on the current - time configuration of the site itself @xmath18 ( so that we can denote it as @xmath65 ) . in other words , the next - time configuration of each site is conditioned only by the current - time configurations of its neighbors . for this reason , we shall denote this case as a purely _ neighbor - conditioned _ dynamics . as mentioned in the introduction , the _ star _ and _ diamond _ approximations of reference @xcite refer to this kind of dynamics . actually , it turns out that , for a generic purely - neighbor - conditioned dynamics , the pq and p approximations degenerate into each other , and both coincide with the star approximation of reference @xcite . similarly , the m and pqr approximations degenerate into each other , and coincide with the diamond approximation . all these connections are detailed in appendices [ subsec : diamond ] and [ app : other_approximations ] . as a further consequence , we can state that in this case the diamond approximation is itself an instance of kikuchi s path probability method . in this section we test our approximations against monte carlo simulations on specific instances of a sis epidemic model on ( sparse ) graphs drawn from well known random ensembles , namely the erds - rnyi ensemble and the random - regular ensemble . in the former case , the degree distribution is poissonian , whereas in the latter case all vertices have the same degree , so that we also denote the two kinds of graphs as _ poissonian _ and _ uniform _ , respectively . we consider a sis model with parallel dynamics , whose definition is as follows . each site variable @xmath0 can take two possible values , @xmath66 , meaning that the @xmath20 site is susceptible or infected , respectively . an infected site @xmath20 has a probability @xmath67 to recover ( i.e. , to become susceptible ) at the next timestep . moreover , a susceptible site @xmath20 has a probability @xmath68 to become infected because of any of its neighbors @xmath69 , if the latter is infected at the current timestep , and @xmath70 otherwise . in other words , the probability of an infection event from @xmath69 to @xmath20 is @xmath71 . since infection events by different neighbors are not mutually exclusive , the _ total probability _ of a given site @xmath20 to become infected can be more easily computed via its complementary probability ( i.e. , the probability that @xmath20 remains susceptible ) , which is equal to the probability that none of the neighbors transmits the contagion . assuming independence among different infection events , the transition probability to a susceptible state can be finally written as @xmath72 whereas , by normalization , the transition probability to an infected state will be @xmath73 we consider two fixed graphs with @xmath74 vertices , a poissonian one with average degree @xmath75 and a uniform one with degree @xmath75 . as far as model parameters are concerned , in both cases we choose a uniform recovery probability , namely @xmath76 , whereas the infection probability is chosen to be homogeneous in the former case ( namely , @xmath77 ) and heterogeneous in the latter ( namely , @xmath68 drawn at random with uniform distribution between @xmath70 and @xmath78 ) . note that , as a result , both systems turn out to be heterogeneous : in the erds - rnyi system , heterogeneity arises from the random graph structure , whereas , in the random - regular system , heterogeneity arises from the randomness of infection probabilities . the initial conditions are chosen in such a way that both systems have on average 1% of infected sites , but with a slight difference . in the poissonian case , we use `` quenched '' initial conditions , that is , 1% of ( randomly drawn ) sites are assigned the initial distribution @xmath79 whereas the remaining 99% are assigned the distribution @xmath80 in the uniform case we use `` annealed '' initial conditions , that is , all sites are assigned the distribution @xmath81 in the simulation , quenched conditions correspond to all histories initialized with exactly the same configuration ( with 1% of infected sites ) drawn at random at the beginning , whereas annealed conditions correspond to a different initial configuration ( still with 1% of infected sites ) drawn at random for each history . note that there is no specific reason for associating a particular type of initial condition to a particular type of graph . we simply observe that different initial conditions give rise to a slightly different ( probabilistic ) short - time behavior of the system , appearing independently of the graph type , and we are meant to analyze whether such a behavior is more or less accurately reproduced by the approximations . for each approximation , each graph and parameters choice , and each site @xmath1 , we can compute the probability of the infected state ( infection probability ) as a function of time @xmath82 . in the simulations , the same quantity @xmath83 is evaluated as an average over @xmath84 histories . the steady - state is easily identified in the approximations , since we observe that , as @xmath44 increases , @xmath82 tends to a well - defined plateau value @xmath85 . conversely , in the simulations @xmath83 exhibits a fluctuating behavior for large @xmath44 , so we compute the steady - state value @xmath86 as an average over @xmath87 timesteps . we then compute the relative errors between theory and simulation ( in the steady state ) as @xmath88 . we display all these data in figure [ fig : ss ] , for each approximation , as a function of the corresponding data for the pqr approximation . note that , as a consequence , pqr data fall on a straight line . more importantly , this way of showing data allows us to argue at a glance that the m approximation is _ systematically _ ( i.e. , for all sites ) better than the pqr approximation , which is in turn systematically better than pq . let us also observe that , in the poissonian graph case ( left panel ) , the pq and p data are so separated that also pq clearly turns out to be systematically better than p. for the uniform graph case ( right panel ) , the boundary between the two data `` clusters '' appears a little more fuzzy , so that one might suspect that , at least for some sites , the p approximation could give a better approximation than pq . nevertheless , plotting as a function of the pq data ( right panel inset ) shows that this is not the case , i.e. , that pq is still _ systematically _ better than p. from figure [ fig : ss ] , one can also assess that the accuracy improvements that one can obtain by `` upgrading '' the approximation level from p , to pq , to pqr are , in relative terms , of the same order of magnitude , whereas a further upgrade to the m approximation achieves a minor improvement . in order to assess also the transient behavior , we display in figure [ fig : wc ] the whole time evolution of the infection probability @xmath82 for a specific site @xmath20 . in particular , we choose the site @xmath20 for which the pqr approximation gives the largest relative error ( worst case ) in the steady state . we can observe that the systematic accuracy improvement , already noticed for the steady state , also takes place in the short - time evolution . such a behavior is indeed common for all sites . we also note that the pqr and m results are almost indistinguishable at this scale , whereas they provide a noticeable improvement over the pq ( pair ) approximation . moreover , we can observe that , in the case of quenched initial conditions ( left panel ) , the infection probability remains sharply zero up to a given timestep ( the @xmath20 site being one of those initially susceptible ) . in such a case , even the simplest p approximation ( mean field ) correctly reproduces the very short time behavior ( including the precise timestep at which the @xmath20 site first exhibits a nonzero infection probability ) , but otherwise it rapidly deviates from the trend predicted by the simulations . one more test case we consider is that of a uniform graph ( still with @xmath89 ) , with homogeneous recovery probability ( @xmath90 ) and , as well , homogeneous infection probability ( @xmath91 ) . initial conditions are chosen to be of the annealed type ( as defined above ) . in such a system we expect a fully homogeneous infection probability ( i.e. , @xmath92 for all @xmath20 ) , so that in the simulation we can evaluate the latter by an extra average over the sites , thus obtaining much less noisy results . in order to enhance the `` noise - reduction '' effect , we also choose a larger graph size , namely @xmath93 . we display the results in figure [ fig : uu ] for the specific cases @xmath94 and @xmath95 . in the former case ( left panel ) , we can observe that almost all the approximations ( except possibly p ) give a rather accurate prediction of the steady - state infection probability . this may be partially ascribed to the fact that in this setting the steady state is expected to be an equilibrium state , at odds with the previously considered cases . it is remarkable that , even though the m , pqr , and pq approximations are all very accurate in the steady state , the pq approximation shows quite a relevant discrepancy during the transient . furthermore , upon decreasing @xmath42 ( right panel ) one can observe that the various approximations exhibit increasingly different results also in the steady state . this is related to the fact that the model approaches its critical point , at which the active ( endemic ) steady state disappears , and the long - time infection probability becomes rigorously zero . in analogy with equilibrium phase transitions , different approximations give different predictions for the critical @xmath42 value , as already pointed out in the continuous - time case @xcite . let us finally compare the different approximations in terms of computational complexity . in figure [ fig : tempi ] ( left panel ) we display the cpu time ( intel core 2 duo ) needed to perform @xmath96 evolution timesteps of a uniform graph with @xmath97 vertices , as a function of the ( fixed ) degree @xmath98 . of course , greater accuracy corresponds to larger computational effort , though not with a simple proportionality . in particular , we note that the m approximation entails a very large increase of the complexity , in spite of the fact that its accuracy is just slightly better than that of pqr ( at least for the specific epidemic model considered here ) . the scaling behavior with respect to @xmath99 is obviously linear , so that we do not need to report detailed results , whereas it turns out to be exponential with respect to @xmath98 . in principle , the complexity scales with the number of configurations of the largest cluster retained in the cvm expansion , so that the p , pq and pqr approximations are all exponential in @xmath98 , whereas the m approximation is exponential in @xmath100 . from the theoretical point of view of computational complexity , this is an irrelevant difference , but from the practical point of view this means that the computational cost of the m approximation may rapidly become similar to that of a simulation , upon increasing the degree , as shown in figure [ fig : tempi ] . on the other hand , the p , pq and pqr approximations present an extra computational advantage , which can be stated as follows . if the transition probabilities can be written as a product of terms depending each on a single neighbor ( or , as well , as a sum of a few terms of this kind ) , simple dynamic programming precautions allow one to reduce the complexity scaling to be linear in the degree . details about this issue are beyond the scope of the current paper , but we only mention the fact that the transition probabilities of the sis model satisfy the above prescription . execution times of the optimized programs are reported in the right panel of figure [ fig : tempi ] . in this paper we have performed a detailed investigation of different types of mean - field - like approximations for stochastic dynamics on graphs , within the unifying framework of a cluster - variational approach . in analogy with its equilibrium counterpart , this approach naturally opens the way to improving the approximation , by a suitable choice of the maximal clusters , which basically define the range of correlations that a given approximation scheme is able to take into account . on the one hand , we have pointed out that the simplest choices of maximal clusters ( namely , the p and pq approximations , according to our nomenclature ) coincide respectively with the ordinary mean - field and the pairwise approximations , previously derived in different contexts without resorting to a variational scheme . on the other hand , we have proposed a slightly more complex cluster choice ( pqr in our nomenclature ) , which provides a generalization of the so - called diamond approximation , to dynamics that are _ not _ purely neighbor - conditioned ( i.e. , such that the transition probability of a given site variable may also depend on the current configuration of the site itself ) . furthermore , we have considered an even larger cluster choice ( m in our nomenclature ) , which turns out to be an instance of kikuchi s path - probability method . we have shown explicitly that , in the case of a purely neighbor - conditioned dynamics , the p ( mean - field ) and pq ( pairwise ) approximations become equivalent , and similarly the m approximation ( path - probability method ) degenerates into the pqr ( generalized diamond ) approximation . we have tested the different approximations on a ( discrete - time ) epidemic model of the sis type ( i.e. , with recurrent dynamics ) on random graphs , possibly with heterogeneous infection parameters . we have compared the results with monte carlo simulations , and demonstrated that increasing the maximal cluster size leads to a systematic improvement of the level of accuracy . such an improvement can obviously be obtained at the cost of a higher computational complexity , but not with a simple proportionality . in particular , the pqr approximation yields nearly the same accuracy as the m approximation , with a considerably lower complexity . we have also confirmed that the pqr approximation yields very good accuracy even in the transient regime , as previously reported for the diamond approximation in the case of kinetic ising models @xcite . let us observe that our choice of a discrete - time dynamics is a bit artificial , as continuous time is the most typical setting of epidemic models ( even though with some exceptions @xcite ) . in our comparative analysis , this choice is meant to point out the difference between the pqr and pq approximations , which we indeed prove to become equivalent in the continuous - time limit . let us finally mention an important alternative class of approximate methods for the same kind of problems , which go under the name of _ dynamic message - passing _ ( dmp ) or the like . basically , these methods rely on the so - called _ dynamic cavity _ equations @xcite , which are exact on tree graphs ( and therefore likely accurate on treelike graphs ) , but require taking into account the whole time evolution of each given site variable . different approximations over the time dimension can then be worked out to make the problem computationally tractable . such alternative approximations include for instance the early `` dynamic cavity '' approximation by aurell and mahmoudi @xcite , the _ markovian closure _ scheme by del ferraro and aurell @xcite , and the very refined ( and computationally demanding ) matrix - product algorithm by barthel and coworkers @xcite . an analogous ( but apparently different ) dmp approach , directly formulated for the continuous - time sis model , has also been put forward recently @xcite . interestingly , the latter method has been reported to perform slightly better than the pairwise ( pq ) approximation , but mainly in the stationary state , whereas it seems to be generally less accurate in transient states . on the other hand , the aforementioned markovian - closure scheme @xcite ( which can in principle be applied to any kind of dynamics falling within our reference model ) has been thoroughly tested @xcite on various ( glauber ) kinetic ising models , and reported to perform ( slightly but quite systematically ) worse than the diamond ( pqr ) approximation . note that , as in this case the dynamics is purely neighbor - conditioned , the pair approximation is no longer a good term of comparison , because it degenerates into the ordinary mean field , as we show in appendix [ subsec : pqandp ] . we have not included in our paper a detailed comparative analysis with respect to dynamic - message - passing approaches , because , as previously mentioned , the main focus of our work is on the unifying framework of the cluster variational method . nevertheless , such an investigation might be in fact worth being performed , so that we shall consider it as a subject for future work . in this appendix we report in detail the method we have used to determine the minimum of the cluster - variational functional for the case of the pqr approximation . as we shall see , this method naturally leads to an `` iterative '' procedure ( the one we have implemented numerically ) , which , starting from the initial conditions , builds up the cluster distributions at subsequent timesteps , up to the eventual steady state . let us first discuss the issue of compatibility constraints , which will be of use in the following . the probability distributions for maximal clusters and subclusters ( which for the pqr approximations have been denoted as @xmath101 ) are of course not independent variational parameters , because they have to satisfy ( besides normalization ) a number of compatibility constraints . by _ compatibility _ we mean that the marginalization of probability distributions of overlapping maximal clusters must give , for the same subcluster , the same distribution . writing down these conditions for all subclusters of all maximal clusters , we obtain a sufficient ( indeed redundant ) set of constraint equations . let us group together these equations depending on the type of maximal cluster involved . we understand that all the equations must hold for @xmath102 . p - type maximal clusters : [ eq : vinc - p ] @xmath103 q - type maximal clusters : [ eq : vinc - q ] @xmath104 r - type maximal clusters : [ eq : vinc - r ] @xmath105 this set of conditions also entails direct compatibility equations between subcluster distributions ( for overlapping subclusters ) . let us write explicitly some of these equations below . from equations and , or equivalently and , we have @xmath106 from equations and , or equivalently and , we have @xmath107 from equations and , or equivalently and , we have @xmath108 from equations and , or equivalently and , we have @xmath109 from equations and , or equivalently and , we have @xmath110 let us finally remark that the largely redundant form , in which we have written the compatibility constraints , is not an end in itself . indeed , it will enable us to get a very compact proof of the fact that our method actually produces compatible solutions . this will be shown , along with the computation algorithm , in section [ subsec : algo ] . as mentioned in section [ subsec : specific_approximations ] , we derive the pqr free - energy functional ( denoted as @xmath111 ) from equation , replacing the entropic term @xmath63 with @xmath112 ( equation ) . taking into account that we are interested in solutions respecting the constraint equations , specifically , , , , , it is possible to rewrite the functional in such a way that the subcluster distributions appear only in the logarithmic terms , that is in the following form @xmath113 = \nonumber \\ & = \sum_{t=0}^{\tau-1 } \sum_{i } \sum_{y_{i},x_{i,\partial i } } { p_{i}}^{(t)}(y_{i},x_{i,\partial i } ) \ln \frac { { p_{i}}^{(t)}(y_{i},x_{i,\partial i } ) } { { w_{i}}^{(t)}(y_{i}|x_{i,\partial i } ) { s_{i}}^{(t)}(x_{i,\partial i } ) } \ , + \nonumber \\ & + \sum_{t=0}^{\tau-1 } \sum_{ij } \sum_{y_{i , j},x_{i , j } } { q_{ij}}^{(t)}(y_{i , j},x_{i , j } ) \ln \frac { { q_{ij}}^{(t)}(y_{i , j},x_{i , j } ) { z_{ij}}^{(t)}(x_{i , j } ) } { { t_{i , ij}}^{(t)}(y_{i},x_{i , j } ) { t_{j , ij}}^{(t)}(y_{j},x_{i , j } ) } \ , + \nonumber \\ & + \sum_{t=0}^{\tau-1 } \sum_{i } \sum_{y_{i,\partial i},x_{i } } { r_{i}}^{(t)}(y_{i,\partial i},x_{i } ) \ln \frac { { r_{i}}^{(t)}(y_{i,\partial i},x_{i } ) { { v_{i}}^{(t)}(y_{i},x_{i})}^{d_{i}-1 } } { \prod_{j \in \partial i } { u_{ij , i}}^{(t)}(y_{i , j},x_{i } ) } \ , , \label{eq : funz - pqr}\end{aligned}\ ] ] where the inner sums run over all possible configurations of p- , q- , and r - clusters , respectively . recall that the terms @xmath114 are transition probabilities , so that by construction we assume they satisfy the following normalization conditions @xmath115 note also that , given the mbius numbers of the pqr approximation ( see table [ tab : superposition_coefficients ] and equation ) , the probability distributions @xmath116 ( i.e. the s - cluster distributions at time @xmath117 ) and @xmath118 ( i.e. the z - cluster distributions at time @xmath117 ) actually do not belong to the entropic term . we have nonetheless included them in the free - energy expression , to take into account the part of energetic term that represents initial conditions , assuming therefore [ eq : iniz - sz ] @xmath119 now , a crucial observation is that the variational functional in the form is made up of a sum of kullback - leibler divergence terms , and therefore that the absolute minimum conditions ( in which every term equals @xmath70 ) are , for @xmath120 , [ eq : algo - pqr ] @xmath121 actually , the minimum condition of a kullback - leibler divergence holds provided its two arguments ( that is in our case the two sides of each equation ) are `` good '' ( i.e. normalized ) probability distributions . assuming of course the normalization of all cluster distributions , it is easy to see that the right - hand sides of equations , and are actually good distributions , if , respectively , the normalization condition and the compatibility conditions and are verified . let us finally observe that equations are somehow `` weak '' , meaning that they do not define a point in the space of variational parameters , but rather a manifold . in order to determine completely the variational parameters , that is the cluster distributions @xmath122 , we have to intersect the aforementioned manifold with that defined by the constraints , that is to solve simultaneously the minimum equations with the compatibility equations , , ( whence also , , , , ) and the normalizations . let us stress the fact that it is not necessary to resort to the lagrange multiplier method to solve the contrained minimum problem , just because it turns out that the intersection between the two manifolds is nonempty . in general , the latter is not a necessary condition for the existence of a constrained minimum . in our specific problem , such a condition arises from the fact that the exponentials ( boltzmann factors ) of the coupling terms are actually ( transition ) probabilities , and therefore that they enjoy the normalization property . in this section we present an iterative procedure , which solves the problem described above . in particular we see that , starting with a set of @xmath123 and @xmath124 distributions satisfying at a given timestep @xmath44 ( initially @xmath117 , with the distributions defined by ) , one can determine all the other distributions @xmath125 and @xmath126 at timestep @xmath44 , as well as @xmath123 and @xmath124 at timestep @xmath13 , satisfying all the required equations at time @xmath44 and at time @xmath127 . moreover , all the computed distributions turn out to be correctly normalized . the steps of the procedure are described in detail below , and summarized as a flow chart in figure [ fig : flow - chart ] . ( 176,50)(-3,-25 ) ( 0,-3)(8,6)@xmath128 ( 8,0)(1,0)15 ( 8,0)(15,5 ) ( 0,-11.5)(8,5)input ( 23,-9)(9,18)@xmath129 ( 32,6)(1,0)21 ( 32,6)(12,5 ) ( 53,3)(8,6)@xmath130 ( 53,-9)(8,6)@xmath131 ( 53,-21)(8,6)@xmath132 ( 32,-6)(1,0)12 ( 44,-6)(1,-1)9 ( 32,-6)(12,5 ) ( 44,-18)(1,1)9 ( 8,-18)(1,0)36 ( 0,-21)(8,6)@xmath131 ( 61,6)(1,-1)6 ( 61,-6)(1,1)6 ( 67,0)(1,0)12 ( 67,0)(12,5 ) ( 79,-9)(9,18)@xmath133 ( 88,6)(1,0)21 ( 88,6)(12,5 ) ( 109,3)(8,6)@xmath134 ( 109,-9)(8,6)@xmath132 ( 109,-21)(11,6)@xmath135 ( 88,-6)(1,0)12 ( 100,-6)(1,-1)9 ( 88,-6)(12,5 ) ( 100,-18)(1,1)9 ( 61,-18)(1,0)39 ( 117,6)(1,-1)6 ( 117,-6)(1,1)6 ( 123,0)(1,0)12 ( 123,0)(12,5 ) ( 135,-9)(9,18)@xmath136 ( 144,0)(1,0)15 ( 144,0)(15,5 ) ( 159,-3)(11,6)@xmath137 ( 159,-21)(11,6)@xmath135 ( 120,-18)(1,0)39 ( 159,-11.5)(11,5)output ( 8,13)(0,9 ) ( 8,17)(53,5)step 2 ( 61,13)(0,9 ) ( 61,17)(56,5)step 3 ( 117,13)(0,9 ) ( 117,17)(53,5)step 4 ( 170,13)(0,9 ) 1 . let us begin , at a given timestep @xmath44 ( initially @xmath117 ) with a set of ( normalized ) s - cluster distributions @xmath138 ( @xmath139 ) and z - cluster distributions @xmath140 ( @xmath141 ) , that we assume to be compatible , i.e. satisfying equations ( which is obviously verified by the initial distributions ) . 2 . we determine the p - cluster distributions at time @xmath44 by means of equations . thanks to , we see that such distributions satisfy equations . moreover , since and hold simultaneously , also turns out to be satisfied . we then determine the following marginals . * t - cluster distributions @xmath142 ( @xmath143 ) by equations . since and hold simultaneously , such distributions also satisfy . * v - cluster distributions @xmath144 ( @xmath139 ) by equations . since and hold simultaneously , also turns out to be satisfied . 3 . we determine the q - cluster distributions at time @xmath44 by means of equations . thanks to , we see that such distributions satisfy equations and . moreover , since and hold simultaneously , also turns out to be satisfied . we then determine the following marginals . * u - cluster distributions @xmath145 ( @xmath143 ) by equations . since and hold simultaneously , such distributions also satisfy . * z - cluster distributions ( at the next timestep ) @xmath146 ( @xmath147 ) by equations . since and hold simultaneously , also turns out to be satisfied . 4 . we determine the r - cluster distributions at time @xmath44 by means of equations . thanks to , we see that such distributions satisfy equations and . moreover , since and hold simultaneously , also turns out to be satisfied . we then determine the following marginals . * s - cluster distributions ( at the next timestep ) @xmath148 ( @xmath139 ) by equations . since and hold simultaneously , such distributions also satisfy at time @xmath13 . 5 . we repeat the procedure from step 2 for the distributions at time @xmath13 . let us stress the fact that , in the above procedure , all the compatibility conditions stated in section [ subsec : comp ] have been either used explicitly to compute the marginals or proved to be satisfied by the computed distributions . in this section we show that , as mentioned in section [ subsec : related_works ] , in the case of a pure neighbor - conditioned dynamics the pqr approximation coincides with the diamond approximation , previously proposed by one of us in reference @xcite . first of all it is useful to define some auxiliary clusters , namely , the b - clusters , made up of two neighbor sites at subsequent timesteps @xmath44 and @xmath13 , for @xmath120 , and the c - clusters , made up of all the neighbors of a given site at a given timestep @xmath44 , for @xmath149 . the related distributions are denoted as @xmath150 these distributions can be written as marginals of maximal cluster distributions , so that they inherit all the required compatibility conditions . we shall use in particular the following relationship @xmath151 which , together with @xmath152 also imply @xmath153 let us now hypothesize that the s- and z - cluster distributions at a given timestep @xmath44 factor in the following fashion [ eq : fatt - ipo ] @xmath154 obviously assuming also that c- and a - distributions are compatible , that is they satisfy @xmath155 retracing the steps of the algorithm described in the previous section , let us show that the hypothesized factorizations are stable along the time evolution ( i.e. , our hypotheses imply the same factorizations at timestep @xmath127 ) , and the evolution equations are identical to those of reference @xcite . 1 . we easily see that the hypotheses and satisfy the compatibility between s- and z - distributions at time @xmath44 . 2 . equation with the factorization reads @xmath156 therefore , equation together with gives @xmath157 where @xmath158 whereas equation together with gives @xmath159 where @xmath160 3 . equation with the factorizations and reads @xmath161 therefore , equation together with gives @xmath162 whereas equation together with gives @xmath163 we thus argue that the z - distributions at time @xmath13 take a factorized form analogous to . equation with the factorizations and reads @xmath164 therefore , equation proves that the s - distributions at time @xmath13 take a factorized form analogous to , namely @xmath165 where @xmath166 note that the `` diamond clusters '' of reference @xcite never appear explicitly in the above discussion . nevertheless , we can see that , plugging equation into and , one obtains evolution equations for b- and a - cluster distributions , which exactly coincide with those one could obtain , in reference @xcite , replacing the diamond cluster distribution ( 26 ) into the marginalization equations ( 25 ) and ( 23 ) . in this appendix we discuss the other approximations we have introduced in the text , namely the pq and p approximations , and the m approximation . we do not go into the same level of detail , as done for the pqr approximation , because the derivations are in fact very much analogous . we only give the final results in terms of the respective computation procedures . the pq approximation can be regarded as a simplification of the pqr approximation , in which the s - cluster distributions are factorized in terms of z - cluster ( nearest - neighbor pair ) and a - cluster ( single site ) distributions , according to @xmath167 as a consequence , one can replace step 4 of the pqr procedure with the previous equation ( evaluated at timestep @xmath13 ) , where the z - cluster distributions have been determined at step 3 , and the a - cluster distributions , compatible with the former , i.e. satisfying @xmath168 can be written by the obvious marginalization . in conclusion , as previously mentioned , the pq approximation turns out to be equivalent to the pairwise approximation , considered by different authors @xcite . such an equivalence might indeed be not fully evident , because in the cited papers the approximations are immediately specialized to a particular model ( namely , the sis model ) and to the continuous - time limit , whereas our treatment is more general , at least under the assumptions of a discrete - time dynamics and a transition probability factorized according to . the p approximation can be regarded as a further simplification , in which also the pair distributions are factorized , as in equation , so that the s - cluster distributions read @xmath169 as a consequence , one can skip also step 3 of the pqr procedure , because there is no need to compute pair distributions at the next timestep , to be able to close the equations . indeed , replacing equation into and then into , one obtains a unique time - evolution equation for the single - site distributions , namely @xmath170 which coincides with the ( quenched ) mean - field approximation @xcite . this argument also proves that , for a pure neighbor - conditioned dynamics , for which holds , the pq approximation degenerates into the p approximation , and therefore that both coincide with the _ star _ approximation of reference @xcite , as previously stated in section [ subsec : related_works ] . the iterative procedure for the m approximation turns out to coincide with that of the pqr approximation until step 3 . step 4 is modified in that the r - cluster distributions are no longer computed by equation , but rather as marginals of the m - cluster distributions , namely @xmath171 the latter being previously computed as @xmath172 in the simpler case of a pure neighbor - conditioned dynamics , we have shown for the pqr approximation that the factorizations , , and hold . these equations remain valid also for the m approximation , because they stay within step 3 of the procedure , so that we can plug them into and then into . taking into account also equation ( still valid ) , we obtain exactly equation . this argument proves that , for a pure neighbor - conditioned dynamics , the m approximation degenerates into the pqr approximation , and therefore that both coincide with the _ diamond _ approximation of reference @xcite , as previously stated in section [ subsec : related_works ] . the approximation schemes we have presented in this paper can be used as well to analyze continuous - time processes ( in a discretized form ) , or equivalently discrete - time processes with sequential dynamics . this can be done by defining the following transition probabilities @xmath173 where it is understood that @xmath174 , whereas @xmath175 denotes a kronecker delta and @xmath176 are suitable conditional probabilities , normalized in the proper way @xmath177 equation can be interpreted as follows . with a large probability ( @xmath178 ) each vertex variable @xmath179 stays in its current value , and with a small probability ( @xmath3 ) it is allowed to undergo a transition , governed by the probabilities @xmath180 . for the case of sequential dynamics , @xmath181 is the probability of choosing one specific vertex out of @xmath99 . alternatively , for the case of continuous - time processes , one can think of @xmath3 as a small time interval , and of @xmath176 as the _ rate _ of the transition @xmath182 ( for @xmath183 ) , which is assumed to depend on the neighborhood configuration @xmath184 . as mentioned in the text , we can show that , in the limit @xmath185 ( continuous - time limit ) , the pqr and pq approximations become equivalent . making use of equations and ( step 2 of the computation algorithm ) with the transition probabilities , and taking into account the compatibility equations , the t - cluster distributions can be written as where we have defined is the t - cluster distribution one would get if the transition probabilities were barely @xmath187 . in the following we use the same notation as well for other cluster types . ] @xmath188 making use of equations and ( step 3 of the computation algorithm ) , still with the transition probabilities , the z - cluster distributions at the next timestep ( now denoted as @xmath189 rather than @xmath13 ) can be written as @xmath190 + \mathcal{o}(\tau^2 ) \ , , \label{eq : ct - zfwd}\ ] ] where we have defined @xmath191 moreover , by and we get @xmath192 where @xmath193 in principle , taking the limit @xmath174 of equations and , one can obtain a system of ordinary differential equations for the pair and site distributions . indeed , equations and can be viewed as discretized forms thereof , which are directly suitable for numerical implementation . let us observe that equations and hold for both the pq and pqr approximations , since steps 2 and 3 of the computational procedure are identical , as mentioned in section [ subsec : pqandp ] . for the pq approximation , these equations are immediately closed , because the @xmath123 distribution ( appearing in the expressions of @xmath194 and @xmath195 ) directly depends on @xmath124 and @xmath196 , according to equation . as we shall see below , the reason why the pq and pqr approximations turn out to be equivalent is that the difference between the @xmath123 distributions for the two cases is @xmath197 for @xmath174 , so that , being multiplied by @xmath3 in both equations and , its contribution becomes irrelevant . let us first consider the pq approximation . equation at time @xmath189 reads @xmath198 plugging equations and into the latter , we easily get @xmath199 as far as the pqr approximation is concerned , the @xmath123 distribution is computed by equation as a marginal of the @xmath200 distribution , which in turn depends on the @xmath201 and @xmath202 distributions via equation ( step 4 of the numerical procedure ) . using equations and ( step 3 of the procedure ) with the @xmath203 distribution expressed by , we can write @xmath204 hence , from equations and , @xmath205 plugging equations and into , we get @xmath206 the latter expression , marginalized via , yields again , which is equivalent to the pq result .

a partial - wave analysis of single - pion photoproduction data has been completed . this study extends from threshold to 2 gev in the laboratory photon energy , focusing mainly on the influence of new measurements and model - dependence in the choice of parameterization employed above the two - pion threshold . results are used to evaluate sum rules and estimate resonance photo - decay amplitudes . these are compared to values obtained in the maid analysis .

nucl-th0205067

meson - nucleon scattering , meson photoproduction , and meson electroproduction have been extensively studied within a comprehensive program exploring the spectroscopy of @xmath0 and @xmath1 resonances . an objective of this program is the determination of all relevant characteristics of these resonances , _ i.e. _ pole positions , widths , principal decay channels , and branching ratios . in order to compare directly with qcd - inspired models , there has also been a considerable effort to find hidden " or missing " resonances , predicted by quark models ( see , for example , the states predicted by capstick and roberts @xcite ) but not yet confirmed . here we will give detailed results from an ongoing analysis of pion photoproduction data . this work complements our studies of pion - nucleon elastic scattering @xcite , both reactions having the same resonance content , and provides the real - photon limit for our pion - electroproduction fits @xcite . resonance characteristics are determined through a two - step procedure . the full database is initially fitted to determine the underlying multipole contributions . those multipoles having resonant contributions are then fitted to a form containing both resonance and background terms . the availability of multipole amplitudes greatly simplifies certain numerical aspects of coupled - channel analysis . the number of fitted multipole amplitudes , associated with a dataset , may be smaller ( than the count of individual data ) by one or two orders of magnitude , and can account for issues associated with statistical / systematic errors , data rejection , and incomplete sets of observables . in general , our partial - wave analyses ( pwa ) have been as model - independent as possible , so as to avoid bias when used in resonance extraction or coupled - channel analysis . however , in the absense of complete experimental information , all multipole analyses above the two - pion production threshold are model - dependent to some degree . this issue will be discussed in section [ sec : ampl ] . the amplitudes from these analyses can be utilized in evaluating contributions to the gerasimov - drell - hearn ( gdh ) sum rule @xcite and sum rules related to the nucleon polarizability . we display our results for these and compare with recent mainz experimental data and predictions . in the next section , we summarize changes in the database since our last published analysis @xcite . results of our multipole analyses , as well as the photo - decay amplitudes for resonances within our energy region , are given in section [ sec : ampl ] . in section [ sec : conc ] , we summarize our findings and consider what improvements can be expected in the future . our three previous pion photoproduction analyses @xcite extended to 1.0 , 1.8 , and 2.0 gev , respectively . the present database @xcite is considerably larger , due mainly to the addition of new data at low to intermediate ( below 800 mev ) energies . in 1994 , bremsstrahlung data comprised over 85% of the existing measurements . these data often suffered from significant uncertainties in normalization that were not completely understood or not quoted . the available tagged - photon data were generally measured in low - statistics experiments , and hence were presented with large energy and angular binning . much of the remaining dataset was comprised of excitation cross sections with no extensive angular range . inconsistencies were obvious almost everywhere that comparisons could be made . the full database has increased by 30% since the publication of ref . @xcite , and is about 40% larger than the set available for the analysis of ref . the majority of these new data have come from tagged - photon facilities at mami ( mainz ) , graal ( grenoble ) , and legs ( brookhaven . ) [ the total database has doubled over the last two decades ( see table [ tbl1 ] . ) ] the distribution of recent ( post-1995 ) @xmath2 and @xmath3 data is given in fig . [ g1 ] ( there are no new @xmath4 and @xmath5 data . ) as the full database contains conflicting results , some of these have been excluded from our fits . we have , however , retained all available data sets ( and labeled these excluded data as flagged " ) so that comparisons can be made through our on - line facility @xcite . data taken before 1960 were not analyzed , nor were those single - angle and single - energy points measured prior to 1970 . some individual data points were also removed from the analysis in order to resolve conflicts or upon authors requests ( since our previous analysis @xcite , we have flagged 950 @xmath2 , 1060 @xmath3 , and 150 @xmath4 bremsstrahlung data taken prior to 1983 . ) some of the data , listed as new , were available in unpublished form at the time of our previous analysis @xcite . a complete description of the database and those data not included in our fits is available from the authors . since 1995 , 87% ( 21% ) of all new @xmath2 ( @xmath3 ) data have been produced at mainz using the mami facility @xcite . these measurements of total and differential cross sections , @xmath6 beam asymmetry , and the gdh - related quantity ( @xmath7 ) have increased the database by a factor of about two over the energy range from the threshold to 800 mev . the angular range of cross sections extends from 10@xmath8 to 170@xmath8 , and thus increases the sensitivity to contributions from higher partial waves . results from other laboratories include low - energy unpolarized @xmath2 total ( 47 data ) and differential ( 198 data ) cross sections measured at sal @xcite , and @xmath3 threshold data ( 45 points ) , covering a range of 2 mev in @xmath9 , produced by the triumf sal collaboration @xcite . at energies spanning the @xmath10 resonance , @xmath6 ( 169 data ) and differential cross section ( 157 data ) for both @xmath2 and @xmath3 channels have been measured by the legs group at bnl @xcite . in the medium - energy range , @xmath3 @xmath6 beam - asymmetry data between 600 and 1500 mev ( 329 data ) have been measured at graal @xcite and @xmath2 @xmath6 data between 500 and 1100 mev ( 158 data ) have been measured at the 4.5 gev yerevan synchrotron @xcite . target asymmetry @xmath11 measurements between 220 and 800 mev for both @xmath2 ( 52 data ) and @xmath3 ( 210 data ) have come from elsa at bonn @xcite . excitation @xmath3 differential cross sections for backward scattering between 290 and 2110 mev , also from the bonn facility , have been replaced by finalized data @xcite , with final versions of other @xmath12 and @xmath13 differential cross sections expected @xcite . further experimental efforts will provide data in the intermediate energy region . above 400 mev , a large amount of new data is expected from clas at hall b of jefferson lab @xcite . differential cross sections associated with the mainz gdh experiment @xcite ( related to the double - polarization quantity @xmath14 @xcite ) should also have an impact on the analysis when combined with 4@xmath8 to 177@xmath8 cross section ( 200 to 790 mev ) and 10@xmath8 to 160@xmath8 @xmath6 beam asymmetry ( 250 to 440 mev ) measurements at mami @xcite . beam asymmetry data for @xmath15 photoproduction below 1100 mev will also be available from graal @xcite . of particular interest are the polarized @xmath15 photoproduction experiments ( including the polarization transfers @xmath16 and @xmath17 from circularly polarized photons to recoil protons ) above 800 mev carried out in hall a of jlab @xcite . from brookhaven , we expect final legs @xmath6 beam asymmetries around the @xmath10 resonance @xcite and new radiative capture cross sections which have been taken at bnl ags using the crystal ball spectrometer ( e913/914 ) at @xmath18 from 400 to 750 mev / c ( e@xmath19 = 430 780 mev ) @xcite . fits to the expanded database were first attempted within the formalism we have used and described previously @xcite . multipoles were parameterized using the form @xmath20 with @xmath21 being the associated elastic pion - nucleon @xmath22matrix , and the terms @xmath23 and @xmath24 being purely phenomenological polynomials with the correct threshold properties . as in our most recent analysis @xcite , some multipoles were allowed an additional overall phase @xmath25 , where the angle @xmath26 was proportional to ( @xmath27 ) . this form satisfied watson s theorem for elastic @xmath28 amplitudes @xcite while exploiting the undetermined phase for @xmath28 inelastic amplitudes . for @xmath21 , we utilized our most recent fit ( sm02 ) to elastic scattering data @xcite . new and precise @xmath6 measurements proved difficult to describe , using this choice of phenomenology . searching for a more successful form , we found an improved description was possible if the dependence on ( @xmath29 ) was additive rather than multiplicative . as a result , we re - fitted the full database , removing the overall phase and instead added a term of the form @xmath30 with @xmath31 and @xmath32 again being energy - dependent polynomials . the resulting energy - dependent solution ( sm02 ) had a @xmath33 of 35296 for 17571 ( @xmath2 , @xmath3 , @xmath4 , and @xmath5 ) data to 2 gev . the overall @xmath33/data was significantly lower than that found in our previously published result ( @xmath33/data = 2.4 ) @xcite . this change is partly a reflection of the the database changes discussed in section [ sec : expt ] . our present and previous energy - dependent solutions are compared in table [ tbl1 ] . as in previous analyses , we used the systematic uncertainty as an overall normalization factor for angular distributions . this renormalization freedom provided a significant improvement for our best fit results as shown in table [ tbl2 ] . in order to see if the inclusion of the full existing database had resulted in a bias towards older and possibly outdated measurements , we compared the fit quality versus measurement date . generally , we found no problem of bias , as illustrated in table [ tbl3 ] , where we have displayed our fit quality over the region covering the n(1535 ) resonance . except for the above mentioned @xmath6 measurements , over this energy region , our overall fit to the most recent measurements ( 1985@xmath34present ) is charactorized by a @xmath33/data of about 1.3 . for all data to 2 gev , the date restriction yields a @xmath33/data near 1.5 . unfortunately , the modern measurements do not completely overlap older data . much of the older data is required in an analysis extending over the full resonance region . the very low energy region is complicated by different thresholds for @xmath2 and @xmath3 final states . while we have obtained a reasonable fit to the available @xmath2 differential and total cross sections ( we fit rather well the new threshold triumf sal @xmath3 cross sections @xcite , and mainz @xmath2 @xmath6 at 160 mev @xcite ) , the multipole amplitudes have no cusp built into the @xmath3 threshold region . both energy - dependent and single - energy solutions ( ses ) were obtained from fits to the combined @xmath2 , @xmath3 , @xmath4 , and @xmath5 databases to 2 gev . in table [ tbl4 ] , we compare the energy - dependent and single - energy results over the energy bins used in these single - energy analyses . also listed are the number of parameters varied in each single - energy solution . a total of 148 parameters were varied in the energy - dependent analysis sm02 . the extended database allowed an increase in the number of ses versus our previous result @xcite over the same energy range to 2 gev . [ g2 ] is a plot of the energy dependent fits sm02 and sm95 over the full energy region . the ses are also shown with uncertainties coming from the error matrix . in the ses fits , initial values for the partial - wave amplitudes and their ( fixed ) energy derivatives were obtained from the energy - dependent solution . a comparison of global and single - energy solutions then serves as a check for structures that could have been `` smoothed over '' in the energy - dependent analysis . partial waves with @xmath35 are displayed , whereas the analysis fitted waves up to @xmath36 . significant deviations from sm95 are visible in multipoles connected to the @xmath28 s- and p - waves , as well as @xmath37 , @xmath38 , and @xmath39 ( for the neutron . ) as mentioned above , the parametrization used in our previous analysis sm95 @xcite did not allow a good to fit recent yerevan @xmath2 @xcite and graal @xmath3 @xcite @xmath6 data ( the critical range extends from 700 to 800 mev . ) for comparison purposes , in table [ tbl5 ] , we compare sm95 , sm02 , and sx99 . the test fit sx99 retains the form used in previous fits ( as in sm95 ) and is applied to the present full database . one can see that recent mainz @xmath3 ( @xmath40 ) @xcite data are also problematic for the sx99 fit , with the most difficult region again covering the 700 to 800 mev range . in fig . [ g3 ] , we display the energy dependence of a differential cross section measurement @xcite , at a fixed angle , displaying rapid variation over the region in question . these mainz data from 560 to 780 mev are reasonably well reproduced ( @xmath41 ) , though not included in our analysis . while the said and maid analyses are qualitatively similar from threshold to 1 gev , some significant differences exist and these have been mentioned in a recent multi - analysis study which fitted to a benchmark dataset @xcite . while some multipoles show significant differences , the photo - decay amplitudes from maid and sm95 are quite similar , with larger differences between the maid and sm02 solutions . the data fit quality shows greater variability and a few cases are given below . the measurement of @xmath6 for @xmath2 at threshold ( 160 mev ) has been discussed in @xcite and is particularly sensitive ( fig . the plotted curves differ mainly in the p - wave multipoles . our ses at 162 mev , which covers 158 @xmath34 165 mev ( see table [ tbl4 ] , ) fits the new mainz data @xcite rather well , while the energy - dependent maid2000 and sm02 solutions are less successful . at higher energies , close to the upper limit of maid , new @xmath3 @xmath6 measurements , shown in fig . [ g5 ] , are clearly problematic . this has been used to suggest a change in the n(1650 ) photo - decay amplitude @xcite . for the forward peaking in @xmath3 unpolarized differential cross sections , displayed in fig . [ g6 ] , the disagreement between said and maid results reaches as much as 30% . in the said fits , some of these problems are resolved once systematic uncertainties are taken in account . the double - polarization quantity @xmath42 , measured in the a2 collaboration gdh experiment @xcite and displayed in fig . [ g7 ] , is well described by both maid2001 and said solution sm02 . at higher energies , deviations become more apparent ( as was also shown in fig . [ g3 ] . ) clas at jlab @xcite , elsa at bonn @xcite , spring8 at hyogo @xcite , and legs at brookhaven @xcite have further programs underway to study this process . in fig . [ g8 ] , we compare our results with the maid analysis @xcite = 1250 mev ) @xcite . ] where more substantial differences are seen for the @xmath43 , @xmath44 , @xmath45 , and @xmath46 multipoles . we have fitted our multipoles using a simple breit - wigner plus background function , as described in ref . we have employed both single - energy and energy - dependent solutions over a variety of energy ranges in order to estimate uncertainties . a listing of our resonance couplings is given in table [ tbl6 ] . here , values for the resonance mass ( @xmath47 ) , full width ( @xmath48 ) , and the decay width to @xmath28 final states ( @xmath49 ) were taken from our elastic @xmath28 analysis @xcite and were not varied in the fits and error estimates . we find that the @xmath43 ( @xmath50 ) multipole is very sensitive to both the database and parametrization in the range associated with the @xmath51 . the range of variation , particularly large for the real part of @xmath43 , is displayed in fig . this sensitivity is not surprising , as the quantity ( @xmath52 ) has a very sharp structure at the @xmath53 threshold . this variability in the multipole amplitude is reflected in the @xmath51 resonance coupling , which we feel is presently too uncertain to quote . other couplings significantly altered using the revised parameterization scheme include the @xmath54 , which is still essentially undetermined , @xmath55 , @xmath56 , and @xmath57 . the @xmath58 has an easily identifiable magnetic but , essentially no electric multipole . multipoles associated with the @xmath59 show very little resonance signature . we were particularly surprised to see a large change in the @xmath60 photo - decay amplitudes . comparison with our sm95 fit shows a significant decrease in cross section at the resonance position . this shift has resulted due to the inclusion of recent mainz @xmath2 measurements @xcite , which are systematically lower ( particularly at backward angles ) than an older set of bonn measurements . more recent mainz fits for maid2001 @xcite give @xmath61 and @xmath62 ( in @xmath63 units ) for @xmath64 and @xmath65 , respectively . these results have also shifted lower and are consistent with our determination . the amplitudes obtained in our analyses can be used to evaluate the single - pion production component of several sum rules . the gdh integral @xcite relates the anomalous magnetic moment @xmath66 , the charge @xmath67 , and nucleon mass @xmath68 to the difference in the total photoabsorption cross sections for circularly polarized photons on longitudinally polarized nucleons @xmath69 where @xmath70 and @xmath71 are the photoabsorption cross sections for the helicity states @xmath72 and @xmath73 , respectively , with @xmath74 being the photon energy . for the proton ( neutron ) target , eq . ( [ 3 ] ) predicts @xmath34205 ( @xmath34233 ) @xmath75 . the running gdh integrals for the proton and neutron are shown in fig . [ g10 ] , where a comparison with maid is also given . the baldin sum rule @xcite relates the sum of electric and magnetic polarizabilities of the nucleon to the total photo - absorption cross section @xmath76 @xmath77 for the proton ( neutron ) target , the recent dispersion calculations by levchuk and lvov give ( @xmath78 [ ( @xmath79 @xcite . for the proton , an independent mainz determination gives @xmath80 @xcite and the legs group quotes @xmath81 @xcite . the isospin averaged nucleon polarizabilities determined by max - lab measurements of compton scattering from the deuteron is ( @xmath82 @xcite . running baldin integrals , with comparisons to maid , are given in fig . [ g11 ] . the forward spin polarizability @xmath83 @xcite is @xmath84 a recent dressed k - matrix model approach for the proton gives @xmath85 @xcite . the legs analysis gives , for a proton target , @xmath86 @xcite . the running integral is shown in fig . [ g12 ] . for charge states @xmath3 and @xmath4 , all three quantities are sensitive to the threshold energy range ( @xmath87 ) , as shown in table [ tbl7 ] . from figs . [ g10]@xmath34[g12 ] , one can see that each integral of the single - pion contribution , based upon the sm02 solution , has essentially converged by 2 gev . experimental data for the gdh and @xmath83 quantities have been obtained from measurements at mami , covering ranges from 200 to 450 mev @xcite and to 800 mev @xcite . in tables [ tbl8 ] and [ tbl9 ] , we show sm02 and maid results for the abovementioned quantities over energy ranges corresponding to measurements . clearly , calculations above 450 mev have to take into account contributions beyond single - pion photoproduction . the single - pion photoproduction database , for proton targets , has increased significantly since the publication of our fit sm95 . the inclusion of these precise new measurements has resulted in a fit with a lower overall @xmath33/data . however , some polarization quantities have been difficult to fit , and these difficulties have prompted an examination of the phenomenological forms we use . by changing the way we extrapolate beyond the two - pion threshold , an improved fit was obtained . this new multipole solution was found to differ significantly from sm95 in a number of partial waves . the largest changes were associated with multipoles connected to @xmath28 resonances with @xmath88 . also quite different was the @xmath89 multipole , for which the associated @xmath28 inelasticity has a sharp increase at the @xmath53 threshold . as might be expected , states with large @xmath28 branching fractions remained stable . this stability held for the @xmath90 as well , though investigations based upon @xmath53 photoproduction , and quantities related to the gdh integral , have suggested a shift in its ratio of photo - decay amplitudes . given the sensitivity of weaker resonances to the choice of phenomenology , we are now attempting to replace the dependence on ( @xmath52 ) with a form more directly connected to the opening of specific channels , such as @xmath53 and @xmath91 . the evaluation of sum rules ( gdh , baldin , and forward spin polarizability ) for a single - pion contribution exhibits convergence by 2 gev . agreement with mainz is now good below 450 mev , with larger deviations at higher energies . in both @xmath28 elastic scattering @xcite and pion photoproduction @xcite , the measurements of precise new single- and double - polarization data have highlighted problems existing in the standard " fits . further polarization measurements will be required to test assumptions implicit in the said and maid programs . the authors express their gratitude to j. ahrens , g. anton , h .- j . arends , r. beck , j. c. bergstrom , j .- bocquet , d. branford , h. m. fischer , k. g. fissum , m. fuchs , r. gilman , h. h. hakopian , d. a. hutcheon , m. khandaker , v. kouznetsov , b. krusche , b. a. mecking , a. omelaenko , p. pedroni , i. preobrajenski , d. rebreyend , g. v. orielly , a. m. sandorfi , c. schaerf , and a. shafi for providing experimental data prior to publication or for clarification of information already published . we also thank l. tiator and s. kamalov for the maid contribution , a. lvov for a discussion of the sum rules , and a. e. kudryavtsev for comments on the @xmath92 problem . this work was supported in part by the u. s. department of energy under grant de fg0299er41110 . the authors acknowledge partial support from jefferson lab , by the southeastern universities research association under doe contract de ac0584er40150 . s. capstick and w. roberts , prog . ( suppl . 2 ) * 45 * , 241 ( 2000 ) . r. a. arndt , i. i. strakovsky , r. l. workman , and m. m pavan , phys . c * 52 * , 2120 ( 1995 ) . r. a. arndt , i. i. strakovsky , and r. l. workman , in : _ proc . of 9th international symposium on meson - nucleon physics and the structure of the nucleon , washington , dc , usa , july 2631 , 2001 _ , eds . h. haberzettl and w. j. briscoe , @xmath28 newslett . * 16 * , 150 ( 2002 ) , eprint nucl - th/0110001 . s. b. gerasimov , phys . at . ( former sov . . phys . ) * 2 * , 430 ( 1966 ) , s. d. drell and a. c. hearn , phys . * 16 * , 908 ( 1966 ) . for an earlier related work , see l. i. lapidus and chou kuang - chao , jetp ( former sov . j. phys . jetp ) * 14 * , 357 ( 1962 ) . r. a. arndt , i. i. strakovsky , and r. l. workman , phys . c * 53 * , 430 ( 1996 ) . r. a. arndt , r. l. workman , z. li , and l. d. roper , phys . c * 42 * , 1853 ( 1990 ) ; _ ibid _ , 1864 ( 1990 ) . z. li , r. a. arndt , l. d. roper , and r. l. workman , phys . c * 47 * , 2759 ( 1993 ) . the full database and numerous pwas can be accessed via a ssh / telnet call to the said facility , with userid : said ( no password ) , or a link to the website . m. fuchs , j. ahrens , g. anton , r. averbeck , r. beck , a. m. bernstein , a. r. gabler , f. hrter , p. d. harty , s. hlavac , b. krusche , i. j. d. mcgregor , v. metag , r. novotny , r. o. owens , j. peise , m. rbig - landau , a. schubert , r. s. simon , h. strher , and v. tries , phys . lett . * b368 * , 20 ( 1996 ) , m. fuchs , ph . d theses , university of giessen , 1995 ( unpublished ) , m. fuchs , private communication , 1996 . m. maccormick , g. audit , n. dhose , l. ghedira , v. isbrt , s. kerhoas , l. y. murphy , g. tamas , p. a. wallace , s. altieri , a. braghieri , p. pedroni , t. pinelli , j. ahrens , r. beck , j. r. m. annand , r. a. crawford , j. d. kellie , i. j. d. macgregor , b. dolbilkin , and a. zabrodin , phys . c * 53 * , 41 ( 1996 ) , p. pedroni , private communication , 1996 . r. beck , h. p. krahn , j. ahrens , h .- j . arends , g. audit , a. braghieri , n. dhose , s. j. hall , v. isbert , j. d. kellie , i. j. d. macgregor , p. pedroni , t. pinelli , g. tamas , th . walcher , and s. wartenberg , phys . lett . * 78 * , 606 ( 1997 ) , h. p. krahn , ph . d. theses , mainz university , 1996 ( unpublished ) , r. beck , private communication , 1997 . f. hrter , ph . d theses , mainz university , kph15/96 , 1996 ( unpublished ) , j. ahrens , private communication , 2002 . m. schneider , ph . d. theses , mainz university , 1994 ( unpublished ) , j. ahrens , private communication , 1998 . b. krusche , j. ahrens , r. beck , m. fuchs , s. j. hall , f. hrter , j. d. kellie , v. metag , m. rbig - landau , and h. strher , eur . phys . j. a * 6 * , 309 ( 1999 ) , b. krusche , private communication , 1999 . j. ahrens , s. altieri , j. r. m. annand , g. anton , h .- j . arends , k. aulenbacher , r. beck , c. bradtke , a. braghieri , n. degrande , n. dhose , h. dutz , s. grtz , p. grabmayr , k. hansen , j. harmsen , d. von harrach , s. hasegawa , t. hasegawa , e. heid , k. helbing , h. holvoet , l. van hoorebeke , n. horikawa , t. iwata , p. jennewein , t. kageya , b. kiel , f. klein , r. kondratiev , k. kossert , j. krimmer , m. lang , b. lannoy , r. leukel , v. lisin , t. matsuda , j. c. mcgeorge , a. meier , d. menze , w. meyer , t. michel , j. naumann , r. o. owens , a. panzeri , p. pedroni , t. pinelli , i. preobrajenski , e. radtke , e. reichert , g. reicherz , ch . rohlof , d. ryckbosch , f. sadiq , m. sauer , b. schoch , m. schumacher , b. seitz , t. speckner , m. steigerwald , n. takabayashi , g. tamas , a. thomas , r. van de vyver , a. wakai , w. weihofen , f. wissmann , f. zapadtka , and g. zeitler , phys . lett . * 84 * , 5950 ( 2000 ) , h .- j . arends and i. preobrajenski , private communication , 2000 , p. pedroni , private communication , 2002 . j. ahrens , s. altieri , j. r. m. annand , g. anton , h .- j . arends , k. aulenbacher , r. beck , c. bradtke , a. braghieri , n. degrande , n. dhose , h. dutz , s. grtz , p. grabmayr , k. hansen , j. harmsen , d. von harrach , s. hasegawa , t. hasegawa , e. heid , k. helbing , h. holvoet , l. van hoorebeke , n. horikawa , t. iwata , p. jennewein , t. kageya , b. kiel , f. klein , r. kondratiev , k. kossert , j. krimmer , m. lang , b. lannoy , r. leukel , v. lisin , t. matsuda , j. c. mcgeorge , a. meier , d. menze , w. meyer , t. michel , j. naumann , r. o. owens , a. panzeri , p. pedroni , t. pinelli , i. preobrajenski , e. radtke , e. reichert , g. reicherz , ch . rohlof , d. ryckbosch , f. sadiq , m. sauer , b. schoch , m. schumacher , b. seitz , t. speckner , m. steigerwald , n. takabayashi , g. tamas , a. thomas , r. van de vyver , a. wakai , w. weihofen , f. wissmann , f. zapadtka , and g. zeitler , phys . 87 * , 022003 ( 2001 ) . r. beck , h. p. krahn , j. ahrens , j. r. m. annand , h .- j . arends , g. audit , a. braghieri , n. dhose , d. drechsel , o. hanstein , j. c. mcgeorge , r. o. owens , p. pedroni , t. pinelli , g. tamas , l. tiator , and th . walcher , phys . c * 61 * , 035204 ( 2000 ) , h. p. krahn , ph . d. theses , mainz university , 1996 ( unpublished ) , r. beck , private communication , 2000 . d. branford , j. a. mackenzie , f. x. lee , j. ahrens , j. r. m. annand , r. beck , g. e. cross , t. davinson , p. grabmayr , s. j. hall , p. d. harty , t. hehl , d. g. johnstone , j. d. kellie , t. lamparter , m. liang , i. j. d. macgregor , j. c. mcgeorge , r. o. owens , m. sauer , r. schneider , a. c. shotter , k. spaeth , d. p. watts , p. j. woods , l. e. wright , and t. t. yau , phys . c * 61 * , 014603 ( 2000 ) , d. branford , private communication , 2000 . a. schmidt , p. achenbach , j. ahrens , h .- j . arends , r. beck , a. m. bernstein , v. hejny , m. kotulla , b. krusche , v. kuhr , r. leukel , i. j. d. macgregor , j. c. mcgeorge , v. metag , v. m. olmos de leon , f. rambo , u. siodlaczek , h. strher , th . walcher , j. weiss , f. wissmann , and m. wolf , phys . lett . * 87 * , 232501 ( 2001 ) , a. schmidt , ph . d. thesis , mainz university , sept . 2001 ( unpublished ) , r. beck , private communication , 2001 . j. c. bergstrom , j. m. vogt , r. igarashi , k. j. keeter , e. l. hallin , g. a. retzlaff , d. m. skopik , and e. c. booth , phys . rev . c * 53 * , 1052 ( 1996 ) , j. c. bergstrom , private communication , 1996 . j. c. bergstrom , r. igarashi , and j. m. vogt , phys . c * 55 * , 2016 ( 1997 ) , j. c. bergstrom , private communication , 1997 . e. korkmaz , n. r. kolb , d. a. hutcheon , g. v. orielly , j. c. bergstrom , g. feldman , d. jordan , a. k. opper , r. e. pywell , b. sawatzky , and d. m. skopik , phys . lett . * 83 * , 3609 ( 1999 ) , g. v. orielly , private communication , 1999 . g. blanpied , m. blecher , a. caracappa , j. r. deininger , c. djalali , g. giordano , k. hicks , s. hoblit , m. khandaker , o. c. kistner , a. kuczewski , f. lincoln , m. lowry , m. lucas , g. matone , l. miceli , b. m. preedom , d. rebreyend , a. m. sandorfi , c. schaerf , r. m. sealock , h. strher , c. e. thorn , s. t. thornton , j. tonnison , c. s. whisnant , h. zhang , and x. zhao , phys . c * 64 * , 025203 ( 2001 ) . differential cross sections for both @xmath13 and @xmath93 channels have been flagged . j. ajaka , o. bartalini , m. battaglieri , v. bellini , j .- bocquet , p. m. calvat , m. capogni , m. castoldi , l. ciciani , a. dangelo , j .- didelez , r. di salvo , m. a. duval , c. gaulard , g. gervino , f. ghio , b. girolami , m. guidal , e. guinault , e. hourany , a. jejcic , i. kilvington , v. kouznetsov , a. lapik , p. levi sandri , a. lleres , d. moricciani , v. nedorezov , l. nicoletti , d. rebreyend , f. renard , m. ripani , n. rudnev , c. schaerf , a. turinge , and a. zucchiatti , phys . lett . * b475 * , 372 ( 2000 ) , v. kouznetsov , private communication , 2000 . o. bartalini , v. billini , j .- p . bocquet , m. capogni , m. castoldi , a. dangelo , j .- didelez , r. di salvo , a. fantini , g. gervino , f. ghio , b. girolami , m. guidal , e. hourany , v. kouznetsov , r. kunne , a. lapik , p. levi sandri , a. lleres , d. moricciani , v. nedorezov , l. nicoletti , d. rebreyend , f. renard , n. roudnev , c. schaerf , m. sperduto , m. sutera , a. turinge , a. zucchiatti , r. a. arndt , w. j. briscoe , i. i. strakovsky , and r. l. workman , submitted to phys . * b * , v. kouznetsov , private communication , 2001 . f. v. adamian , a. yu . buniatian , g. s. frangulian , p. i. galumian , v. h. grabski , a. v. hairapetian , h. h. hakopian , v. k. hoktanian , j. v. manukian , a. m. sirunian , a. h. vartapetian , h. h. vartapetian , v. g. volchinsky , r. a. arndt , i. i. strakovsky , and r. l. workman , phys . c * 63 * , 054606 ( 2001 ) . h. dutz , d. kraemer , b. zucht , k. h. althoff , g. anton , h .- j . arends , w. beulertz , a. bock , m. breuer , r. gehring , m. gemander , s. grtz , k. helbing , j. hey , w. meyer , g. noldeke , r. paulsen , g. reicherz , a. thomas , and s. wartenberg , nucl . * a601 * , 319 ( 1996 ) , g. anton , private communication , 1996 . this is a final version of preliminary t @xmath94 data used in previous analyses . a. bock , g. anton , w. beulertz , chr . bradtke , h. dutz , r. gehring , s. grtz , k. helbing , j. hey , w. meyer , m. pluckthun , g. reicherz , l. szr , m. breuer , j .- didelez , and p. hoffmann - rothe , phys . lett . * 81 * , 534 ( 1998 ) , g. anton , private communication , 1996 . h. w. dannhausen , e. j. durwen , h. m. fischer , m. leneke , w. niehaus , and f. takasaki , eur . j. a * 11 * , 441 ( 2001 ) . this final version of preliminary @xmath94 differential cross sections by h. w. dannhausen , ph . d. theses , bonn university , bonn ir7729 , 1977 ( unpublished ) and m. leneke , ph . d. theses , bonn university , bonn ir7714 , 1977 ( unpublished ) used in previous analyses . e. j. durwen , ph . d. theses , bonn university , bonn ir807 , 1980 ( unpublished ) . the final version of these @xmath12 differential cross sections will submit , h. m. fischer , private communication , 2001 . u. schfer , ph . d. theses , bonn university , bonn ir8121 , 1981 ( unpublished ) . these @xmath12 differential cross sections have been removed from the database because they were taken in a first test of a new spectrometer , h. m. fischer , private communication , 2001 . w. heise , ph . d. theses , bonn university , bonn ir8806 , 1988 ( unpublished ) . the final version of these @xmath12 differential cross sections will submit , h. m. fischer , private communication , 2001 . p. zenz , ph . d. theses , bonn university , bonn ir8812 , 1988 ( unpublished ) . the final version of these @xmath13 differential cross sections will submit , h. m. fischer , private communication , 2001 . w. j. briscoe , j. ficenec , d. jenkins _ et al . _ cebaf proposal , e94103 , 1994 . i. preobrajenski , in : _ proc . of 9th international symposium on meson - nucleon physics and the structure of the nucleon , washington , dc , usa , july 2631 , 2001 _ , eds . h. haberzettl and w. j. briscoe , @xmath95 newslett . * 16 * , 308 ( 2001 ) , i. preobrajenski , ph . d. thesis , mainz university , sept . 2001 ( unpublished ) . i. s. barker , a. donnachie , and j. k. storrow , nucl . b95 * , 347 ( 1975 ) . r. leukel , ph . d. thesis , mainz university , sept . 2001 ( unpublished ) . d. rebreyend , private communication , 1999 . r. j. holt , r. gilman _ cebaf proposal , e94012 , 1994 , r. gilman , private communication , 2000 . a. m. sandorfi , private communication , 1995 . the preliminary version of @xmath4 @xmath6 has been used in previous analyses . a. shafi for the crystal ball collaboration , bull . soc . * 44 * , 43 ( 1999 ) . k. watson , phys . rev . * 95 * , 228 ( 1954 ) . r. a. arndt , r. l. workman , i. i. strakovsky , and m. m. pavan , in preparation , eprint nucl - th/9807087 , . mainz fits are available at the maid website . see also s. s. kamalov , s. n. yang , d. drechsel , o. hanstein , and l. tiator , phys . c * 64 * , 032201 ( 2001 ) . the unitary isobar model is developed at mainz ( hereafter called maid ) , d. drechsel , o. hanstein , s. s. kamalov , and l. tiator , nucl . * a645 * , 145 ( 1999 ) . maid2000 refers to the feb . 2001 version of the maid solution , maid2001 refers to the nov . 2001 version of the maid solution from s. kamalov . j. ahrens , s. altieri , j. r. m. annand , g. anton , h .- j . arends , k. aulenbacher , r. beck , c. bradtke , a. braghieri , n. degrande , n. dhose , d. drechsel , h. dutz , s. grtz , p. grabmayr , k. hansen , j. harmsen , d. von harrach , s. hasegawa , t. hasegawa , e. heid , k. heibling , h. holvt , l. van hoorebeke , n. norikawa , t. iwata , o. jahn , p. jennewein , t. kageya , s. kamalov , b. kiel , f. klein , r. kondratiev , k. kossert , j. krimmer , m. lang , b. lannoy , r. leukel , v. lisin , t. matsuda , j. c. mcgeorge , a. meier , d. menze , w. meyer , t. michel , j. naumann , a. panzeri , p. pedroni , t. pinelli , i. preobrajenski , e. radtke , e. reichert , g. reicherz , ch . rohlof , g. rosner , d. ryckbosch , m. sauer , b. schoch , m. schumacher , b. seitz , t. speckner , n. takabayashi , g. tamas , a. thomas , l. tiator , r. van de vyver , a. wakai , w. weihofen , f. wissmann , f. zapadtka , and g. zeitler , eprint hep ex/0203006 , submitted to phys . this set is a part of ref . @xcite and has not been used in our analysis . r. a. arndt , i. aznauryan , r. m. davidson , d. drechsel , o. hanstein , s. s. kamalov , a. s. omelaenko , i. i. strakovsky , l. tiator , r. l. workman , and s. n. yang , summary of the partial wave analysis group of the brag workshop , mainz , march 56 , 2001 , in : _ proceedings of nstar2001 , mainz , germany , march 710 , 2001 _ , eds . d. drechsel and l. tiator , world scientific , p. 467 ; eprint nucl - th/0106059 . s. s. kamalov , g .- y . chen , s. n. yang , d. drechsel , and l. tiator , phys . b522 * , 27 ( 2001 ) . o. hanstein , d. drechsel , and l. tiator , phys . b385 * , 45 ( 1996 ) . p. j. bussey , c. raine , j. g. rutherglen , p. s. l. booth , l. j. carroll , g. r. court , p. r. daniel , a. w. edwards , r. gamet , c. j. hardwick , p. j. hayman , j. r. holt , j. n. jackson , j. h. norem , w. h. range , f. h. combley , w. galbraith , v. h. rajaratnam , and c. sutton , nucl . phys . * b154 * , 205 ( 1979 ) . j. alspector , d. fox , d. luckey , c. nelson , l. s. osborne , g. tarnopolsky , z. bar - yam , j. de pagter , j. dowd , w. kern , and s. m. matin , phys . * 28 * , 1403 ( 1972 ) . c. betourne , j. c. bizot , j. p. perez - y - jorba , d. treille , and w. schmidt , phys . rev . * 172 * , 1343 ( 1968 ) . s. d. ecklund and r. l. walker , phys . rev . * 159 * , 1195 ( 1967 ) . j. h. boyden , ph . d. thesis , ucla , 1961 ( unpublished ) . j. r. kilner , ph . d. thesis , ucla , 1962 ( unpublished ) . g. buschhorn , j. carroll , r. d. eandi , p. heide , r. hbner , w. kern , u. ktz , p. schmser , and h. j. skronn , phys . lett . * 18 * , 571 ( 1967 ) . d. sober , g. zeitler , in : _ proc . of 9th international symposium on meson - nucleon physics and the structure of the nucleon , washington , dc , usa , july 2631 , 2001 _ , eds . h. haberzettl and w. j. briscoe , @xmath96 newslett . * 16 * , 311 ( 2001 ) . t. iwata , in : _ proc . of the gerasimov - drell - hearn sum rule and the nucleon spin structure in the resonance region , mainz , germany , june 1417 , 2000 _ , eds . d. drechsel and l. tiator , p. 289 , + . a. lehman , _ proceedings of 9th international conference on the structure of baryons ( baryons2002 ) , newport news , va , usa , march 38 , 2002 _ , to be published . l. tiator , talk given at 9th international symposium on meson - nucleon physics and the structure of the nucleon , washington , dc , usa , july 2631 , 2001 , private communication , 2001 . s. s. kamalov , private communication , 2002 . r. l. crawford and w. t. morton , nucl . * b211 * , 1 ( 1983 ) , particle data group , phys . * b239 * , 1 ( 1990 ) . r. l. crawford , in : _ proceedings of nstar2001 , mainz , germany , march 710 , 2001 _ , eds . d. drechsel and l. tiator , world scientific , p. 163 . d. drechsel , o. hanstein , s. s. kamalov , and l. tiator , nucl . * a645 * , 145 ( 1999 ) . d. e. groom _ et al . _ , _ review of particle physics _ , j. c * 15 * , 1 ( 2000 ) , 2001 off - year partial update for 2002 edition available on the pdg site . s. capstick , phys . d * 46 * , 2864 ( 1992 ) . a. m. baldin , nucl . phys . * 18 * , 310 ( 1960 ) , l. i. lapidus , jetp ( former sov . j. phys . jetp ) * 16 * , 964 ( 1963 ) . m. i. levchuk and a. i. lvov , nucl . a674 * , 449 ( 2000 ) . v. olmos de leon , f. wissmann , p. achenbach , j. ahrens , h. j. arends , r. beck , p. d. harty , v. hejny , p. jennewein , m. kotulla , b. krusche , v. kuhr , r. leukel , j. c. mcgeorge , v. metag , r. novotny , a. polonski , f. rambo , a. schmidt , m. schumacher , u. siodlaczek , h. strher , a. thomas , j. weiss , and m. wolf , eur . j. a * 10 * , 207 ( 2001 ) . m. lundin , j .- o . adler , t. glebe , k. fissum , k. hansen , l. isaksson , o. kaltschmidt , m. karlsson , k. kossert , m. i. levchuk , p. lilja , b. lindner , a. i. lvov , b. nilsson , d. e. oner , c. poech , s. proff , a. sandell , b. schrder , m. schumacher , and d. a. sims , eprint nucl - ex/0204014 , submitted to phys . s. ragusa , phys . d * 47 * , 3757 ( 1993 ) . s. kondratyuk and o. scholten , phys . c * 65 * , 038201 ( 2002 ) . i. g. alekseev , v. p. kanavets , b. v. morozov , d. n. svirida , s. p. kruglov , a. a. kulbardis , v. v. sumachev , r. a. arndt , i. i. strakovsky , and r. l. workman , phys . c * 55 * , 2049 ( 1997 ) . .comparison of present ( sm02 ) and previous ( sm95 @xcite , sp93 @xcite , and sp89 @xcite ) energy - dependent partial - wave analyses of charged and neutral pion photoproduction data . the @xmath97 values for the previous solutions correspond to our published results . @xmath98 is the number parameters varied in the fit . [ cols="^,^,^,^,^,^,^ " , ] energy - angle distribution of recent ( post-1995 ) data : ( a ) unpolarized @xmath2 , ( b ) polarized @xmath2 , ( c ) unpolarized @xmath3 , ( d ) polarized @xmath3 . ( a , c ) total cross sections and ( b , d ) ( @xmath99 ) are plotted at zero degrees . partial - wave amplitudes ( l@xmath100 ) from threshold to @xmath9 = 2 gev . solid ( dashed ) curves give the real ( imaginary ) parts of amplitudes corresponding to the sm02 solution . the real ( imaginary ) parts of single - energy solutions are plotted as filled ( open ) circles . the previous sm95 solution @xcite is plotted with long dash - dotted ( real part ) and short dash - dotted ( imaginary part ) lines . plotted are the multipole amplitudes ( a ) @xmath101 , ( b ) @xmath102 , ( c ) @xmath103 , ( d ) @xmath104 , ( e ) @xmath105 , ( f ) @xmath106 , ( g ) @xmath107 , ( h ) @xmath108 , ( i ) @xmath109 , ( j ) @xmath110 , ( k ) @xmath111 , ( l ) @xmath112 , ( m ) @xmath113 , ( n ) @xmath114 , ( o ) @xmath115 , ( p ) @xmath116 , ( q ) @xmath117 , ( r ) @xmath118 , ( s ) @xmath119 , ( t ) @xmath120 , ( u ) @xmath121 , ( v ) @xmath122 , ( w ) @xmath123 , and ( x ) @xmath124 . the subscript p ( n ) denotes a proton ( neutron ) target . photon asymmetry for @xmath15 photoproduction on the proton at 159.5 mev . data are from mainz ( solid circles ) @xcite . plotted are the sm02 ( solid line ) , the 162 mev - ses ( 158 @xmath34 165 mev ) fit associated with sm02 ( dotted lines represent uncertainties of the ses fit ) and the maid2000 results ( dash - dotted ) @xcite . @xmath6 beam asymmetry for @xmath3 at 1100 mev . black circles show graal results @xcite , open circles indicate the results of the daresbury group @xcite , open triangles indicate the results from saclay @xcite . the solid ( dash - dotted ) line represents the sm02 ( maid2001 @xcite ) solution . forward ( @xmath128 ) differential cross section for @xmath129 as a function of energy . experimental data for the range of @xmath130 are from orsay @xcite ( black circles ) , slac @xcite ( open circles ) , @xcite ( open triangles ) , @xcite ( black square ) , and desy @xcite ( black diamonds . ) the solid ( dash - dotted ) line represents the sm02 ( maid2001 @xcite ) solution . difference of the total cross sections for the helicity states 1/2 and 3/2 . ( a ) @xmath42 and ( b ) @xmath131 . the solid ( dash - dotted ) line represents the sm02 ( maid2001 @xcite ) solution . experimental data are from mainz @xcite . selected partial - wave amplitudes to @xmath9 = 1250 mev . solid ( dashed ) curves give the real ( imaginary ) parts of amplitudes corresponding to the sm02 solution . the recent maid2001 solution @xcite is plotted with long dash - dotted ( real part ) and short dash - dotted ( imaginary part ) lines . plotted are the multipole amplitudes ( a ) @xmath43 [ @xmath101 ] , ( b ) @xmath44 [ @xmath106 ] , ( c ) @xmath45 [ @xmath110 ] , and ( d ) @xmath46 [ @xmath113 ] . the subscript p ( n ) denotes a proton ( neutron ) target . @xmath43 multipole for 600 to 1200 mev . plotted are ( a ) real part and ( b ) imaginary part . the sm02 ( sx99 ) solution is plotted with a solid ( dashed ) line and previous sm95 solution @xcite with a dash - dotted line .

this paper extends our previous study of planet / brown dwarf accretion by giant stars to solar mass stars located on the red giant branch . the model assumes that the planet is dissipated at the bottom of the convective envelope of the giant star . the giant s evolution is then followed in detail . we analyze the effects of different accretion rates and different initial conditions . the computations indicate that the accretion process is accompanied by a substantial expansion of the star , and in the case of high accretion rates , hot bottom burning can be activated . the possible observational signatures that accompany the engulfing of a planet are also extensively investigated . they include : the ejection of a shell and a subsequent phase of ir emission , an increase in theli surface abundance and a potential stellar metallicity enrichment , spin - up of the star due to the deposition of orbital angular momentum , the possible generation of magnetic fields and a related x - ray activity due to the development of shear at the base of the convective envelope , and the effects on the morphology of the horizontal branch in globular clusters . we propose that the ir excess and high li abundance observed in 4 - 8% of the g and k giants originate from the accretion of a giant planet , a brown dwarf or a very low - mass star . -1 cm -0.65 cm -0.65 cm 2h # 1 accretion - planetary systems - stars : low - mass , brown dwarfs - stars : late type - stars : evolution - stars : mass loss - stars : chemically peculiar - stars : rotation - x - ray : stars - infrared : stars

astro-ph9905235

planets and brown dwarfs have now been detected around several main sequence stars ( e.g. mayor & queloz 1995 , butler & marcy 1996 , cochran et al . 1997 , rebolo et al . 1995 , 1996 , basri et al . 1996 , marcy et al . 1998 , butler et al . 1998 , delfosse et al . 1998 ) , and strikingly , a fraction of these `` hot jupiters '' orbit very close to the central star , at less than 1 au . this proximity to the central star poses a theoretical problem regarding the mechanism that brought these planets to these orbits , but it also raises the question of their fate . indeed , rasio et al . ( 1996 ) showed that the orbit of 51 peg s companion is tidally unstable and that the planet will ultimately be engulfed in the stellar envelope . the problem we address here is that of the effects of the accretion of a giant planet or a brown dwarf on the structure and evolution of solar mass stars on the red giant branch . this study is the continuation of our previous work ( siess & livio 1998 ) , in which we investigated the swallowing of massive planets / brown dwarfs by a 3 agb star . the general scenario for the accretion process is as follows : planets in close orbits ( as observed ) will be engulfed in the envelope of giant stars as the latter evolve away from the main sequence . due to viscous and tidal forces , angular momentum is imparted to the envelope and the planet spirals - in . previous studies of the spiralling - in process ( e.g. livio & soker 1984 ; soker , harpaz & livio 1984 ) have revealed that depending on the mass of the planet / brown dwarf three different evolutionary paths can take place . if the mass of the planet is larger than some critical value @xmath3 , the planet can accrete a fraction of the envelope mass . on the other hand , lower mass planets will either evaporate or collide with the stellar core . for the specific stellar model used in their computation , livio and soker ( 1984 ) estimated @xmath4 . however , these results must be taken with great caution since the physical processes were treated only approximatively . we also emphasize that such computations are very sensitive to the stellar model , notably to the size and mass of the convective envelope . more recently , sandquist et al . ( 1998 ) performed 3d hydrodynamical simulations of jupiter - like planets impacting onto solar type main sequence stars . they also found that depending on the initial stellar model , planets can be evaporated or survive the crossing of the convective zone . the present work has been motivated by the fact that planets surrounding low mass stars are more likely to be engulfed in the star s envelope during the red giant branch ( rgb ) phase than during the asymptotic giant branch ( agb ) phase . indeed , if we compare ( for different stars ) the ratio of the maximum radius a star can reach during the rgb to its maximum radius on the agb phase , this ratio is much closer to one for low mass stars ( @xmath5 ) than more massive stars ( e.g. soker 1998a ) . consequently , we expect that in most cases low mass stars will interact with close low - mass binary companions on the rgb . in the present paper , we used a spherically symmetric stellar evolution code to follow the evolution of a solar type star that accretes a giant planet or a brown dwarf on the red giant branch . in @xmath6[conditions ] , we describe briefly the computational approach and the initial models . then , in @xmath6[struc ] , we analyze the effects of accretion on the structure and the evolution of the star . we present results for different mass accretion rates and different initial stellar models on the rgb . the observational signatures expected from this process are investigated in detail in @xmath6[signature ] . a discussion and conclusions follow . to investigate the effects of planet accretion by solar type stars , we used 3 different evolutionary models of the sun corresponding to 3 different points on the rgb . these models are characterized by their mass , which is equal to @xmath7 , 0.9 and 0.8 and they are referred to as model # a , # b and # c , respectively . the main properties of the initial models are summarized in table [ tab1 ] . the general rgb structure is a degenerated he core , surrounded by a thin hydrogen burning shell ( hbs ) that provides most of the luminosity . a thin radiative layer separates the hbs from the large convective envelope . the star was assumed to lose mass as prescribed by reimer s law ( 1975 ) at a rate given by @xmath8 where the parameter @xmath9 in our computations , and @xmath10 , @xmath11 , and @xmath12 are given in solar units . the input physics for the accretion process has been previously described by siess and livio ( 1998 ) . briefly , to determine the locus of the dissipation process we estimated the virial temperature of the brown dwarf and the critical radius at which tidal effects lead to a total disruption of the brown dwarf . for characteristic values of the mass and radius for the brown dwarf , we found a temperature of a few million k and a critical radius of the order of @xmath13 . from the numbers presented in table [ tab1 ] , we can localize the accretion process at the bottom of the convective envelope near the hbs . to estimate the accretion rate , we determined the orbital decay timescale of the brown dwarf in the vicinity of the dissipation region . this timescale roughly corresponds to the time required by the brown dwarf to cross the dissipation region and thus represents the rate at which mass is added there . for typical stellar parameters , we inferred mass accretion rates of the order of @xmath14 , depending mainly on the mass of the brown dwarf . .physical properties of the initial models [ cols="<,^,^,^,^,^,^ " , ] @xmath15for these models , we were not able to follow reliably the evolution and the computations were stopped at the indicated accreted mass the deposition of mass in the star is accompanied by the release of a large amount of gravitational energy that acts to expand the envelope ( fig . [ evol5 ] ) . however , because of the long timescale of heat diffusion , the expansion is delayed , especially for the low luminosity stars ( cases a0 and a1 ) . as the radius increases , the star also becomes more luminous . according to reimer s law , these modifications to the structure significantly increase the mass loss rate , and we expect some changes in the subsequent evolution and stellar environment ( see @xmath6[shell ] and @xmath6[morpho ] ) . the expansion proceeds until a thermal equilibrium is reached , in which the rate of energy loss by the surface balances the rate of energy deposition by the accretion process . depending on the initial model , the return to contraction occurs after @xmath16 , @xmath17 and @xmath18yr , in cases a0 , b0 and c0 , respectively . this timescale corresponds roughly to the time of heat diffusion ( thermal timescale ) of the expanding part of the star and it is approximatively given by @xmath19 , where @xmath20 is the mass of the envelope and is the accretion luminosity . the latter is given by @xmath21 , where is the mass accretion rate and @xmath22 and @xmath23 the mass and radial coordinates of the accretion zone . it should be noted that represents an upper limit to the accretion luminosity since energy is lost during the spiralling - in process to heating and spinning up the envelope . however , throughout most of the envelope , the orbital energy disspation rate represents a small fraction of the intrinsic stellar luminosity and thus we do not expect that the stellar structure will be significantly affected until the planet reaches the accretion zone . finally , we should mention that the energy deposition might not be local ( in the accretion region ) . for example , the star could be tidally excited , oscillations may be initiated and then damped elsewhere in the envelope . also , in this common envelope phase , magnetic fields might be generated in the accretion region ( see @xmath6[xray ] ) and energy could be dissipated at the surface of the star or in the corona ( e.g. regos & tout 1995 ) . during the envelope expansion , the opacity @xmath24 increases in the accretion region and rapidly the radiative gradient becomes larger than the adiabatic one ( @xmath25 ) ; thus the convective envelope deepens ( fig . [ kippen ] ) . thereafter , as the hbs advances in mass and reaches the cool regions that have been previously expanded , the nuclear energy production rate decreases significantly . as an order of magnitude , during most of the accretion phase , the luminosity due to h burning represents only a few percent of the total stellar luminosity . the accretion luminosity thus supports the convective envelope almost entirely and the accretion region acts as a source shell that prevents an efficient h burning . to see how the structure is affected by the accretion process , we have presented in fig . [ prof1 ] the profiles of several variables for case a0 , but they are globally similar in the other models . in the accretion region where a large amount of energy is deposited , the luminosity and temperature rise . but , due to the relatively long timescale associated with heat diffusion , energy accumulates locally and both profiles now present a bump . the heat excess finally diffuses when the star has sufficiently expanded and a monotonically increasing luminosity profile is restored . however , the temperature profile still presents a local maximum that persists throughout the accretion phase . this is due to the fact that this region undergoes a compressional heating from the mass accumulated above it and also from the expanding region surrounding the core . indeed , the luminosity @xmath26 is negative just behind the luminosity jump , indicating a region in expansion ( @xmath27 ) . the temperature is thus compelled to rise and , as we shall see in @xmath6[relax ] , this situation can give rise to a thermal instability . the temperature gradient also becomes progressively steeper beyond the bump ( fig . [ prof1 ] ) as energy accumulates . this contributes to the shrinkage of the hbs and to the subsequent drop in the nuclear luminosity . in the hbs , prior to accretion , the nuclear energy production rate is mainly supplied by the cno bi - cycle . more precisely , the most important reactions ( and their contribution ) are c(p,@xmath28)n ( @xmath29 ) , n(p,@xmath28)o ( @xmath30),n(p,@xmath31)c ( @xmath32 ) and o(@xmath33)n ( @xmath32 ) . however , because the hbs advances in mass into a cooler region , the pp1 branch progressively takes over the nuclear energy production and during the accretion phase , the reactions 2he(@xmath34,2p)@xmath35 and c(p,@xmath28)n contribute @xmath36% and @xmath37% of the total nuclear energy production , respectively . on the other hand , in the region of maximum temperature , c is now efficiently burnt through the c(@xmath31,n)o reaction . this reaction is highly energetic and provides up to @xmath38 of the global nuclear energy production ( which still remains very low , however ) . in fig . [ prof2 ] , we show the evolution of the nuclear activity for case b0 . first , c burning is ignited in the vicinity of the temperature peak ( dotted line ) . then , as it is depleted and converted into o , the nuclear energy production rate progressively decreases near the temperature peak ( short - dashed and long - dashed curves ) , and finally we end up with a configuration in which c is burnt through c(@xmath35,n)o around the temperature peak and through c(p,@xmath28)n above the bump in the hbs . in case c0 , the temperature in the peak reaches @xmath39k and he is ignited off - center in a non degenerate shell ( we will discuss this special case further in @xmath6[relax ] ) . for higher accretion rates , the situation is somewhat different because of the appearance of hot bottom burning ( hbb ) . indeed , contrary to the lower accretion rate case , the larger energy release contributes to a further increase in the opacity and the convective envelope deepens even more . it finally reaches the hbs where its advance stops ( fig . [ struc4 ] ) . our computations indicate that hbb is activated approximately 200 , 25 and 15 yr after the onset of accretion in cases a2 , b2 and c2 respectively . the temperature at the base of the convective envelope rises suddenly from @xmath40 to about @xmath41 million kelvin , depending on the initial model . however , this temperature is still too low to give rise to a rich nucleosynthesis . the onset of hbb entails a significant and fast expansion of the envelope as heat is directly transferred to the convective zone . moreover , the arrival of he from the convective envelope and its destruction through the reaction 2he(@xmath28,2p)@xmath35 increases temporarily the nuclear luminosity in the hbs . during the hbb phase , li is slowly depleted because of the competing effects of li production through the reactions @xmath35(he,@xmath28)be(@xmath42)li and li destruction through the reaction li(p,@xmath35)@xmath35 ( see also @xmath6[chem ] ) . finally , in cases b2 and c2 we were not able to follow reliably the computations . the star s tendency to expand and increase its luminosity made the computations intractable . at that point the envelope is probably ejected and the star enters a proto - planetary nebula phase . after the end of the accretion phase , the envelope rapidly contracts because of the sudden suppression of accretion energy . however , in the models in which the duration of the accretion process is shorter than the thermal timescale associated with the envelope ( as in cases b1 and c1 for example ) , the expansion still proceeds for a while after the end of mass addition . thereafter , the gravitational energy released during the contraction phase heats up the central parts of the star and the hbs re - ignites . during that period of contraction , the convective envelope retreats towards the surface and hbb disappears . the contraction rate is very fast , especially for stars having a shallower convective envelope . the computations indicate that in a few thousand years , the radius decreases by a factor of @xmath40 which , we suspect , could initiate stellar pulsations . when the hbs finally provides most of the stellar luminosity ( @xmath43% ) , the contraction ends and the star resumes a standard evolution , climbing the red giant branch until he flashes at its tip . interestingly , the star is now more luminous and has a larger radius than in the standard evolution despite the small amount of added mass ( fig . [ evol5 ] ) . the enhanced mass loss rate that follows can affect the subsequent evolution ( see @xmath6[morpho ] ) . in all our simulations , except for case c0 , the temperature `` bump '' dissipated as the gravitational energy released during the contraction phase heated up the stellar interior . however in this particular model , in which the accretion process is maintained for @xmath44yr at a rate equal to @xmath45 , the temperature in the peak reaches @xmath39k and he ignites off - center in a non degenerate shell . the ignition of he through the @xmath46 reactions is very energetic and this gives rise to a thermal instability ( fig . [ pulse ] ) . the geometrical thinness and the large opacity of the he burning shell ( hebs ) are responsible for the thermal runaway that leads to the development of a convective tongue . the release of energy due to he burning is used to expand the layers above and secondarily to heat up the core . the hbs is pushed outward into a cooler region and it extinguishes . the remaining energy excess is then absorbed by the envelope which subsequently expands . during the thermal pulse , the nuclear energy production comes from 3@xmath35 ( @xmath47% ) , n(@xmath48)f ( @xmath49 ) and , to a lesser extent , from o(@xmath50)ne and o(@xmath51)n ( @xmath52 each ) . the thermal pulse is also responsible for the further destruction of h through c(p,@xmath28)n . after a significant expansion that allows the star to get rid of the energy excess , contraction resumes . however , due to the extremely low h content left after the pulse ( the h mass fraction @xmath53 ) , the hbs can not ignite before a new thermal instability is triggered . a new he shell flash thus occurs deeper inside the star where both the temperature and he abundance are higher . as evolution proceeds , the pulse duration increases but the maximum luminosity due to he burning decreases mainly because of the progressive lifting of the core degeneracy as the temperature increases . during the subsequent evolution , the star experiences relatively short phases of expansion following the thermal pulses and longer phases of contraction the deepening of the pulse accounts for the smaller and smaller increase in radius following the he shell flashes since most of the released energy is absorbed in the radiative layers before reaching the convective envelope . after the 12@xmath54 pulse , the hbs ignites precisely at the mass coordinate corresponding to the top of the first convective tongue in which the h abundance is high ( @xmath55 ) . the h luminosity then increases and the thermal pulse finally reaches the center of the star . because of the previous thermal instabilities , the degeneracy has been lifted in the interior and consequently he does not flash . helium is now burnt at the center and the star will spend the next @xmath56yr on the horizontal branch . in this section , we attempt to give a broad overview of the different observational signatures that might be expected from the accretion of a planet or a brown dwarf by a giant star on the red giant branch . in general , effects may be expected in at least six different areas , which we describe in some detail below . the modifications to the structure during the accretion process increase the mass loss rate and may lead to the ejection of shells . they also have potential effects on the morphology of the horizontal branch . the deposition of angular momentum during the spiralling - in process can also alter the stellar structure by increasing the rotation rate and this may also produce x - ray emission . finally , the mixing of the dissipated planet material in the envelope can change the surface chemical abundances . all of these effects may have important observational consequences but , as we shall see in the following sections , it is more the combination of several signatures , such as a high rotation rate and/or a large ir excess and/or a high lithium abundance that can identify an accretion event less ambiguously , rather than the detection of one particular signature . during the red giant phase , chemical changes are mainly the result of `` dilution '' effects due to the deepening of the convective envelope but light elements such as li , be and b can also be burnt as the convective envelope approaches its deepest extent . the first dredge up reduces the li abundance by more than 2 orders of magnitude . there is also a decrease in the c / c ratio by a factor of 3 , changing from about 83 to 26 . the first dredge up also accounts for the large enrichment of he as a by product of h burning . with our assumed chemical composition for giant planets and brown dwarfs ( from anders & grevesse 1989 ) , the deposition of accreted matter , mostly in the convective envelope , will principally modify the chemical abundances of the light elements that were previously depleted during the first dredge up . in that respect , li will be the most affected . our computations indicate that , depending on the accreted mass , the li abundance can be increased by more than 2 orders of magnitude , making this element easily detectable at the surface of the star [ where by definition log@xmath57h@xmath58 ] ( see also @xmath6[lirich ] ) . as an illustration , we present in fig . [ dilu ] the li surface abundances resulting from our computations and compare them with what is expected from simple dilution effects . in the dilution hypothesis , we assume that the envelope mass @xmath20 is held constant during the accretion process and that no nuclear burning is taking place in the convective zone . the mass fraction @xmath59 of an element @xmath60 is then simply given by @xmath61 where @xmath62 is the mass accreted into the envelope , and @xmath63 and @xmath64 are the mass fraction of element @xmath60 in the envelope and in the accreted matter , respectively . note that this expression is general in the sense that @xmath59 can either represent the h mass fraction @xmath65 , the he mass fraction @xmath66 or the metallicity @xmath67 . figure [ dilu ] shows that the simulations follow relatively well the theoretical curves defined by eq . ( [ eqli ] ) . the slight deviations from this relation are mainly due to nuclear burning that causes li depletion . note that in case a2 , li is severely depleted due to the long lasting hbb . the deposition of accreted matter will also increase the be and b chemical abundances , but to a lower extent , since the effects of dilution during the first dredge up are smaller for these elements . a modest increase in the c / c ratio can also take place . the effects of planet accretion on the stellar metallicity are very sensitive to the amount and chemical composition of the accreted material and also to the mass of the convective envelope ( which depends on the evolutionary status of the star ) . if the metal content of the giant planet is higher than the one adopted here , the metallicity can be further enhanced . secondly , if the planet is engulfed while the star is still on the main sequence for example , the convective envelope is then very shallow ( @xmath68 for the sun ) , and consequently the stellar photospheric composition can be substantially modified ( eq . [ eqli ] ) . recent determinations of the chemical composition of the main sequence stars hosting planets ( gonzales 1997 , 1998 ) indeed show that in some cases the metallicity is abnormally high , suggesting that proto - planetary material or planets have been accreted ( e.g. lin et al . 1996 , laughlin & adams 1997 ) . more recently , sandquist et al . ( 1998 ) showed that a jupiter - like planet can be evaporated in the convective envelope of the sun . they also inferred that in some cases a metal enrichment as high as 50% can be obtained , but the exact value depends greatly on the mass of the convective zone and on the chemical composition and interior structure of the giant planet . all the above suggests that if the presence of planets around solar type stars is rather common , and if some of these planets are indeed in close orbits , then certain chemical peculiarities in giants stars may be the consequence of the accretion of planets . during the spiralling - in process , angular momentum is imparted to the envelope . as a consequence , the stellar rotation rate will be substantially increased . more quantitatively , if all the orbital angular momentum of the planet is entirely deposited and redistributed in the envelope , i.e. if we assume that the convective envelope is rotating rigidly , then the ratio of the envelope angular velocity @xmath69 to the critical velocity ( surface keplerian angular velocity ) @xmath70 is given by @xmath71 where @xmath72 is the orbital separation of the brown dwarf , @xmath73 the gyration radius and @xmath74 the mass of the envelope . equation ( [ omega ] ) indicates that only massive planets or brown dwarfs ( @xmath75 ) can spin - up the giant envelope to a significant fraction of the keplerian velocity ( @xmath76% ) . the spin - up would be most effective at the tip of the giant branch , where both @xmath20 and @xmath73 are smaller . measurements of the projected rotational velocity of luminous class iii giants indicate a sudden drop in near spectral type g0 iii ( gray 1989 ) . the same discontinuity is also observed in subgiants at spectral type f8 iv ( gray & nagar 1985 ) and on the main sequence near mid - f stars ( kraft 1970 ) . these observations thus delineate in the hr diagram a rotational dividing line ( rdl ) where slow rotators ( @xmath77 ) populate the cool side ( de medeiros et al . the strong braking at the rotational discontinuity occurs very rapidly , perhaps on a time scale as short as @xmath78 yr or less ( ayres et al . 1998 ) , and probably results from an efficient magnetic braking attributed to a dynamo action ( gray 1981 , 1982 ) . in the `` rotostat '' hypothesis ( gray 1986 ) , angular momentum is dredged up from the rapidly rotating core by the deepening of the convective envelope . this activates the dynamo and then an efficient magnetic braking takes place . in this model , the rotation rate is controlled in a self - regulatory way by the balance between the arrival of angular momentum from the core and the strong braking that ensues , and consequently , the rotational velocity is maintained constant . however , a few stars do not conform to this scheme and still rotate rapidly beyond the rdl , challenging the rotostat hypothesis . for example the case of 7 boo = hr 5225 remains unexplained ( gray 1989 , 1991 ) . this single star of spectral type g5iii has a moderate lithium abundance log@xmath79(li)=1.2 and its rotational velocity (= 14.1 ) is three times higher than the average . this isolated case could fit into a scenario in which the giant star has recently accreted a planet and its surface layers have not yet spun down . the moderate lithium abundance would suggest the accretion of a relatively small planet . more generally , de medeiros et al . ( 1996a ) found that among a sample of @xmath80 stars , less than @xmath81% of the late g and k giants located to the right of the rdl have @xmath82 . on the other side ( left ) of the rdl , fk comae - type stars show extreme rotation velocities , ( @xmath83 ) . the general explanation for their fast rotation assumes that these single g or k giants are undergoing ( or have just completed ) coalescence with a close binary companion ( e.g. webbink 1977 , bopp & stencel 1981 , livio & soker 1988 , welty and ramsey 1994 ) . other scenarios have also been proposed and they generally invoke tidal synchronization in a very close binary , normal evolution from a rapidly rotating upper main sequence star , or the dredge up of angular momentum ( e.g. simon & drake 1989 ) . but , as pointed out by rucinski ( 1990 ) , fk coma has a few times more ( @xmath84 ) angular momentum than a typical rapidly rotating main sequence star . therefore , some angular momentum must have been added ( accreted ) to the star . generally fk comae - like stars lack photospheric lithium detection . however in some cases , e.g. hd 36705 ( rucinski 1985 ) , hd 33798 ( fekel & marschall 1991 ) or 1e 1751 + 7046 ( ambruster et al . 1997 ) , a high lithium abundance has been observed ( see also @xmath6[lirich ] ) . these observations clearly support our accretion scenario since they combine two different outcomes of the accretion process : a high rotation rate and a high li abundance . however , different scenarii have also been proposed and among them the possibility that these stars were initially fast rotating a or f type stars that have rapidly evolved in the hertzsprung gap . the suggestion is then that they recently became giants and are now depleting lithium due to the deepening of the convective envelope ( fekel 1988 ) . it has also been suggested ( fekel & marschall 1991 ) that angular momentum and cno processed material can be dredged up from a rapidly rotating core as the convective zone deepens . in this scenario li can be produced and burnt through the chain he(@xmath48)be(@xmath85)li(@xmath86)he and , if on its way to the surface li is not entirely depleted , then we may account for temporarily high rotation together with enhanced lithium abundance . however , this hypothesis would imply that essentially every star arriving to the rgb should go through a phase of fast rotation and high li abundance . the paucity of such candidates poses a severe problem to this scenario and can be regarded as additional evidence against the rotostat hypothesis . to conclude this section , we feel that observations support the coalescence hypothesis for the explanation of fk comae - like stars and we stress that rapidly rotating stars may also have accreted angular momentum from the swallowing of a giant planet or a brown dwarf ( see also vitenbroek et al . the planet accretion scenario can also account for enhanced li abundances and in that respect , hd 33798 and hd 36705 are good candidates for such an event . finally , we would like to point out that fast rotation velocities , of the order of 15 - 40 , have also been observed in horizontal branch stars ( e.g. peterson et al . 1995 , cohen & mccarthy 1997 ) . notably , peterson et al . ( 1983 ) already mentioned that the deposition , during the red giant phase , of the orbital angular momentum of a planet of a few jupiter masses could account for the fast rotational velocities observed in horizontal branch stars . some agb stars also exhibit high rotation rates ( e.g. v hydrae ) . for these exceptional stars , the rapid rotation rate is thought to be due to spinning up by a companion that has been swallowed or is in a common envelope phase ( barnbaum et al . more generally , the deposition of angular momentum in the envelope of evolved stars by a companion ( even substellar ) is frequently invoked to explain the axisymmetrical morphology of bipolar nebulae ( e.g. iben & livio 1993 , livio 1997 ) . if the accretion of angular momentum can cause the star to rotate differentially then differential rotation and convection can generate a magnetic dynamo . the magnetic field can then confine the plasma in the stellar chromosphere and produce x - ray emission activity . to see if this scenario is plausible , let us first examine the observational situation . observations indicate the existence of an x - ray dividing line in the hr diagram ( xdl also called coronal dividing line ) which delimits giant stars with ( to the left ) and without coronal emission characteristic of high temperature gas ( @xmath87k ) . in the hr diagram , this line is approximatively vertical at spectral type @xmath88k3 and it spans the luminosity classes ii , iii and iv ( haisch et al . 1991 , 1992 ) . note that bright giants of luminosity class ii and ib turn out to be x - ray sources if exposed sufficiently deeply ( reimers et al . 1996 ) . on the right side of the xdl , the vast majority of the stars exhibit a very low x - ray emission ( at least below the actual detection threshold ) whereas a large spread of x - ray luminosities is observed for g and k giants of earlier spectral type than k3 ( strassmeier et al . 1994 , hnsch et al . the x - ray emission originates from the stellar corona and requires a strong magnetic field in order to confine the hot plasma , since the thermal velocity of the particles exceeds the escape velocity . changes in the magnetic field configuration ( due to changes in the dynamo ) are generally invoked to account for the sudden drop in x - ray emission at the xdl ( e.g. rosner et al . 1991 , 1995 ) but the precise reasons why the dynamo is modified remain unclear . recently hnsch and schrder ( 1996 ) presented a `` revised '' xdl that seems to follow the evolutionary track of a 1.2 star . if this is confirmed , then evolutionary effects may account for this demarcation line . low mass stars , on the right of the xdl , are generally slow rotators on the main sequence and consequently they sustain a lower magnetic activity . conversely , more massive stars ( @xmath89 ) have faster rotation rates on the main sequence and therefore they can develop an efficient dynamo and produce a higher x - ray emission . more recently , hnsch et al . ( 1998 ) looked for x - ray emission in the `` forbidden '' region beyond the xdl and they detected 4 intrinsically x - ray bright m - type giants : 15 tri , hr 5512 , 42 her and hr 7547 . no evidence for binarity exists yet for 15 tri and hr 5512 but even if the presence of a companion is confirmed , it can not easily explain the x - ray emission which remains quite unusual . in a rosat survey of a complete , volume limited sample of late type stars , hnsch et al . ( 1996 ) also pointed out the unusual behavior of @xmath90 sgr = hr 6859 , an apparently single , slowly rotating but strong x - ray emitting k3 iii giant . another puzzling case is that of hr 1362 ( strassmeier et al . 1990 ) . despite its low rotation rate this single g8 iv giant presents an x - ray emission that is more than one order of magnitude higher than that expected from the empirical rotation versus activity trend . the strong emission of hr 1362 led stepien ( 1993 ) to suggest that it might represent an evolved ap star that has preserved its strong magnetic field . on the other side ( left ) of the xdl , several unusually bright x - ray sources have also been detected . they are often referred to as @xmath91 ceti - like stars , based on the prototype @xmath91 ceti ( hnsch et al . 1996 , maggio et al . 1998 ) , which exhibits an extremely high x - ray luminosity and a slow rotation rate ( @xmath92 ) . these observations are puzzling since it is difficult to reconcile the high x - ray emission with the slow rotation rate . below we investigate the physical effects related to the accretion of a planet / brown dwarf in terms of the generation of magnetic field and x - ray emission . suppose that an initially slowly rotating star undergoes an accretion event . as the planet spirals in , it will deposit angular momentum in the envelope and finally dissipate close to the core . however , the resultant profile of angular velocity @xmath93 may not be uniform throughout the envelope but rather may be steeper close to the dissipation region , where the gravitational potential increases abruptly . to quantify this , at least approximatively , we follow livio and soker ( 1988 ) and estimate the parameter @xmath94 defined by @xmath95 where @xmath96 represents the decay timescale of the planet s orbit , and @xmath97 is the spin - up timescale of the envelope . the decay timescale is given by @xmath98 where @xmath99 the mass interior to @xmath72 and @xmath100 is the gravitational drag force that causes the spiralling - in . its expression is approximately given by @xmath101 where @xmath102 is a factor of order @xmath103 for supersonic flows ( e.g. shima et al . 1985 ) , @xmath104 is the accretion radius , @xmath105 the relative velocity between the planet and the envelope and @xmath106 the local density . finally , the spin - up timescale @xmath107 can be approximated by @xmath108 where @xmath109 is the moment of inertia of the envelope interior to radius @xmath72 . in the computations for our presented models , we assumed that the relative velocity @xmath105 is keplerian and impose @xmath110 as in livio & soker ( 1988 ) . figure [ beta ] depicts the profiles of @xmath94 for the 3 initial models . although these profiles are crude estimates , they suggest that a strong differential rotation is likely to take place at the base of the convective envelope , in the regions where @xmath111 . the shear can then sustain a dynamo action and lead to a high x - ray emission ( e.g. regos & tout 1995 ) . furthermore , the profiles of @xmath94 indicate that for less evolved configurations the outer layers are not spun - up , and therefore that it is possible to obtain x - ray emission from apparently slowly rotating stars as observed among @xmath91 ceti - like stars . it is difficult to estimate for how long the slow rotation will be maintained at the surface of the star since such a study would require to take into account the angular momentum transport in the convective envelope coupled with the presence of a strong magnetic field . additionally the outcome depends on the evolutionary status of the star . we expect however that the spin up the envelope will take longer in younger and less centrally condensed stars for which @xmath112 in a larger fraction of the envelope . in some configurations magnetic braking may not allow a significant spin - up of the outer layers . some indications that differential rotation may persist in the interior of stars are suggested by observations . indeed , in the hr diagram , the rotational ( @xmath6[rotation ] ) and x - ray dividing lines are shifted with respect to one another . thus , stars located between these two lines present both a magnetic activity and a slow rotation rate . if this shift between the dividing lines corresponds effectively to an evolutionary sequence , this indicates either that the rotational braking timescale is shorter than the decay timescale associated with the dynamo activity or , alternatively , that differential rotation is still maintained in the stellar interior and sustains a dynamo activity . to summarize this section , the deposition of angular momentum by a spiralling - in planet / brown dwarf can produce a strong differential rotation at the base of the convective envelope . we then suspect that in this region a magnetic field can be generated via an @xmath113 dynamo and that this may ultimately result in an enhanced x - ray activity . note that the magnetic field can also be used to brake the star . finally , we would like to mention that the shear may trigger some instabilities that could give rise to extra mixing processes ( see @xmath6[lirich ] ) . the increase in the mass loss rate during the accretion phase can lead to the formation of a shell around the star , and that shell can be expected to emit in the infrared ( ir ) . more quantitatively , the mass loss rate can be increased by a factor of 2 to @xmath114 and a mass of the order @xmath115 can be ejected . these numbers depend however on the evolutionary status of the star , the duration of the accretion event and the accretion rate . figure [ mloss ] demonstrates the very strong dependence of the mass loss rate on the accretion rate and on the initial model . the higher the accretion rate , the larger is the increase in the mass loss rate . similarly , the more evolved the star , i.e. the lower its surface gravity , the easier it is to remove material from its surface . figure [ mloss ] also indicates that the rate of increase in is faster for older stars . this is an illustration of the thermal response of the star since is inversely proportional to the kelvin helmholtz timescale ( @xmath116 ) . finally , we would like to emphasize that in the high accretion rate cases ( cases b2 and c2 ) , even a relatively small planet of a few jupiter masses can increase the mass loss rate by about 2 orders of magnitude in a few hundred years ! it is worth noting that in the estimates of the mass loss rate , the effects of rotation have been neglected . the deposition of the orbital angular momentum into the envelope can lower the effective surface gravity and facilitate mass ejection ; it can also influence of geometry of the ejection . in particular , if the rotational velocity of the star exceeds some fraction of the break - up speed , then the ejected material could form an equatorially compressed outflow ( e.g. bjorkman & cassinelli 1993 , owocki et al . 1994 , livio 1997 ) . the peculiar circumstellar environment of u equulei could be consistent with such a scenario ( barnbaum et al . 1996 ) . the ir emission associated with giant stars is not easy to explain ( zuckerman et al . 1995 ) , while in main sequence stars ( vega - like stars ) the emission most likely arises from the dust associated with remnants of the star formation process or with small bodies like comets or planetesimals ( e.g. aumann et al . 1984 , telesco & knacke 1991 , backman & paresce 1993 ) . in an extensive survey of luminosity class iii stars , plets et al . ( 1997 ) report that @xmath117% of the giants in their sample have ir excesses . they conclude that the ir emission is unlikely to be due to a present - day mass loss episode because , as they argue , @xmath118 in a standard evolution , mass loss is predominantly efficient during a he shell flash so one would expect the stars to be more clustered at higher luminosities in the hr diagram and @xmath119 the dust emission is too cold for a recent ejection . they thus attribute the ir excess to a reheated dusty disk , as in vega - like stars . instead , we propose that the ejection of a shell resulting from an accretion event can account for the ir emission . indeed , this scenario does not imply a clustering of stars at the tip of the rgb since the accretion event can , in principle , take place any time once the star leaves the main sequence . early engulfing could also account for the cold dust emission . moreover , the ir excess is often too large to be attributed to a vega - like phenomenon , since during the main sequence phase , most of the circumstellar debris will probably be evaporated ( zuckerman et al . . an important test of our scenario can be provided by measuring the li abundance for example , since this element should , at some level , accompany the accretion process ( see @xmath6[lirich ] ) . + the detection of a detached shell around the j - type carbon star y canum venaticorum ( izumiura et al . 1996 ) is also interesting since this star is probably in an early evolutionary stage , on the rgb or on the he core - burning stage . the mass loss rate at the time of shell formation and the mass of the shell are estimated to be in the range @xmath120and @xmath121 , respectively . the present mass loss rate estimated from co observations ( e.g. olofsson et al . 1993 ) is about @xmath122 , two orders of magnitude smaller than the past value ! therefore , if the rgb status of y cvn is confirmed , the existence of a massive shell around this star may be a signature of an accretion event , since no other straightforward model can account for these observations . in this context , the massive shell and high mass loss rate at the time of ejection would suggest a very high accretion rate ( @xmath123 ) . shells have also been observed around many agb stars ( e.g. olofsson et al . 1992 , 1996 , bujarrabal & cernicharo 1994 ) , but such structures can result from episodic high mass - loss events triggered by helium shell flashes ( e.g. olofsson et al . 1990 , izumiura et al . it is therefore difficult to identify the signature of an accretion event , although the presence of a companion is frequently invoked for the shaping of planetary nebulae ( e.g. soker 1997 , livio 1997 ) . finally , we should mention that in model c0 the star experiences a series of thermal pulses before reaching the horizontal branch . from , the simulations we can estimate that during every interpulse , a mass of the order of @xmath124 is ejected from the surface of the giant . these episodic ejections would generate distinct shells , separated from each other by a few thousand years . however , the probability that a massive planet would be engulfed at the short - lived tip of the giant branch is rather small . the distributions of stars along the horizontal branch ( hb ) of globular clusters varies from one cluster to another . in some clusters , the population of hb stars exhibits an extended blue tail while others contain only a red hb . it was first recognized by sandage and whitney ( 1967 ) that the metallicity is the first parameter influencing the morphology of the hb . generally , more metal - rich clusters have red hbs , while metal - poor clusters have bluer hbs . but , for globular clusters of the same metallicity , the hb morphology can vary widely . thus some `` second parameter '' other than metallicity must be affecting the hb morphology . among potential second parameters , it has been shown that age ( e.g. lee et al . 1994 ) , mass loss ( e.g. dcruz et al . 1996 ) , he mixing ( sweigart 1997 ) , cno enhancement ( dorman et al . 1991 ) , core rotation , or structural properties of the globular clusters such as concentration and density can also affect the hb morphology . nevertheless the discovery of gaps in the blue tail of the hb ( e.g. walker 1992 , catelan et al . 1997 , ferraro et al . 1997 ) is still considered puzzling . recently , soker ( 1998b ) suggested that planetary systems can influence the hb morphologies . he proposed that the interaction of a rgb star with a giant planet / brown dwarf can affect the mass loss rate and thus the subsequent location of the star on the hb . by depositing angular momentum and energy into the envelope , additional mass can be removed from the star during the rgb . the `` peeled off '' star will then have a higher effective temperature and will appear bluer in the hr diagram ( e.g. castellani & castellani 1993 , dorman et al . in particular , soker ( 1998b ) argued that , depending whether the planet survives the spiralling - in , collides with the stellar core or evaporates in the envelope , one can account for the 3 gaps observed by sosin et al . ( 1997 ) in the blue tail of ngc 2808 . our study introduces a quantitative element into the scenario proposed by soker . our computations indeed indicate that a significant increase in the mass loss rate accompanies the accretion process , especially if the star is evolved and the accretion rate is high . the simulations also show that , when the star has relaxed , the mass loss rate remains higher than in a standard evolution . we estimated that when a massive planet is accreted ( not a brown dwarf ) , the mass loss rate is enhanced by a factor of @xmath125 after the termination of the accretion process . this is equivalent to setting reimer s parameter @xmath126 to a value close to @xmath127 ( instead of 0.5 ) in a standard evolution . if we now refer to the computations of castellani & castellani ( 1993 ) for such values for @xmath126 , we conclude that stars accreting a planet at the bottom of the rgb , will actually be extremely blue on the horizontal branch . on the other hand , if the accretion occurs later , the effects are difficult to quantify because of the strong sensitivity of our computations to the accretion rate and the planet mass . the effects can be moderate , as in cases c2 or b2 , or dramatic , as in cases b3 or c4 . the possibility that a planet accretion would modify the evolutionary age of the star and thus affect the morphology of the hb is improbable . indeed our initial models are relatively old ( @xmath128myr ) and the duration of the red giant phase is relatively short compared to its age . also , the accretion of a planet does not modify substantially the mass of the star therefore we do nt expect the stellar evolutionary timescales to be significantly modified . sweigart ( 1997 ) suggested that an extra mixing driven by internal rotation could transport helium from the hbs into the convective envelope . this would increase the rgb tip luminosity and hence the mass loss , leading to the formation of blue hb stars . sweigart and catelan ( 1998 ) also showed that _ only _ non - canonical mechanisms invoking deep helium mixing , internal rotation or a high initial helium abundance can reproduce the distribution of the blue hb stars in ngc 6388 and ngc 6441 . however , these authors did not specify the source of the angular momentum required for the rotation scenario nor how the deep he - mixing process is activated . the accretion scenario can provide some support to these non - canonical processes , since it both offers a natural source of angular momentum for the core and it can be responsible for the generation of strong instabilities as the planet collides with the core ( @xmath6[xray ] ) . note also that some rapid rotators are observed in the blue side of the hb ( e.g. harris & mcclure 1985 ) . concerning the gaps , recent observations ( ferraro et al . 1998 ) indicate that they are not randomly distributed along the hb but rather located at similar positions in the color - magnitude diagram . aside from the star - planet interaction hypothesis ( soker 1998b ) , it has been suggested that some poorly specified physical processes must operate during the rgb to increase the mass loss rate ( ferraro et al . 1998 ) or that gaps are a consequence of a bimodal or multimodal mass distribution of the globular clusters ( e.g. catelan et al . 1998 ) . let us therefore return to the planet accretion scenario and analyze the implications of this assumption . we have seen that depending on the evolutionary status of the star at the time of engulfing ( i.e. on the initial planet separation ) , different amounts of mass can be lost and thus different positions in the hb can be reached . assuming for the moment that planets form preferentially around solar - type stars ( as it has been observed so far ) , the similarity in the positions of the gaps would suggest that planets orbit at preferred distances to the star . a theory suggesting that this is indeed the case exists ( notalle 1996 , notalle et al . 1997 and references therein ) . if this is true , planets would be swallowed only when the solar - type star reaches a certain radius ( or equivalently an evolutionary status ) , determined approximatively by the orbital separation of the planet . in this hypothesis , the accretion process would modify selectively the mass loss rate and thus the location of the star on the hb . + however , the question of why the star - planet interaction process acts only sporadically and does not appear as a general phenomenon affecting the morphology of all the clusters would remain puzzling . to attempt to answer this question , one needs a better understanding of the processes involved in planetary formation . for example , metallicity and other globular cluster properties such as density may be important factors that can influence the formation of planets and they are still poorly understood . to conclude this section , the accretion of planets by giant stars can potentially influence the mass loss rate and thus modify the position of the star on the hb . however , it is not clear at present that the gaps observed in the blue tail of some globular clusters are really the signature of such accretion events . the detection of lithium in the spectra of g and k giants ( e.g. wallerstein & sneden 1982 , brown et al . 1989 , gratton & dantona 1989 , pilachowski et al . 1990 , castilho et al . 1995 , de la reza & da silva 1995 , castilho et al . 1998 ) poses a problem concerning the origin of this element in these stars since , in standard evolution , lithium is entirely depleted during the first dredge up . several mechanisms have been suggested to explain the li enrichment and we refer to gratton and dantona ( 1989 ) for a more complete discussion . briefly , it has been proposed for example that li could be preserved in the atmosphere of giant stars , but the low c / c ratio observed in these stars ( e.g. pilachowski et al . 1990 , de la reza et al . 1995 ) indicates that deep mixing ( and thus li burning ) has occurred . this represents therefore a failure of this scenario . gratton and dantona ( 1989 ) considered the possibility that the giant s envelope could be contaminated by a nova exploding companion . however , the non detection of a white dwarf in the vicinity of li - rich giants ( e.g. de la reza & da silva 1995 ) weakens this hypothesis . the possibility that li could be produced by the cameron - fowler mechanism ( 1971 ) has also been investigated . this mechanism , which consists of the rapid dredge up of be and li production through the be(@xmath129)li reaction , is known to operate in agb stars of 4 - 6 for which the convective timescale is short compared to the be destruction timescale ( sackmann & boothroyd 1992 ) . unfortunately , this process is inefficient in low luminosity stars , unless one invokes some extra mixing processes ( we will return to this point later ) . finally , and we will develop this point below , several authors ( e.g. alexander 1967 , brown et al . 1989 , gratton and dantona 1989 ) proposed that the engulfing of a planet or a brown dwarf could be responsible for the high li abundance . based on iras observations of g and k giants , de la reza et al . ( 1996 , 1997 ) concluded that _ almost all _ the li rich g and k giants have an ir excess compatible with the presence of a circumstellar shell . to account for these observations , they constructed a model in which li is produced inside the star by an extra mixing process , the `` cool bottom burning '' ( wasserburg et al . 1995 ) . this process allows the fast transport of be from the hbs into the convective envelope where be then decays to produce li . furthermore , they assumed that the mechanism for li enrichment is accompanied by a sudden increase in the mass loss rate and thereby the ejection of a shell . following the evolution of a thin expanding circumstellar shell , they could reasonably account for the ir emission and li abundance distribution of g and k giants in a color - color diagram . however , de la reza et al . remained relatively vague concerning the mechanism that is responsible for mass loss ; they assumed that it is a consequence of the li enrichment process . instead , we propose here that the accretion of a planet or a brown dwarf can naturally and consistently explain all of the observed features . we describe our scenario in the next paragraph and illustrate it in fig . [ dela ] which is an adaptation of fig . 2 of de la reza et al . ( 1996 ) . figure [ dela ] represents a color - color diagram based on the iras density fluxes at 12 , 25 and 60 @xmath130 m . in this diagram , de la reza et al . ( 1996 , hereafter ddd ) identified 3 regions , populated respectively by normal k giants with no ir excess ( region @xmath131 ) , li rich giants with a large ir excess ( region @xmath132 ) , and relatively li rich giants with a moderate ir excess ( region @xmath133 ) . when the accretion process starts , li is progressively deposited in the stellar envelope and the mass loss rate increases . rapidly , a shell is ejected and the giant moves quickly from region i to region ii . in their shell model , ddd found that several hundred years are required to reach region @xmath132 . this timescale gives an estimate of the duration of the accretion process . then , li starts to be depleted in the envelope while the shell expands and cools down . the 25@xmath130 m flux decreases and the star now settles in region @xmath133 . ddd estimated that it takes approximatively @xmath134yr to move from region @xmath132 to @xmath133 . finally , in about @xmath135yr , the circumstellar shell dissipates , li is entirely depleted and the star returns to region @xmath131 as a normal k giant . assuming a wind velocity of 2 and a mass loss rate of @xmath136 , ddd found that @xmath137 was ejected in the shell , in agreement with our models . in a second publication , de la reza et al . ( 1997 ) completed their study by presenting new observations . they showed that li - rich and li - poor stars are mixed in the color - color diagram . they interpreted these observations as being due to differences in the li destruction timescales . in our model , this is explained naturally if the amount of li deposited into the star , or equivalently the mass of the planet , is different from one star to another . the mechanism that leads to the li enrichment must also account for its rapid depletion , which takes place in @xmath78yr . in our computations without diffusion , li is maintained at the surface of the star but charbonnel ( 1995 ) showed that this element can be rapidly depleted in red giants due to non - standard mixing processes . in particular , the model developed by zahn ( 1992 ) showed that the efficiency of the mixing is mainly determined by the loss of angular momentum via a stellar wind and , even in the absence of a high mass loss rate , the process takes place where there is a strong gradient in @xmath69 . he also pointed out that additional mixing is expected in shell burning regions . since all of these conditions are satisfied in the context of the accretion process ( increase in the mass loss rate , deposition of angular momentum and a strong gradient of @xmath69 in the dissipation region as described in @xmath6[xray ] ) and as a part of the fact that the star is on the rgb ( the presence of the hbs ) , we have every reason to believe that the mechanism proposed by zahn ( 1992 ) , and that successfully explains some peculiar anomalies in red giant stars ( charbonnel 1995 ) , could be responsible for the fast li depletion following a short accretion phase . moreover , observations of the c / c ratio in some li rich giants ( da silva & de la reza 1995 ) indicate that some extra mixing is operating since the carbon isotope ratio in these stars is lower than the standard value . + our scenario does encounter some difficulties in explaining the very high li abundances [ log@xmath138(li)@xmath139 found in some giants ( e.g. gratton & dantona 1989 , de la reza & da silva 1995 , fekel et al . 1996 ) . this would require the deposition of a huge amount of lithium that can only be achieved either by the accretion of numerous planets such that @xmath140 or if the li abundance in the brown dwarf is extremely high ( @xmath141 for @xmath142 ) . the accretion scenario does not provide a straightforward explanation but we can imagine that when the planet reaches the central region , the dissipation process may generate some instabilities ( due to the shear ) that could , in turn , trigger the `` cool bottom burning '' invoked by de la reza et al . ( 1996 , 1997 ) . the accretion process would then lead to some li production ( as in more massive agb stars ) . de medeiros et al . ( 1996a ) reported that the majority of li k giants are normal rotators ( @xmath143 ) although some exceptions exist [ e.g. hd 233517 ( fekel et al . 1996 ) , hd 9746 , hd 25893 , hd 31993 , hd 33798 , hd 203251 ( fekel & balachandran 1993 ) , 1e 1751 + 7046 ( ambruster et al . however , as pointed out in @xmath6[rotation ] , high angular velocities ( @xmath37% percent of the keplerian velocity ) can only be achieved if a sufficiently massive planet ( @xmath144 ) is accreted . for a typical 1.0 rgb star , with a radius @xmath145 , the deposition of the orbital angular momentum of a 5 @xmath146 planet will `` only '' increase the star s rotational velocity to @xmath147 [ eq . ( [ omega ] ) ] . consequently , as long as the mass of the engulfed body is low ( a planet rather than a brown dwarf ) , the observed angular momentum of li rich giants is compatible with our scenario . we should also note that due to the increased mass loss , magnetic braking could also be operating ( e.g. leonard & livio 1995 ) . finally , let us point out that _ all the li rich giants are single stars _ , hence according to the prevailing common wisdom , they could possess planetary systems . to summarize this section , we propose that the accretion of a planet or a brown dwarf is responsible for both the ir excess and the high li abundance present in almost all single li rich g and k giants . in this paper , we investigated the effects of the accretion of a planet or a brown dwarf by a solar type star located on the red giant branch . our computations show that the deposition of mass ( potential energy ) produces a substantial expansion of the envelope and , for higher accretion rates , nuclear burning at the base of the convective envelope can take place . the simulations show that the results are a sensitive function of the accretion rate and the structure of the star . generally , the higher the accretion rate and/or the more evolved the star , the larger is the expansion . after the end of the accretion process , the star relaxes and resumes a standard evolution . however , in one simulation ( case c0 ) , the star avoided the he flash but instead developed thermal instabilities . for this model , after a phase of thermal pulses , he burning was finally ignited at the center of the star . we discussed the possible observational signatures of the accretion of a planet / brown dwarf . in particular ( 1 ) we showed that the rotational velocity of the star can be substantially increased , provided that the accreted planet / brown dwarf is sufficiently massive . in this respect the accretion of a brown dwarf could account for the fast rotation of fk comae stars and more generally , the deposition of angular momentum via this process can explain unusually fast rotation rates . ( 2 ) investigating the x - ray properties of giant stars , we showed that a strong differential rotation may take place at the bottom of the convective envelope . the presence of shear in this region can turn - on the dynamo activity and this may ultimately generate x - ray emission . we showed that under some conditions the surface layers of the star could rotate slowly while a rapid rotation is maintained in the core . ( 3 ) we also showed that the accretion scenario can lead to the ejection of shells whose mass is in the range @xmath115 . the ir excesses observed in @xmath1488% of luminosity class iii stars could in part be due to this process . ( 4 ) we argued that the increase in the mass loss rate can also have indirect effects on the morphology of the horizontal branch and in this respect , the accretion process is a potential second parameter candidate . ( 5 ) we showed that the deposition of planet material in the convective envelope can efficiently increase the li surface abundance and more generally enhance the stellar metallicity of giant stars . ( 6 ) finally , on a related topic , we proposed that the accretion of a planet / brown dwarf can explain consistently both the ir excess and the high li abundance observed only in single red giant stars . in terms of physical processes , we suggested that planet / brown dwarf accretion may provide the triggering for a variety of processes . for example , the disruption of the planet and the associated instabilities could increase the he abundance in the giant s envelope and thus affect the morphology of the hb as proposed by sweigart ( 1997 , @xmath6[morpho ] ) . they could also be responsible for the production of lithium through the mechanism proposed by wasserburg et al . ( 1995 , @xmath6[lirich ] ) . finally , the rapid contraction of the star at the end of the accretion process could generate some pulsations . observations indicate that several of potential different signatures of an accretion event , e.g. a fast rotation rate , a significant x - ray emission , a high li abundance or a large ir excess , do not always occur simultaneously . as we have shown , this is not surprising . first , depending on the mass of the planet , the outcomes could be different . for example , a small planet that evaporates in the envelope before reaching its base will perturb the structure only slightly . in such a case , no shell will be ejected but the deposition of angular momentum and matter in the surface layers could still produce a relatively rapid rotation and a high lithium abundance , as observed in some chromospherically active giants ( fekel & balachandran 1993 ) . secondly , the timescales associated with the various phenomena ( ir excess , x - ray activity , rotational braking , particle transport ) can be different , and thus , not all the signatures will be observed at the same time . for example , if the spin - down timescale is longer than the timescale associated with the dissipation of the ejected shell , one can account for the absence of ir excess in some fast rotators . thirdly , the consequences of the accretion process depend on the evolutionary status of the star . giant stars located at the bottom of the rgb have much deeper convective envelopes than near the tip of the giant branch . thus , if we consider for example the deposition of li in the envelope , it is more difficult to obtain high li abundances in younger red giants ( it is also be more difficult to eject a shell ) . finally , our simulations show that the results are sensitive to the physics of the dissipation process , and more specifically , to the accretion rate and the location of the accretion process . at present , both these physical parameters are rather uncertain , and a better understanding of the dissipation process is definitively required . as a last point , we can attempt to use the expected observational signatures to infer some estimate for the frequency of planets around solar - type stars . in a search for lithium in 644 g and k giants , brown et al . ( 1989 ) found that 2% in the stars of their sample have log@xmath138(li)@xmath149 and 4% have log@xmath138(li)@xmath150 . if we adopt for our definition of li rich giants the value log@xmath138(li)@xmath151 , we find that @xmath117% of the stars surveyed by brown et al . belong to this group . in a more recent survey , castilho et al . ( 1998 ) report the detection of 5 li rich and 6 moderately li rich giants among 164 stars , indicating a proportion of li rich giants of @xmath152% . the study of plets et al . ( 1997 ) indicates that 8% of luminosity class iii giants have unexpected ir excess . finally , based on rotational velocity measurements of @xmath80 g and k giants , de medeiros et al . ( 1996a ) found that @xmath153 of late g and k giants located beyond the rotational dividing line are fast rotators with @xmath154 . thus , if the ir excess and/or high li abundance and/or unusually high rotational velocity among g and k giants result from the accretion of a planet / brown dwarf , the above numbers suggest that at least 4 - 8% of the single , solar - type stars must have a low mass companion , either a brown dwarf or a planet . if we were to take seriously the frequency of planets inferred by soker ( 1997 ) to account for the shaping of planetary nebulae ( @xmath155 of all progenitors of masses @xmath156 ) , our deduced frequency would indicate that either a large fraction of the planets do not interact during the red giant phase or , more probably , that the observational signatures of the accretion event are very short - lived compared to stellar evolutionary timescales . indeed , because of the large radii reached by solar type stars during the rgb , some planets will most likely be engulfed during the early giant phase rather than on the agb ( although jupiter will not be engulfed by our own sun ) . on the other hand , if we compare the timescales associated with both the li and ir excesses ( @xmath157yr ) to the red giant lifetime ( between @xmath158 and @xmath159yr for giants with masses between 1.0 and 2.5 ) , we conclude that the probability of detecting an accretion event is such that we are likely to miss most candidates . ls acknowledges support from the director s discretionary research fund at stsci and thanks the stsci for its hospitality . this work has been supported in part by nasa grants nagw-2678 , g005.52200 and g005.44000 . the computations presented in this paper were performed at the `` centre intensif de calcul de lobservatoire de grenoble '' .

we compare the distribution of diffuse intracluster light detected in the virgo cluster via broadband imaging with that inferred from searches for intracluster planetary nebulae ( ipne ) . we find a rough correspondence on large scales ( 100 kpc ) between the two , but with very large scatter ( 1.3 mag / arcsec ) . on smaller scales ( 1 10 kpc ) , the presence or absence of correlation is clearly dependent on the underlying surface brightness . on these scales , we find a correlation in regions of higher surface brightness ( .5ex27 ) which are dominated by the halos of large galaxies such as m87 , m86 , and m84 . in those cases , we are likely tracing pne associated with galaxies rather than true ipne . in true intracluster fields , at lower surface brightness , the correlation between luminosity and ipn candidates is much weaker . while a correlation between broadband light and ipne is expected based on stellar populations , a variety of statistical , physical , and methodological effects can act to wash out this correlation and explain the lack of a strong correlation at lower surface brightness found here . a significant complication comes from the stochastic nature of the pn population combined with contamination of the ipn catalogs due to photometric errors and background emission line objects . if we attribute the lack of a strong observed correlation solely to the effects of contamination , our monte carlo analysis shows that our results are mostly consistent with a `` ipne - follow - light '' model if the ipn catalogs are contaminated by 40% . further complications arise from the line - of - sight depth of virgo and uncertainty in the stellar populations of the icl , both of which may contribute to the slight systematic differences seen between the ipn - inferred surface brightnesses and those derived from our deep surface photometry .

0906.4703

the diffuse intracluster light ( icl ) that permeates clusters of galaxies has become a valuable tool for studying galaxy cluster evolution . this icl arises from a variety of sources , and is likely dominated by tidal stripping of galaxies during interactions both within the cluster and within infalling groups ( merritt 1984 ; moore et al . 1996 ; rudick , mihos , & mcbride 2006 ; murante et al . 2008 ) , as well as stripping by the cluster potential itself for galaxies passing near the cluster core ( byrd & valtonen 1990 ; gnedin 2003 ) . recent models of galaxy cluster accretion and evolution show that @xmath2% of the luminosity of clusters may be found in the icl ( e.g. , willman et al . 2004 ; murante et al . 2004 ; sommer - larsen et al . 2005 ; rudick et al . 2006 ; purcell et al . 2007 ) . deep imaging of galaxy clusters has begun to reveal this icl component ( bernstein et al . 1995 ; feldmeier et al . 2002 , 2004 ; mihos et al . 2005 ; zibetti et al . 2005 ; gonzalez et al . 2005 ) , with some suggestion of a correlation between cluster mass and icl fraction ( ciardullo et al . 2004 ; krick & bernstein 2007 ) . of particular interest is the dynamical information contained within the icl . as material is stripped from its host galaxies , long tidal streams are formed ( gregg & west 1998 ; calcneo - roldn et al . 2000 ; feldmeier et al . 2004 ; mihos et al . 2005 ) ; these coherent streams can later be destroyed during a subsequent cluster accretion event , contributing to a smooth diffuse icl component ( e.g. , rudick et al . 2006 , 2008 ) . the morphology of the diffuse icl on cluster scales thus contains information on the recent accretion history of galaxy clusters . on smaller scales , the icl can trace the dynamical history of individual galaxies , as extended tidal debris can highlight past interactions , either with the cluster potential or with other galaxies ( e.g. , schweizer 1980 ; malin 1994 ; weil et al . 1997 ; katsiyannis et al . 1998 ; feldmeier et al . 2004 ; janowiecki et al . in preparation ) . along with diffuse light , the icl can be studied through discrete tracers such as globular clusters ( west et al . 1995 ; williams et al . 2007a ) , red giants ( ferguson , tanvir , & von hippel 1998 ; durrell et al . 2002 ; williams et al . 2007b ) , and planetary nebulae ( arnaboldi et al . 1996 ; feldmeier et al . 1998 , 2004 ; aguerri et al . intracluster planetary nebulae ( ipne ) are of particular interest here , since they are believed to trace the stellar luminosity of galaxies and as emission line objects they permit study of the kinematics of the icl . ipn kinematics have been studied in both the virgo ( arnaboldi et al . 2004 ) and coma ( gerhard et al . 2005 ) clusters , and yield interesting constraints on kinematic substructure . larger samples of ipn velocities would yield information on the degree of relaxation of the icl , its kinematic connection to the cluster galaxies , and also provide further constraints on the cluster mass distribution ( gerhard et al . 2007 ; sommer - larsen et al . 2005 ; willman et al . however , existing surveys for ipne have largely been done `` blind , '' without any of the foreknowledge of the underlying properties of the diffuse light provided by recent broadband imaging surveys ( e.g. , mihos et al . 2005 ) . a strong correlation between broadband light and ipn density would then open up the capability of using deep imaging to maximize the efficiency of followup ipn searches . if well determined , the connection between planetary nebulae and broadband luminosity in the icl also can provide constraints on its stellar populations . the ratio @xmath3 depends on the stellar populations of the icl ( see , e.g. , buzzoni , arnaboldi , & corradi 2006 ) , and in galaxies shows a correlation with galaxy type ( peimbert 1990 ; ciardullo et al . 2005 ) , wherein massive metal - rich ellipticals have a much lower value of @xmath4 than do low - luminosity , metal - poor galaxies . measurements of @xmath4 can test various models for the formation of the icl , as icl formed from the mergers of massive galaxies during the formation of a cluster ( murante et al . 2008 ) would show lower values of @xmath4 than icl formed from the ongoing disruption of low luminosity dwarfs orbiting in the cluster potential ( moore et al . 1996 ) . the data now exist to do this detailed comparison between the icl measured by broad band light and ipne in the virgo cluster . we ( mihos et al . 2005 ) have conducted a deep wide - field imaging survey of the virgo core down to a @xmath5 limiting surface brightness of @xmath6 , while narrowband surveys ( feldmeier et al . 2004 ; hereafter f04 ; aguerri et al . 2005 ; hereafter a05 ) are discovering significant numbers of ipn candidates in virgo . many of these ipn fields overlap with our imaging fields , allowing us to do direct comparison between the different icl tracers . these tracers have very different systematic uncertainties : broad band surface photometry is sensitive to sky subtraction , flat field quality , and scattered light control , while the ipn measurements rely on statistics of faint point source detection and contamination from background emission line objects . while the comparison will be complicated by these different systematic uncertainties , it also gives us a new opportunity to assess the consistency of the two methods in detecting the diffuse intracluster light in galaxy clusters . the optical imaging comes from our ongoing survey for icl in the virgo cluster using the 0.6 m burrell schmidt at kitt peak . figure 1 shows our image of the central 2.5 square degrees of the virgo core ( see mihos et al . 2005 for details ) . this image has been constructed from co - adding 72 individual 900s images , each dithered by up to half a degree . the dithering pattern means that the total exposure time across the field varies . the median number of exposures per pixel in the final mosaic is 31 . the exposure time is highest in the inner 17% of the image , where there are at least 40 exposures per pixel ; 73% of the pixels have at least 20 exposures , and the minimum exposure per pixel ( which defines the edges of the mosaic ) is set at 5 . each individual image was flattened using a night sky flat built from 127 offset sky pointings . the sky pointings were taken bracketing the virgo images in both time and position , to minimize flat fielding variations due to telescope flexure and illumination . all images were taken in washington m and transformed to johnson v assuming a color of b - v=1 typical of the old stellar populations thought to represent intracluster light . the uncertainty in the icl surface brightness introduced by the photometric transform is small for example , adopting an extremely and unrealistically blue icl color of b - v=0 changes the inferred surface brightness by only 0.18 magnitudes . sky subtraction is problematic when dealing with an image which subtends only a small fraction of the cluster area . to obtain an reliable estimate of sky , we first need to subtract the wings of bright stars from the image . we use long exposures of @xmath4 leo to construct the stellar point spread function out to 30 , and then subtract the extended wings of stars out a radius where they contribute 0.1 adu ( @xmath7 ) to the local sky . we also mask stars completely inside the radius where the scaled stellar psf is brighter than 3 adu . iraf s objmask task is then run on the star subtracted image to mask all _ extended _ sources 2.5@xmath8 above a locally - measured sky value . the image is then spatially rebinned into `` superpixels '' which are 32x32 pixels in size ; each superpixel is given an intensity equal to the mode of its individual pixels . regions are then selected to represent `` true sky '' these regions are chosen to be far from bright galaxies , and typically make up @xmath0 15% of each individual image . for each image , an absolute sky value is then calculated from the mode of the superpixel intensities in these pure sky regions and subtracted off . depending on varying airmass , time , and hour angle , these sky levels varied from 1100 - 1500 adu ( @xmath9 ) . to make the final mosaic , we then register each image and use the `` true sky '' regions to simultaneously and iteratively fit residual sky planes to the individual images , constrained to minimize the frame - to - frame deviations in overlapping regions of images . after registering and medianing the individual images into the final mosaic , one additional residual sky plane is fit and subtracted from the mosaic . it is important to note that this process will over - subtract any icl which is smoothly distributed on scales _ larger _ than the areal coverage of the image , approximately 400 kpc on a side . to construct the error model for the dataset , we follow the procedure outlined by morrison et al . ( 1997 ) , described in more detail in the appendix . in brief , we expect a random error of approximately 2 adu per pixel mag / arcsec@xmath1 . ] from a combination of readnoise , small - scale flat fielding errors , and sky photon noise . this uncertainty should become negligible when measuring the diffuse icl over large scales ( @xmath1030 ) , where many pixels contribute to the measurement . on these larger scales , however , the error model is dominated by uncertainty in the sky background and large scale flat - fielding , which we estimate to be approximately 0.5 adu by examining the residual variance in boxes @xmath11 on a side in regions lacking galaxies or diffuse light . this places the @xmath5 limiting depth to our surface photometry at @xmath12 . cccccccccc 3 & 27.0 & 20.0 & 46 & 26.5 & f04 & 26.4 & 26.7 & 27.0 & 35.3 + 4 & 26.6 & 16.6 & 4 & 27.4 & f04 & 27.6 & 27.7 & 27.8 & 34.2 + 7 & 26.8 & 18.3 & 32 & 28.4 & f04 & 27.2 & 27.3 & 27.5 & 45.1 + core & 27.2 & 22.0 & 77 & 28.6 & a05 & 27.8 & 27.9 & 28.1 & 35.2 + fcj & 27.0 & 20.0 & 20 & 27.4 & a05 & 26.4 & 26.7 & 27.0 & 35.3 + lpc & 27.5 & 25.2 & 14 & 31.7 & a05 & 28.1 & 28.2 & 28.4 & 16.6 + sub & 28.1 & 33.1 & 36 & 28.5 & a05 & 26.6 & 26.9 & 27.3 & 37.7 + we compare the broadband diffuse light measured in our virgo imaging to that inferred from compilations of recent virgo ipn surveys by f04 and a05 . catalogs for the latter datasets were kindly provided by ortwin gerhard and collaborators . these surveys were conducted on a variety of telescopes , have different areal coverage and selection effects , and go to different limiting magnitudes ; combining these catalogs into a homogeneous dataset is problematic . instead , our goal is to compare the ipn - inferred icl surface brightnesses averaged over each field ( 3 ) , and then examine the correlation between individual ipn candidates and broadband light within individual fields ( 4 ) . one major issue is that these catalogs are catalogs of ipn _ candidates _ detected via a two filter method . ipn candidates are identified as point sources that are present in a narrowband `` on - band '' image that includes the redshifted @xmath13 5007 [ oiii ] line , but disappears in a spectrally - adjacent `` off - band '' image that excludes the line . unfortunately , such a search technique can also include contaminating objects , and the contamination in the ipn catalogs can be significant . this contamination comes in two forms . the first is background emission line objects , most notably high - redshift galaxies . at a redshift of @xmath14 , ly@xmath4 falls into the narrowband [ oiii ] filters used to detect ipne , such that a fraction of the ipn candidates are in fact background ly@xmath4 emitting galaxies . estimates of this contamination fraction from narrowband imaging of blank fields ( ciardullo et al . 2002 ) put the fraction of background emitters at @xmath0 20% . this fraction can vary significantly from field to field , however , depending on the fluctuations in the background distribution of the ly@xmath4 emitters . a second source of contamination can occur if the off - band image has a shallow limiting magnitude and joint photometric errors in the on - band and off - band images lead to sources being erroneously classified as emission - line objects ( the so - called `` spillover effect , '' see a05 ) . combined , these effects can lead to significant contamination of the catalogs : arnaboldi et al . ( 2004 ) have performed follow - up spectroscopy of ipn candidates in the fcj , core , and sub fields , and achieve a confirmation rate of 84% , 32% , and 72% , respectively , for ipn candidates brighter than @xmath15 . we address the issue of contamination in more depth in 4 and 5 . a second concern is the uniqueness of the ipn candidate lists . both the field 3 and field fcj candidates were extracted from the same imaging dataset , but by two different groups using somewhat different methods for extracting the emission line candidates ( see f04 and a05 for details ) . an examination of the candidates in the two fields shows markedly different samples . of the 46 field 3 and 19 fcj candidates brighter than @xmath16 , only 12 candidates are common to both catalogs . spectroscopic follow - up of the fcj candidates has been conducted by arnaboldi et al . ( 2004 ) , and the overlap of the spectroscopically - confirmed candidates is much better : 10/11 spectroscopically confirmed fcj candidates also show up in the field 3 catalog . nonetheless , without full spectroscopic follow - up of all the ipn candidates in survey fields , we limit our analysis to the existing photometric ipn candidate lists . with these caveats in mind , we show in figure 2 the ipn candidates from f04 and a05 overlaid on the broadband image from mihos et al . a visual examination of figure 2 suggests that the correlation between pne candidates and diffuse light depends on the underlying surface brightness the connection is stronger near the luminous virgo ellipticals , where their halo light dominates the fields , and weak or absent in regions of low surface brightness in truly intracluster fields . for example field 3/fcj , which is dominated by halo light from m87 , shows a clear gradient in pne density from the ne to the sw , along the direction of m87 s luminosity gradient . many of these pne are likely associated with m87 itself rather than being true intracluster pne ( see , e.g. , the discussion in f04 ) ; indeed , followup spectroscopy of 15 ipn candidates in the fcj field by arnaboldi et al . ( 2004 ) shows that 12 of the 15 candidates have a narrow velocity distribution ( @xmath17 km / s ) that can be associated with m87 s halo . similarly , the majority of ipn candidates in field sub are very clearly concentrated around the giant ellipticals m86 and m84 . in other fields further away from the luminous ellipticals and at much lower surface brightness , the correspondence between ipn candidates and broad band light is less clear . field 4 only has four ipn candidates brighter than the limiting magnitude ; it is hard to make any inferences about this field with so few candidates . field 7 is more complex ; there are candidate ipne around the galaxy pair ngc 4435/8 , as well as along the eastern edge of the field , where they may be associated with tidal streams coming from m87 s halo . in the core field , there are many ipn candidates but no obvious connection to the galaxies . field lpc is located in a region of our image that shows the least diffuse light , and indeed has only a small number of ipn candidates . if anything , however , these candidates appear anti - correlated with the broadband light , as they lie in the southwest ( lower right ) portion of the field while the m86/m84 complex lies just off the northeast ( upper left ) corner of the field . as a comparison , we show in figure 3 _ simulated _ pne catalogs for each survey field , based on the assumption that pne follow the broadband light detected in our deep imaging . to construct these catalogs , we build for each field a master pne list under the assumption that the number of pne in any given pixel is proportional to the pixel intensity ( @xmath18 ) . to avoid pne being assigned to noise spikes , the images are all median smoothed with a 9x9 pixel box before the ipn catalog is generated , and we do not generate pne in regions of high surface brightness ( at @xmath19 for all fields but sub ; in sub , pne candidates were found at higher surface brightness , and we choose the cutoff for this field to be @xmath20 . ) we choose the constant of proportionality @xmath21 so that a master list is generated of @xmath22 pne , which ensures that master catalog contains potential ipn candidates even at low surface brightnesses . we then randomly select from this master catalog the same number of ipne as found in the actual narrowband imaging data from f04 and a05 . the random catalog shown in figure 3 is typical of the catalogs generated this way . examining figure 3 , the connection between the icl and the simulated ipne is clearly stronger than in the real data . many simulated pne are found near the bright ellipticals in fields 3 , fcj , and sub , but now the correlation with ngc 4435/8 is much stronger in fields 7 and core . the simulated ipne in the eastern ( left ) half of field 7 more obviously lie on top of the m87 streams , and the simulated ipne in field lpc are concentrated in the northeast ( upper left ) portion of the image , towards the m86/m84 pair . the fact that these spatial biases are evident in the simulated dataset shows that the weaker bias seen in the real dataset is likely not simply a result of small number statistics . visually , therefore , the correlation between the ipn candidates and the diffuse broadband light is loose at best . in regions of higher surface brightness which encompass the extended halos of bright ellipticals field 3/fcj and core the ipn candidates do seem to correlate with the galaxy light . at lower surface brightness in regions which are more truly intracluster , the results are more ambiguous field 7 suggests a possible correlation between the diffuse light and ipn candidates , but this not seen in fields core and lpc . because of the lower surface brightness in the intracluster fields , much of this lack of correlation may be due to fractional contamination being higher in these fields . furthermore , other factors may contribute as well to wash out the correlation , such as the depth of the virgo cluster causing ipne on the far side of virgo to be lost from the magnitude - limited ipn samples , or systematic variations in the value of @xmath4 across virgo . we address each of these possibilities in more detail below . however , because visual impressions are easily misled by the eye s desire to find patterns in the data , we first must develop a more quantitative measure of the correlation between ipne and the diffuse light . ideally , one could do this comparison by combining the various ipn catalogs into a single master catalog , and calculating the surface density of ipn candidates in regions of a given surface brightness across the cluster as whole . unfortunately , this test is problematic for a number of reasons one being the heterogeneity of the existing ipn datasets . because the datasets all have different limiting magnitudes , contamination fractions , and selection criteria , we can not combine them together to generate a homogeneous `` master list '' of ipn candidates on which to make the suggested test . for example , at a given surface brightness , the ipn counts in one field may be low compared to another simply because the first field did not go as deep . alternatively , counts in one field could be high compared to another if the first field had more contaminants . in principle , while a homogeneous master catalog can not be built , we could attempt the test within individual fields . here , however , we are limited both by the small number of ipn candidates per field , and , at lower surface brightness , uncertainty in the absolute surface photometry due to sky subtraction . in fields 3/fcj and sub , however , we have a means to attempt this test in a somewhat more model - dependent fashion . these fields are dominated by the halos of the elliptical galaxies m87 ( for field 3/fcj ) and m84/m86 ( for field sub ) , and using our deep imaging we can use iraf s ellipse task to construct model luminosity profiles for these ellipticals ( janowiecki et al . in preparation ) . we then define the areas of the field covered by isophotes of a given surface brightness , and calculate the ipn surface density in these areas as a function of surface brightness . since the isophotal models are spatially smooth and extend to high surface brightness , this should provide a demonstration of how well the ipn candidates follow the galaxy luminosity distribution . the results are presented in figure 4 , where the dotted line shows the expectation if pne surface density follows the stellar light profile of the galaxies . the correlation between the galaxy light and pne surface density is clear , reminiscent of other studies of the pne distribution around luminous spheriods such as m104 ( ford et al . 1996 ) and ngc 5128 ( hui et al . 1993 ) . however , the analysis in figure 4 only tests the correlation between ipn density and the light profile of the giant ellipticals . it does not probe truly intracluster regions at low surface brightness where the luminosity distribution is spatially complex , nor does it factor in the contribution of fainter galaxies which may lie in the field . to test the correlation between ipn candidates and diffuse light more generally in all the ipn survey fields , we need to employ an alternative set of quantitative tests to explore the correlation between ipn candidates and broad band light . first , in 3 we compare the _ absolute _ level of surface brightnesses measured by our deep imaging with that inferred from the ipne surveys , averaged over each survey field . then , in 4 we examine the _ relative _ excess of light around ipn candidates in individual fields . the first approach deals with the problems of heterogeneity in the icl data by relying on the statistical corrections for completeness and contamination that f04 and a05 have adopted for their datasets , allowing us to make the proper absolute comparison between different survey fields . the second approach deals with the problems of systematic uncertainty in the surface brightness data by making a relative measure of flux excess rather than an absolute measurement of surface brightness . if ipne faithfully trace the diffuse light , that should manifest as an excess of light in regions around ipn candidates , compared to randomly sampled regions throughout a given field . we begin with an absolute comparison of the icl measured in our deep broadband imaging to that inferred from the ipn measurements by f04 and a05 over larger scales ( 60150 kpc ) . as discussed in 2 , given the differences in the various ipn survey fields and catalog extraction techniques , combining the various ipn datasets is rather problematic . rather than attempting to create a master ipn catalog , we adopt the analyses of f04 and a05 , which convert ipn density to inferred broad - band surface brightness using statistical corrections for incompleteness and contamination which are appropriate to the individual survey fields . we also adopt a value of @xmath4 ( @xmath23 ) for the conversion which is standardized across all ipn fields ( see below ) . this then gives us a measure of the absolute level of icl in each survey field , which we can then compare to our own broadband imaging . we also note that the variation between the analysis by f04 and a05 of field 3/fcj can then be viewed as an estimate of the systematic uncertainty in the ipn - inferred surface brightness . as mentioned previously , one needs to adopt a value for the parameter @xmath4 to convert ipn counts to broadband surface brightness . the exact value of @xmath4 depends on how far down the planetary nebular luminosity function ( pnlf ) one detects pne , as well as the stellar population mix of the icl ( buzzoni et al . 2006 ) : more luminous , metal - rich galaxies have lower values of @xmath4 than do low luminosity metal - poor systems ( ciardullo et al . 2005 ) . for simplicity , we adopt ipn - inferred surface brightnesses calculated using the value of @xmath4 determined from studies of individual intracluster stars in virgo by durrell et al . ( 2004 ) , @xmath24 pne / l@xmath25 , where the `` 2.5 '' refers to the number of pne 2.5 magnitudes below m * , the high - luminosity cutoff in the pnlf ( @xmath26 ; ciardullo et al . this relatively high value of @xmath4 measured by durrell et al . ( 2004 ) for the virgo icl is reflective of low luminosity systems ( @xmath27 ) , but is measured only over a small field ( @xmath0 4 arcmin@xmath1 ) halfway between m87 and m86 . over the larger fields surveyed by the ipn surveys and our deep imaging , there may be significant and systematic differences in the stellar populations and therefore the value of @xmath4 in virgo s icl . we return to the uncertainty in @xmath4 in 5 below . we do make one more adjustment to the a05 analysis . a05 quote their inferred surface brightnesses in , while both our imaging and the f04 results work in . we therefore transform the a05 results from to by adopting a b - v color of 1.2 , consistent with the choice of bolometric correction ( bc=@xmath28 ) adopted for the stellar populations in their work . to measure the diffuse intracluster light in our fields , we must first mask out foreground stars and background galaxies ( large , bright virgo galaxies are handled in a separate step , described below ) . we mask using the prescription described in feldmeier et al . ( 2002 ) , and refer the reader there for more details ; here we will walk briefly through the steps , using the field 7 to illustrate the process ( figure 5 ) . in brief , we start by manually masking all extremely bright stars , as well as their diffraction spikes and column bleeds ( figure 5b ) . we then run a ring median on the image and subtract off the ring median image from the raw image ; producing an image where the large galaxies have essentially been removed . we use sextractor ( bertin & arnouts 1996 ) to identify sources on this median - subtracted image , using a very aggressive search : all objects 0.6@xmath8 above sky and at least 3 pixels in size , creating a preliminary mask catalog which consists of both real objects and faint positive noise spikes . we next need to filter the noise spikes out of the mask catalog , since masking only positive noise spikes , but not the corresponding negative noise spikes , would bias our final measurement of the icl to artificially low values . we do this masking in a statistical sense , by multiplying the median - subtracted image by @xmath29 to create a `` negative image '' and rerunning sextractor with the same search parameters the negative noise features it identifies should have the same distribution as the positive noise features in the original image , and we can use this distribution to statistically remove the noise features from the mask catalog . specifically , as a function of source magnitude , we define a `` noise fraction '' ( simply the ratio of the source counts detected in the negative image to those in the actual image ) and then , for each source in the actual image , we use rejection sampling ( based on the magnitude - dependent noise fraction ) to determine whether or not to keep it in our final sextractor masking catalog . we add this mask to the bright star mask to produce the second level of masking ( figure 5c ) . the sextractor mask does a good job of masking small galaxies and non - saturated stars , but as it is generated from a median - subtracted image , it fails to capture the large galaxies which disappear during a median subtraction . it also spuriously identifies as discrete sources bright spots ( either real or noise ) in the inner regions of galaxies . to overcome this , we add one last layer of masking , which masks out the high surface brightness regions of galaxies . this mask also qualitatively captures an areal selection effect wherein ipn studies generally do not concentrate on the inner regions of galaxies where the ipn detection is difficult against the high surface brightness of the galaxy . to create this final mask , we mask all pixels brighter than a given surface brightness threshold , taken to vary from = 25.5 to = 26.5 . figure 5d-5f show the mask at these varying surface brightness levels ; at = 25.5 , light is clearly `` leaking out '' from the large galaxies , while at = 26.5 , the mask appears too aggressive , blocking out both diffuse light ( for example , the faint tidal features extending south from ngc 4425 near bottom right ) and individual noise spikes . while our preference from an examination of these masks is to place most emphasis on the results using the = 26 mask , we use the results of the analysis using the = 25.5 and = 26.5 masks as a conservative estimate of our uncertainty . given the masked image , we then define the photometric surface brightness of the diffuse light in each field to be the mean surface brightness of the unmasked pixels below the surface brightness threshold . since ipne are expected to be an unbiased tracer of luminosity , this simple definition of the icl surface brightness should connect to the ipn - inferred surface brightness measurements most directly . in figure 6 we plot a comparison of the photometric and ipn - inferred icl surface brightness measurements for the six fields shown in figure 1 . an examination of figure 6 shows a rough correspondence between the two measures of icl . field 3/fcj shows the highest inferred surface brightnesses using both techniques , and it is clear from figure 1 that this field largely covers the extended stellar halo of m87 . a large fraction of the pne in this field are likely associated with m87 rather than being true intracluster objects . indeed , followup spectroscopy of 15 ipn candidates in the fcj field by arnaboldi et al . ( 2004 ) shows that 12 of the 15 candidates have a narrow velocity distribution ( @xmath17 km / s ) that can be associated with m87 s halo . field sub also has significant overlap with the halos of bright galaxies , most notably the ellipticals m84 and m86 . in this field , many of the pne candidates lie close to bright galaxies and have a velocity structure which is closely correlated with the galaxy velocities ( arnaboldi et al . 2004 ) , again suggesting that we are not probing a true intracluster population here . fields 7 , 4 , and core all have lower photometrically - inferred surface brightnesses , and correspond to regions which have more intracluster area . the bright interacting pair ngc 4435/38 falls in both field 7 and field core , but covers a relatively small fraction of the total area of these fields . the smaller field 4 contains one low surface brightness dwarf ( vcc 1052 ; bingelli , sandage , & tammann 1985 ) , and no other galaxies of note . in both surface brightness and ipne , the lpc field has the faintest inferred intracluster light . lying south of the m84/m86 complex , this field contains no bright galaxies and overlaps only minimally with the extended diffuse light around m84/m86 . this field also shows the largest disagreement between the photometric and ipn - inferred surface brightness , our measurement is approximately 3.5 mag / arcsec@xmath1 brighter than that inferred from the ipn data . as noted previously , the ipn candidates in this field are also found largely to the southwest portion of the field , away from the complex of galaxies just north of the field . so while lpc has the least amount of icl measured both by broad band light and ipn studies , the marked disagreement in both the absolute level of diffuse light and its spatial distribution makes the comparison between the two methods less than satisfactory . one caveat is that our broad - band imaging only partially covers the full lpc field , such that we are biased towards the high surface brightness portion of the field closer to the luminous ellipticals . however , if we presume that there is _ no _ icl present in the portion of the field we do not image , the net surface brightness we would measure for the entire field would drop only by 0.5 mag / arcsec@xmath1 , still leaving us with a significant discrepancy . an examination of figure 6 shows a tendency for the ipn - inferred surface brightness to be systematically fainter than the photometric surface brightness . again , the ipn - inferred surface brightness is based on the durrell et al . ( 2004 ) value of @xmath30 , and a systematic offset between the two surface brightnesses could come from the use of an incorrect value for @xmath4 . if instead we demand that the data scatter around a line of equality and _ fit _ for a value of @xmath31 , we get @xmath32 , lower than the durrell et al . value . such values of @xmath31 are similar to those measured in luminous ellipticals ( ciardullo et al . 2005 ) , and may suggest an icl which originates from stellar populations stripped from virgo s luminous ellipticals . however , the scatter in the relationship shown is quite large , and the systematic effects present in both datasets ( see 4 ) complicate this simple analysis . the previous section looked at the correlation between the absolute levels of icl measured by broad band imaging and ipn surveys on the @xmath0 100 kpc scales . however , measuring the correlation between icl and ipne on these scales blurs together different environments an individual ipn field may simultaneously contain the outskirts of luminous galaxies , faint tidal streams , and true intracluster space . the variation in environment may make it hard to identify a clean connection between the diffuse light and the ipne , and so we wish also to probe the connection on smaller scales ( @xmath33 15 kpc ) , where individual galaxies or tidal streams can be isolated . to examine the correlation between ipn candidates and diffuse broadband light on these smaller scales , we look _ within _ individual survey fields , and measure the amount of _ excess _ luminosity around ipn candidates ( compared to random positions in the field ) . if the ipn candidates trace the broadband luminosity , this should manifest as a luminosity excess around the candidates , and since it is a relative measurement , it is more robust against systematics of sky subtraction at low surface brightness , where we are most interested in testing the correlation . our quantitative technique explores the luminosity excess in boxes of varying angular scale that contain one or more ipn candidates , compared to randomly scattered boxes . for a field with ipn candidates , we randomly sample the field using boxes of a given size , until we have boxes each containing at least one ipn candidate ; the median pixel intensity @xmath34 ( in adu ) is then calculated using _ all _ unmasked pixels in these boxes . an identical procedure is used to measure the median pixel intensity in randomly placed boxes ( @xmath35 ) . we then define the value @xmath36 to be the difference in median pixel intensity between these two samples : @xmath37 . we repeat this process 40 times , and calculate the median and upper and lower quartile of the @xmath38 s over these 40 trials . for these trials , we choose sampling boxes which range in size from 3x3 pixels ( 4.35or 0.3 kpc on a side ) to 131x131 pixels ( 190or 14.6 kpc on a side ) to look for correlations over a range of physical scales . in this analysis , we also restrict ourselves only to ipn candidates brighter than the limiting magnitude of each field given by f04 and a05 . one concern with such an analysis is a selection effect for detecting ipn candidates : individual pne will be hard to distinguish against a high surface brightness background such as the inner regions of bright galaxies . this places a limiting surface brightness above which ipn candidates will not be detectable . the exact surface brightness limit will depend on the observational details and analysis of the different ipn surveys and are hard to uniquely determine . nonetheless , we can mimic such a selection function by masking all pixels brighter than a certain limiting magnitude in our imaging data before doing our analysis . an examination of figure 2 shows that for most fields except sub , nearly all ipn candidates are found at surface brightnesses @xmath10 25 mag / arcsec@xmath1 ; for sub , this surface brightness limit is a magnitude brighter , @xmath10 24 mag / arcsec@xmath1 . accordingly , we apply surface brightness masks to the data at these levels to mimic the ipn surface brightness selection effect . the correlations for each field are shown in figure 7a , where the errorbars represent the upper and lower quartile of the measurements . we see the strongest correlations in the fields near m87 ( fields 3 and fcj ) and m86/m84 ( field sub ) , which have on average higher surface brightness . near m87 , the luminosity excess is stronger on small scales , and slowly declines towards zero on larger scales . this trend is seen in both field 3 and fcj , although the significance is lower in field fcj due to the smaller number of ipn candidates detected in the analysis by a05 . clearly in this field , the ipn candidates are tracing m87 s diffuse halo , and indeed may not be truly _ intracluster _ objects , but reflect instead the bound stellar population around m87 . as mentioned in 3 , this suggestion is borne out by the kinematic study of ipne in this field by arnaboldi et al . ( 2004 ) , which reveal a velocity distribution to the ipne which is similar to m87 itself . in field sub , which contains m86 , m84 , and several other luminous galaxies , the correlation is even stronger than in field 3/fcj , both in amplitude ( since ipne are found deeper inside galaxies in this field ) and in the decline with angular scale . in this field , the ipn candidates are almost all found near m86 and m84 , and again are likely to be associated with the giant ellipticals themselves , rather than being intracluster objects . in the fields further away from large galaxies ( fields 4 , 7 , core , and lpc ) , where the average surface brightness is much lower , we find no correlation between the local broadband surface brightness and the ipn candidates at any spatial scale . on the smallest scale , it is possible in principle to detect the ipn candidates themselves ( the brightest pne would have @xmath39 , yielding a peak intensity of @xmath0 0.9 adu in our image ) , but we see no evidence of this in the data . there are small fluctuations ( @xmath40 adu ) at these scales in some fields , but we find no evidence of point sources when we co - add the images around the location of the ipn candidates . this fact , combined with the fact that the signal does not depend on the number of ipn candidates , and is in fact negative in field lpc , leads us to believe that fluctuations on these scales are simply due to statistical noise . in the case of field lpc , we actually see a flux _ deficit _ around ipn candidates . while there are not many ipn candidates in this field , this flux deficit does appear to be statistically meaningful . this again is a manifestation of light and ipn candidates being anti - correlated in this field : randomly placed boxes sample the broad band light in an unbiased fashion , while boxes that contain ipn candidates are biased to the southern portion of the field , away from the diffuse light surrounding the complex of galaxies near ( and including ) m86 and m84 . these differences lead to the anti - correlation seen in the lpc field . in summary , we see correlations between the ipn candidates and the broadband light in higher surface brightness regions near the luminous ellipticals in virgo , but in true intracluster fields at low surface brightness there seems to be no small - scale correlation visible at all , save for the anti - correlation seen in lpc . one concern with this analysis remains the very different areas and surface densities of ipn candidates in the different fields . given the relatively small number of ipn candidates , and the likely contamination of the candidate list by background emission line galaxies , is it likely that we would able to _ detect _ the correlation between broadband light and ipn candidates in these datasets ? we address this question by creating an artificial sample of ipn candidates from the broad band image , under the explicit assumption that pne trace light ( as described in 2 ) , and that contaminating objects are uniformly distributed across the field . then , for a given contamination fraction @xmath41 , we randomly draw @xmath42 pne from the catalog ( where @xmath43 is the number of candidate ipne actually observed in each field ) , and add another @xmath44 pne randomly distributed across the image . we then analyze this artificial dataset in the same way that the real candidate lists were analyzed , to see what correlation signal we would detect . for each of the 40 trials we select a new random sample of ipne from the master list , so that our results are not simply reflecting the distribution of a single realization of modeled ipne . an example of a single realization of ipn candidates under the pne - follows - light model ( with no contaminant ) was shown in figure 3 , while the statistical analysis of the modeled ipn datasets is shown in figure 7b . for each field we show curves for the artificial datasets given contamination fractions of 0% , 20% , 40% , 60% , 80% , and 100% , with the true ( observed ) correlations overlaid . as expected , contamination acts to reduce the correlation signal quite quickly . the only field consistent with no contamination is field sub , which is not surprising since its candidate ipne were detected using the two - line method which significantly reduces contamination . in field 3 and fcj , the data is consistent with a model where ipne follow light ( in this case , the extended halo of m87 ) , assuming a contamination fraction of @xmath0 40% , somewhat higher than the value of 20% adopted by f04 and reported by arnaboldi et al . ( 2004 ) from follow - up spectroscopy of a subset of candidates . in fields 7 and lpc , the model shows that a small signal might be detectable , but at a very low significance level that is quickly washed out by even a modest amount of contamination . given the contamination that is known to exist in these fields , it is no surprise that we see no correlation , although the _ anti - correlation _ seen in field lpc is still unexplained . with more ipn candidates , the correlation signal in core should have been detectable if the contamination was moderate ; unfortunately , spectroscopic follow - up has shown a very low confirmation rate for the ipn candidates ( 32% ; arnaboldi et al . if the contamination rate is as high as 70% , the correlation signal in this field is clearly undetectable . and finally , as expected for field 4 , there are simply not enough ipn candidates to distinguish any correlation we show the results for this field simply for completeness sake . analyzing the mock ipn catalog moderates somewhat the disconnect seen between the broad band light and the ipn candidate catalogs . where we expect a strong signal and modest contamination i.e. , near the giant ellipticals we do in fact recover the correlation ( fields 3 , fcj , and sub ) . in the intracluster fields , the signal is much weaker and contamination acts to wash out the correlation . since pne are formed in the late stages of the evolution of low mass stars , they should trace the stellar mass of galaxies . in galaxy clusters , this means that ipne should , in principle , trace the diffuse intracluster light formed from stars stripped from their host galaxies . we have compared the icl detected in broad band light with candidate ipne detected in narrowband imaging surveys , and find only a rather rough correspondence between the two . the strength or absence of the correlation clearly depends on local surface brightness . at higher surface brightness , near the luminous ellipticals m87 , m86 , and m84 , the correspondence is strong , as evidenced by fields 3 , fcj , and sub . in truly intracluster fields where the average surface brightness is lower , however , the correlation is much weaker , on both large ( @xmath0 100 kpc ) and small ( @xmath0 1 - 10 kpc ) scales . unfortunately , while the expectation that ipne trace light is well - motivated , many effects , both systematic and statistical , can act to cloud the connection . we consider each in turn now . _ ipn catalog contamination _ the biggest complication to the analysis is that of contamination in the ipn catalogs , as described extensively in the previous sections . in the field - averaged absolute comparison detailed in 3 , the ipn - inferred surface brightnesses taken from f04 and a05 factor in this contamination statistically . clearly , however , this correction can be significant and can vary both from field to field ( due to differences in limiting magnitude and off - band subtraction ; see f04 and a05 for details ) and likely within individual fields due to clustering of background sources ( e.g. , ciardullo et al . 2002 ) . contamination is even more problematic when correlating the light on small scales with individual ipn candidates , as was done in 4 . our monte - carlo modeling of the effects of contamination shows that contamination fractions of @xmath0 40% are needed to explain discrepancy between the data and the `` pne - follow - light '' model , a number consistent with the rates of contamination inferred from the extant spectroscopic followup that exists for several of the fields ( see the discussion in 2.2 and in arnaboldi et al . the only field consistent with the pure pne - follow - light model is field sub , and the ipn catalog for this field was generated using the two - line method ( okamura et al . 2002 ) , which significantly cuts out contamination from background sources and photometric errors . we also note also that a population of contaminants ( real or artificial ) that is distributely uniformly across the fields could easily give rise to the fact that the correlation we detect is a strong function of surface brightness . if pne do follow light , a constant density of contaminants means that the contamination fraction will increase at lower surface brightness , leading to a correlation at high surface brightness that washes out as lower surface brightness regions are probed . _ the effects of stellar populations _ another possible complication is variation in the parameter @xmath45 , which converts pne number density to broadband surface brightness . this parameter is sensitive to the underlying stellar population ( buzzoni et al . 2006 ) ; measurements of @xmath4 within individual galaxies show a scatter of a factor of five between systems , with some suggestion that the scatter is linked to the metallicity of the host system ( ciardullo et al . 2005 ) . in converting the pne measurements to equivalent v - band surface brightness , we have adopted the value of @xmath4 obtained by durrell et al . ( 2004 ) which is typical of low luminosity systems . on average , the ipn - inferred surface brightnesses are fainter than the photometrically derived surface brightnesses by about one mag / arcsec@xmath1 , a discrepancy would could be accounted for by adopting a value of @xmath46 instead of the durrell et al . ( 2004 ) value of @xmath47 . this would suggest that the icl in the virgo fields may arise from material associated with or stripped from virgo s giant ellipticals . indeed , three of the fields lie close to the luminous ellipticals m87 , m86 , and m84 ( i.e. , the subaru and 3/fcj fields ) , consistent with this idea . more globally in virgo , however , the exact value of @xmath4 could show systematic and complex variations depending on the progenitor population of galaxies contributing to the icl , and it may be naive to expect one value of @xmath4 to accurately describe the entirety of the virgo icl population . one might hope with sufficient broadband data and uncontaminated ipn catalogs to invert the problem and map out the variations in icl stellar populations across the cluster , but clearly the existing data shown here is not sufficient for such a task . the stochastic nature of the ipn population statistics also affects the detected correlations . the number of ipne per unit stellar luminosity ( given by @xmath4 ) is low , such that quantitative measurements will be affected by the poisson noise in the number of detected ipne . for example , adopting the value of @xmath4 measured for the icl in virgo by durrell et al . ( 2004 ) and a color of @xmath48 for the icl , the tidal stream emanating from m87 should only have @xmath0 5 pne within 1 magnitude of m * spread over its 100 kpc extent . within the fields analyzed here , there are between 10 and 100 candidate ipne per field , but given the high contamination fraction in some fields , the expected number of _ true _ ipne is more like a few to 50 per field . with such small numbers of ipne , the statistical fluctuations due to poisson noise in the number counts will be large . _ ipn spatial distribution _ one potentially important systematic difference between broadband imaging surveys for diffuse icl and searches for ipne relates to the line - of - sight depth of the icl . as surface brightness is a distance - independent quantity , our deep imaging will detect icl over the entire depth of the virgo cluster . as the ipn surveys detect point source objects down to a given limiting magnitude , at a certain depth these surveys will begin to miss ipne . the virgo cluster has complex substructure and studies of its three dimensional structure have suggested significant depth . while much of this structure is on scales larger than the so - called virgo `` cloud a '' which our data probes , even the core of virgo has evidence of substructure in its depth . for example , jerjen et al . 2003 use sbf distances to early - type galaxies to show that the distance distribution is broad , with a front - to - back depth for the cluster of 5.6 mpc ( defined as @xmath49 in the line - of - sight distances ) . moreover , the distribution is bimodal , with peaks at 16 and 18.5 mpc associated with the m87 and m86 subclusters , respectively . more recently , mei et al . ( 2007 ) measure a smaller depth of the virgo core of 2.4 @xmath50 0.4 mpc . however , the depth of the cluster defined by the spiral population may be much larger , as the spiral population in virgo is more spatially extended than is the early - type galaxy population ( binggeli , tammann , & sandage 1987 ) . indeed , while cepheid distances to virgo spirals cluster around a central value of 15 mpc ( freedman et al . 2001 ) , outliers do exist , as shown by cepheid distance to ngc 4639 ( d=25 mpc ; saha et al . 1997 ) . the limiting depth of the ipn surveys depends both on the survey limiting magnitude and the luminosity of the ipne . the planetary nebula luminosity function shows a sharp break at @xmath51 ( ciardullo et al . 2002 ) ; very few pne are found brighter than this value . at a canonical virgo distance of 16 mpc , this corresponds to an apparent magnitude of @xmath52 , such that all surveys can detect the brightest ipn candidates at least to the distance of virgo . however , because of the sharp cutoff in the pnlf at @xmath53 , one needs to go at least half a magnitude down the luminosity function to get appreciable samples of ipne . as a result , the limiting magnitude of several of the ipn fields means that only ipne in the _ near side _ of the cluster will be detected . table 1 shows the limiting @xmath53 depth of each field ; going half a magnitude fainter will reduce this depth by 20% . the problem is particularly severe for fields 4 and 7 , which have the shallowest limiting depth . other fields have gone deeper and should detect ipne further along the line of sight through the cluster , but all will be systematically biased towards detecting ipne on the near side of virgo . in contrast to the ipn surveys , our deep broadband imaging will probe the full depth of the cluster and therefore can reveal icl that is too distant to have detectable ipne . this effect would result in photometric surface brightnesses which are systematically _ brighter _ than those inferred from ipn counts . again , figure 6 shows a slight systematic trend in exactly this sense , but the scatter is large , and there is no obvious correlation with limiting depth of the field . for example , the fairly shallow field 4 shows agreement between the two methods , while the deeper sub and lpc fields show large discrepancies . deeper ipn imaging will be needed to test this issue further . _ other methodological uncertainties _ certainly the uncertainties on these measurements are large ; both techniques push the limits of the available data , and there are large and differing systematic uncertainties in each technique . indeed , even _ within _ each technique the uncertainties are large . for example , depending on the adopted surface brightness threshold for masking , our data yield icl surface brightnesses which can vary by @xmath0 1 mag / arcsec@xmath1 ( see table 1 ) . similarly , different analyses of the _ same _ imaging data for field 3/fcj by f04 and a05 using different detection criteria yield inferred surface brightnesses that differ again by @xmath0 1 mag / arcsec@xmath1 . one last systematic effect to consider is that of over - subtraction of the icl during the sky subtraction phase of the broadband imaging data reduction . during this process , we fit planes to the sky and then subtract the fitted planes ; if there is diffuse , structureless icl component in the virgo cluster core which subtends a size comparable to the scale of our image ( @xmath54 or 420 kpc on a side ) , we will have no way of distinguishing this from the night sky , and it will be subtracted off . in this way , our broadband icl estimates could be systematically low . however , for a cluster with such significant substructure in its galaxy population , it seems unlikely that a substantial amount of the icl would be smoothly distributed over such a broad scale . furthermore , the amount of light would be substantial : spread out over a 420x420 kpc field , a smooth icl component at @xmath55 would have a total luminosity of @xmath56 , five times brighter than the total luminosity of m87 , m86 , and m84 _ combined_. this would skew the icl fraction of the cluster to extraordinarily high values , much higher than expected from simulations of cluster collapse . summing up the luminosity of all vcc galaxies ( binggeli et al . 1985 ) in our imaging field , we obtain a total blue luminosity of @xmath57 ; adopting an icl color of b - v=1 , this would imply a _ missing _ icl fraction of @xmath0 60% , not factoring in the icl that we actually do detect in our imaging . this suggests that if any icl was removed artificially during the sky subtraction stage , it must be at much lower surface brightnesses than we are detecting in our deep imaging . given these uncertainties , it is perhaps encouraging that we recover even the weak correlations discussed in 3 and 4 . the hope of using deep surface photometry to act as a finding chart for ipn surveys remains valid ; however , the ipn surveys need to be both deeper ( to uncover more ipne and improve the poisson statistics ) and rid of contaminants . with single - line narrow band imaging , these two needs work at cross purposes , as contamination gets worse at fainter levels ( e.g. , ciardullo et al . unfortunately , the hope of constraining icl stellar populations through a mapping of @xmath4 across the cluster is clearly unfeasible without resolving these issues . this task will not prove easy . however , there are a number of observationally feasible ways to reduce the amount of contamination from current and future ipn surveys . imaging observations of ipn candidates in better seeing than the original surveys ( the average seeing of all of the ipn fields is 1.25 ) will remove a significant fraction of contaminating objects . at distances of @xmath58 16 mpc , all ipne should remain spatially unresolved , while background contaminating objects such as bright ly@xmath4 galaxies at z = 3.1 have half - light diameters as large as 0.6(gronwall et al . 2008 , in prep ) . therefore , narrow - band [ o iii ] imaging with significantly better ambient seeing , imaging systems with adaptive optics , or imaging systems that use orthogonal transfer ccds ( jacoby et al . 2002 ; howell et al . 2003 ) will be extremely helpful in removing contaminants . although ly@xmath4 galaxies generally have high observed equivalent widths ( @xmath10 80 ) , the density of ly@xmath4 galaxies decreases exponentially as the equivalent width increases ( gronwall et al . thus , the deeper the off - band imaging is in an ipn survey , the higher equivalent width limit will be obtained , and the fewer contaminants will enter the sample . similarly , using a third image as a `` veto '' filter would also reduce the possibility of contamination all genuine ipne should also emit in other emission lines , the brightest of which are the @xmath13 4959 [ o iii ] emission line and the h@xmath4 + [ n ii ] complex . unfortunately , both of these emission line regions are @xmath58 3 times fainter than the @xmath13 5007 [ o iii ] line ( storey & zeippen 2000 ; ciardullo et al . 2002 ) , making confirmatory observations more difficult . however , it is worth noting that the sub field observations used deep h@xmath4 + [ n ii ] imaging , and we have found that this field is consistent with no contamination ( 4 ) . however , most definitive way to completely eliminate contamination of ipn catalogs is still follow - up spectroscopic observations . ipne will have narrow emission lines compared to contaminating sources ( freeman et al . 2000 ; arnaboldi et al . 2004 ) , the @xmath13 4959 [ o iii ] should always be present , and there should be no sign of continuum emission . despite some significant efforts ( freeman et al . 2000 ; arnaboldi et al . 2004 ; arnaboldi et al . 2008 ; doherty et al . 2008 ) , the fraction of ipn candidates that have been spectroscopically observed is still quite small . with the growing availability of high throughput , wide - field , multi - object spectrographs on large telescopes , this situation should improve . since the radial velocities of ipne are of considerable interest for dynamical studies of the intracluster light , such follow - up observations will have multiple benefits . on the broadband imaging side , going deeper not only will prove challenging technically , but will also quickly run into a fundamental background : the backscattering of galactic light by dust in the milky way at high galactic latitude . this scattered light shows up in deep imaging as diffuse light similar in morphology to the diffuse icl ( e.g. , sandage 1976 ; de vries & le poole 1985 ; witt et al . 2008 ; janowiecki et al . in preparation ) . this high galactic latitude dust can be traced by its thermal emission in the far infrared , and dust maps constructed from iras imaging ( schlegel , finkbeiner , & davis 1998 ) show that the virgo core fortuitously sits in a hole in the galactic cirrus , such that the deep surface photometry shown in figure 1 is relatively unimpacted by this dust . however , surrounding regions have significantly more cirrus and detecting and quantifying the icl in those regions will be much more difficult . for example , in our imaging of fields to the north and southeast of the the field shown in figure 1 , we clearly see cirrus at surface brightness levels of @xmath59 mag / arcsec@xmath1(janowiecki et al . in preparation ) . to correct for the contamination of the images by the galactic cirrus , we need both a high resolution , wide field map of the dust , as well as a good model for the backscattering of the galactic light ( see witt et al . 2008 ) . existing dust maps from infrared surveys suffer from poor spatial resolution ( @xmath0 6 in the maps of schlegel et al . 1998 ) ; while _ spitzer _ provides higher spatial resolution than iras ( @xmath0 1 ) , its small field of view makes deep imaging over a wide area such as the virgo cluster prohibitively expensive . as such , the galactic cirrus represents a hard floor to deep broadband surveys for intracluster light . to go deeper , one must rely on star counts ( e.g. , ferguson et al . 1998 ; ferguson et al . 2002 ; durrell et al . 2002 ; williams et al . 2007 ) which at the distance of virgo can currently be done only over small angular scales using space - based telescopes . again , mapping the icl across the face of virgo this way will be prohibitive . in summary , we find a modest correlation on large ( @xmath0 100 kpc ) scales between the broad - band diffuse light in virgo and the inferred density of ipne . there is appreciable scatter to this correlation , and the correlation depends appreciably on the underlying surface brightness . in regions of higher surface brightness near luminous ellipticals , the connection is evident , and we are likely seeing a pne population which simply traces the luminous profile of the the connection shows appreciable scatter on large scales , and no correlation on smaller ( @xmath0 10 kpc ) scales . a number of observational and physical effects act to wash out the expected `` ipne - follow - light '' correlation signature , the most pernicious of which are contamination the ipn catalogs and the small number statistics that arise from the inherently small number of pne per unit luminosity in any stellar population . indeed , given these complications , it is encouraging that we recover even the weak correlations seen in our data , and our results remain consistent with the expectation that ipne trace the underlying broadband icl . improvements on this analysis will need to come through improved ipn samples which go deeper ( to increase sample sizes ) _ and _ simultaneously reduce the contamination due to background emission sources and faint stellar sources . we thank ortwin gerhard , magda arnaboldi , and their collaborators for providing their ipn catalogs . this work is supported by nsf grants ast-9876143 and ast-0607526 ( jcm ) , ast-0098435 ( hlm ) , and ast-0302030 ( jjf ) , and by research corporation cottrell scholarships to jcm and hlm . we quantify the photometric uncertainties in our broadband imaging , following the prescription of morrison et al . 1997 , and direct the reader there for more detailed descriptions of the process . the first set of considerations are processes that introduce uncorrelated noise on a pixel - by - pixel basis : readnoise , photon noise , and small - scale flat fielding errors . for the site ccd used in the observations , the gain was 2.8e@xmath60/adu and the readout noise was 7.5e@xmath60 , or 2.6 adu . since the icl is a small fraction ( @xmath331% ) of the night sky , photon noise is dominated by sky photons . for a typical sky background of 1250 adu , we expect photon noise of 21 adu per pixel in a single image . photon noise in the individual blank sky images will also translate into uncertainty in the resulting flat field . this uncertainty will be mediated in the final image by our dithering pattern each pixel in the final image comes from a different pixel of the flat field ( each with independent errors ) . the final error is given by @xmath61 where @xmath62 is the number of counts in the final , combined master sky flat image , g is the gain , @xmath63 is the number of individual sky flats used to make the master sky flat . for a typical sky value of 1250 adu , and 127 individual sky images , this corresponds to 2.3 adu . these processes will sum in quadrature to give an uncertainty in a single virgo image of 21.4 adu per pixel . since the final mosaic consists of a median of many individual images , this noise will scale with image number as @xmath64 , where @xmath65 is the number of individual images that contribute to the final mosaic . due to our large dithering pattern , this number varies across the mosaic . for most of the ipn fields , @xmath0 35 images contribute to that portion of the mosaic ( see table 1 ) ; field 7 has the most ( @xmath0 45 images ) , and lpc the fewest ( @xmath0 17 images ) . for @xmath66 , the random error per pixel becomes 4.5 adu , while for @xmath67 it becomes 6.6 adu . when measuring the icl over large fields , such as the ipn survey boxes shown in figure 1 , these errors drop as @xmath68 and become negligible when compared to the uncertainty due to large scale flat fielding and sky background variations ( discussed below ) . however , when measuring the icl on small scales ( as done in 4 ) , the measurement boxes are smaller and the uncertainty can be significant . the smallest boxes we analyze are 3x3 pixels in size ; this corresponds to an uncertainty of 1.4 ( 2.2 ) adu for 40 ( 17 ) combined images . over larger scales , there are two dominant sources of uncertainty which do not easily `` root n '' away . the first is large scale flat fielding errors , predominantly due to variation in sky brightness and the unsubtracted extended wings of stars in the blank sky images . the second is night sky variations in the object images themselves , which can vary both randomly and systematically ( with , for example , hour angle , time of night , or airmass ) . to quantify these combined uncertainties , we identify regions of blank sky ( predominantly in the southwest , north , and eastern portions of the image ) and measure the residual flux and uncertainty in 100x100 pixel boxes . these boxes show a mean residual sky background of 0.1 adu and a variation of 0.5 adu , which we adopt as our measure of the large scale uncertainty of our data . since these effects do not `` root n '' away , they represent the floor to our error model .

the rayleigh product channel model is useful in capturing the performance degradation due to rank deficiency of mimo channels . in this paper , such a performance degradation is investigated via the channel outage probability assuming slowly varying channel with delay - constrained decoding . using techniques of free probability theory , the asymptotic variance of channel capacity is derived when the dimensions of the channel matrices approach infinity . in this asymptotic regime , the channel capacity is rigorously proven to be gaussian distributed . using the obtained results , a fundamental tradeoff between multiplexing gain and diversity gain of rayleigh product channels can be characterized by closed - form expression at any finite signal - to - noise ratio . numerical results are provided to compare the relative outage performance between rayleigh product channels and conventional rayleigh mimo channels . zheng : outage capacity of rayleigh product channels : a free probability approach central limit theorem ; finite - snr diversity - multiplexing tradeoff ; free probability theory ; mimo ; outage capacity ; rayleigh product channels .

1502.05516

multi - input multi - output ( mimo ) wireless communications have received considerable attention since it is seen as the most promising way to increase link level capacity . extensive works have focused on the performance of mimo channels assuming a rich scattering environment . therein , the presumed models include full - rank independent rayleigh or rician mimo channels . however , in certain environments the propagation may be subject to structural limits of fading channels caused by either insufficient scattering @xcite or the so - called keyhole effect @xcite . these channels exhibit rank deficiency compared to the independent rayleigh and rician models . the mimo model that captures these effects is referred to as the double - scattering channel @xcite . it is characterized by a matrix product involving three deterministic matrices ( i.e. , transmit , receiver , and scatterer correlation matrices ) , and two statistically independent complex gaussian matrices . in a typical office environment , empirical measurements have been used to demonstrate the validity of the double - scattering channel model @xcite . there exist a number of studies concerning the information - theoretic quantities of the double - scattering channels . derived an upper bound for the ergodic capacity ( * ? ? ? iii.3 ) and an exact expression for a single keyhole channel ( * ? ? ? the diversity - multiplexing tradeoff of the double - scattering channel was obtained in @xcite . the authors in @xcite investigated the asymptotic rayleigh - limit when one of the matrix dimensions approaches infinity . in this case , the double - scattering model reduces to an equivalent rayleigh mimo channel . furthermore , if all matrix dimensions are large , the ergodic capacity has been obtained in @xcite via numerical integration . recently , an asymptotic expression for ergodic capacity of the double - scattering channels was derived in @xcite . moreover , authors in @xcite derived the ergodic mutual information for finite dimensional channel matrices . however , all the above results are valid for ergodic channels , where each codeword has infinite length . for many practical communication systems such as wlans @xcite , the channels , albeit random , are slowly varying and the encoding / decoding process is subject to a delay constraint with moderate target packet error rates around @xmath0@xmath1 . the fading channel seen by each codeword are therefore non - ergodic . in this case , the ergodic capacity has no physical significance , whereas the outage capacity is a more relevant performance metric @xcite . in literature , the outage capacity has been studied for conventional rayleigh mimo channels @xcite as well as for rician mimo channels @xcite via various random matrix techniques . to the best of our knowledge , the outage capacity for the double - scattering channel has not been addressed in the most general form . it turns out to be a difficult random matrix theory problem . to gain insights into the outage behavior of the double - scattering channel , we consider a simplified channel model involving a product of two statistically independent complex gaussian matrices , also known as the rayleigh product channel . this channel model corresponds to the scenario where the antenna elements as well as the scattering objects are sufficiently separated and there is no spatial correlation at antenna arrays or between scatterers . to characterize the capacity fluctuations , we use the free probability theory for large dimensional random matrices @xcite . by utilizing the second order cauchy transform and machinery , we derive a compact expression for the asymptotic variance of the capacity of the rayleigh product channel . we further show that the channel capacity distribution is asymptotically gaussian by proving a central limit theorem ( clt ) for the linear spectral statistics ( lss ) of the rayleigh product ensemble . this result generalizes the clt for correlated wishart random matrices @xcite and the clt for rayleigh product ensembles from polynomial lss to generic analytic functions @xcite . the capacity distribution is then utilized to study the corresponding finite signal - to - noise - ratio ( snr ) diversity - multiplexing tradeoff ( dmt ) . the derived results in this paper are formally valid when the dimensions of the channel matrices grow to infinity . however , numerical simulations show that they serve as good approximations when the numbers of antennas and scatterers are comparable to practical systems . the rest of the paper is organized as follows . in section [ secmodel ] , we give the channel model , the signal model as well as the mimo capacity formulation . in section [ secmoment ] , we study the second order eigenvalue fluctuations and the asymptotic capacity variance . the second order cauchy transform of rayleigh product ensembles is derived in section [ seccumu ] . the clt of the capacity of rayleigh product channels is proved in section [ secclt ] . based on this result , the approximations for outage probability and the finite - snr dmt are calculated . in section [ secconc ] , we conclude the main findings of the paper . proofs of the technical results are provided in the appendices . _ notations . _ throughout the paper , vectors are represented by lower - case bold - face letters , and matrices are represented by upper - case bold - face letters . the complex vector field with length @xmath2 is denoted as @xmath3 . we use @xmath4 to denote the zero - mean complex gaussian vector with covariance matrix @xmath5 and @xmath6 is an @xmath7 identity matrix . the superscript @xmath8 denotes the matrix conjugate - transpose operation and @xmath9 is matrix transpose . we denote @xmath10 as the complex conjugate operator . denote @xmath11 as the trace of @xmath7 matrix @xmath5 and @xmath12 as the normalized trace @xmath13 . the notation @xmath14 $ ] denotes the expectation , and @xmath15 denotes the matrix determinant . consider a discrete - time , baseband mimo system with @xmath16 transmit and @xmath17 receive antennas . the channel is assumed to follow the rayleigh product fading with @xmath18 scattering objects , as shown in fig . [ figdoublescatter ] . the channels between the @xmath19-th scatterer and transmit antennas are denoted by vector @xmath20 $ ] , and the channels between receive antennas and the @xmath19-th scatterer are denoted by vector @xmath21 $ ] . the end - to - end equivalent channel matrix @xmath22 is given by @xmath23 where @xmath24^{\dag}$ ] and @xmath25^{\dag}$ ] . we assume @xmath26 and @xmath27 , where @xmath28 and @xmath29 , @xmath30 , are statistically independent . the channel @xmath22 is thus modeled as a product of two independent complex gaussian random matrices . in line with @xcite , the channel @xmath22 is normalized by the constant @xmath31 so that the total energy of the channel is equal to an awgn channel with an array gain @xmath32 = \sum_{i , j } { \mathbb{e}}[|h_{ij}|^2 ] = t$ ] . transmit antennas , @xmath17 receive antennas , and @xmath18 scatterers.,width=432 ] the presence of independent gaussian matrices @xmath33 and @xmath34 in ( [ eqh ] ) requires two independent and richly - scattered environments , where the scattering happens between the @xmath18 scatterers / keyholes and transmit and receive arrays , respectively . this requires the existence of a large number of independently reflected and scattered paths around the antenna arrays @xcite . the two environments are connected only via the @xmath18 scatterers / keyholes . by controlling the number @xmath18 , the rayleigh product channel ( [ eqh ] ) embraces a general family of mimo fading channel , spanning from the degenerate keyhole channel @xmath35 @xcite to the full - rank rayleigh mimo channel @xmath36 with fixed @xmath17 and @xmath16 @xcite . mller and hofstetter @xcite have shown that the number of significant scatterers is around ten in a typical office building with an @xmath37 antenna configuration . measurement results in @xcite indicate that the effective rank of a @xmath38 keyhole channel depends on the sizes of scatterer / keyhole at different transmission frequencies . in general , the number of separable scattering objects depends on the number of antenna elements since a larger array increases the spatial resolution @xcite . note that the model ( [ eqh ] ) also describes the mimo relay channels when assuming noiseless relays @xcite . the channel output vector @xmath39 , at a given time instance , equals @xmath40 where @xmath41 is the transmit vector that follows the complex gaussian distribution @xmath42 with @xmath43 $ ] . the additive noise @xmath44 is modeled as an _ . _ complex gaussian vector @xmath45 . in this paper , we have adopted the following assumptions : 1 . the channel state information ( csi ) is perfectly known at the receiver but not at the transmitter.[a1 ] 2 . the channel @xmath22 is frequency flat and quasi - static . it remains constant for certain symbol durations and takes independently a new value for each coherence time.[a2 ] 3 . delay - constrained encoding / decoding . the encoded transmit message has a finite block length and spread in time over no more than a maximum allowable decoding delay . we assume the length of a coding block is equal to one independently faded interval.[a3 ] under a[a1 ] , telatar @xcite has shown that the channel capacity is achieved when the transmitted symbols are independent across antennas and the power is equally allocated , i.e. @xmath46 and @xmath47 denotes the signal to noise ratio ( snr ) per received antenna . the instantaneous capacity of the mimo channel ( [ eqy ] ) in nats / sec / hz is given by @xmath48 where @xmath49 , @xmath50 , refer to the eigenvalues of @xmath51 . for the hermitian matrix @xmath52 , we find it convenient to introduce the empirical spectral distribution ( esd ) defined as @xmath53 where @xmath54 denotes the indicator function . by letting @xmath55 , the channel capacity @xmath56 can be rewritten in terms of @xmath57 as @xmath58 as the channel matrix @xmath22 is random , the instantaneous capacity ( [ eqmi ] ) is also a random variable . without csi at the transmitter , there is a non - zero probability , independent of the code length , that the channel capacity ( [ eqmi ] ) falls below any positive rate . due to the assumptions a[a2 ] and a[a3 ] , the error probability corresponding to this rate can not be decreased exponentially with the code length @xcite . in this case , no reliable transmission is possible and the performance can not be evaluated using the ergodic capacity . instead , the fundamental performance limit of such a system is best explained with the capacity versus outage tradeoff , characterized by the cumulative distribution function ( cdf ) of @xmath56 . given a fixed rate @xmath59 , the outage probability is defined as the probability that capacity @xmath56 is less than @xmath59 , i.e. @xmath60 where @xmath61 denotes the cdf of @xmath56 . when the cdf @xmath61 is monotonically increasing , the outage capacity for a given probability @xmath62 is obtained as @xmath63 the outage probability ( [ eqpout ] ) is achievable @xcite in the sense that for any @xmath64 , there exists a code of sufficiently large block length for which the packet error rate is upper - bounded by @xmath65 . thus , outage capacity provides useful insights on the operational performance of a delay - constrained coded system . outage probability is also a meaningful metric to characterize the performance of some contemporary communication systems @xcite , where timely csi is available at the transmitters . from this viewpoint , the complementary outage probability @xmath66 can be interpreted as the percentage of time that a transmission takes place at given rate @xmath59 under perfect link adaptation . in this section , we first review the convergence of empirical eigenvalue distribution of the hermitian matrix @xmath67 when matrix dimensions grow to infinity . the capacity per receive antenna is shown to converge to a deterministic value and expressed by a known result in @xcite . then , we study the global fluctuation of eigenvalues around the limiting distribution by deriving a closed - form expression for the the second order cauchy transform of @xmath52 . this result is utilized to obtain the asymptotic variance of the channel capacity . for an @xmath68 hermitian random matrix @xmath5 , we assume that its esd @xmath69 converges to a non - random limiting distribution @xmath70 as @xmath71 . such a convergence is alternatively established in @xcite via the convergence of resolvent @xmath72 to the first order cauchy transform @xmath73 , defined as @xmath74 here , @xmath75 and @xmath76 denotes the support of @xmath70 . due to this limiting behavior of eigenvalues , the normalized linear spectral statistics , such as normalized capacity @xmath77 , converges to a non - random limit as @xmath71 for a wide class of matrix ensembles @xcite . in the following , the limit @xmath78 denotes the asymptotic regime , @xmath79 in the asymptotic regime ( [ eqdim ] ) , silverstein @xcite shows that the esd @xmath80 converges almost surely to a non - random cdf @xmath81 and its cauchy transform @xmath82 is the solution to @xmath83 where @xmath84 and @xmath85 is the well - known marenko - pastur distribution @xcite . the integration range in ( [ eqgq ] ) is @xmath86 $ ] . using multiplicative free convolution , mller has shown in @xcite that @xmath87 in the asymptotic regime ( [ eqdim ] ) and @xmath82 satisfies the cubic equation @xmath88 if @xmath89 , @xmath90 admits the formal power series expansion @xmath91 where @xmath92 and @xmath93 is the @xmath2-th free moment of @xmath52 , defined as @xmath94.\ ] ] a concept closely related to the cauchy transform @xmath95 is the @xmath96 , defined as a functional @xcite @xmath97 if @xmath89 , @xmath96 has the formal power series representation @xmath98 where @xmath99 is the @xmath2-th free cumulant of @xmath52 . as will be shown in section [ seccumu ] , the free cumulant sequence @xmath100 and its generating function @xmath96 serve as the key analytical tools in the proof of proposition [ propcauchy2 ] . note that when the matrix dimensions are finite , the computation of @xmath93 involves non - trivial summations over all partitions of integer @xmath2 ( * ? ? ? * eq . ( 27 ) ) . this complicated expression makes it challenging to obtain an explicit expression for the free cumulant @xmath99 . as the matrix dimensions approach infinity , the calculation of @xmath99 is much simplified , involving only the so - called non - crossing permutations over integers , see lemma [ lemmacumu ] and appendix [ appxnc ] for a detailed discussion . using random matrix theory techniques , authors in @xcite prove that the capacity per receive antenna of the rayleigh product channel converges to an asymptotic limit such that @xmath101 . the asymptotic capacity @xmath102 is given by an explicit closed - form expression , which is summarized in the following proposition . ( asymptotic capacity @xcite)[propmu ] when @xmath103 , the asymptotic capacity per receive antenna @xmath104 in the regime ( [ eqdim ] ) is given by @xmath105 where @xmath106 is the unique solution to @xmath107 such that @xmath108 . although the asymptotic capacity @xmath109 grows to infinity in the asymptotic regime ( [ eqdim ] ) , it serves as a tight approximation to the mean capacity @xmath110 $ ] with finite matrix dimensions as shown in @xcite . in the following , we will also use @xmath109 as the approximated @xmath110 $ ] whenever it is clear from the context . as @xmath111 in the asymptotic regime ( [ eqdim ] ) , the asymptotic capacity per - received antenna @xmath102 can be formulated by replacing @xmath112 in ( [ eqmi ] ) with @xmath113 . using an integral identity ( * ? ? ? * eq . ( 1.14 ) ) , we utilize an amenable form of asymptotic capacity , which is useful in the following discussion , namely @xmath114 the complex integral on the right hand side of ( [ eqcontourint ] ) is over any positively oriented closed contour @xmath115 enclosing the support @xmath116 and on which @xmath117 is analytic . for the instantaneous channel capacity ( [ eqmi ] ) , there exists a similar integral expression as in ( [ eqcontourint ] ) with cauchy transform @xmath95 replaced with the resolvent @xmath118 . to see this , let the contour @xmath115 be selected according to ( [ eqcontourint ] ) and apply cauchy s integral formula on @xmath119 , it follows that the instantaneous capacity ( [ eqmi ] ) becomes @xmath120 exchange the integrations and recall the definition of resolvent ( [ eqresolvent ] ) , we obtain @xmath121 let us now consider the variance of capacity @xmath56 , defined as @xmath122)^2]$ ] . replacing @xmath56 with ( [ eqmib ] ) and @xmath110 $ ] with ( [ eqcontourint ] ) , the variance @xmath123 can be rewritten as @xmath124,\ ] ] where @xmath125 , the contours @xmath126 and @xmath127 are non - overlapping and are taken in the same way as in ( [ eqcontourint ] ) . after interchanging the expectation and integrations , we have @xmath128 where @xmath129 $ ] is the covariance function of matrix resolvent scaled by the matrix dimension @xmath17 . in the context of free probability theory , this covariance function is known as the second order cauchy transform @xcite and is denoted as @xmath130 . the rest of this section is devoted to derive the second order cauchy transform of @xmath52 and the asymptotic variance of capacity @xmath123 by using a recent result from free probability theory . namely , by the framework of the second order freeness @xcite , the second order cauchy transform @xmath131 exists if @xmath52 has a second order limiting distribution according to the following definition : [ defsecdist ] let @xmath132 be an @xmath68 random matrix . we say that it has a second order limiting distribution if for all @xmath133 the moments @xmath134 and the limits @xmath135 exist and if for all @xmath136 and all @xmath137 , @xmath138 where @xmath139 denotes the @xmath59-th classic cumulant . as @xmath140 and @xmath141 are independent complex wishart matrices , they are unitarily invariant and their second order limiting distributions exist ( * ? ? ? it follows from ( * ? ? ? * eq . ( 29 ) ) and the cyclic invariant property of matrix trace that @xmath52 also has the second order limiting distribution . the second order cauchy transform @xmath131 is given by the functional ( * ? ? ? ( 53 ) ) @xmath142 where @xmath143 denotes the second order of @xmath52 . similar to the first order case , if @xmath144 and @xmath145 , @xmath131 and @xmath143 have formal power series representations @xmath146 in literature , the covariance function @xmath147 for the wishart type @xmath68 random matrix @xmath148 has been studied in @xcite , where @xmath149 is a non - random hermitian matrix and @xmath150 is a gaussian like random matrix with _ _ entries . therein , the correlation function of @xmath151 has the form @xmath152 and it is subsequently used to derive an asymptotic variance of rayleigh mimo capacity in @xcite . note that the second term of the right hand side of ( [ eqseccauchy ] ) is exactly the same as ( [ eqgb ] ) by replacing @xmath151 with @xmath153 and assuming @xmath154 non - random . therefore , the fluctuation of capacity @xmath123 of rayleigh product channels has a distinct functional structure from the rayleigh mimo channels , see ( [ eqvar ] ) and ( [ eqseccauchy ] ) . the increased fluctuation is due to a non - zero @xmath143 . closed - form expressions of @xmath131 and @xmath143 are summarized in the following proposition : [ propcauchy2 ] the second order cauchy transform of @xmath52 is given by ( [ eqseccauchy ] ) with @xmath155 where @xmath156 is the cauchy transform of a marenko - pastur distribution with the parameter @xmath157 as in ( [ eqdim ] ) @xmath158 the proof of proposition [ propcauchy2 ] depends on the combinatorial structure of cumulants @xmath100 and @xmath159 and is given in detail in section [ seccumu ] . substitute ( [ eqseccauchy ] ) into ( [ eqvar ] ) and denote @xmath160 , then the asymptotic variance @xmath123 is rewritten as @xmath161 in the general setting , it is difficult to further simplify the double integral ( [ eqvarc ] ) . however , when the transmitter and receiver have equal number of antennas , i.e. @xmath162 , a compact expression for @xmath123 can be obtained . the results are summarized in the following proposition . [ propvar ] when @xmath103 , the asymptotic variance @xmath123 is given by @xmath163 where @xmath164 is the solution of the cubic equation @xmath165 the proof of proposition [ propvar ] is in appendix [ appxasyvar ] . using cardano s formula to solve the cubic equation ( [ eqcubic ] ) , the explicit expressions for the roots are @xmath166 where @xmath167 for general values of @xmath47 and @xmath157 , it is not straightforward to gain insights based on the variance expressions ( [ eqvars ] ) , ( [ eqt1])-([equ ] ) . however , in the high snr regime with @xmath168 , the asymptotic variance @xmath123 is characterized by explicit expressions and the behavior of capacity fluctuation can be understood . [ corollary1 ] in the high snr regime @xmath168 , the asymptotic variance @xmath123 is given by ( [ eqvars ] ) with @xmath169 approximated by @xmath170 for a fixed snr , the variance of channel capacity is highest when the number of scattering objects equals to the number of antennas . the proof of corollary [ corollary1 ] is in appendix [ appxcorollary1 ] . the asymptotic variance @xmath123 is derived with the assumption that the dimensions of matrices are large . however , @xmath171 serves as a good approximation for the variance of capacity even when the matrix dimensions are comparable to realistic mimo systems . in fig . [ figvar ] , we plot the variance of channel capacity as a function of the number of scattering objects @xmath18 for @xmath172 and @xmath37 mimo systems . the asymptotic variance is calculated by proposition [ propvar ] at snrs @xmath47 ranging from @xmath173 db to @xmath174 db with a step size of @xmath175 db . the analytical calculations are compared with monte carlo simulations , where each curve is generated by @xmath176 independent channel realizations . we also plot the asymptotic variance of a conventional rayleigh mimo channel using ( * ? ? ? * eq . ( 13 ) ) . [ figvar ] shows that the asymptotic variance achieves a good agreement with the simulations for a wide range of snrs and numbers of scatterings , especially in the low snr regime . it is only when @xmath177 db that there are observable gaps between analytical and simulation curves . the asymptotic variance for a @xmath37 mimo system remains a better approximation than that of a @xmath172 system , as expected . in the high snr regime , see fig . [ figvar ] ( a ) , there exists a peak value for the variance of capacity when @xmath178 . as the snr @xmath47 increases , the peak of the variance occurs at a fixed value @xmath179 ( @xmath180 ) , which is in line with our prediction in corollary [ corollary1 ] . this is analogous to the observations in @xcite that the capacity variance of the conventional rayleigh mimo channel is largest when @xmath181 . on the other hand , the variance is monotonically decreasing in the low snr regime , see fig . [ figvar ] ( b ) . as the number of scatterers becomes large , we also observe that the capacity variance of the rayleigh product channel approaches a limit . this limit is set by the variance of conventional rayleigh mimo channel with the same number of antennas . this agrees with the results in @xcite , where the multi - keyhole channel converges to a rayleigh mimo channel when the number of scatterers is large . [ figvarh ] [ figvarl ] this section is devoted to the proof of proposition [ propcauchy2 ] , which relies on the knowledge of free cumulants of @xmath52 . let us recall from ( [ eqrz ] ) and ( [ eqrxy ] ) that @xmath96 and @xmath143 are generating functions for the cumulant sequences @xmath100 and @xmath159 , respectively . we will first deduce the combinatorial descriptions of @xmath99 and @xmath182 . these results reveal that the cumulant sequences of @xmath52 have the same combinatorial structures as the moment sequences of a marenko - pastur distribution . namely , @xmath143 can be obtained based on known results . the notations and terminologies used in the formulation of lemma [ lemmacumu ] and in the proof of proposition [ propcauchy2 ] are given in appendix [ appxnc ] . [ lemmacumu ] for integers @xmath133 , the first order free cumulant @xmath99 of matrix @xmath52 is given by @xmath183 and the second order free cumulant @xmath182 is given by @xmath184 where @xmath185 and @xmath186 denote the set of non - crossing permutation in disc and annular sense . the proof of lemma [ lemmacumu ] is in appendix [ appxlemmacumu ] . by comparing ( [ eqcumu1 ] ) with the @xmath2-th moment @xmath187 of @xmath188 ( * ? ? ? * eq . ( 7.3 ) ) , we have @xmath189 with @xmath190 . the cumulant @xmath99 can be viewed as the scaled version of the moment @xmath187 , where the normalized trace has the normalization factor @xmath17 instead of the actual matrix dimension @xmath18 . similarly , by comparing ( [ eqcumu2 ] ) with the second order moment @xmath191 of @xmath188 ( * ? ? ? * eq . ( 7.5 ) ) , we have the second order cumulant @xmath192 , where no normalization is needed as in definition [ defsecdist ] . note that the normalization for the first order moments can be arbitrarily chosen without affecting the underlying combinatorial structures provided that the normalization factor grows at the same rate as the matrix dimensions . the functional relations between moments and cumulants as well as their generating functions @xmath90 , @xmath193 are therefore preserved . for notational simplicity , it is convenient to work with the properly scaled moment sequences @xmath194 and @xmath195 . as @xmath187 is the @xmath2-th moment of a marenko - pastur distribution with parameter @xmath157 , by the moment - cumulant relation @xcite , the corresponding cumulant @xmath196 equals @xmath157 . in addition , it follows from the second order moment - cumulant relation @xcite that the moment @xmath191 can be expressed in terms of @xmath196 as well as the corresponding second order cumulant @xmath197 @xmath198 on the right hand side of ( [ eqsecmocumu ] ) , @xmath199 contains @xmath200 orbits and the @xmath201-th orbit contains @xmath202 elements with @xmath203 . in the second summation , @xmath204 and @xmath205 contain @xmath200 and @xmath206 orbits , respectively . the @xmath201-th orbit of @xmath207 contains @xmath208 elements with @xmath209 , and @xmath210-th orbit of @xmath211 contains @xmath212 elements with @xmath213 . the partition @xmath214 is composed of elements from the @xmath215-th orbit of @xmath207 and the @xmath216-th orbit of @xmath211 and corresponds to the second order cumulant @xmath217 . inserting @xmath196 into ( [ eqsecmocumu ] ) and comparing with ( [ eqcumu2 ] ) , we notice that @xmath191 is entirely determined by the summation over non - crossing annular permutation and ( [ eqsecmocumu ] ) is only valid when the second order cumulant @xmath197 is zero . in summary , the cumulant sequences are @xmath218 apply the functional relation ( * ? ? ? * eq . ( 52 ) ) @xmath219 where @xmath220 due to ( [ eqkpu ] ) , the formal power series @xmath221 and @xmath222 , and ( [ eqmxy ] ) becomes @xmath223 comparing ( [ eqrxy ] ) with ( [ eqmxyb ] ) , we obtain @xmath224 . this completes the proof of proposition [ propcauchy2 ] . in this section , we prove a central limit theorem for the linear spectral statistics of the matrix @xmath51 and show that the cdf of channel capacity @xmath56 is asymptotically gaussian as the matrix dimensions grow to infinity . this result generalizes the well - known clt for the correlated wishart matrix @xcite . together with the asymptotic mean and variance of capacity calculated in propositions [ propmu ] and [ propvar ] , the gaussian convergence of capacity @xmath56 gives a compact yet accurate approximation for the outage capacity . in addition , the approximative cdf of @xmath56 is useful to analyze the diversity - multiplexing tradeoff of rayleigh product channels in the finite snr regime . let @xmath225 , we are interested in the distribution of random variable @xmath226 using the integral identities ( [ eqcontourint ] ) and ( [ eqmib ] ) , we can rewrite ( [ eqmuh ] ) as @xmath227 where @xmath115 is the closed contour selected as in ( [ eqcontourint ] ) . in the following proposition , we prove that @xmath228 and @xmath229 with @xmath230 are jointly gaussian distributed in the asymptotic regime ( [ eqdim ] ) . [ propclt ] in the asymptotic regime ( [ eqdim ] ) , @xmath231 forms a tight sequence ( see , e.g. @xcite ) on a closed contour @xmath115 enclosing the support of @xmath81 , and @xmath232 converges weakly to a gaussian process on the complex plane . the proof of proposition [ propclt ] is in appendix [ appxclt ] . by proposition [ propclt ] , the asymptotic gaussianity of ( [ eqmug ] ) follows from the fact that the riemann sum corresponding to this integral has jointly gaussian summands and the sum of which can only be gaussian . proposition [ propclt ] generalizes the clt of lss for wishart type random matrices involving one deterministic correlation matrix and one random matrix with _ i.i.d . _ entries @xcite . when both matrices @xmath34 and @xmath33 are random and independent , @xmath233 can be decomposed into two random processes , see ( [ eqgr ] ) in appendix [ appxclt ] . both random processes are asymptotically gaussian and each is governed by lemma [ lemmacovbai ] . as already discussed in section [ secvar ] , the induced fluctuation of lss is characterized by both the first order cauchy transform and the second order . this is different from the wishart random matrices , where the corresponding @xmath233 only involves one asymptotic gaussian process and the fluctuation of lss is solely determined by the first order cauchy transform . this makes the clt of rayleigh product ensembles distinct from the one in @xcite . together with the mean @xmath109 and variance @xmath123 in propositions [ propmu ] and [ propvar ] , an analytical gaussian approximation to the capacity distribution of the rayleigh product channel is available . this result can not be directly derived based on the existing results in @xcite . note that the clt of lss for the biorthogonal ensembles , such as the rayleigh product ensemble , was proved by breuer and duits @xcite for polynomial functions @xmath234 . however , it is not clear how to extend this result to generic analytic functions @xmath234 such as the channel capacity @xmath235 . recently , the clt for the product of two real and square random matrices was proved by gtze , naumov , and tikhomirov @xcite for smooth function @xmath234 . rayleigh product channel . solid line : gaussian approximation ( [ eqpoutg ] ) ; markers : simulations ; dashed line : cdf of @xmath172 mimo capacity with independent rayleigh fading.,width=384 ] let @xmath236 denote the error function , the gaussian approximation to the cdf of channel capacity @xmath56 is @xmath237 and thus the outage capacity is @xmath238 based on ( [ eqpoutg ] ) and ( [ eqcoutg ] ) , the outage behavior of the rayleigh product channel can be understood . in fig . [ figpout ] the impact of the number of scatterers @xmath18 as well as the received snr @xmath47 is studied , where a @xmath172 mimo system is considered in the presence of @xmath239 scattering objects . we plot the approximative outage probability as well as the empirical one obtained by monte carlo simulations . the outage probabilities are evaluated at snrs @xmath240 and @xmath241 db . as a comparison , we also plot the outage probability of a @xmath172 rayleigh mimo channel with independent fading entries . as the number of scatterers @xmath18 increases , the outage capacity at a given probability level rapidly increases until @xmath18 is equal to the number of antennas , which is especially visible when snr is large . in this range , the rank of the channel matrix is limited by the number of scatterers and increasing the scatterers effectively improves the rank of channel matrix . when @xmath242 , the matrix rank is limited by the number of antennas and the improvement of outage capacity is relatively slow . yet , the outage probability curve approaches to a limit , which corresponds to the outage probability of a conventional rayleigh mimo channel as predicted by @xcite . in fig . [ figcout ] we examine the impact of number of antennas on the outage capacity . we plot the approximative @xmath243 outage capacity ( [ eqcoutg ] ) as a function of received snr @xmath47 . assume the number of transmit and receive antennas @xmath244 while fixing the number of scatterers @xmath245 . as expected , it is seen that the outage capacity of the rayleigh product channel is lower than the conventional rayleigh mimo channel due to the presence of a finite number of scatterers . in the high snr regime , the outage capacity curves of both channels attain the same slope when @xmath246 , which suggests that the capacity scales at the same rate as the limiting rayleigh mimo channel . on the other hand , when @xmath247 there is an increasing gap between the two channels as @xmath47 increases . finally , it is observed from fig . [ figpout ] and [ figcout ] that the gaussian approximation ( [ eqpoutg ] ) and ( [ eqcoutg ] ) is reasonably accurate for a wide range of parameter settings . outage capacity of rayleigh product channels with @xmath245 scattering objects and equal numbers of antennas . solid line : gaussian quantile approximation ( [ eqcoutg ] ) ; markers : simulations ; dashed line : outage capacity of conventional rayleigh mimo channels with equal number of antennas.,width=384 ] the concept of dmt was originally proposed in @xcite to characterize the diversity gain , which is related to link reliability , and the multiplexing gain , which is related to spectral efficiency . the dmt indicates that both types of performance gains can be obtained simultaneously while satisfying a fundamental tradeoff . the operational interpretation of the dmt framework is via the existence of universal codes , which are tradeoff optimal in the high snr regime @xcite . in space - time code design @xcite , dmt represents a useful analytical tool to characterize the asymptotic performance of codes . however , the asymptotic tradeoff is a too optimistic upper bound to estimate the operational performance at realistic snrs . recent works have shown that codes optimized at high snr may not be optimal at low or moderate snr . motivated by these facts , narasimhan @xcite proposed a finite - snr dmt framework , which characterizes the non - asymptotic dmt . there , he studied the finite dmt for the correlated rayleigh and rician mimo channels at realistic snr levels . under the assumptions of slow fading and capacity achieving codes with rate @xmath59 , the multiplexing gain @xmath248 of a mimo channel is defined according to ( * ? ? ? * eq . ( 21 ) ) as @xmath249 where @xmath250 . the multiplexing gain provides an indication of the sensitivity of rate adaptation strategy as the snr changes . when the applied codes have a higher multiplexing gain , the rate adaptation tends to respond more dramatically to the snr variations . at a fixed multiplexing gain , the finite - snr diversity gain @xmath251 is defined as the negative slope of the log - log plot of outage probability @xmath252 at rate @xmath253 versus snr @xmath47 , @xmath254 at a particular snr @xmath47 and multiplexing gain @xmath248 , the diversity gain ( [ eqdiv ] ) provides an estimate of the additional snr needed to reduce the outage probability by a certain amount . using the derived outage probability ( [ eqpoutg ] ) , the finite - snr dmt can be obtained for the rayleigh product channel . [ propdmt ] when @xmath103 , the finite - snr dmt of rayleigh product channels can be approximated by @xmath255 where @xmath256 with @xmath109 and @xmath123 calculated by propositions [ propmu ] and [ propvar ] . the proof of proposition [ propdmt ] follows by substituting ( [ eqpoutg ] ) into ( [ eqdiv ] ) . note that the approximation ( [ eqdivapprox ] ) is tight in the asymptotic regime ( [ eqdim ] ) . this is because the approximation error is induced from ( [ eqpoutg ] ) . [ figdmtt2s2 ] shows the finite - snr dmt of a @xmath257 rayleigh product channel with @xmath258 scatterers . the approximated tradeoff curves are generated by ( [ eqdivapprox ] ) at snrs @xmath259 db and @xmath260 db . compared to the monte carlo simulations , the proposed approximation yield close estimate for the mimo diversity gain . as @xmath248 approaches the maximum multiplexing gain , the discrepancies between the approximation and simulation curves decrease . when @xmath261 antennas are used , the mimo channel achieves improved channel diversity for a given multiplexing gain as shown in fig . [ figdmtt4s2 ] . in both figures , we have also plotted the asymptotic dmt of rayleigh product channels according to ( * ? ? ? * eq . ( 8) ) , when snr @xmath47 approaches infinity . it is clear that the asymptotic results significantly overestimate the channel diversity at the considered operational snr levels , which justifies the usefulness of the proposed approximation ( [ eqdivapprox ] ) . we studied the outage probability of rayleigh product channels , which explicitly model the rank deficiency effect . using free probability theory , the asymptotic variance of channel capacity is calculated for large channel matrix and becomes exact when matrix dimensions approach infinity . compared to the conventional rayleigh mimo channels , the rayleigh product channels induce a higher capacity fluctuation , which is determined by the second order of the channel matrix . we have proved that the channel capacity is asymptotically gaussian by establishing a clt of a relevant linear spectral statistics . numerical results show that the proposed gaussian approximation is reasonably accurate for realistic channel dimensions . results have been utilized to characterize the tradeoff between diversity and multiplexing of rayleigh product channels , while the asymptotic tradeoff for large snr may be an over - optimistic estimate . z. zheng and j. hmlinen are supported by the academy of finland ( grant 284634 ) . l. wei and j. corander are supported by the academy of finland ( grant 251170 ) . r. speicher is supported by the erc advanced grant ( ncdfp 339760 ) . r. mller is supported by the alexander von humboldt foundation . we would like to thank o. arizmendi and a. soshnikov for helpful discussions . let @xmath262 denote the complement of the support @xmath116 on the real axis . it is shown in @xcite that for a given open interval @xmath263 , the function @xmath90 is continuous , real , and decreasing . this is also true for @xmath264 with @xmath265 . therefore , there exists an inverse function @xmath266 continuous , real , and decreasing over @xmath267 . we choose the contour @xmath126 to be inside of @xmath127 such that they both cross real - axis in the intervals @xmath268 and @xmath269 , where @xmath270 denotes the right end - point of the support @xmath116 . by substitutions @xmath271 and @xmath272 , the integral ( [ eqvarc ] ) can be alternatively integrated over contours @xmath273 and @xmath274 as @xmath275 where @xmath276 the transformed contours @xmath273 and @xmath274 cross the real - axis in the intervals @xmath277 and @xmath278 , where @xmath279 the inverse function @xmath280 is calculated via ( [ eqgp ] ) as @xmath281 we obtain the inverse function @xmath282 by solving the quadratic equation ( [ eqcubicgq ] ) in @xmath283 as @xmath284 where the minus sign is taken by the requirement @xmath285 @xcite . substituting ( [ eqgpinv ] ) and ( [ eqgqinv ] ) into ( [ eqkinner ] ) and applying integration by parts , we can rewrite @xmath286 as @xmath287 where @xmath169 , @xmath288 , and @xmath289 are the three roots of the cubic equation @xmath290 and @xmath169 denotes the real solution such that @xmath291 . the integrand of ( [ eqinner ] ) has two simple poles at @xmath292 and @xmath293 in @xmath273 , and by applying the residue theorem , the integral @xmath286 becomes @xmath294 substituting ( [ eqinnerb ] ) into ( [ eqvard ] ) , the variance @xmath171 can be therefore expressed as @xmath295 the second integral in ( [ eqsigd ] ) has an anti - derivative @xmath296 , which is single - valued over @xmath274 and therefore vanishes due to cauchy s theorem . applying the residue theorem to the first integral in ( [ eqsigd ] ) , we obtain @xmath297 the proof is completed by the fact that @xmath298 . when @xmath299 , @xmath300 and @xmath301 can be expanded at @xmath302 as @xmath303 the real and negative solution @xmath169 of ( [ eqcubic ] ) corresponds to @xmath304 and it follows from ( [ eqt1 ] ) that @xmath305 when @xmath306 , the asymptotic expansion of @xmath300 at @xmath302 yields @xmath307 where we took the principle value @xmath308 . in this case , the real and negative solution @xmath169 of ( [ eqcubic ] ) corresponds to @xmath309 . inserting ( [ equl ] ) into ( [ eqt2 ] ) , we have @xmath310 inserting ( [ eqwrr ] ) into ( [ eqvars ] ) and taking derivative with respect to @xmath157 , we obtain @xmath311 therefore , when @xmath299 the asymptotic variance @xmath123 is a monotonically decreasing function of @xmath157 . when @xmath306 , the variance @xmath123 is a monotonically increasing function of @xmath157 , where the derivative @xmath312 with @xmath169 given by ( [ eqwrl ] ) . to sum up , the asymptotic variance is maximum when @xmath157 approach @xmath313 from both sides of the axis . this completes the proof of corollary [ corollary1 ] . let us introduce the main combinatorial objects , the non - crossing disc and annular permutations , and the related notations , which are used in lemma [ lemmacumu ] and in the proof of proposition [ propcauchy2 ] . we refer the readers to @xcite for a comprehensive description of the non - crossing permutations . for a positive integer @xmath2 , we denote the set @xmath314 as @xmath315 $ ] . let @xmath316 denote the set of all partitions of @xmath315 $ ] . given a partition @xmath317 , we have @xmath318 , where @xmath319 , called blocks of @xmath320 , are non - empty disjoint subset of @xmath315 $ ] , i.e. @xmath321 , \mbox { and } b_i \cap b_j=\emptyset\ \mbox{for}\ i\neq j$ ] . given two partitions @xmath322 , we have @xmath323 if and only if every block of @xmath207 is contained in a block of @xmath211 and denote @xmath324 the largest partition over @xmath315 $ ] . we say a partition @xmath320 is non - crossing in disc sense if there does not exist @xmath325 , @xmath326 , and @xmath327 , such that @xmath328 and @xmath329 . a non - crossing disc partition @xmath317 can be visualized as follows : draw the points @xmath330 clockwise around the boundary of a disc and connect the points belonging to the same block with a convex hull . the partition @xmath320 is non - crossing if the convex hulls are pairwise disjoint . a concept closely related to the partition is the set permutation . let @xmath331 denote the set of all permutations over @xmath315 $ ] . given a permutation @xmath332 , we have @xmath333 such that @xmath315 $ ] is decomposed into @xmath215 orbits and @xmath334 is the @xmath201-th orbit of @xmath335 containing @xmath19 elements . for two elements @xmath336 belong to the same orbit @xmath337 , there exists an integer @xmath338 such that @xmath339 . for instance , if @xmath340 , it maps the elements as @xmath341 , @xmath342 , @xmath343 , @xmath344 , and @xmath345 . the notion @xmath346 is used as the number of orbits of @xmath335 . we say a permutation is standard in disc sense if for every orbit @xmath347 of @xmath335 , there is @xmath348 . a standard disc permutation @xmath335 has an induced partition @xmath320 , where each block of @xmath320 contains the same elements as the corresponding orbit of @xmath335 . in addition , if the partition induced by standard permutation @xmath335 is non - crossing in disc sense , @xmath335 is a non - crossing disc permutation and we denote the set of all non - crossing disc permutation on @xmath315 $ ] as @xmath185 . let @xmath349 the forward cyclic permutation of @xmath315 $ ] . a permutation @xmath332 satisfy the so - called geodesic condition as @xmath350 where @xmath351 or alternative @xmath352 is the composite permutation by first applying @xmath353 and then @xmath354 . the equality in ( [ eqdiscgeo ] ) only holds when @xmath335 is non - crossing disc permutation . the geodesic condition can be intuitively viewed as the triangular inequality for the cayley graph of permutation group @xmath331 @xcite . let @xmath355 , the distance between @xmath356 and @xmath357 in cayley graph of @xmath331 amounts to @xmath358 . the inequality ( [ eqdiscgeo ] ) can be rewritten in terms of cayley distance as @xmath359 , where @xmath360 is the identity permutation . the condition that permutation @xmath335 is non - crossing is equivalent that @xmath335 lies on the geodesic connecting @xmath360 and @xmath353 in the cayley graph . let us consider another set of permutations @xmath361 , illustrated via topological drawing in the @xmath362-annular sense . instead of placing @xmath363 points on the boundary of one disc , we will use two concentric circles . the points @xmath364 are placed clockwise on the external circle and the points @xmath365 are placed counter - clockwise on the internal circle . the annulus between the two circles are referred to as @xmath362-annulus . given a permutation @xmath366 , it is visualized by drawing curves within the @xmath362-annulus , which connect the elements of each orbit , respectively . let @xmath367 an orbit of @xmath335 with @xmath19 elements . the corresponding curve connects @xmath368 to @xmath369 , then @xmath369 to @xmath370 , , then @xmath371 to @xmath368 such that : 1 ) it does not intersect with itself ; 2 ) it encloses a region completely contained in @xmath362-annulus ; 3 ) it goes clockwise around the region . we say a permutation @xmath335 is @xmath362-connected if there is at least one orbit of @xmath335 contains elements on both circles , otherwise @xmath335 is @xmath362-disconnected . in addition , @xmath366 is standard in @xmath362-annular sense if each orbit of @xmath335 satisfies either of the following conditions : 1 . given an orbit @xmath372 such that @xmath373 or @xmath374 . the elements of @xmath375 , upon cyclic permutations , can be sorted in increasing order . 2 . @xmath376 and @xmath377 . we have @xmath378 , where @xmath379 and @xmath380 . both sequences @xmath381 and @xmath382 , upon cyclic permutations , can be sorted in increasing order , respectively . we say a permutation @xmath366 is non - crossing in @xmath362-annular sense if @xmath335 is standard and the regions enclosed by every orbits of @xmath335 are not overlapping in the annular visualization described above . we denote @xmath186 as the set of non - crossing @xmath362-annular permutations . finally , according to ( * ? ? ? 6.1 ) , a permutation @xmath366 and @xmath362-connected satisfies a geodesic condition in the @xmath362-annular sense as @xmath383 where @xmath384 and the equality only holds when @xmath385 . the proof relies on a known combinatorial identity of the moments of gaussian random variables , which is stated below . ( wick s lemma @xcite).[lemmawick ] let @xmath386 denote _ i.i.d . _ complex gaussian random variables with zero mean and unit variance . 1 . let @xmath387 be positive integers such that @xmath388 , and consider two functions @xmath389\rightarrow [ t]$ ] and @xmath390\rightarrow [ t]$ ] . then @xmath391=0.\ ] ] 2 . let @xmath2 be a positive integer and consider two functions @xmath392\rightarrow [ t]$ ] . then @xmath393=\mbox{card}\left\{\tau\in{\mathcal{s}}_n|\alpha=\beta\circ\tau\right\},\ ] ] where @xmath394 denotes the cardinality . denote @xmath395 with the entry @xmath396 given by @xmath397 in light of ( * ? ? ? * th . 2.12 ) , the second order free cumulant of @xmath52 can be expressed in terms of classic joint cumulants of entries @xmath396 as @xmath398 where the vector @xmath399 $ ] can be any distinct choice of @xmath400 . for a partition @xmath317 , we define @xmath401 as a block of @xmath320 with @xmath19 elements . the expectation over the blocks of partition @xmath320 is defined as @xmath402 = \prod_{\pi_i\subset\pi } { \mathbb{e}}\left[a_{\pi_i(1)}\cdots a_{\pi_i(s)}\right].\ ] ] using the cumulant - moment relations ( * ? ? ? * eq . ( 10 ) ) , @xmath403 can be written as a sum of @xmath404 $ ] for all @xmath405 , namely @xmath406\,\mbox{m\"ob}_{{\mathcal{p}}_{m+n}}(\pi,1_{m+n}),\end{aligned}\ ] ] where @xmath407 denotes the function @xcite on @xmath408 , which satisfies @xmath409 inserting ( [ eqqij ] ) into ( [ eqkmn ] ) and applying lemma [ lemmawick ] , we see that for a given partition @xmath320 the multiplicative moment @xmath404 $ ] is non - zero only when the partition @xmath320 takes the forms @xmath410 , @xmath411 . the corresponding function can be calculated via ( [ eqmobius ] ) as @xmath412 it follows from ( [ eqkmn ] ) and ( [ eqmobius2 ] ) that the cumulant @xmath413 equals @xmath414-{\mathbb{e}}_{\pi^{(2)}}[{\mathbf{q}}_{m , n}]$ ] . denote the permutation @xmath415 . substituting @xmath416 , @xmath417 , into ( [ eqqij ] ) , we can express @xmath418 $ ] as @xmath419.\end{aligned}\ ] ] it is convenient to introduce the functions @xmath420\rightarrow [ s]$ ] , @xmath421\rightarrow [ t]$ ] , @xmath422\rightarrow [ s]$ ] . due to the independence between @xmath423 and @xmath424 , we can rewrite ( [ eqpi1 ] ) as @xmath425\,{\mathbb{e}}\left[\prod_{t=1}^{m+n } \psi_{c\left(\eta_0^{-1}(t)\right)i(t)}\prod_{t=1}^{m+n}\bar{\psi}_{a(t)i(t)}\right].\end{aligned}\ ] ] since the indexes @xmath426 are distinct , by lemma [ lemmawick ] the second expectation in ( [ eqpi1b ] ) is non - zero only when @xmath427 . the first expectation in ( [ eqpi1b ] ) is calculated by ( [ eqwick ] ) with @xmath428 and @xmath429 . for a given permutation @xmath366 , the summands of ( [ eqpi1b ] ) should fulfill @xmath430 and @xmath431 to be able to contribute to the summation . to summerize , @xmath418 $ ] is expressed as @xmath432=\frac{1}{(r s)^{m+n}}\sum_{b , c}\mbox{card}\left\{\tau\in{\mathcal{s}}_{m+n}\mid c\circ(\tau^{-1}\eta_0)=c , b\circ\tau = b\right\}.\label{eqprodmoment1a}\end{aligned}\ ] ] interchange summation and cardinality operations in ( [ eqprodmoment1a ] ) and write ( [ eqprodmoment1a ] ) as a sum over the permutation @xmath335 , @xmath433=\frac{1}{(r s)^{m+n}}\sum_{\tau\in{\mathcal{s}}_{m+n}}\mbox{card}\left\{(b , c)\mid c\circ(\tau^{-1}\eta_0)=c , b\circ\tau = b\right\}.\ ] ] the condition @xmath434 is equivalent to requiring @xmath435 to be constant on the orbits of @xmath436 . for a given permutation @xmath436 , there are @xmath437 ways to choose indexes @xmath435 . similarly , the condition @xmath438 is equivalent to requiring @xmath439 to be constant on the orbits of @xmath335 and there are @xmath440 ways to choose indexes b. as a result , ( [ eqprodmoment1b ] ) equals @xmath441=\frac{1}{(r s)^{m+n}}\sum_{\tau\in{\mathcal{s}}_{m+n}}s^{\#(\tau^{-1}\eta_0 ) } t^{\#(\tau)}.\ ] ] following the same procedures as in ( [ eqpi1])-([eqprodmoment1c ] ) , we obtain @xmath442 $ ] and @xmath443 $ ] as @xmath444 & = \frac{1}{(r s)^m } \sum_{\tau_1\in{\mathcal{s}}_m } s^{\#(\tau_1^{-1}\eta_1 ) } t^{\#(\tau_1)},\label{eqprodmoment2}\\ { \mathbb{e}}_{\pi_2^{(2)}}[{\mathbf{q}}_{m , n } ] & = \frac{1}{(r s)^n } \sum_{\tau_2\in{\mathcal{s}}_n } s^{\#(\tau_2^{-1}\eta_2 ) } t^{\#(\tau_2)},\label{eqprodmoment3}\end{aligned}\ ] ] where the permutations @xmath445 and @xmath446 . we multiply ( [ eqprodmoment2 ] ) with ( [ eqprodmoment3 ] ) and combine the permutations @xmath356 and @xmath357 to form a new permutation @xmath447 , where @xmath448 is homogeneous to @xmath357 with @xmath201-th element relabeled as @xmath449 . note that @xmath450 , where partitions @xmath451 and @xmath452 are induced by @xmath335 and @xmath453 , respectively . the new permutation @xmath335 is therefore @xmath362-disconnected , namely @xmath454\,{\mathbb{e}}_{\pi_2^{(2)}}[{\mathbf{q}}_{m , n } ] = \frac{1}{(r s)^{m+n}}\sum_{\substack{\tau\in{\mathcal{s}}_{m+n},\\ ( m , n)\operatorname{-disconnected}}}s^{\#(\tau^{-1}\eta_0 ) } t^{\#(\tau)}.\ ] ] inserting @xmath414-{\mathbb{e}}_{\pi^{(2)}}[{\mathbf{q}}_{m , n}]$ ] into ( [ eqcumu2b ] ) , we obtain @xmath455 according to ( [ eqanngeo ] ) , the exponent @xmath456 for @xmath366 and is @xmath362-connected . in addition , the equality holds only when @xmath335 is non - crossing in the @xmath362-annular sense . let @xmath36 , all terms in the summation with crossing permutation vanish and for @xmath385 , @xmath457 cancels with @xmath458 . the derivation for the first order cumulants follows similarly . denote @xmath459 $ ] as the expected resolvent of @xmath52 , which is averaged over the ensembles of @xmath34 . as matrix @xmath188 is random , @xmath460 is also a random variable and is the solution of ( [ eqgq ] ) with @xmath85 replaced by its empirical version @xmath461 , namely @xmath462 we divide @xmath233 into two parts as @xmath463 the proof of asymptotic gaussianity of @xmath233 follows in two steps , showing the asymptotic gaussianity of @xmath464 and @xmath465 , respectively . the proof then boils down to a direct application of bai and silverstein s lemma ( * ? ? ? * lemma 1.1 ) : 1 . @xmath150 is an @xmath467 complex random matrix with _ _ entries , @xmath468 = 0 $ ] , @xmath469 = 1 $ ] , and @xmath470 = 2 $ ] ; 2 . @xmath149 is a non - random hermitian nonnegative definite matrix and its esd @xmath471 converges weakly to a non - random limiting distribution @xmath472 . let @xmath473 and @xmath474 a positive contour enclosing the support of @xmath151 , then the sequence @xmath475 is tight on the contour @xmath474 , and @xmath476 converges weakly to a gaussian process on the complex plane with @xmath477 = 0 $ ] and @xmath478 given by ( [ eqgb ] ) . conditioned on @xmath188 , it is straightforward to verify that the complex gaussian matrix @xmath34 fulfills the assumption 1 ) . the esd @xmath461 converges to the marenko - pastur distribution and therefore fulfills the assumption 2 ) . furthermore , @xmath460 is , by definition , the average of @xmath479 . it thus follows from lemma [ lemmacovbai ] that @xmath464 given @xmath188 converges to a gaussian process on the complex plane with @xmath480=0 $ ] and @xmath481 by ( [ eqgqt ] ) , we have @xmath482 subtracting ( [ eqgq ] ) from ( [ eqgqtb ] ) yields @xmath483 where @xmath484 . by definition of cauchy transform , we have @xmath485 by inserting ( [ eqgpq ] ) into ( [ eqgqdiff ] ) and multiplying @xmath17 on both sides of ( [ eqgqdiff ] ) , we obtain @xmath486 . in the asymptotic regime ( [ eqdim ] ) , @xmath487 converges to @xmath82 and @xmath435 converges to @xmath488 . it follows from lemma [ lemmacovbai ] ( with an @xmath489 matrix @xmath490 and @xmath149 being an identity matrix ) that @xmath491 converges to a centered gaussian process . note that the covariance ( [ eqcovgr1 ] ) of @xmath464 is independent of @xmath188 and the randomness of @xmath465 only comes from @xmath188 , which makes @xmath464 and @xmath465 independent of each other . combining the above arguments , @xmath492 is asymptotically a sum of two independent gaussian processes and therefore @xmath233 is also a gaussian process . d. chizhik , g. j. foschini , m. j. gans , and r. a. valenzuela , keyholes , correlations , and capacities of multielement transmit and receive antennas , _ ieee trans . wireless commun . _ , vol . 1 , no . 2 , pp . 361 - 368 , apr . 2002 . g. levin and s. loyka , from multi - keyholes to measure of correlation and power imbalance in mimo channels : outage capacity analysis , _ ieee trans . inf . theory _ , 57 , no . 6 , pp . 3515 - 3529 , june 2011 . a. l. moustakas , s. h. simon , and a. m. sengupta , mimo capacity through correlated channels in the presence of correlated interferers and noise : a ( not so ) large @xmath493 analysis , _ ieee trans . inf . theory _ , 2545 - 2561 , oct . b. m. hochwald , t. l. marzetta , and v. tarokh , multiple - antenna channel hardening and its implications for rate feedback and scheduling , _ ieee trans . inf . theory _ , 1893 - 1909 , sept . 2004 . p. kazakopoulos , p. mertikopoulos , a. l. moustakas , and g. caire , living at the edge : a large deviations approach to the outage mimo capacity , _ ieee trans . theory , _ vol . 1984 - 2007 , apr . 2011 . s. h. simon , a. l. moustakas , and l. marinelli , capacity and character expansions : moment - generating function and other exact results for mimo correlated channels , _ ieee trans . inf . theory _ 5336 - 5351 , dec . 2006 . e. riegler and g. taricco , asymptotic statistics of the mutual information for spatially correlated rician fading mimo channels with interference , _ ieee trans . inf . theory _ , 4 , pp . 1542 - 1559 , apr . 2010 . a. lozano and n. jindal , are yesterday s information - theoretic fading models and performance metrics adequate for the analysis of today s wireless systems ? _ ieee commun . mag . 210 - 217 , nov . 2012 . p. elia , k. r. kumar , s. a. pawar , p. v. kumar , and h .- f . lu , explicit space - time codes achieving the diversity - multiplexing gain tradeoff , _ ieee trans . inf . theory _ , vol . 3869 - 3884 , sept .

binary particle coagulation can be modelled as the repeated random process of the combination of two particles to form a third . the kinetics can be represented by population rate equations based on a mean field assumption , according to which the rate of aggregation is taken to be proportional to the product of the _ mean _ populations of the two participants . this can be a poor approximation when the mean populations are small . however , using the poisson representation it is possible to derive a set of rate equations that go beyond mean field theory , describing pseudo - populations that are continuous , noisy and complex , but where averaging over the noise and initial conditions gives the mean of the physical population . such an approach is explored for the simple case of a size - independent rate of coagulation between particles . analytical results are compared with numerical computations and with results derived by other means . in the numerical work we encounter instabilities that can be eliminated using a suitable ` gauge ' transformation of the problem [ p. d. drummond , eur . phys . j. b38 , 617 ( 2004 ) ] which we show to be equivalent to the application of the cameron - martin - girsanov formula describing a shift in a probability measure . the cost of such a procedure is to introduce additional statistical noise into the numerical results , but we identify an optimised gauge transformation where this difficulty is minimal for the main properties of interest . for more complicated systems , such an approach is likely to be computationally cheaper than monte carlo simulation .

1212.2232

the process of coagulation or aggregation is responsible for the coarsening of a size distribution of particles suspended in gaseous or liquid media . in essence the phenomenon consists of a sequence of statistically independent events where two ( or possibly more ) particles unite , perhaps as a result of collision , to create a composite particle , and each event reduces the number of particles in the system @xcite . this has further consequences such as colloidal precipitation or accelerated rainfall from clouds @xcite . fragmentation can be present as well @xcite , but precipitation processes are typically dominated by irreversible agglomeration . the phenomenon is familiar and yet it can present some surprises , an example of which was presented by lushnikov @xcite in an exact study of coagulation kinetics driven by various choices of aggregation rates . in the later stages of a process where the binary coagulation rate is proportional to the product of the masses of the participants , for example , a single particle emerges with a mass representing a sizable fraction of that of the entire system @xcite . this has been termed a gelation event and standard kinetic models of coagulation are unable to account for the phenomenon , for the simple reason that they are designed to describe systems consisting of very large populations of particles in each size class . they rely on a so - called mean field approximation , though this is not always clearly recognised . when small particle populations play a key role in the kinetics , different approaches become necessary @xcite , the most common being monte carlo modelling @xcite . this paper investigates the utility of a rate equation model of kinetic processes that is able to capture small population effects . the markovian master equations that describe coagulation may be transformed mathematically into a problem in the dynamics of continuous stochastic variables acted upon by complex noise . the solutions to the transformed dynamics may be related to the evolving statistical properties of the populations in the physical system . the recasting of the problem into one that concerns complex ` pseudo - populations ' can be done in two distinct ways , either using methods of operator algebra similar to those employed in quantum field theories @xcite , or through the so - called poisson representation of the populations @xcite . the physical problem concerning the stochastic evolution of a set of ( integer - valued ) populations is replaced by the task of solving and averaging a set of stochastic differential equations @xcite . the advantage of such a transformation over monte carlo simulation of the process @xcite emerges for cases where the particles come in many species or sizes , since the configuration space of the populations , and hence the computational cost increases very rapidly , but in order to illustrate the approach we study a very simple example of coagulation , where the aggregation rates do not depend on the masses of the participants . this is very different from the cases that exhibit gelation , studied by lushnikov and others , and we might not expect major deviations from a mean field approach , but nevertheless it is possible to demonstrate the analytic procedure , and compare the accuracy of numerical pseudo - population dynamics and averaging with respect to other approaches @xcite , in order to form a view on the usefulness of the approach . in section [ sec : analytic - model - of ] the master equations describing the basic problem of @xmath0 aggregation are developed and then the poisson representation is introduced and used to derive the equivalent stochastic dynamics problem and associated averaging scheme . the exact solution for the evolving complex pseudo - population is obtained and its properties established . in section [ sec : numerical - approach . ] a parallel numerical study of the stochastic problem is discussed . inherent instabilities in the dynamics may be eliminated using the cameron - martin - girsanov formula , or equivalently through a so - called gauge transformation introduced by drummond @xcite . however , this comes with the introduction of diffusive dynamics for a subsidiary variable that for extended evolution times would seem to require high computational cost for accuracy . nevertheless , with a certain choice of transformation , termed an optimised gauge , we can ensure that the coagulation process is completed before this diffusive noise becomes apparent . we comment on the procedures and prospects for their further use in more complicated models in section [ sec : discussion ] . we consider the dynamics of a population of particles of a species @xmath1 undergoing binary coagulation @xmath2 . the distinction between particles of different mass is ignored and the rate at which aggregation events takes place is a constant , making this one of the simplest cases to study . the evolution of @xmath3 , the probability at time @xmath4 that @xmath5 particles remain , is described by a set of master equations of the form @xmath6 the first term corresponds to the rate of gain in the probability of observing population @xmath5 as a result of an aggregation event in circumstances where there are @xmath7 particles , properly weighted by the number of particle pairs in such an initial state and the probability @xmath8 of an event per unit time and per pair . the second term corresponds to the rate of loss of probability due to aggregation starting from a population equal to @xmath5 . multiplying the master equations by @xmath5 and summing over @xmath5 it can be shown that @xmath9 where statistical averages are written @xmath10 . if we take the view that @xmath11 , which corresponds to the neglect of fluctuations in population ( which is why the procedure is called a mean field approximation ) , and consider a large mean population @xmath12 , such that only the first term on the right hand side in eq . ( [ eq:101 ] ) is retained , we can write @xmath13 and this can be integrated to give @xmath14 where @xmath15 is the initial mean population . this is the classical smoluchowski solution to this type of coagulation kinetics @xcite . if the second term in eq . ( [ eq:101 ] ) is retained , but again neglecting fluctuations , then the integration yields @xmath16 which ought to be more accurate than eq . ( [ eq:102 ] ) , especially in the limit @xmath17 when it goes to unity rather than zero . rather than making the mean field approximation @xmath18 , we could generate an evolution equation for @xmath19 by multiplying the master equation by @xmath20 and summing , namely @xmath21 but now the third moment appears on the right hand side . an equation for the evolution of this moment would involve the fourth moment : this is a hierarchy problem that often arises in kinetic theory . a closure condition such as @xmath22 could be imposed , but the accuracy of such assumptions is questionable . similarly , the numerous master equations for the @xmath23 in a more general model where there are different categories of species @xmath24 with populations @xmath25 would reduce under a mean field approximation to @xmath26 where the coefficients @xmath27 quantify the rate of aggregation of the form @xmath28 between species @xmath29 and @xmath30 , with @xmath29 representing species mass , for example . if we should need a treatment beyond the mean field approximation , then for relatively simple stochastic processes we could use monte carlo simulations to extract the relevant statistical properties , but for more complex coagulation problems the use of this technique can become computationally expensive . the transformation of the problem we wish to investigate is of interest since it reduces the master equations to the form of eq . ( [ eq:101 ] ) and more generally eq . ( [ eq:105 ] ) , but supplemented by a noise term on the right hand side . this is not the same as inserting noise to represent a random source term in the population dynamics , nor is it an additional representation of stochasticity in the coagulation kinetics . it is a noise term that creates the statistical correlations in the populations that are neglected when the mean field approximation is taken . it turns out that the dynamics of simple binary coagulation may be modelled by a pseudo - population @xmath31 that evolves according to the equation @xmath32 where @xmath33 is a noise with certain statistical properties . this may be compared with eq . ( [ eq:101a ] ) . the notable aspect of this representation is that the noise is complex , such that the variable @xmath31 is generally complex as well . averages are to be taken over the noise history to make a connection with the dynamics of a real population . nevertheless , the solution of eq . ( [ eq:106 ] ) can be a simpler task than that posed by eq . ( [ eq:100 ] ) . we next describe how this transformation may be achieved . gardiner and chaturvedi @xcite _ _ outlined a method for transforming master equations into a fokker - planck equation by representing probability distributions as integrals of weighted complex poisson distributions . the probability of finding @xmath5 particles at time @xmath4 may be written @xmath34 this is a superposition of poisson distributions , over a closed contour of complex mean values @xmath31 , with an evolving weighting function @xmath35 which has initial value @xmath36 such that if the contour @xmath37 includes the origin , the initial condition @xmath38 is recovered from eq . ( [ eq:107 ] ) . there are two particular initial conditions of interest . if the initial population is known to be @xmath39 then @xmath40 and @xmath41 whereas if @xmath38 is a poisson distribution with mean @xmath42 then we might use @xmath43 alternatively , we could employ a poisson representation that involves a 2-d integration over the entire complex plane , namely @xmath44 which is particularly convenient if @xmath3 is initially a poisson distribution since we would then use @xmath45 . substituting the poisson representation of the probabilities given in eqs . ( [ eq:107 ] ) or ( [ eq:110a ] ) into eq . ( [ eq:100 ] ) leads to an evolution equation for @xmath35 : @xmath46 this emerges as long as integration boundary terms can be dropped , which could potentially be a problem for the 2-d integration scheme ( [ eq:110a ] ) but is not an issue for the closed contour integral representation ( [ eq:107 ] ) @xcite . but since a stochastic problem described by a fokker - planck equation such as ( [ eq:111 ] ) can be recast as one involving a stochastic differential equation ( sde ) , the properties of the distribution @xmath35 can be reconstructed by solving @xmath47 where @xmath48 is an increment in a wiener process with properties @xmath49 and @xmath50 , the brackets representing an average over the noise . this ito - type sde takes the promised form of eq . ( [ eq:106 ] ) . note that the noise term is complex since the second derivative in eq . ( [ eq:111 ] ) has a negative coefficient . the equivalence between approaches is such that the average of a function @xmath51 weighted according to the solution @xmath35 with initial condition @xmath52 ( namely the green s function of the fokker - planck equation ) is equal to the average of the same function of a stochastic variable @xmath53 evolved over a noise history @xmath54 with initial condition @xmath55 . both averages will be denoted with angled brackets and subscript in the form @xmath56 with time dependence understood . we next demonstrate that @xmath53 can be determined explicitly for this problem . consider @xmath57 with stochastic variables evolving according to @xmath58 and @xmath59 . by ito s lemma we have @xmath60 so @xmath61 with the aim of reducing this to a deterministic form , we choose @xmath62 , in which case @xmath63 we set @xmath64 and @xmath65 , in which case the sde for @xmath66 integrates to give @xmath67 , having chosen initial condition @xmath68 . furthermore if @xmath69 then @xmath70 as desired , and @xmath71 and this integrates to @xmath72 since @xmath73 and @xmath74 , we obtain an explicit solution to the stochastic dynamics : @xmath75 we next show that an average of this stochastic variable can be related to @xmath76 , the particle population averaged over the stochasticity of the _ _ physical _ _ coagulation process . using eq . ( [ eq:107 ] ) we can write @xmath77 noting that the final expression differs from the average @xmath56 introduced earlier , since the initial weighting @xmath78 takes the general form of eq . ( [ eq:108 ] ) instead of @xmath79 , and where @xmath80 for example , if @xmath81 , then @xmath82 , and for @xmath83 we have @xmath84 . a similar construction might be made using the 2-d integration scheme ( [ eq:110a ] ) . as for the state probabilities @xmath3 , we note that @xmath85 . taking @xmath86 we find that the corresponding @xmath51 according to eq . ( [ eq:121 ] ) is @xmath87 and hence @xmath88 as noted earlier , the fokker - planck equation ( [ eq:111 ] ) , written as @xmath89 , has a green s function satisfying @xmath90 for @xmath91 with initial condition @xmath92 and we can write @xmath93 where the contour should pass through the point @xmath94 such that the initial condition @xmath95 is satisfied at @xmath96 . furthermore , the evolution of @xmath35 , starting from an initial distribution @xmath78 that is non - zero only at points @xmath97 that define an origin - encircling contour @xmath37 , can be constructed from the superposition @xmath98 so that from eq . ( [ eq:120 ] ) : @xmath99 and we deduce using ( [ eq:122 ] ) that @xmath100 which is intuitively understood as a superposition of the stochastic evolution of a function of the pseudo - population @xmath53 that evolves from points on a contour of initial values @xmath97 , weighted by the function @xmath101 , given in eq . ( [ eq:108 ] ) , and then averaged over the noise present in its sde eq . ( [ eq:112 ] ) or in its solution ( [ eq:119 ] ) . this picture is illustrated in figure [ fig : sketch - of - the ] . in the complex plane according to eq . ( [ eq:112 ] ) , starting from point @xmath97 on a contour that encircles the origin . an average over all end - points , three examples of which are shown as filled circles , corresponds to the quantity @xmath102 . [ fig : sketch - of - the ] ] of course , we could equally well employ the representation @xmath103 together with the associated 2-d integral versions of eqs . ( [ eq:122 ] ) and ( [ eq:123 ] ) to obtain an analogous result @xmath104which is particularly convenient for a initial poisson distribution of population , such that @xmath105 , where @xmath42 is the initial mean . we now establish the statistical properties of the stochastic variable @xmath53 . our strategy will be to represent @xmath102 , the average of eq . ( [ eq:119 ] ) over the noise for a given initial value @xmath97 , as a ( formal ) series in positive powers of @xmath97 . the first term is straightforward to identity : we write @xmath106 and then employ the stochastic integral identity @xmath107 such that @xmath108 the proof of eq . ( [ eq:126a ] ) follows from considering the stochastic evolution of @xmath109 where @xmath110 . ito s lemma states that @xmath111 , where the primes indicate differentiation . averaging leads to @xmath112 and hence to @xmath113 since @xmath114 . the formal series is written @xmath115 with @xmath116 and where @xmath117 we focus our attention on the evolution of the mean particle population @xmath76 for cases where there are @xmath39 particles at @xmath96 . the initial distribution is @xmath40 , so we use eqs . ( [ eq:125 ] ) and ( [ eq:109 ] ) to obtain @xmath118 and by inserting ( [ eq:128 ] ) and expanding the integrand we find that @xmath119 using standard calculus of residues . notice that the series for @xmath76 is finite even though the series for @xmath102 is infinite . we have already established that @xmath120 in eq . ( [ eq:127 ] ) . in order to evaluate @xmath121 we need to consider @xmath122 which in more explicit form reads @xmath123 in which the weighting is a product of the gaussian probabilities for generating a value @xmath124 of the wiener process at time @xmath125 , and a value @xmath54 at the later time @xmath4 given the earlier value . writing @xmath126 where @xmath127 we find this factorises as follows : @xmath128 using eq . ( [ eq:126a ] ) , and hence @xmath129 with @xmath130 the averaging rapidly becomes more laborious . consider next @xmath131 . we need to evaluate @xmath132\right\rangle \\ & = 2\int_{0}^{t}ds_{1}\int_{0}^{s_{1}}ds_{2}e^{\kappa(t+s_{1}+s_{2})}\\ & \times\left\langle \exp\left[i(2\kappa)^{1/2}\left(w_{t}-w_{s_{1}}+2(w_{s_{1}}-w_{s_{2}})+3w_{s_{2}}\right)\right]\right\rangle , \end{aligned } \label{eq:138}\ ] ] where an ordering @xmath133 has been imposed by the choice of integration limits , the change in which is accounted for by inserting the prefactor of two . ( [ eq:126a ] ) this reduces to @xmath134 which becomes @xmath135.\label{eq:140}\end{aligned}\ ] ] similarly , we can show that @xmath136 and the pattern that emerges is @xmath137 in order to proceed further , we notice that the @xmath138 are related to one another by repeated integration of the nested integrals . we define @xmath139 such that @xmath140 and @xmath141 we evaluate @xmath142 in detail in order to relate it to @xmath143 and @xmath144 : @xmath145 on the basis of an analysis of further cases , we conjecture that the pattern is @xmath146}\\ & + \frac{m_{n-3}}{n[n+(n-1)][n+(n-1)+(n-2)]}\\ & \cdots+(-1)^{n}\frac{m_{1}}{n\left[n+(n-1)\right]\cdots\left[n+(n-1)+\cdots+2\right]}\\ & \!\!+(-1)^{n+1}\frac{1}{n\left[n+(n-1)\right]\cdots\left[n+(n-1)+\cdots+2\right]}\\ & \!\!\times\!\int_{0}^{2\kappa t}\!\!\ ! ds_{1}\exp[-(n+[n-1]+\cdots+1)s_{1 } ] . \end{aligned } \label{eq:146}\ ] ] this may be simplified since @xmath147 , a triangular number , and @xmath148\\ & = \frac{1}{2}n(n+1)-\frac{1}{2}(n - j)(n - j+1)=\frac{j}{2}\left(2n+1-j\right ) , \end{aligned } \label{eq:147}\ ] ] is a difference of triangular numbers , such that @xmath149}m_{n-2}+\cdots\nonumber \\ & & + ( -1)^{l+1}\left(\prod_{j=1}^{l}\frac{j}{2}\left(2n+1-j\right)\right)^{-1}m_{n - l}+\cdots\nonumber \\ & & + ( -1)^{n+1}\left(\prod_{j=1}^{n}\frac{j}{2}\left(2n+1-j\right)\right)^{-1}m_{0}(n),\label{eq:148}\end{aligned}\ ] ] where we define @xmath150.\label{eq:149}\ ] ] next we note that @xmath151=\frac{l!(2n)!}{2^{l}(2n - l ) ! } , \end{aligned } \label{eq:150}\ ] ] so finally @xmath152 in agreement with eqs . ( [ eq:144 ] ) and ( [ eq:145 ] ) , and where the @xmath153dependence of @xmath154 appearing in the final term is understood . this provides a complete recursive solution to the stochastic problem . evaluation of the @xmath155 by integration of eq . ( [ eq:143 ] ) using _ mathematica _ @xcite for @xmath156 up to twenty confirms this relation and the conjectured pattern that relates them . from the previous section we can establish that @xmath120 and @xmath157 and we use these and eq . ( [ eq:132 ] ) to obtain the exact time dependence of the mean population for initial population @xmath39 from 1 to 4 . for @xmath158 , we obtain @xmath159 which is clearly what we would expect . for @xmath160 we find that @xmath161 and this can be checked by solving the underlying master equations for this simple case . they are @xmath162 and @xmath163 with initial conditions @xmath164 and @xmath165 , giving solutions @xmath166 and @xmath167 such that @xmath168 . for @xmath169 we get @xmath170 which may also be checked by solving the master equations directly . finally for @xmath171 we obtain @xmath172 all these solutions satisfy the initial condition @xmath173 at @xmath96 and tend to unity as @xmath174 , as required . these results should be compared with those of barzykin and tachiya @xcite who found an exact solution to the master equations for this problem using a generating function approach . they obtained @xmath175 and our calculations are consistent with this expression . we now go on to study the variance of the particle population . in order to calculate the variance in the population we need the second moment @xmath19 , which is equivalent to the quantity @xmath176 . we could expand @xmath177 as a formal power series , just as we did for @xmath31 , but instead we exploit a short cut due to the relationship @xmath178 that arises from eq . ( [ eq:112 ] ) . we hence obtain from eq . ( [ eq:132 ] ) @xmath179 and using eqs . ( [ eq:152 ] ) we can construct the time derivatives , giving @xmath180 together with @xmath181 which puts us in a position to calculate the variance @xmath182 . considering first the trivial case @xmath158 , we deduce @xmath183 and @xmath184 as we would expect since there is no evolution . for @xmath160 , @xmath185=1 + 3\exp\left(-2\kappa t\right)$ ] . this has the correct limits of @xmath186 and 1 for @xmath96 and @xmath174 , respectively . the variance is then @xmath187 . this also has the correct limits of zero at both @xmath96 and @xmath174 . it goes through a maximum at @xmath188 or @xmath189 . the maximum variance is then @xmath190 . we can check the behaviour of the variance for this case using the previously derived solutions to the master equation for @xmath160 . we have @xmath191 as required . we employ _ mathematica _ @xcite to calculate higher order coefficients @xmath155 and hence the @xmath192 in eq . ( [ eq:132 ] ) . we illustrate this with the time - dependent mean and standard deviation for the case with @xmath193 , shown in figure [ fig : mean-(red)-and ] . the exact form of @xmath76 obtained from the analysis is @xmath194 which is consistent with the barzykin and tachiya expression @xcite . ( [ eq:103 ] ) happens to account reasonably well for the evolution of the mean population , as shown in figure [ fig : mean-(red)-and ] , but of course the mean field approximation upon which it is based can not produce the standard deviation , which is also shown . and with @xmath195 . the long dashed curve is the mean population according to eq . ( [ eq:103 ] ) . [ fig : mean-(red)-and ] ] it is of interest to determine the value of @xmath102 as @xmath17 . consider , using ito s lemma , @xmath196 and insert eq . ( [ eq:112 ] ) and average such that @xmath197 assuming that as @xmath174 all the @xmath198 become time independent , we deduce that @xmath199 in this limit , in which case all moments for @xmath200 tend to zero , for all initial conditions . next , we note that @xmath201 using eq . ( [ eq:112 ] ) , such that @xmath202 , and imposing the initial condition we obtain @xmath203 for all @xmath4 . since @xmath204 as @xmath17 , due to the vanishing of moments with @xmath200 , we conclude that the stochastic dynamics of @xmath53 give rise to @xmath205 in the late time limit . we shall use this result to check the outcomes of numerical calculations in a later section . the result makes sense because the mean population arising from a poisson distribution with mean @xmath42 at @xmath96 is equal to @xmath206 . intuitively , all initial situations sampled from such a distribution lead to a final population of unity as @xmath17 except for the case of a initial population of zero , the probability of which is @xmath207 . the probability of an initial situation that leads to a final population of unity is therefore @xmath208 . the mean population as @xmath17 is @xmath209 . we can confirm this result by considering kinetics starting from a definite initial population @xmath39 . from eq . ( [ eq:131 ] ) we write @xmath210 demonstrating that the asymptotic mean population is unity , unless @xmath211 in which case it is zero . we can determine the asymptotic behaviour of a system of coagulating species in which there is also an injection rate @xmath30 . the underlying master equation is @xmath212 with the first term deleted for the case @xmath213 . the evolution of the corresponding pseudo - population is given by @xmath214 this time we find that @xmath215 so that @xmath216 and @xmath217 , with the proportionality factor depending on the initial condition . we also have @xmath218 which in a stationary state implies , for @xmath219 , that @xmath220 this is reminiscent of a recurrence relation @xcite for modified bessel functions , @xmath221 , when written in the form @xmath222 , which suggests that @xmath223 for @xmath224 . since with @xmath225 we have @xmath226 , the proportionality factor is @xmath227^{-1}$ ] . notice that we do not label these stationary averages according to @xmath97 since memory of the initial condition is lost for this case . in the final stationary state of a source enhanced coagulating system we therefore expect the mean population to be @xmath228 which is similar to analysis performed elsewhere @xcite . note the limit of this result when @xmath229 does not correspond to eq . ( [ eq:167a ] ) since dependence on the initial condition is retained when @xmath230 . we conclude our consideration of analytical approaches by noting that the coagulation problem can be studied in a variety of ways , and that introducing complex stochastically evolving pseudo - populations , averaged over the noise and over a complex contour of initial values ( or over the entire plane ) , might seem very elaborate compared with the direct analytic solution to the master equations , for example . our purpose , however , is to establish explicitly that the approach works for a simple case . our attention now turns to the numerical solution of the stochastic dynamics and averaging procedure to determine whether this can be performed efficiently for the model . establishing this would suggest that a wider set of stochastic problems based on master equations might be amenable to solution using these techniques . an analytical solution to the sdes corresponding to more complicated coagulation schemes , for example those involving multiple species and modelled in the mean field approximation with rate equations such as ( [ eq:105 ] ) , is unlikely to be available . in these cases our approach is implemented through a numerical solution of the sdes for the relevant pseudo - populations @xmath231 , followed by averaging over the noise and a suitable weighting of the results over the initial values @xmath232 . we test the feasibility of this strategy for the simple @xmath0 model . we solve the sde ( [ eq:112 ] ) numerically using a c++ code . in figure [ fig : solution - to - sde ] , we plot @xmath233 against @xmath4 for @xmath234 , 2 and 3 , averaging @xmath31 over @xmath235 trajectories driven by independent realisations of the wiener process based on @xmath236 timesteps of length @xmath237 , and with @xmath195 . also shown are analytical results based on the formal series developed in section [ sub : noise - averaged - pseudo - population ] , truncated at the 12th power of @xmath97 . this representation of the analytical solution appears to be accurate for values of @xmath97 up to three . for the largest value of @xmath4 shown , the series approximates well to values @xmath238 , in agreement with the analysis in section [ sub : averaged - pseudo - population - as ] . in contrast , he 12th order series representation with @xmath239 deviates noticeably from the expected asymptotic behaviour for this range of @xmath4 , indicating a series truncation error . for small @xmath4 the outcomes resemble the mean field result eq . ( [ eq:102 ] ) , as we suggest they should in appendix a. however , the numerical results do not appear to be very satisfactory . firstly , the averages are noisy , suggesting that @xmath235 realisations are too few in number to obtain statistical accuracy , though this can of course be increased . secondly , they do not seem to possess the correct limits at large times : they all seem to tend towards unity . thirdly , they often exhibit sharp peaks , which are sometimes large enough to cause the numerical simulation to fail , as is seen in the case for @xmath240 , where a negative spike at @xmath241 causes the calculation to crash . these instabilities , even if not terminal , have a disproportionate effect on the statistics when the number of realisations is limited . such instabilities have been encountered before in simulations @xcite , and in the next section we examine their origin and consider how they can be avoided . ) , averaged over @xmath235 realisations of the noise with a timestep of @xmath237 and @xmath195 , and with initial values @xmath234 , 2 and 3 , compared with series representations of @xmath102 ( dashed lines ) including terms up to 12th order in @xmath97 , as derived in section [ sub : noise - averaged - pseudo - population ] . [ fig : solution - to - sde ] ] the origin of the large fluctuations in the value of @xmath53 can be understood by considering the deterministic contribution to the evolution of @xmath53 on the complex plane according to eq . ( [ eq:112 ] ) . this is shown schematically by block arrows in figure [ fig : drift - directions - over ] , representing the magnitude and orientation of the complex quantity @xmath242 . were it not for stochastic noise , a trajectory that started with a positive value of re@xmath31 would remain in the right hand side of the complex plane and indeed would drift asymptotically towards the origin . the picture is quite different for a negative initial value of re@xmath31 : there is a component of drift away from the origin , though in most cases the path ultimately moves into the right hand side of the plane and from there to the origin . the exception is for trajectories starting on the negative real axis , which asymptotically move towards @xmath243 . this pattern of drift has the potential for instability . of course stochastic dynamics is also driven by noise , and an example trajectory is illustrated in figure [ fig : drift - directions - over ] that for much of the time evolves in a rather well - behaved fashion over the complex plane , forming a crescent shaped trace situated to the right hand side of the origin . the noise maintains the trajectory away from the deterministic attractor at the origin . however , noise is also the undoing of this stability : if the trajectory wanders into the left hand side of the plane due to the stochastic term in eq . ( [ eq:112 ] ) , then a tendency to drift towards @xmath244 can set in . this is also shown in figure [ fig : drift - directions - over ] : at some point a fluctuation causes @xmath53 to move towards the left , away from the crescent pattern . in this example , quite a significant excursion towards negative real values occurs before noise deflects the trajectory sufficiently away from the real axis , allowing the underlying drift to take @xmath53 in an anticlockwise fashion back towards the right hand side of the plane . once there , the @xmath53 settles again into the previous pattern . the intervals spent within the crescent correspond to those periods in figure [ fig : solution - to - sde ] where fluctuations are small , whilst excursions into the left hand plane seem to be responsible for the spikes @xcite . across the complex plane brought about by the deterministic term @xmath245 in the sde ( [ eq:112 ] ) . the effect of this pattern is seen in an example realisation of the evolution . an excursion from the quasistable crescent shaped region near the origin towards negative re@xmath246 , and the return path in an anticlockwise direction , is responsible for a spike in the average of @xmath31 shown in figure [ fig : solution - to - sde ] . [ fig : drift - directions - over ] ] these instabilities are not desirable and action should be taken to eliminate them . they are less common if the timestep in the numerical simulation is reduced , since this lowers the likelihood of a noise - driven jump into the right hand side of the complex plane , but this increases the computational expense of the approach . we therefore need a mathematical scheme that can suppress the dangerous drift pattern in the stochastic dynamics while retaining the statistical properties of solutions to the sde . drummond @xcite proposed a scheme for the elimination of such instabilities that took the form of a modification to the fundamental poisson representation through the introduction of a weighting parameter @xmath247 , chosen to evolve according to @xmath248 with initial condition @xmath249 , and where @xmath250 is an arbitrary function . a new variable @xmath251 is introduced , subject to the same initial condition imposed on @xmath31 , and evolving according to @xmath252 and drummond showed that the @xmath253-weighted average of @xmath251 over the noise is the same as the corresponding average of @xmath31 : @xmath254 however , since @xmath251 and @xmath31 evolve according to different sdes , they need not suffer from the same instabilities . drummond termed this the ` gauging away ' _ _ of the instabilities of the original sde , the terminology suggested by recognising that eq . ( [ eq:171 ] ) possesses an invariance with respect to different choices of @xmath250 . the approach is in fact equivalent to a transformation of the probability measure in stochastic calculus , and can be perhaps be most easily understood as an application of the cameron - martin - girsanov formula @xcite in stochastic calculus , as we describe in more detail in appendix b. we first explore one of drummond s gauge functions @xmath255 that has the capacity to tame the instability in the coagulation problem . consider the choice @xmath256 which leads to the sdes @xmath257 and @xmath258 the crucial difference between eqs . ( [ eq:173 ] ) and ( [ eq:112 ] ) is that the drift term in the sde for @xmath251 is always directed towards the origin . it lacks the ability to produce an excursion towards @xmath259 . meanwhile , @xmath247 evolves diffusively in the complex plane , with no drift and hence @xmath260 . as long as @xmath251 is real , the increment @xmath261 vanishes . because of this feature , drummond called this choice of @xmath250 a _ minimal _ gauge function , and regarded it as a natural choice for cases where @xmath251 is expected to take real values for most of its history . we have studied this reformulation of the coagulation problem and confirmed numerically that @xmath262 does not suffer from instabilities of the kind experienced by @xmath53 , and that the @xmath253-weighted average of @xmath251 appears to agree with various analytic results expected for @xmath102 , as shown in figure [ fig :- against - time , ] . nevertheless , we notice that the statistical uncertainty in the average of @xmath247 grows as time progresses , and this introduces a decline in accuracy , for a given number of realisations . we next address the reasons for this . according to eqs . ( [ eq:173 ] ) and ( [ eq:174 ] ) , with @xmath234 , 2 and 3 , for the same timestep and number of realisations as in figure [ fig : solution - to - sde ] . the expected analytic behaviour is indicated , and in comparison with figure [ fig : solution - to - sde ] the quality of the numerical results is much improved , which is a consequence of the gauging procedure . the evolution of re@xmath263 for the case with @xmath234 is shown in the inset , to illustrate its slight deviation from the expected value of unity as @xmath4 increases , due to sampling errors . [ fig :- against - time , ] ] the diffusive behaviour of the weight function @xmath253 can be best demonstrated by considering the evolution of the mean of @xmath264 . we have @xmath265 and hence @xmath266 such that , using ito s lemma @xmath267 and so @xmath268 which indicates that the mean square modulus of @xmath253 increases monotonically , whatever choice of function @xmath250 is made . this result is illustrated schematically in figure [ fig :- evolution.- ] : the mean of @xmath253 is unity for all @xmath4 , but its increasing mean square modulus suggests that the distribution of @xmath253 spreads out . as a consequence , the extraction of the statistical properties of the problem will require an ever larger number of realisations as @xmath4 increases . this is the price to pay , it seems , for the taming of the instabilities in the original sde . over the complex plane as time proceeds . the common centre of the dark ( early time ) and lighter ( late time ) density plots lies at @xmath269 . the diffusive behaviour suggests that the statistics in the sampling of @xmath253 and hence the quality of @xmath253-weighted averages will suffer as time progresses . [ fig :- evolution.- ] ] nevertheless , the choices available to us in the gauging procedure , or equivalently the freedom to shift the probability measure according to the cameron - martin - girsanov formula , allow us to optimise the quality of the numerical estimates of @xmath270 . consider a new gauge function labelled by the real variable @xmath271 : @xmath272 which differs from eq . ( [ eq:172 ] ) by an imaginary constant . it leads to the sdes @xmath273 and @xmath274 with the property @xmath275 . for @xmath276 this has the effect of strengthening the drift towards the origin in the sde for @xmath251 . it is revealing to study the evolution of the square modulus of @xmath251 under this gauge . we write @xmath277dt\nonumber \\ & = -\kappa\vert\phi^{\prime}\vert^{2}\left(r+\vert\phi^{\prime}\vert\right)dt - i(2\kappa)^{1/2}\vert\phi^{\prime}\vert^{2}dw_{t}\nonumber \\ & \quad-\kappa\vert\phi^{\prime}\vert^{2}\left(r+\vert\phi^{\prime}\vert\right)dt+i(2\kappa)^{1/2}\vert\phi^{\prime}\vert^{2}dw_{t}+2\kappa\vert\phi^{\prime}\vert^{2}dt,\label{eq:179a}\end{aligned}\ ] ] so @xmath278 and thus @xmath279 evolves deterministically towards an asymptotic value of @xmath280 , if @xmath281 , or zero if @xmath282 . it is also revealing to consider the sde for @xmath283 explicitly : @xmath284 such that as @xmath17 the stochastic term tends towards either @xmath285 + 1\right)\omega dw_{t},\label{eq:179c2}\ ] ] if @xmath281 , or @xmath286 if @xmath282 . furthermore , the evolution of the average square modulus of @xmath253 is described by @xmath287 which tends to @xmath288 - 1\vert^{2}\vert\omega\vert^{2}\rangle dt$ ] if @xmath281 , or @xmath289 if @xmath282 . this analysis indicates that there are advantages in making specific choices of @xmath271 . if we choose @xmath290 , eq . ( [ eq:179c ] ) shows that we can eliminate the noise term controlling the asymptotic time evolution of @xmath283 . on the other hand , it is the diffusive behaviour of @xmath253 that characterises the asymptotic quality of the statistics of other quantities , so it is advisable to try to control this too . the strength of the diffusion depends on the quantity @xmath291 - 1\vert^{2}=(1-r)^{2}-2(1-r)\cos{\rm arg}(\phi^{\prime})+1 $ ] for @xmath281 , and @xmath292 for @xmath282 . the behaviour is difficult to specify in detail , but for @xmath282 it is clear that the asymptotic rate of diffusive spread of @xmath253 increases with @xmath271 , such that large @xmath271 is disadvantageous . we have explored gauges specified by values of @xmath271 between zero and two . as @xmath271 increases , the diffusive behaviour of @xmath253 becomes more marked while the noise in the evolution of @xmath283 reduces . we find that the choice @xmath293 strikes a balance by suppressing noise in the evolution of @xmath294 , at the price of more significant statistical deviations of re@xmath263 from the expected value of unity . this is illustrated in figure [ fig : efficacy ] which is to be compared with figure [ fig :- against - time , ] . the shift in gauge seems to have transferred the statistical noise from @xmath295 into @xmath263 . we suggest that @xmath293 specifies an optimised gauge for this stochastic problem . gauged according to the improved procedure described by eqs . ( [ eq:178 ] ) and ( [ eq:179 ] ) with @xmath293 , and for initial conditions @xmath234 , 2 and 3 . this should be compared with the results from eqs . ( [ eq:173 ] ) and ( [ eq:174 ] ) in figure [ fig :- against - time , ] , obtained with the same timestep and number of realisations but using @xmath296 . the evolution as @xmath4 increases is less affected by statistical noise and corresponds more closely to the expected analytical behaviour . the evolution of @xmath253 is more noisy , however , as indicated in the inset . [ fig : efficacy ] ] using the gauge function ( [ eq:177 ] ) with @xmath293 we can study further aspects of the pseudo - population dynamics that were investigated analytically in section [ sec : analytic - model - of ] . for particle coagulation starting from a poisson distribution with mean @xmath42 the averages are simply given by quantities @xmath105 . in figure [ fig : probabilities . ] we plot the evolution of a selection of state probabilities @xmath3 for @xmath297 , derived from eq . ( [ eq:78 ] ) , together with the solutions to eq . ( [ eq:100 ] ) for a similar case with the initial poisson distribution truncated at @xmath298 for convenience . the timestep and number of realisations are the same as in earlier cases . the correspondence is very good , and the noise does not significantly distort the results over the time scale for the completion of the coagulation . at @xmath96 , comparing numerical evaluations of eq . ( [ eq:78 ] ) , gauged through eq . ( [ eq:177 ] ) with @xmath293 , with solutions to the master equations ( [ eq:100]).[fig : probabilities . ] ] finally , we examine source enhanced coagulation modelled by eq . ( [ eq:72 ] ) , with the use of the optimised gauge . the quantities @xmath275 and @xmath299 are plotted for @xmath300 in figure [ fig : source - enhanced ] and they correspond well with the analytical results for late times obtained in section [ sub : source - enhanced - coagulation ] . the initial condition is a poisson distribution with @xmath301 . furthermore , to indicate that not all quantities are statistically well behaved , we also show the average of @xmath302 , which ought to equal @xmath303 ) for this case . clearly more realisations would be needed to reproduce this behaviour accurately with the prevailing choice of gauge . and @xmath304 according to the source enhanced coagulation kinetics embodied in eq . ( [ eq:72 ] ) , with @xmath300 , initial condition @xmath305 , and @xmath235 realisations with time step @xmath237 . at late times they correspond well to the stationary state values @xmath306 and @xmath307 derived in section [ sub : source - enhanced - coagulation ] . the quantity @xmath308 is also compared with the expected behaviour @xmath303 ) , to illustrate that not every variable has acceptable statistics for this number of realisations of the stochastic evolution . [ fig : source - enhanced ] ] we have outlined in this paper how to solve a simple coagulation problem using some rather elaborate analytical and numerical methods , retrieving results that were previously either known , or computable from the master equations . the effort has not been misdirected , however , since our aim has been to establish that the approach can be implemented successfully for a well characterised example problem . our real interest lies in more complicated cases of coagulation such as those with multiple species that agglomerate at a variety of rates . in such systems we expect to find behaviour that arises due to statistical fluctuations around low mean populations that can not be captured in the usual mean field formulation . we plan to address such cases in further work . specifically , we employ a poisson representation @xcite of the probability @xmath3 that there should be @xmath5 particles present at time @xmath4 in the simple case of @xmath0 kinetics . the poisson representation is a superposition of poisson distributions with complex means . the description can be cast as a problem in the stochastic dynamics of a complex pseudo - population , the average of which over the noise and the initial condition is related to the average of the physical particle population . analytical work gives a series expansion of this average in powers of the initial value of the pseudo - population , and this can be employed to recover known results @xcite . the development of the series involves the evaluation of multiple integrals of functions of the wiener process . we have also identified some exact results valid in the limit @xmath17 , and in appendix a give an approximate result for small @xmath309 that bears a resemblance to a mean field solution to the kinetics . we have discussed the evaluation of averages of general functions of the population , such as higher moments and the standard deviation . we have also evaluated moments of the pseudo - population in the stationary state of a coagulating system enhanced by a constant injection rate of new particles . analytical work can only be performed in certain circumstances , and more generally the numerical solution of the stochastic differential equations ( sdes ) is necessary , followed by averaging over noise and initial condition . unfortunately , numerical instabilities make this problematic , as has been noted previously @xcite . however , we follow drummond @xcite in identifying an sde for a pseudo - population that is free of instabilities , that is to be used together with a weighting procedure that reproduces the statistics of the desired system . in appendix b we have shown that this ` gauging ' scheme is equivalent to an adaptive shift in the probability measure for the stochastic variable in the sde , and that the weighting procedure is an application of the cameron - martin - girsanov formula . this interpretation arguably makes the gauging procedure a little more intuitive . we have identified a choice of gauge , and associated transformation of the problem , that provides the statistical information in a rather optimised fashion . with this approach , the growth with time in the statistical uncertainty that is inherent with gauging can be adequately controlled , such that we need to generate relatively few realisations of the evolution , requiring a rather slight computational effort . a number of the analytically derived results have been reproduced using this numerical approach . of course monte carlo simulation of coagulation events taking place in an evolving population of particles would be an obvious alternative numerical method for studying this system . it is readily implemented , for example using the gillespie algorithm @xcite , and has the capacity to include population fluctuation effects , allowing us to explore various processes of interest . it is an approach that is probably easier to grasp than the methods we have outlined , and more computationally efficient for the problem under consideration here . nevertheless , we believe that pseudo - population methods will prove to be more efficient than monte carlo in coagulation problems involving multiple species . since monte carlo is equivalent to a solution of the master equations , the point can be made by enumerating how many of these equations there might be . if we wish to study the evolution of clusters of monomers that have size dependent agglomeration properties , then each size must be treated as a distinct species . if we consider cluster sizes ranging from 1 to @xmath310 monomers , then the master equations will describe the evolution of probabilities @xmath311 where @xmath312 denotes the population of clusters of size @xmath29 . if the number of possible values of each population is allowed to lie between 0 and @xmath313 , then the number of elementary population distributions that must be considered is @xmath314 . it is reasonable to say that this number could be rather large and will grow faster than @xmath310 . the solution to such a large number of coupled odes , or the equivalent monte carlo simulation , would be daunting . on the other hand , the stochastic approach would require the solution to just @xmath310 sdes for pseudo - populations @xmath231 ; these are analogues of the mean populations of clusters of size @xmath29 . the sdes would bear some resemblance to the smoluchowski coagulation equations for mean populations in the mean field , fluctuation - free limit , given in eq . ( [ eq:105 ] ) . analytic solution to these equations might not be possible , but numerical solution is not difficult , especially if we have techniques such as gauging at our disposal to avoid some of the pitfalls we have identified . the drawback is that averaging of the sdes over a variety of noise histories and perhaps initial conditions is necessary . nevertheless , the task is linear in @xmath310 and must eventually become more efficient than direct solution to the master equations for large enough @xmath310 . it is only when the mean of a population becomes small that deviations from mean field behaviour emerge , and so a hybrid approach might be possible whereby mean field rate equations are used to model the early stages of coagulation , going over to pseudo - population rate equations when the mean population becomes small . this is possible since the mean field rate equations have a close resemblance to the evolution equations of the pseudo - populations , which is an important conceptual connection . specifically , we could integrate the equations for the @xmath231 with the neglect of the noise when the modulus of @xmath231 is large , starting them off at the initial mean populations of clusters of size @xmath29 , such that they remain real , and introduce noise , and average over it , only when the modulus becomes small . this is to be investigated . in conclusion , we have made conceptual and numerical developments of a method for kinetic modelling that was introduced some decades ago @xcite but that appears not to have been fully exploited . we believe that in spite of some complexity in the formulation , the approach possesses considerable intuitive value , and that it has the capability to treat systems with interesting statistical properties for which alternative methods are inappropriate or expensive . we intend to explore these possibilities in further studies . funding from the leverhulme trust through grant f/07/134/bv is gratefully acknowledged . we have seen that if we employ the representation eq . ( [ eq:125 ] ) , the set of initial values of @xmath97 need only form a closed contour around the origin in the complex plane . we now demonstrate that for the purposes of computation there is an optimal contour , depending on the initial population in the problem @xmath39 , and that @xmath315 where @xmath316 is given by eq . ( [ eq:109 ] ) , reduces to eq . ( [ eq:102 ] ) for situations with a definite initial population , at least for small @xmath309 . contour integrals of the form @xmath317dz$ ] may be evaluated by the method of steepest descents about saddle points in @xmath318 defined by locations @xmath319 in the complex plane where @xmath320 . we represent the integrand near a saddle point as @xmath321 and integrate along a contour through @xmath322 on which the imaginary part of @xmath323 is zero and the real part is negative . if the contour is arranged such that it passes over saddle points along the line of steepest descent on the surface of the modulus of the integrand , and elsewhere follows valleys where the modulus of the integrand is small , then the contour integral may be approximated by a set of gaussian integrations about each saddle point position . we write @xmath324 and consider @xmath325 . in this limit @xmath326 , according to eq . ( [ eq:142 ] ) , and hence @xmath327 such that @xmath328 . inserting this into the exponent in ( [ eq : a2 ] ) , we seek to identify saddle points in the remaining part , namely @xmath329 . solving @xmath330 identifies a single saddle point of the integrand at a location @xmath331 . the modulus of @xmath332 has a minimum at this point and the real axis is the path of steepest ascent . the path of steepest descent passes through the saddle point perpendicular to the real axis . the modulus of the integrand , for @xmath333 , is shown in figure [ fig : modulus - of- ] . for @xmath333 , indicating the saddle point at @xmath239 : the path of steepest descent through this point lies parallel to the imaginary axis.[fig : modulus - of- ] ] clearly , it is sensible to place the contour @xmath37 along the path of steepest descent through the saddle point and then to complete the circuit around the origin through regions where the modulus of the integrand is as small as possible . the contour integral can then be approximated by the contribution along the straight line parallel to the imaginary axis through @xmath334 writing @xmath335 and with @xmath336 we have @xmath337 using stirling s approximation @xmath338 which is quite accurate even for @xmath39 as low as unity . finally , the expansion ( [ eq:128 ] ) with the @xmath339 valid for @xmath325 may be written @xmath340 making eq . ( [ eq : a3 ] ) and hence eq . ( [ eq : a2 ] ) consistent with the mean field approximation @xmath341 for small @xmath309 . the equivalence between the statistical properties of the gauged stochastic variable @xmath251 , when suitably weighted , and those of the ungauged variable @xmath31 , derived by drummond @xcite and explored in section [ sub : elimination - of - instabilities ] , is a consequence of some fundamental rules in stochastic calculus that are expressed by the cameron - martin - girsanov formula @xcite . an sde such as @xmath342 is a statement of a connection between @xmath343 and a stochastic variable @xmath48 with certain statistical properties . when we introduce expectation values such as @xmath344 we are implicitly defining a probability distribution or _ _ measure _ _ over the values taken by the variable @xmath48 . normally the notation @xmath48 represents an increment in a wiener process , the continuum limit of a symmetric 1-d random walk in discrete space and time , and the implication is that the probability distribution of values of @xmath48 is gaussian with zero mean and variance equal to @xmath345 . but how might the expectation value of the increment @xmath48 change if we were to evaluate it with respect to a different probability distribution ? for example , what if it were distributed according to a shifted gaussian proportional to @xmath346 where @xmath347 is the non - zero mean of @xmath48 making it no longer correspond to a wiener process ? let us take the drift in the mean of @xmath48 under such a shifted gaussian distribution to be proportional to the time elapsed , so we may write @xmath348 and @xmath349 . the new probability distribution , based on an asymmetric random walk and denoted @xmath350 , is indicated through a suffix on the expectation value . we write @xmath351 in the old measure , which we denote @xmath352 , and according to which @xmath48 is indeed an increment in a wiener process . note that @xmath353 and @xmath354 since the variance of @xmath48 under measure @xmath350 is the same as that under @xmath352 : we have shifted the gaussian probability distribution for @xmath48 but not changed its width . hence , if we define @xmath355 then @xmath356 and @xmath357 : we can identify a wiener process that operates under probability measure @xmath350 , and relate it to a wiener process under measure @xmath352 . the sde for @xmath66 now reads @xmath358 and we can choose to evaluate expectation values under measure @xmath359 for which @xmath360 , or measure @xmath361 for which @xmath356 . the expectation value @xmath362 is the solution to a problem that differs from the one initially posed , since the drift term in the sde has been changed from @xmath363 to @xmath364 . however , the point is that it is possible to establish a link between expectation values under the two different measures . the average of @xmath343 under measure @xmath352 is equal to a _ weighted _ average of @xmath343 under @xmath350 . formally , we can write @xmath365 where @xmath366 is the radon - nikodym derivative of probability measure @xmath352 with respect to @xmath350 . furthermore , the cameron - martin - girsanov formula states that we can write @xmath367 which is just a ratio of the gaussian distributions @xmath368 $ ] and @xmath369 $ ] . we now define a quantity @xmath253 through the relation @xmath370 such that @xmath371 and using ito s lemma , we show that @xmath253 evolves according to @xmath372 we see from eqs . ( [ eq:303]-[eq:307 ] ) that averages of the increment @xmath343 over differently distributed random increments can be related to one another , and that the quantity @xmath261 describes the connection . in order for this to make sense mathematically , two conditions must be met . the first is that no process that has non - zero probability under @xmath352 should be impossible under @xmath350 , and vice versa . the second , known as novikov s condition , requires that @xmath373 . we now invert the point of view , and instead of considering a single variable treated according to two different averaging procedures , we relate the evolution of two different variables under the same averaging . explicitly , we consider variable @xmath66 evolving as @xmath374 and another variable @xmath375 that evolves according to @xmath376 , with @xmath377 . the above results imply that we can write @xmath378 which resembles a combination of eqs . ( [ eq:303 ] ) and ( [ eq:304 ] ) . we shall demonstrate its validity shortly . note that there is no need to indicate the probability measure on the brackets , or further label the increment @xmath48 , since here it is a standard wiener increment with zero mean in the probability distribution under consideration . the inserted factor accounts for the difference in drift in the evolution of @xmath66 and @xmath375 and may also be written @xmath379 with @xmath380 or equivalently @xmath381 . now consider the following : @xmath387 where we use a discrete representation of the time integration . we have recognised that if @xmath388 then @xmath389 such that by repetition of this step , the expression reduces to @xmath390dx_{j}^{\prime}\right\rangle $ ] . now , since @xmath391 we can write @xmath392 and with the insertion of @xmath376 , we conclude that @xmath393 such that @xmath394 and since @xmath395 this means that @xmath396 thus we have shown that if we wish to evaluate the quantity @xmath397 generated by sde @xmath374 , we could instead solve the sdes @xmath398 with initial conditions @xmath399 , @xmath249 , and with the function @xmath400 taking arbitrary form subject to novikov s condition , and then use the results to evaluate the equivalent quantity @xmath401 . clearly , this is identical to drummond s gauging scheme . 41ifxundefined [ 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 @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1002/aic.690420612 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) \doibase 10.1051/jphys:019850046090146900 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1007/bf01014349 [ * * , ( ) ] @noop _ _ ( , ) link:\doibase 10.1140/epjb / e2004 - 00157 - 2 [ * * , ( ) ] link:\doibase 10.1021/jp104471e [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) \doibase 10.1051/0004 - 6361:20010961 [ * * , ( ) ] @noop ( ) @noop _ _ ( , ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , )

the large hadron collider is shown to have great scope for a light charged higgs discovery , in the context of the-violating type - i two higgs doublet model . this scenario with similar masses of and was suggested by the puzzling departure from charged current lepton universality found in the lep data . with the lightest neutral higgs mass set to , the charged - neutral higgs associated production mechanism can cause a significant excess in the events over a vast range of as long as the higgs mixing pattern avoids a few limiting cases . thanks to the low mass , the charged higgs loop can play a striking role in neutral higgs decays into , thereby compensating for a suppressed gluon - gluon fusion rate . the effect of scalar pseudo - scalar mixing on loop - induced higgs signals is also discussed .

1205.2692

there are many motivations to introduce an extended higgs sector . examples are , including obsolete ones , fine - tuning in the weak - scale higgs mass @xcite suggesting supersymmetry @xcite , @xmath0-violation @xcite , grand unification @xcite , extended gauge symmetry @xcite , strong @xmath0 problem @xcite , string theory @xcite , vacuum stability @xcite , light fermion mass / weak - scale hierarchy @xcite , fermion flavour structure @xcite , neutrino mass @xcite , dark matter ( see e.g. @xcite ) , cosmic rays @xcite , baryogenesis @xcite , inflation @xcite , novel collider phenomenology @xcite , as well as anomalies seen in precision ( see e.g. @xcite ) and accelerator ( see e.g. @xcite ) experiments . in this article , we entertain the possibility that an intriguing outcome from the lep experiment is in fact pointing to an extended higgs sector , which may be directly inspected at the large hadron collider ( lhc ) . specifically , we take as our framework the two higgs doublet model ( 2hdm ) of type - i with @xmath7 in the vicinity of the @xmath2-boson mass . the two main points of this paper are the following . ( 1 ) the charged higgs produced in association with the lightest neutral higgs can be discovered as an excess in the @xmath8 process at the lhc . moreover , this could happen even at the early stage with @xmath9 in a substantial portion of the parameter space . ( 2 ) with the lightest neutral higgs mass assumed to be @xmath10 , the charged higgs one - loop effect can enhance @xmath11 enough to produce the excess observed at the recent lhc analyses @xcite . this enhancement is particularly welcome since the @xmath12 production at hadron colliders is generically suppressed due to higgs mixing . this light charged higgs scenario was previously advocated in @xcite , whose contents we recap here . the lep elecroweak working group has performed an analysis of @xmath2 decay branching ratios . in their report , @xmath13 appears to be higher than @xmath14 and @xmath15 , to the level of 2.8 standard deviations @xcite . should this difference be real , it would violate the charged current lepton universality , that is predicted by the gauge invariance of the standard model ( sm ) and has been confirmed in many indirect tests . as a resolution of this puzzle , one of the authors pointed out that a significant portion of the apparent lepton non - universality can be attributed to production of light charged higgs pairs which predominantly decay to the @xmath16 final states @xcite . one of the most natural ways to realise a @xmath1 that has mass @xmath17 and is thus light enough to be on - shell produced at lep , is to employ the type - i two higgs doublet model ( 2hdmi ) . in this model , all the @xmath1fermion couplings are suppressed like @xmath18 , the ratio of the two vacuum expectation values ( vevs ) . with the choice of a high enough @xmath5 , this enables @xmath1 to escape from the severe flavour changing neutral current ( fcnc ) constraints as well as any process that probes the @xmath1fermion interaction such as @xmath19 , @xmath20 , and @xmath21 . presuming the above lep anomaly to be real , the following points make this scenario more appealing . first , invoking a light charge higgs ( discussed also in @xcite ) is the only proposed solution that does not spoil the other precision lepton universality tests using lepton or meson decays @xcite . second , the requirement of a light charged higgs plus @xmath22 singles out type - i out of the four types of 2hdms @xcite in which tree - level fcnc is forbidden by a @xmath23 symmetry @xcite . note that these assumptions already fix many aspects of the model that otherwise offer more than one option , i.e.the yukawa coupling structure , the charged higgs mass , and the viable range of @xmath5 . these combine to predict a distinct set of collider physics signatures . in this paper , we shall consider @xmath0-violation in the higgs sector , which can in general occur in a class of 2hdms . in this way , we can explore higgs mixing patterns which received relatively less attention . a practical advantage is that one can analyse both the scalar and the pseudo - scalar higgses in a unified manner : switching between the two sectors reduces to tuning the higgs mixing angles , which determine the composition of each neutral mass eigenstate . within the context of the minimal supersymmetric standard model ( mssm ) , there are earlier studies of @xmath0-violating higgs phenomenology . in particular , the general @xmath0-violating potential and mass spectrum can be found in the paper of a. pilaftsis and c. wagner @xcite , where the importance of the scalar pseudo - scalar transitions in higgs phenomenology was emphasised . in the same article , the couplings of the charged higgs boson to @xmath0-violating higgs bosons were also discussed . this article is organised as follows . we begin by giving an overview of the model in the next section . section [ sec : strategy ] is a summary of our search strategy , which leads to the main phenomenological results presented in section [ sec : res ] . section [ sec : h1 ] is devoted to the diphoton signal of the lightest neutral higgs . we deliver our conclusion in section [ sec : conclusions ] . in addition , we present : the ghost lagrangian for our setup of 2hdm in appendix [ appe : b ] , the relation between our parametrisation and perturbativity in appendix [ appe : perturbativity ] , and differential cross - sections of our main process in appendix [ appe : diff ] . let us give a brief description of the type - i 2hdm that we have chosen . the type of a 2hdm refers to one of the multiple ways to organise the yukawa couplings . there are four types in which quarks or leptons of each charge couple to only one higgs doublet @xcite , so that neutral higgses are naturally prevented from mediating fcnc @xcite . to this end , we impose the softly broken @xmath23 symmetry , under which the two higgs doublets @xmath24 transform as @xmath25 whereas the fermion fields remain unchanged . this allows only the following yukawa couplings , @xmath26 between @xmath27 and fermions , so all the fermions acquire mass from the vev of @xmath27 @xcite . the scalar potential including the soft @xmath23-breaking terms reads @xmath28 \\ & -\frac{1}{2}\left\{m_{11}^2(\phi_1^\dagger\phi_1 ) \!+\!\left[m_{12}^2 ( \phi_1^\dagger\phi_2)\!+\!{\rm h.c.}\right ] \!+\!m_{22}^2(\phi_2^\dagger\phi_2)\right\}. \end{split}\ ] ] in unitary gauge , the higgs doublet fields are expanded around the minimum like @xmath29 , \quad \phi_2 = \left [ \begin{array}{c } -i { \cos\!\beta}h^+ \\ \frac{1}{\sqrt{2 } } { ( v_2 + \eta_2 + i { \cos\!\beta}\,\eta_3 ) } \end{array } \right ] .\ ] ] one can take a basis of @xmath24 such that @xmath30 and @xmath31 are both real . their quadrature sum is subject to the constraint , @xmath32 , where @xmath33 is the sm higgs vev . their ratio is parametrised by @xmath34 in terms of the pseudo - scalar and charged higgs mixing angle @xmath35 . we introduce @xmath0-violation in the higgs sector by assuming @xmath36 and @xmath37 , which enables the neutral scalar and pseudo - scalar components to mix together . the real symmetric mass - squared matrix @xmath38 with respect to the basis @xmath39 , is diagonalised by the mass eigenstates @xmath40 like @xmath41 using a @xmath42 orthogonal matrix @xmath43 . it is parametrised in terms of the mixing angles @xmath44 in the form , @xmath45 where @xmath46 and @xmath47 . the neutral higgs mass eigenvalues are assumed to be in the order , @xmath48 this determines the following physical domain of the mixing angles @xcite : @xmath49 the two @xmath0-violating mass matrix elements are related by @xcite @xmath50 which translates into the constraint , @xmath51 we shall remain in the parameter volume that is compatible with the mass ordering ( [ eq : mass ordering ] ) in conjunction with ( [ eq : m3 ] ) , in order to avoid double counting of physically identical parameter sets . two more dimensionful quantities can be derived from the scalar potential : the charged higgs mass , @xmath52 and the auxiliary parameter , @xmath53 that sets the mass scale of charged / neutral higgs particles except @xmath12 . we have the eight physical quantities determined from the higgs potential , @xmath54 in this work , we opt to invert this dependency and express the parameters appearing in ( [ eq : pot ] ) in terms of those in ( [ eq : input ] ) @xcite . the mass parameters are then given by @xmath55 with @xmath56 and @xmath57 fixed by ( [ eq : mu ] ) . the quartic couplings are [ eq : lambda ] @xmath58 , \end{split } \\ \begin{split } \lambda_2&=\frac{1}{s_\beta^2v^2 } [ s_1 ^ 2c_2 ^ 2m_1 ^ 2 + ( c_1c_3-s_1s_2s_3)^2m_2 ^ 2 \\ & + ( c_1s_3+s_1s_2c_3)^2m_3 ^ 2-c_\beta^2\mu^2 ] , \end{split } \\ \begin{split } \lambda_3&=\frac{1}{c_\beta s_\beta v^2 } \{c_1s_1[c_2 ^ 2m_1 ^ 2+(s_2 ^ 2s_3 ^ 2-c_3 ^ 2)m_2 ^ 2 \\ & + ( s_2 ^ 2c_3 ^ 2-s_3 ^ 2)m_3 ^ 2 ] + s_2c_3s_3(c_1 ^ 2-s_1 ^ 2)(m_3 ^ 2-m_2 ^ 2)\ } \\ & + \frac{1}{v^2}(2m_{h^\pm}^2-\mu^2 ) , \end{split } \\ \lambda_4&=\frac{1}{v^2 } [ s_2 ^ 2m_1 ^ 2+c_2 ^ 2s_3 ^ 2m_2 ^ 2+c_2 ^ 2c_3 ^ 2m_3 ^ 2 + \mu^2 - 2m_{h^\pm}^2 ] , \\ { \mathrm{re}}\lambda_5&=\frac{1}{v^2 } ( -s_2 ^ 2m_1 ^ 2-c_2 ^ 2s_3 ^ 2m_2 ^ 2-c_2 ^ 2c_3 ^ 2m_3 ^ 2+\mu^2 ) , \\ \begin{split } \label{eq : lambda5i } { \mathrm{im}}\lambda_5&=\frac{-1}{c_\beta s_\beta v^2 } \{c_\beta[c_1c_2s_2m_1 ^ 2-c_2s_3(c_1s_2s_3+s_1c_3)m_2 ^ 2 \\ & + c_2c_3(s_1s_3-c_1s_2c_3)m_3 ^ 2 ] + s_\beta[s_1c_2s_2m_1 ^ 2 \\ & + c_2s_3(c_1c_3\!-\!s_1s_2s_3)m_2 ^ 2 \!-\!c_2c_3(c_1s_3\!+\!s_1s_2c_3)m_3 ^ 2]\ } , \end{split}\end{aligned}\ ] ] where @xmath59 , @xmath60 . in this approach , the solutions for @xmath61 may turn out to be non - perturbatively large , unless one takes care to set the input parameters in ( [ eq : input ] ) to sensible values . to avoid entering a nonsensical regime , we shall check that the sizes of the higgs quartic couplings meet the conditions , @xmath62 ) , within the parameter space explored in this work . the interactions of higgs bosons with other particles are described by the lagrangian , @xmath63 with @xmath64 from ( [ eq : yukawas ] ) . .feynman rules . the momentum directions are indicated in figure [ fig : hhc via w ] . the chirality projection matrices are defined by @xmath65 . [ cols="^,^ " , ] one expands the covariant derivatives , replaces @xmath24 by ( [ eq : higgs expansion ] ) , and diagonalises the fermion and higgs mass terms , to arrive at the feynman rules in table [ tab : feynman rules ] . they will be used to calculate and interpret the results that we present in the following sections . a remark is in order regarding the parametrisation in the @xmath0-conserving limit . one can enforce @xmath0-conservation simply by setting @xmath66 . this eliminates mixing between the @xmath0-odd and the @xmath0-even degrees - of - freedom , as can be seen in ( [ eq : rotation matrix ] ) . the mixing between @xmath67 and @xmath68 with the angle @xmath69 , results in the lighter and the heavier @xmath0-even neutral eigenstates , @xmath70 and @xmath71 . the @xmath0-odd neutral higgs @xmath72 is identified with @xmath73 . in this case , the relation ( [ eq : rel m13-m23 ] ) becomes trivial and therefore it no longer constrains @xmath74 to be determined as a function of the other parameters . as a result , we are left with two less mixing angles and one more free mass parameter in the @xmath0-conserving higgs sector than in the @xmath0-violating case . conventionally , these seven free parameters are chosen to be : @xmath75 @xcite . having established the framework , one should fix the range of parameters to study . for future references , we first collect the values which were chosen for the reasons explained below : @xmath76 this set shall be used by default throughout the analysis unless specified otherwise , in combination with one of the two higgs mixing configurations , @xmath77 the charged higgs mass is more or less predetermined by the lep data . the lower bound is provided by the direct search , and the upper limit comes from the requirement that the @xmath1 pair production should restore lepton universality in leptonic @xmath2 decays . we take @xmath7 which reduces the apparent discrepancy between @xmath78 and the average of @xmath79 and @xmath80 by 1 @xmath81 @xcite . the latest update from the opal collaboration reports the direct search limit , @xmath82 , for @xmath83 at 95% confidence level ( cl ) @xcite . in view of the recent evidences from atlas @xcite and cms @xcite , we set the lightest neutral higgs mass to @xmath3 . the heavier higgs mass parameter is taken to be fairly higher than @xmath84 . non - degenerate mass spectrum is needed for @xmath0-violating higgs mixing . the last dimensionful parameter @xmath85 is set to a representative value of the weak scale order . the chosen @xmath86 is a mass at which the sm higgs production is strongly constrained at the lhc . not far from there lies also @xmath87 . in this range , the rates of the @xmath88 and @xmath89 channels are limited to be roughly below half their sm predictions @xcite . in our model , the upper bounds must be reinterpreted as arising from processes mediated by @xmath90 and @xmath91 whose interactions are different from those of the sm higgs . for this , we require that @xmath92 a potentially important constraint is provided by the @xmath93 parameter @xcite . with the higgs mass spectrum assumed in this work , one can check that the one - loop correction from the higgs bosons to @xmath93 is small enough to pass the electroweak precision tests @xcite . for safety of a light charged higgs , it is crucial to choose a high enough @xmath5 . the strongest bound is given by @xmath94 , which requires @xmath95 @xcite . this range of @xmath5 also suppresses @xmath96 below 4% . this maximal top - quark branching fraction coincides roughly with the numerical values of the 95% cl limits from the light charged higgs searches at the lhc @xcite . however , these lhc constraints on @xmath97 are based on the assumption that @xmath98 , and therefore become weaker in our setup where @xmath99 . in principle , one could use any @xmath5 larger than the fcnc bound , as long as it lets the charged higgs decay inside the detector . in practice , it is bounded from above by perturbativity through ( [ eq : lambda ] ) . obviously , this upper limit is a function of the other parameters . in appendix [ appe : perturbativity ] , it is shown that @xmath85 can always be adjusted so that perturbativity holds . one should not regard this as a fine - tuning but rather an artifact of the employed parametrisation . if one started by setting the parameters appearing in ( [ eq : pot ] ) and then derived the quantities in ( [ eq : input ] ) , then one could trivially satisfy perturbativity by keeping @xmath61 inside the range ( [ eq : perturbativity ] ) . even in this reverse approach , one could naturally obtain a high @xmath5 , as one can see from the example of inert doublet model where an exact @xmath23 symmetry leads to infinite @xmath5 @xcite . among the three remaining higgs mixing angles , we scan @xmath100 and @xmath101 which determine the composition of the lightest neutral higgs . in ( [ eq : rotation matrix ] ) , one can notice that @xmath102 is the fraction of the @xmath0-odd component in @xmath12 . we fix @xmath103 to @xmath104 to have an intermediate level of @xmath0-violation . for illustrative purposes , we select two benchmark points , labelled @xmath105 and @xmath106 in ( [ eq : benchmark points ] ) . the fraction of @xmath72 in @xmath12 is chosen to be small at @xmath105 and large at @xmath106 . given the model and the input parameters , we consider which charged higgs production mechanism is most useful . we first calculate the cross - sections of the following standard channels : @xmath107 the last two entries are meant to imply their conjugate processes as well . the final state in ( [ eq : tbhc ] ) can be made through either top pair production or gluon - gluon fusion @xcite . + in figures [ fig : sp ] , we plot the cross - sections against @xmath5 for the two benchmark points . one observes the general tendency that each production rate is depleted for high @xmath5 . as one can expect from table [ tab : feynman rules ] , processes ( [ eq : thc ] ) and ( [ eq : tbhc ] ) die out as @xmath5 grows , while the others approach their individual asymptotic cross - sections even though many of them are small . ( 180,120)(0,-60 ) ( 25,35)(50,0 ) ( 25,-35)(50,0 ) ( 50,0)(100,0)34.5 ( 140,40)(100,0)4 ( 100,0)(140,-40)4 ( 75,8)[b]@xmath2 ( 127,12)[l]@xmath108 ( 127,-12)[l]@xmath109 ( 148,40)[l]@xmath110 ( 148,-40)[l]@xmath111 ( 17,35)[r]@xmath112 ( 17,-35)[r]@xmath113 an outstanding exception is the production associated with a neutral higgs ( [ eq : hhc ] ) , which is not suppressed even for high @xmath5 . the diagrams for this process are shown in figure [ fig : hhc via w ] , which scale according to the @xmath114@xmath1@xmath2 vertices in table [ tab : feynman rules ] . in the @xmath12@xmath1@xmath2 interaction , the pseudo - scalar coupling is proportional to @xmath102 and independent of @xmath35 , and the scalar coupling is proportional to @xmath115 in the limit of @xmath116 . combining these two behaviours , one can expect that the @xmath117 production becomes efficient unless @xmath118 is large and @xmath119 is small . the results plotted in figures [ fig : sp ] do not depend on @xmath85 . therefore , they remain valid even if one adjusts @xmath85 for the perturbativity of @xmath61 in the high @xmath5 regime . this means that the production mechanism ( [ eq : hhc ] ) is of the most general interest among the displayed channels , and we shall focus on it in what follows . note that the same type of diagram was previously considered in the contexts of the mssm @xcite and a fermiophobic higgs scenario @xcite . the next step should be to select the decay products of @xmath1 and @xmath12 . the two dominant branching ratios of @xmath111 are @xmath120 as long as @xmath121 , beyond which the charged higgs decays mediated by a virtual top becomes non - negligible @xcite . these branching fractions are essentially independent of any of the input parameters . this is a notable feature of the 2hdmi , stemming from the universal scaling behaviour of the @xmath1fermion couplings : they are all proportional to @xmath122 , as shown in table [ tab : feynman rules ] . as we are mainly interested in a charged higgs that is light enough to have been produced at lep energies , we can regard ( [ eq : brs of h+ ] ) as good approximations and simply choose the decay product with the highest rate , @xmath16 . + as for @xmath12 , we plot its branching fractions in the vicinity of @xmath123 , in figures [ fig : brh1 ] . one finds that the @xmath124 mode is dominant at both benchmark points . this tends to be the case unless both @xmath125 and @xmath102 are vanishingly small . at the point where @xmath126 , the tree - level @xmath12fermion coupling vanishes , causing its fermiophobia . considering these dominant decay modes of @xmath1 and @xmath12 , we are going to concentrate on the final state , @xmath4 for numerical simulation . one can also notice that the branching fractions of the @xmath6 and @xmath127 modes can differ significantly from their individual sm values . albeit not directly related to the charged higgs search , these changes greatly affect the phenomenology of @xmath12 . we shall come back to this issue in section [ sec : h1 ] . the main process of our concern is @xmath128 , followed by the decays , @xmath129 and @xmath130 at the lhc . among these three subprocesses , the charged higgs production and decay revealed their outstanding stability under variation of @xmath5 over a wide range , in the preceding section . we shall envisage two experimental conditions at the lhc , which we call the `` early '' and the `` full '' searches . the `` early search '' denotes the centre - of - mass energy of the current run , @xmath131 , and the integrated luminosity of @xmath132 . we assume that the `` full search '' will deliver @xmath133 at @xmath134 . the kinematics of the @xmath4 state can be fully reconstructed with high efficiency as it has only one neutrino . we assume a @xmath135-reconstruction efficiency @xmath136 and a @xmath137-tagging efficiency @xmath138 , plus the following set of selection cuts : @xmath139 which we choose considering the operation of atlas . assuming these cuts , we calculated the sm background . the cross - section is @xmath140 at @xmath131 and @xmath141 at @xmath134 . along with the associated production of @xmath142 , we also study a similar process with @xmath1 replaced by @xmath2 , i.e.@xmath143 . as we shall see , the parameter space covered by this higgs - strahlung is complementary to the region accessible to the @xmath142 state . we select the final state , @xmath144 with @xmath145 , and assume a @xmath2-reconstruction efficiency @xmath136 . the cuts in ( [ eq : cuts ] ) are used again , and the same cut on @xmath135 is applied on the transverse momenta of the light leptons , @xmath146 . using these parameters , we obtain the higgs - less sm background cross - section to be @xmath147 at @xmath131 and @xmath148 at @xmath134 . we used lanhep @xcite to implement the model , and calchep @xcite plus the cteq6 m parton distribution functions @xcite , to perform the phenomenological analysis . the passarino - veltman functions appearing in the effective vertices of @xmath114@xmath149@xmath149 and @xmath114@xmath150@xmath150 are calculated by calling the looptools package @xcite . we first check the rates of our main processes with @xmath151 . in this @xmath0-conserving case , one has the lighter and the heavier @xmath0-even higgses , @xmath152 and @xmath153 , and the @xmath0-odd higgs , @xmath154 . the production cross - sections of @xmath70 , @xmath71 , and @xmath72 , each in association with a charged higgs , are shown in figure [ fig : cpc xs vs angles ] as functions of @xmath155 . for each type of neutral higgs , there are three curves that respectively represent @xmath156 , @xmath111 , and @xmath1 productions in the order of increasing thickness . the higher @xmath111 rate relative to @xmath156 reflects the fact that the lhc is a @xmath157 collider , not @xmath158 . the shape of each curve is essentially determined by the dependence of the corresponding @xmath2@xmath1@xmath114 coupling on @xmath100 and @xmath35 , presented in table [ tab : feynman rules ] . these patterns will be useful for understanding the simulation results to follow . from the plot , one can expect @xmath159 events that occur through a charged - neutral higgs associated production with the `` full luminosity '' at @xmath160 , unless the mixing angle is such that the given channel is highly suppressed . our analysis is of relevance to a light charged higgs , even if one is not interested in a scenario where charged higgs bosons have been on - shell produced at lep energies . in this regard , it is legitimate to consider changes that would be caused by varying @xmath7 . obviously , the above production rates will be reduced as @xmath7 grows , as is shown in figure [ fig : cpc xs vs mhc ] . we take the range of @xmath7 up to @xmath161 , beyond which the changed higgs decay pattern undergoes a qualitative change @xcite , thereby invalidating our assumptions given in ( [ eq : brs of h+ ] ) . for @xmath162 , each cross - section decreases roughly down to half the value in panel [ fig : cpc xs vs angles ] . these curves illustrate the phase - space suppression that takes place for higher @xmath7 . we emphasise that this reduction of statistics is the only essential information that one needs to understand how the significances would change in the following quantitative analysis with @xmath163 , if one took a different @xmath7 . for the simulation of charged higgs search , we turn on @xmath0-violation allowing for mixing among all the three neutral higgses , @xmath70 , @xmath71 , and @xmath72 . in figures [ fig : significance cpv ] , the contours of the significance @xmath164 are shown for each process , which have been evaluated by means of the method described in @xcite . there are two complications in comparison to the @xmath0-conserving limit . one is that the additional mixing between the @xmath0-odd and the @xmath0-even components affects the significance of each process . the other is that @xmath165 are no longer independent of one another but constrained by ( [ eq : m3 ] ) . in combination with the latter relation , the mass ordering ( [ eq : mass ordering ] ) excludes the upper - right and the lower - left rectangular regions from the parameter domain . another theoretical constraint that rules out the upper - left and the lower - right triangular regions , is perturbativity ( [ eq : perturbativity ] ) . we have also checked numerically that imposing perturbativity guarantees the tree - level unitarity in higgs - higgs scattering in the parameter space under consideration . in addition to the above theoretical requirements , we impose the experimental constraints in ( [ eq : h2 exclusion ] ) in order to respect the current higgs production limits from the lhc . this rules out the dark - shadowed strip on the plane . the bounds on @xmath90 and @xmath91 apply as long as their masses are in the range between 160 and 600 gev . all these conditions being fulfilled , the question to ask is how much portion of the remaining unshaded region the lhc can cover . as in the preceding plot , the @xmath2@xmath1@xmath12 coupling largely determines the shape of the @xmath142 contours . as long as one moves along the horizontal axis , @xmath12 is purely @xmath0-even , and therefore its interactions not involving other neutral higgses reduce to those of @xmath70 in the @xmath0-conserving higgs sector . and @xmath91 do not reduce to those of @xmath71 and @xmath72 due to non - vanishing @xmath103 . ] for instance , the @xmath2@xmath1@xmath12 coupling vanishes at the point @xmath166 , as is the case in figure [ fig : cpc xs vs angles ] . this explains why the significance decreases as one approaches this point . it should be meaningful to interpret the interesting behaviour of @xmath164 around the point @xmath167 , although it belongs to the @xmath168 exclusion strip . for instance , there might be a way to circumvent the constraint ( [ eq : h2 exclusion ] ) by taking a different set of parameters . in the neighbourhood of this point , one finds a sudden drop of significance even though the @xmath2@xmath1@xmath12 coupling stays nearly maximal . this is because the @xmath12@xmath137@xmath169 coupling is suppressed there as can be seen in table [ tab : feynman rules ] . at the origin , @xmath12 becomes fermiophobic , and its dominant decay modes are @xmath170 and @xmath171 . the fermiophobia of @xmath12 generically gives rise to an enhanced @xmath172 without the charged higgs loop contribution . including the @xmath1-loop , the diphoton branching fraction can range from zero up to @xmath173 . [ we give a related discussion in section [ sec : h1 ] . ] therefore , the possibility of observing the @xmath142 state in the central low - significance hole via an alternative decay channel of @xmath12 such as the @xmath6 mode , is highly dependent on the other model parameters . as one departs away from the horizontal axis , @xmath174 , @xmath12 acquires more of the @xmath0-odd component . this reinforces the production rate of @xmath142 and thereby the significance , since the @xmath2@xmath1@xmath72 coupling is not suppressed for any @xmath100 as can be seen in figure [ fig : cpc xs vs angles ] . overall , figures [ fig : significance cpv ] demonstrate that there is an ample portion of the parameter space in which the lhc has a good chance to discover the @xmath142 production . obviously , the chance is substantially better in the `` full luminosity '' case , shown in figure [ overlap_14tev ] . with the exception of the narrow corners around both ends of the horizontal axis and the hole encircling the fermiophobic point , the discovery of @xmath142 through the @xmath4 final state should be possible at any point in the unshaded physically sensible domain defined by ( [ eq : mass ordering ] ) , ( [ eq : perturbativity ] ) , and ( [ eq : h2 exclusion ] ) . in the same figures is presented the @xmath12-strahlung process , which could also be detected depending on the mixing angles and the lhc operation condition . one can understand the shape of the @xmath175 contours by looking at the @xmath12@xmath176@xmath177 coupling in table [ tab : feynman rules ] . in figure [ overlap_07tev ] for `` early searches '' , one finds no common portion of the parameter space with @xmath178 for both processes . this setup predicts that it is not possible to discover both channels during the first data taking period in the @xmath179 final states . should this occur , this scenario will have to be modified to survive . in figure [ overlap_14tev ] , one finds a much broader region in which @xmath12-strahlung could be discovered . part of this region is shadowed in light grey where the `` full lhc '' could observe both this process and @xmath142 production . this section is devoted to phenomenology of the lightest neutral higgs . in particular , we show that the charged higgs loop can exert a marked influence on the @xmath181 decay mode , which is of the utmost importance in the neutral higgs search at the lhc . we shall consider the following standard higgs production mechanisms : @xmath182 in figures [ fig : np ] , we plot the cross - sections of these channels at @xmath131 against @xmath5 for each of the two benchmark points . we have explicitly verified that all these production mechanisms are mostly unaffected in the range of @xmath84 between @xmath183 and @xmath184 and in the neighbourhood of the chosen @xmath7 . all the production rates are lower than those in the sm . this is because of the suppression of the couplings in table [ tab : feynman rules ] . combining production and decay , we obtain an intriguing result : @xmath12 may have escaped from the past and present searches due to the suppression of both its production and the branching ratios to massive vector boson pairs . as this scenario generically predicts lower rates into fermion pairs , it could be disfavoured by the recent hint in the @xmath124 mode from the tevatron @xcite if it persists . obviously , one can ease the reduction of @xmath12 production by taking the higgs mixing pattern to the sm - like limit , @xmath185 . in this case , the charged - neutral higgs associated production will be suppressed as shown in figures [ fig : significance cpv ] . recall that we are assuming a hierarchical neutral higgs mass spectrum and that all the experimental evidences from the lhc and the tevatron are caused by the lightest of the three neutral higgs particles , which is also supposed to be produced in association with @xmath1 . relaxing one or more of these hypotheses may lead to a different consequence , although we do not explore this possibility in this paper . on the other hand , branching fractions of the two loop - induced decay modes , @xmath186 and @xmath187 , have possibilities of being enhanced as well as diminished , depending on the input parameters . if they are increased , the suppressed higgs production can be alleviated . the decay into the digluon final state arises from the fermion loop , dominated by the top - quark contribution . although the same type of loop is present also in the sm , its value can be different due to the variation of the @xmath12@xmath188@xmath189 coupling which is determined by the higgs mixing angles as in table [ tab : feynman rules ] . in particular , an interesting role is played by the @xmath0-odd component @xmath72 in @xmath12 that could enhance the digluon branching fraction . its effects are twofold : ( a ) @xmath72 does not couple to a vector boson pair unlike the @xmath0-even higgs @xmath70 ; ( b ) the digluon to difermion partial width ratio of @xmath72 is higher than the corresponding quantity of @xmath70 , i.e.@xmath190 when @xmath191 . this effect stems from the different lorentz structures of the scalar and the pseudo - scalar couplings with fermions . the potential enhancement of the @xmath127 decay mode might be of interest at a future lepton collider experiment . the decay amplitude of @xmath186 has two more components , the @xmath2- and the @xmath1-loops , in addition to the fermion loop . in the sm , the dominant contribution comes from the @xmath2-loop . like the fermion loop , the @xmath2-loop can vary as a function of @xmath192 . one might attempt to increase the diphoton branching fraction by invoking the same mechanism as with the digluon mode . as @xmath12 becomes pseudo - scalar - like , the fermion loop alone does receive the relative enhancement that is given by ( [ eq : agg / hgg ] ) with @xmath127 replaced by @xmath6 and each numerator including only the fermion loop . however , this kills the @xmath2-loop at the same time , which is the dominant contribution in the sm . therefore , a high @xmath0-odd fraction in @xmath12 is not really helpful in increasing @xmath193 . nevertheless , it is known that the charged higgs loop can greatly affect the @xmath6 mode when @xmath7 is low @xcite . in particular , this mechanism can lead to a high enough diphoton rate to be consistent with the excess surrounding the higgs mass of @xmath3 recently observed by atlas @xcite and cms @xcite . the @xmath12@xmath111@xmath156 coupling depends on more parameters than the other loops , in particular on @xmath85 , with which one can play in order to engineer the amplitude . the maximal permitted size of the @xmath1-contribution is mostly controlled by perturbativity and unitarity . in passing , we remark that change of @xmath85 affects also @xmath194 and therefore the region excluded by ( [ eq : h2 exclusion ] ) , as we will see in the following plots . + to further study this potential variation , we plot the branching fraction of @xmath186 in figures [ h1-aa ] and [ h1-aa mu=180 ] . indeed , it can deviate substantially from its sm value depending on the mixing angles . one finds a valley between the two curves labelled `` a '' on each of the two plots , along which the branching ratio is highly suppressed . across this valley , the @xmath186 amplitude flips its sign . as the plots show , the contours are deformed if @xmath85 is changed , since they depend on the @xmath1-loop . one can also notice that part of the `` d '' contour of a high branching fraction lies on the verge of the perturbativity border , demonstrating that a sizeable @xmath1-loop contribution involves large quartic higgs couplings . in conjunction with gluon - gluon fusion , this brings us to the cross - section of @xmath195 in figures [ gg - aa ] and [ gg - aa mu=180 ] . with our crude leading - order approximation , it is estimated to be about @xmath196 in the sm . in the entire region of panel [ gg - aa ] , the cross - section is smaller than its sm value due to the suppressed @xmath12 production . nevertheless , one still has the option to change @xmath85 that affects the charged higgs loop . for instance , the cross - section at @xmath197 is approximately @xmath198 for @xmath199 . at this point , the @xmath2- and the @xmath1-loops interfere destructively . taking @xmath200 instead , one can make the interference constructive by flipping the sign of the @xmath1-contribution . this pushes the cross - section up above the sm prediction , as displayed in figure [ gg - aa mu=180 ] . in this plot , the rate of @xmath195 can be comparable to its sm value even when the mixing angles are fairly removed from the sm - like point , @xmath185 . in combination with figures [ fig : significance cpv ] , this exhibits the exciting possibility that the lhc can discover a light charged higgs in the foreseeable future while the recent evidences for higgs decays to @xmath6 are partly due to the charged higgs loop effect . as one can expect , the charged higgs loop effect weakens for high @xmath7 . for instance , @xmath193 for @xmath162 is roughly half its value for @xmath163 , when one keeps @xmath197 and @xmath200 . we demonstrated the superb potential of the lhc to discover a light charged higgs close to the @xmath2-boson mass in the @xmath0-violating 2hdmi . for this purpose , we exploited the property of the @xmath201 process that its cross - section is stable against variation of @xmath5 , which can be anywhere above the lower bound placed by the most stringent fcnc constraints . we set the higgs masses to their favoured values , @xmath202 and @xmath203 , and found that the lhc can spot an excess in the @xmath4 events , except in small corners and holes of the parameter space . we also examined the @xmath12-strahlung channel from @xmath2 , which offers access to the complementary parameter volume to that of @xmath204 . with regard to @xmath0-violation , we observed interesting consequences of the scalar pseudo - scalar mixing . in particular , @xmath205 can be considerably affected by the composition of @xmath12 . this result might be relevant to higgs scrutiny at prospective linear collider experiments . motivated by the recent evidences for higgs production at the lhc , we studied @xmath206 . due to the higgs mixing matrix elements appearing at interaction vertices , the production of @xmath12 is generically suppressed in comparison to the sm . nonetheless , the @xmath1-loop amplitude can make a constructive interference with the standard @xmath2-loop to raise @xmath207 . depending on the parameter choice , this effect can be significant enough to increase the diphoton production rate from gluon - gluon fusion up to or even above the sm prediction . alternatively , one can opt to enhance the @xmath12 production itself by letting the higgs mixing pattern approach the sm - like limit . in this case , the other processes with @xmath12 going into the fermion and the massive vector boson pairs are also restored , the price to pay being suppression of the @xmath204 rate . related to this last point , it may be beneficial to study further possibilities of the light charged higgs scenario in which one makes assumptions different from those underlying the present article . gmp would like to thank per osland and alexander pukhov for helpful discussions . jp thanks eung jin chun for valuable comments and kazuki sakurai for the help with access to the literature . we acknowledge financial support from german research foundation dfg through grant no . sto876/21 and bmbf . assuming the t hooft - feynman gauge , we choose a basis for the @xmath208 by expanding the higgs doublets as @xmath209 where @xmath210(@xmath211 ) is the goldstone boson of @xmath212(@xmath213 ) . we remark that the @xmath73 pseudo - scalar field is orthogonal to the neutral goldstone boson @xmath211 . as for the goldstone boson mass spectrum , it is possible to find a convenient way to write the mass matrix . having each @xmath214 the same group representation of the sm higgs , following the notation of @xcite , in the gauge - goldstone matrix follows from the four gauge bosons @xmath215 , @xmath213 , and the three goldstone bosons @xmath216 . ] basis we find the following representation of the co - variant derivative : @xmath217 while the vector boson ( and ghost ) mass matrix is @xmath218 , the goldstones mass matrix is : @xmath219 therefore we get @xmath220 the mass matrix in equation ( [ gb_mass_matrix ] ) shows that the goldstones have a mass that is equivalent to the sm - ones , as expected since the gauge sector has not been extended . as we have already intimated , the ghost mass matrix and interactions are defined by means of the same matrix @xmath221 via @xmath222 notice that the @xmath223 and the @xmath224 of equation ( [ gb_mass_matrix ] ) have different numbers of zero - eigenvalues , but their non - zero eigenvalues are in a one - to - one correspondence ; furthermore , the eigenvalues of the gauge - fixing mass matrix are the same of the gauge boson mass matrix , as expected . then , the ghost lagrangian is defined as @xmath225 c^b\ , \nonumber \\ & - & \bar{c}^a \left [ \mathcal{d}_2^a \cdot \left ( \mathcal{d}_2^b + \mathcal{s}_2^b \right)^t \right ] c^b\ , , \end{aligned}\ ] ] where the matrices @xmath226 represent the link between the fluctuations ( goldstones ) of the higgses around their vevs and the gauge bosons ; a convenient combination of scalar fields . consistently , the sum of the two contributions disarms these connections : ghost fields do not link to charged higgs . ] way to write this matrices is @xmath227 @xmath228 where @xmath229 finally , the ghost fields ( @xmath230 ) read as @xmath231 where @xmath232 in order to properly realise the high @xmath233 scenario , we must carefully consider the perturbativity of the couplings . firstly , from equation ( [ eq : m3 ] ) we see that the @xmath234 approximation leads to the following definition of the heaviest neutral scalar mass : @xmath235 from equations ( [ eq : lambda ] ) we can conclude that all the couplings are finite and perturbatively small except for @xmath236 and @xmath237 . is apparently divergent , by plugging the definition of @xmath238 in ( [ eq : lambda5i ] ) it is easy to see that the result is always finite . if @xmath30 exactly vanishes , one can rotate away the phase of @xmath239 to render @xmath0-conservation manifest @xcite . ] in fact , for @xmath240 or equivalently @xmath241 and @xmath242 , the two potentially diverging couplings become : @xmath243 , \\ \lambda_3&\simeq\frac{m_1 ^ 2-m_2 ^ 2}{v^2\cos{\beta}}\ , \frac{r_{22 } r_{12}}{r_{33}}.\end{aligned}\ ] ] hence , no matter how large @xmath5 is , we can always keep @xmath236 and @xmath237 perturbatively small by choosing @xmath244 respectively . this shows that there is no strict upper limit on @xmath5 from perturbativity . however , we do not make use of this mechanism for drawing our conclusions in the main part of the present paper . with respect to @xmath245 of the @xmath137-quark and the @xmath135-lepton at the indicated centre - of - mass energies , for the benchmark point @xmath105 . each curve is the sum of @xmath246 and that for the conjugate final state . the shadowed range ( red online ) shows the selection cuts set for the signal - to - background analysis.,scaledwidth=48.0% ] in figure [ fig : spectra ] , we show the @xmath245 spectra for the @xmath137-quark and the @xmath135-lepton , extracted from the process @xmath247 at the lhc . they were evaluated at the benchmark point @xmath105 . in the @xmath248 distributions , one finds that the selection cuts in ( [ eq : cuts ] ) act on regions with suppressed differential cross - sections . this means that we retain the major portion of events at our disposal even after applying the cuts . even in the @xmath249 distributions , the excluded area below each curve is relatively small compared to the total cross - section . s. weinberg , _ gauge hierarchies _ , _ phys . _ * b82 * ( 1979 ) 387 ; m. veltman , _ the infrared ultraviolet connection _ , _ acta phys . _ * b12 * ( 1981 ) 437 ; c. h. llewellyn smith and g. g. ross , _ the real gauge hierarchy problem _ , _ phys . _ * b105 * ( 1981 ) 38 . see e.g. j. l. rosner , _ an @xmath251 interpretation of an @xmath252 event _ , _ phys . _ * d55 * ( 1997 ) 3143 [ http://arxiv.org/abs/hep-ph/9607467[hep-ph/9607467 ] ] ; s. king , s. moretti , and r. nevzorov , _ exceptional supersymmetric standard model _ , _ phys . _ * b634 * ( 2006 ) 278 , [ http://arxiv.org/abs/hep-ph/0511256[hep-ph/0511256 ] ] . see e.g.l . basso , s. moretti , and g. m. pruna , _ theoretical constraints on the couplings of non - exotic minimal @xmath253 bosons _ , _ jhep _ * 1108 * ( 2011 ) 122 , [ http://arxiv.org/abs/1106.4762[arxiv:1106.4762 ] ] . see e.g.m . gonderinger , y. li , h. patel , and m. j. ramsey - musolf , _ vacuum stability , perturbativity , and scalar singlet dark matter _ , _ jhep _ * 01 * ( 2010 ) 053 , [ http://arxiv.org/abs/0910.3167[arxiv:0910.3167 ] ] . see e.g.e . ma , _ naturally small seesaw neutrino mass with no new physics beyond the tev scale _ , _ phys . _ * 86 * ( 2001 ) 25022504 , [ http://arxiv.org/abs/hep-ph/0011121[hep-ph/0011121 ] ] ; s. m. davidson and h. e. logan , _ dirac neutrinos from a second higgs doublet _ , _ phys . _ * d80 * ( 2009 ) 095008 , [ http://arxiv.org/abs/0906.3335[arxiv:0906.3335 ] ] . r. barbieri , l. j. hall , and v. s. rychkov , _ improved naturalness with a heavy higgs : an alternative road to lhc physics _ , _ phys . _ * d74 * ( 2006 ) 015007 , [ http://arxiv.org/abs/hep-ph/0603188[hep-ph/0603188 ] ] ; m. gustafsson , e. lundstrom , l. bergstrom , and j. edsjo , _ significant gamma lines from inert higgs dark matter _ , _ phys . lett . _ * 99 * ( 2007 ) 041301 , [ http://arxiv.org/abs/astro-ph/0703512[astro-ph/0703512 ] ] . see e.g.j .- o . gong , h. m. lee , and s. k. kang , _ inflation and dark matter in two higgs doublet models _ , http://arxiv.org/abs/1202.0288[arxiv:1202.0288 ] . see e.g.r . schabinger and j. d. wells , _ a minimal spontaneously broken hidden sector and its impact on higgs boson physics at the large hadron collider _ , _ phys . _ * d72 * ( 2005 ) 093007 , [ http://arxiv.org/abs/hep-ph/0509209[hep-ph/0509209 ] ] . cheung , c .- h . chou , and o. c. kong , _ muon anomalous magnetic moment , two higgs doublet model , and supersymmetry _ , _ phys . rev . _ * d64 * ( 2001 ) 111301 , [ http://arxiv.org/abs/hep-ph/0103183[hep-ph/0103183 ] ] . a. j. buras , m. v. carlucci , s. gori , and g. isidori , _ higgs - mediated fcncs : natural flavour conservation vs. minimal flavour violation _ , _ jhep _ * 1010 * ( 2010 ) 009 , [ http://arxiv.org/abs/1005.5310[arxiv:1005.5310 ] ] . p. ko , y. omura , and c. yu , _ chiral @xmath254 flavor models and flavored higgs doublets : the top fb asymmetry and the @xmath255 _ , _ jhep _ * 1201 * ( 2012 ) 147 , [ http://arxiv.org/abs/1108.4005[arxiv:1108.4005 ] ] . , g. aad _ _ , _ search for the standard model higgs boson in the diphoton decay channel with @xmath256 fb@xmath257 of @xmath157 collisions at @xmath258 tev with atlas _ , _ phys . _ * 108 * ( 2012 ) 111803 , [ http://arxiv.org/abs/1202.1414[arxiv:1202.1414 ] ] . , g. aad _ _ , _ combined search for the standard model higgs boson using up to 4.9 fb@xmath257 of @xmath157 collision data at @xmath259 tev with the atlas detector at the lhc _ , _ phys . _ * b710 * ( 2012 ) 49 , [ http://arxiv.org/abs/1202.1408[arxiv:1202.1408 ] ] . r. dermisek , _ light charged higgs and lepton universality in @xmath2 boson decays _ , http://arxiv.org/abs/0807.2135[arxiv:0807.2135 ] ; r. dermisek , _ light charged and @xmath0-odd higgses in mssm - like models _ , _ aip conf . _ * 1078 * ( 2009 ) 226 , [ http://arxiv.org/abs/0809.3545[arxiv:0809.3545 ] ] ; k. j. bae , r. dermisek , d. kim , h. d. kim , and j .- h . light higgs scenario in bmssm and lep precision data _ , http://arxiv.org/abs/1001.0623[arxiv:1001.0623 ] . a. filipuzzi , j. portoles , and m. gonzalez - alonso , _ @xmath260 flavor symmetry and lepton universality violation in @xmath261 _ , http://arxiv.org/abs/1203.2092[arxiv:1203.2092 ] . see e.g.v . d. barger , j. hewett , and r. phillips , _ new constraints on the charged higgs sector in two higgs doublet models _ , _ phys . _ * d41 * ( 1990 ) 3421 . a. pilaftsis and c. e. m. wagner , _ higgs bosons in the minimal supersymmetric standard model with explicit @xmath0 violation _ , _ nucl . _ * b553 * ( 1999 ) 3 , [ http://arxiv.org/abs/hep-ph/9902371[hep-ph/9902371 ] ] . s. l. glashow and s. weinberg , _ natural conservation laws for neutral currents _ , _ phys . _ * d15 * ( 1977 ) 1958 . a. w. el kaffas , w. khater , o. m. ogreid , and p. osland , _ consistency of the two higgs doublet model and @xmath0 violation in top production at the lhc _ , _ nucl . _ * b775 * ( 2007 ) 45 , [ http://arxiv.org/abs/hep-ph/0605142[hep-ph/0605142 ] ] . see e.g.s . kanemura , y. okada , e. senaha , and c .- p . yuan , _ higgs coupling constants as a probe of new physics _ , _ phys . _ * d70 * ( 2004 ) 115002 , [ http://arxiv.org/abs/hep-ph/0408364[hep-ph/0408364 ] ] . atlas collaboration , _ an update to the combined search for the standard model higgs boson with the atlas detector at the lhc using up to 4.9 fb@xmath257 of @xmath157 collision data at @xmath259 tev _ , http://cdsweb.cern.ch/record/1430033[atlas-conf-2012-019 ] ; cms collaboration , m. pieri _ et al . _ , _ searches for the standard model higgs boson at cms _ , http://arxiv.org/abs/1205.2907[arxiv:1205.2907 ] . h. e. haber and h. e. logan , _ radiative corrections to the @xmath265 vertex and constraints on extended higgs sectors _ , _ phys . rev . _ * d62 * , 015011 ( 2000 ) [ http://arxiv.org/abs/hep-ph/9909335[hep-ph/9909335 ] ] . d. toussaint , _ renormalization effects from superheavy higgs particles _ , _ phys . _ * d18 * , 1626 ( 1978 ) ; s. bertolini , _ quantum effects in a two higgs doublet model of the electroweak interactions _ , _ nucl . phys . _ * b272 * , 77 ( 1986 ) ; w. hollik , _ nonstandard higgs bosons in @xmath266 radiative corrections _ , _ z. phys . _ * c32 * , 291 ( 1986 ) ; _ radiative corrections with two higgs doublets at lep / slc and hera _ , _ z. phys . _ * c37 * , 569 ( 1988 ) . atlas collaboration , g. aad _ _ , _ search for charged higgs bosons decaying via @xmath267 in top quark pair events using @xmath157 collision data at @xmath259 tev with the atlas detector _ , http://arxiv.org/abs/1204.2760[arxiv:1204.2760 ] ; cms collaboration , s. chatrchyan _ et al . _ , _ search for a light charged higgs boson in top quark decays in @xmath157 collisions at @xmath259 tev _ , http://arxiv.org/abs/1205.5736[arxiv:1205.5736 ] . j. l. diaz - cruz and o. a. sampayo , _ contribution of gluon fusion to the production of charged higgs at hadron colliders _ , _ phys . _ * d50 * ( 1994 ) 6820 . a. djouadi , w. kilian , m. muhlleitner , and p. zerwas , _ production of neutral higgs boson pairs at lhc _ , _ eur . j. _ * c10 * ( 1999 ) 45 , [ http://arxiv.org/abs/hep-ph/9904287[hep-ph/9904287 ] ] . a. g. akeroyd and m. a. diaz , _ searching for a light fermiophobic higgs boson at the tevatron _ , _ phys . _ * d67 * ( 2003 ) 095007 , [ http://arxiv.org/abs/hep-ph/0301203[hep-ph/0301203 ] ] ; a. akeroyd , m. a. diaz , and f. j. pacheco , _ double fermiophobic higgs boson production at the cern lhc and lc _ , _ phys . _ * d70 * ( 2004 ) 075002 , [ http://arxiv.org/abs/hep-ph/0312231[hep-ph/0312231 ] ] . i. f. ginzburg , _ light charged higgs at lhc _ , http://arxiv.org/abs/1205.5890[arxiv:1205.5890 ] . a. v. semenov , _ lanhep : a package for automatic generation of feynman rules in gauge models _ , http://arxiv.org/abs/hep-ph/9608488[hep-ph/9608488 ] . a. pukhov , _ calchep 2.3 : mssm , structure functions , event generation , batchs , and generation of matrix elements for other packages _ , http://arxiv.org/abs/hep-ph/0412191[hep-ph/0412191 ] ; http://theory.sinp.msu.ru/~pukhov/calchep.html[http://theory.sinp.msu.ru/ pukhov / calchep.html ] . p. m. nadolsky , h .- lai , q .- h . cao , j. huston , j. pumplin , d. stump , w .- k . tung and c .- p . yuan , _ implications of cteq global analysis for collider observables _ , _ phys . _ * d78 * ( 2008 ) 013004 [ http://arxiv.org/abs/0802.0007[arxiv:0802.0007 ] ] . t. hahn and m. perez - victoria , _ automatized one loop calculations in four - dimensions and @xmath268-dimensions _ , _ comput . _ * 118 * ( 1999 ) 153 , [ http://arxiv.org/abs/hep-ph/9807565[hep-ph/9807565 ] ] . s. i. bityukov and n. v. krasnikov , _ on the observability of a signal above background _ , _ nucl . instrum . _ * a452 * ( 2000 ) 518 ; s. bityukov , s. erofeeva , n. krasnikov , and a. nikitenko , _ program for evaluation of significance , confidence intervals and limits by direct calculation of probabilities _ , in _ statistical problems in particle physics , astrophysics and cosmology : proceedings of phystat05 , oxford , uk , 1215 sep 2005 _ ( l. lyons and m. k. nel , eds . ) , pp . 106107 , imperial college press , 2005 . , _ combined cdf and d0 search for standard model higgs boson production with up to 10.0 fb@xmath257 of data _ , http://arxiv.org/abs/1203.3774[arxiv:1203.3774 ] . n. bernal , d. lopez - val , and j. sola , _ single higgs - boson production through @xmath6 scattering within the general 2hdm _ , _ phys . _ * b677 * ( 2009 ) 39 , [ http://arxiv.org/abs/0903.4978[arxiv:0903.4978 ] ] ; d. lopez - val and j. sola , _ single higgs - boson production at a photon - photon collider : general 2hdm versus mssm _ , _ phys . _ * b702 * ( 2011 ) 246 , [ http://arxiv.org/abs/1106.3226[arxiv:1106.3226 ] ] . a. arhrib , r. benbrik , and n. gaur , _ @xmath270 in inert higgs doublet model _ , http://arxiv.org/abs/1201.2644[arxiv:1201.2644 ] . m. e. peskin and d. v. schroeder , _ an introduction to quantum field theory _ , addison - wesley , 1995 .

we discuss the low - energy limit of three - orbital kondo - lattice and hubbard models describing orbitals on a triangular lattice near half - filling . we analyze how very flat single - particle bands with non - trivial topological character , a chern number , arise both in the limit of infinite on - site interactions as well as in more realistic regimes . exact diagonalization is then used to investigate an effective one - orbital spinless - fermion model at fractional fillings including nearest - neighbor interaction ; it reveals signatures of fractional chern insulator ( fci ) states for several filling fractions . in addition to indications based on energies , e.g. flux insertion and fractional statistics of quasiholes , chern numbers are obtained . it is shown that fci states are robust against disorder in the underlying magnetic texture that defines the topological character of the band . we also investigate competition between a fci state and a charge density wave ( cdw ) and discuss the effects of particle - hole asymmetry and fermi - surface nesting . fci states turn out to be rather robust and do not require very flat bands , but can also arise when filling or an absence of fermi - surface nesting disfavor the competing cdw . nevertheless , very flat bands allow fci states to be induced by weaker interactions than those needed for more dispersive bands .

1208.3481

quantum liquids are among the most sought after states of matter . one celebrated class of such quantum liquids , the fractional quantum - hall ( fqh ) states , @xcite has been a locus of scientific attention for almost three decades . initially introduced to explain the fractional quantum - hall effect ( fqhe ) observed in semiconductor devices , @xcite they describe interacting electrons constrained in two dimensions and subject to a strong perpendicular magnetic field . fractionally filling a magnetic landau level ( ll ) then yields an incompressible fqh liquid and gives rise to a precisely quantized hall conductivity . in addition , the quasiparticles of these states obey anyonic statistics , @xcite which can be either abelian or non - abelian , the latter fulfilling an essential condition for fault - tolerant quantum computation . @xcite an alternative route to quantum hall liquids is via tightly bound electrons moving in a magnetic texture . the simplest such magnetic texture is a uniform magnetic field perpendicular to the plane of the system . for non - interacting , tightly bound electrons , this gives rise to an integer quantum - hall ( iqh ) state and leads to hofstadter s fractal energy spectrum , @xcite whereas in a fractionally filled tight - binding model of electrons interacting via a screened coulomb repulsion , the realization of a fqh state is possible . @xcite haldane noted @xcite that an external magnetic field is not the only viable path to iqh states and introduced a honeycomb - lattice model , with complex hoppings that break time - reversal invariance , and defined the first integer chern insulator ( ci ) , in which the total magnetic field through the unit cell averages to zero . in the meantime , mechanisms leading to complex hoppings with the necessary properties have been identified . one realization may be found in strongly spin - orbit coupled semiconductor materials that are ferromagnetically ordered . @xcite another possibility arises through the coupling of itinerant electrons to localized magnetic moments , @xcite for example the kondo - lattice model on the triangular lattice supports a non - trivial magnetic texture , which induces an integer - quantized hall conductivity of the itinerant electrons . @xcite a considerable body of recent research has addressed the question whether the lattice counterpart of the fqhe can be observed when topologically non - trivial bands , called chern bands , are fractionally filled and electrons are interacting . several numerical studies using exact diagonalization techniques have convincingly established the existence of laughlin - series states in an number of different models on various lattices . @xcite these systems have since been called fractional chern insulators ( fci ) . very recent work has reported fci states beyond laughlin fractions . @xcite in addition to reproducing the known fqhe on a lattice and potentially at higher temperatures , cis also offer the intriguing possibility of a chern number higher than one , a departure from the analogy with lls . recent studies have explored this direction by constructing models that have higher chern numbers and studying possible fci states . @xcite from the analytical side , the problem of fci states has been approached by a careful study of emergent translational symmetries @xcite and many - body trial wave functions . @xcite others have examined the algebraic properties of the density operators projected onto one chern band and made a comparison with the girvin - macdonald - platzman algebra that is satisfied by the lowest ll density operators in the continuum fqhe . @xcite in numerical studies of fci states , one typically starts by adding inter - site interactions to topologically nontrivial but non - interacting ci models . as potential realizations of the non - interacting and nearly flat `` parent '' bands , cold atoms , @xcite oxide heterostructures , @xcite strained graphene , @xcite and strongly correlated multi - orbital models for layered oxides @xcite have been proposed . in the present paper , we build upon this last approach and thus focus our attention on a strongly correlated model on the triangular lattice . it was shown that on the mean - field level and near half - filling , a magnetically ordered ci with a very flat single - particle band emerges . doping this nearly flat band to fractional fillings was shown to give rise to fci states within the framework of an effective single - orbital model . here , we give a detailed account of the mapping onto the effective single - orbital model and show how such topologically nontrivial and nearly flat bands emerge in kondo - lattice and hubbard models both for the limit of infinite onsite interactions and for more realistic intermediate interaction strength . we then provide extensive numerical evidence , based on both eigenvalue spectra ( e.g. , ground - state degeneracy and spectral flow ) and eigenstate properties ( e.g. , many - body berry curvature and hall conductivity ) , for the existence of fci states in this model by using exact diagonalization . we discuss the robustness of fci states against disorder originating from single - site defects in the magnetic ordering , which will always occur in realistic situations . another very relevant issue in the context of fci states is their competition with other phases , @xcite e.g. symmetry - broken states such as a charge - density wave ( cdw ) . we study this issue by using a method that does not project onto the nearly flat chern band and thus keeps the effects of dispersion . the model considered in this work allows for a careful study of the competition between finite dispersion and interactions . we map out a phase diagram for filling fractions @xmath3 and @xmath4 . in the latter case , filling permits a commensurate cdw , whereas in the former it does not . the cdw is favored by fermi - surface ( fs ) nesting , and we accordingly find the fci at @xmath4 to be far more stable when the bands are poorly nested . this paper is organized as follows : in sec . [ sec : kondo_hubb ] , we extend the discussion of flat and topologically nontrivial bands arising in @xmath0 models and derive the effective model . numerical results on the model are presented in sec . [ sec : results ] , where we first focus on information obtained from eigenenergies , namely gaps and flux insertion ( sec . [ sec : spectrum ] ) and fractional statistics of charged excitations in fci states based on a recently introduced state - counting argument ( sec . [ sec : stat ] ) . @xcite we then add information obtained from the eigenstates : the hall conductivity , which allows us to address the impact of impurities in sec . [ sec : disorder ] and the static charge - structure factor , which allows us to discuss the competition with the cdw at filling @xmath4 in sec . [ sec : phdiag ] . we conclude with some remarks summarizing the main points in sec . [ sec : conclusions ] in ref . , it was shown that both @xmath5 and @xmath0 orbital manifolds in octahedral coordination can reduce the bandwidth of topologically nontrivial bands . for a schematic illustration of orbital degeneracy in @xmath6-electron systems see fig . [ fig : oct_orb ] . in particular , this was discussed for the spin - chiral phase arising in kondo - lattice models on the triangular lattice@xcite at quarter and three - quarter fillings . while the flat band of interest mixes both the @xmath7 and the @xmath8 orbitals in the @xmath5 case , it is dominated by a particular orbital state in the @xmath0 manifold , the @xmath9 state , which allowed the straightforward mapping onto an effective one - band model . @xcite in this section , we first review the basic setup of the orbital symmetries in sec . [ sec : orbs ] and the modifications of the hopping through the magnetic order in sec . [ sec : berry ] , as we believe that it may be of value to the reader . after this presentation of the physical ingredients at work here , we discuss how the nearly flat @xmath9 band arises due to effective longer - range hoppings . after the simplest scenario of infinite hund s rule coupling in sec . [ sec : jinf ] , we go to the more realistic multi - orbital model with finite interactions in sec . [ sec : jfinite ] , and finally decide on a relatively simple effective model in sec . [ sec : eff_model ] , which nevertheless captures several important features . in many transition - metal ( tm ) compounds , the local symmetry around a tm ion is cubic , with ligand oxygens forming an octahedron , as depicted in fig . [ fig : oct ] . this splits the degeneracy between the @xmath6 levels , because the two @xmath5 orbitals point toward the negatively charged oxygens , while the three @xmath0 levels have their lobes in between . consequently , the energy of @xmath5 levels is higher . depending on the total electron filling , the valence states may be found in either manifold . we are here discussing the situation where the three @xmath0 levels share @xmath10 to @xmath11 electrons and the @xmath5 levels are empty . furthermore , we consider the case of a layered triangular lattice , as can be realized in compounds of the form abo@xmath12 . in this geometry , the octahedra are edge sharing and electrons ( or holes ) can hop from one tm ion to its neighbor either through direct overlap or via the ligand oxygens . the hopping symmetries can be most easily worked out using the usual basis functions for the @xmath0 states , @xmath13 , @xmath14 , and @xmath15 @xcite and following refs . . considering hopping for bonds along the @xmath16 direction and choosing the local coordinate system such that this corresponds to the @xmath17 direction in the @xmath18-@xmath19 plane , one finds that direct hopping @xmath20 is only relevant for the @xmath21 orbital and conserves orbital flavor . due to the @xmath22 angle of the tm - o - tm bond , oxygen - mediated hopping @xmath23 is , on the other hand , mostly via the oxygen-@xmath24 orbital and mediates processes between @xmath25 and @xmath26 states , thereby always _ changing _ orbital flavor . hoppings along the other two , symmetry - related , directions @xmath27 and @xmath28 are obtained by symmetry transformations . these hoppings can then be expressed in orbital- and direction- dependent matrix elements @xmath29 , where @xmath30 and @xmath31 denote orbitals ( @xmath25 , @xmath26 , and @xmath21 ) and @xmath32 the direction . they are given by @xmath33 for nn bonds along the three directions @xmath16 , @xmath27 , @xmath28 . the two hopping processes are expected to be of comparable strength , but with @xmath34 for @xmath35 elements , and will typically have opposite sign . @xcite if the width of a triangular layer made of octahedra is compressed ( extended ) , the energy of the highly symmetric orbital state @xmath36 is raised ( lowered ) with respect to the remaining orbital doublet ( @xmath37 ) , see fig . [ fig : split ] for illustration . this energy shift can be written as @xmath38 and depends on the jahn - teller effect as well as on the lattice . @xcite especially for large splitting between @xmath9 and @xmath37 states , which may also be enhanced through onsite coulomb interactions , see sec . [ sec : jfinite ] , it is more appropriate to use a basis that reflects the triangular lattice symmetry . we thus go over into the @xmath39 basis , which is done via @xcite @xmath40 the transformed hopping matrices @xmath41 are then obtained from eq . ( [ eq : hopp_xyz ] ) as @xmath42 where @xmath43 and @xmath44 . observe that the intra - orbital hopping of the @xmath9 state is the same in all three lattice directions , as expected for @xmath9 symmetry . however , we also see that hopping elements mix all three orbitals . in addition to an orbital degree of freedom , we now consider coupling to a localized spin @xmath45 , modelled by a kondo - lattice model , where the kinetic energy is given by hopping elements @xmath29 taken from the matrices eq . ( [ eq : hopp_xyz ] ) or eq . ( [ eq : hopp_a1g_eg ] ) . this situation is described by @xmath46 where @xmath47 denotes nn bonds along the three directions @xmath32 , @xmath30 and @xmath31 are orbital indices , @xmath48 ( @xmath49 ) annihilates ( creates ) an electron with spin @xmath50 in orbital @xmath30 at site @xmath51 , and @xmath52 is the corresponding vector of orbital electronic spin operators . @xmath53 couples the itinerant electrons to a generic localized spin @xmath45 , the origin of which is left unspecified for the moment , but will be discussed extensively later . ( it will turn out to be the spin degree of freedom of the @xmath0 electrons themselves , as in ref . . ) the coupling is assumed to be fm , as one would expect from hunds - rule coupling . however , we are furthermore going to consider @xmath54 as a _ classical _ spin , in which case afm coupling to @xmath45 would lead to the equivalent results . for classical spins and large @xmath53 , it is convenient to go over to a _ local _ spin - quantization axis , where `` @xmath55 '' ( ` @xmath56' ) refers to parallel ( antiparallel ) orientation of the electron s spin to the local axis . this simplifies the kondo term to @xmath57 where @xmath58 ( @xmath59 ) is the electron density at site @xmath60 in orbital @xmath30 with spin ( anti- ) parallel to the localized spin . this local spin definition is particularly convenient when going to the limit of large @xmath53 , where one immediately finds the low - energy states as given by only `` @xmath55 '' electrons . on the other hand , the fact that the spin - quantization axis is not the same at all sites implies that the hopping no longer conserves the new spin . instead , hopping acquires as spin - dependent factor @xmath61 , @xcite with @xmath62 where @xmath63 and @xmath64 , @xmath65 and the set of angles @xmath66 and @xmath67 are the polar and azimuthal angles corresponding to @xmath68 , respectively . as one can see , these effective hoppings can become complex , and it has been shown that non - coplanar spin configurations can endow the electronic bands with a nontrivial topology.@xcite additionally , the itinerant electrons mediate an interaction between the localized spins , which typically competes with antiferromagnetic spin - spin interactions ; on frustrated lattices , this competition can resolve itself in non - coplanar and thus topologically nontrivial phases.@xcite the interplay of the orbital symmetries summarized in sec . [ sec : orbs ] with the berry phases of sec . [ sec : berry ] was discussed in ref . for the limit of infinite hund s rule coupling to classical localized spins , corresponding to the double exchange model . in this case , one only keeps the @xmath55 electrons parallel to the local spin - quantization axis and electrons effectively become spinless fermions . for the chiral spin pattern in fig . [ fig : chiral ] , which has been found as the ground state of triangular kondo - lattice models,@xcite the berry phases between the four sites of the magnetic unit cell can be parametrized as @xmath69 it turns out that these hoppings can in fact be written in a two - site unit cell ( containing sites 1 and 2 ) due to an internal symmetry of the four - site pattern , @xcite see also the effective model eq . ( [ eq : hkin ] ) below . + combining the phases ( [ eq : hopp_chiral ] ) with the hoppings given by eqs . ( [ eq : hopp_xyz ] ) or ( [ eq : hopp_a1g_eg ] ) and the crystal - field splitting eq . ( [ eq : jt_xyz ] ) still gives a non - interacting model that can be easily solved in momentum space . one finds that large @xmath70 , see eq . ( [ eq : jt_xyz ] ) , strongly reduces the dispersion of one subband . @xcite this can also be seen in fig . [ fig : bands_kondo ] , which shows the one - particle energies obtained on a cylinder . figure [ fig : bands_kondo ] also reveals the edge states crossing some gaps , indicating the topologically nontrivial nature of these bands . calculating chern numbers @xmath71 corroborates this and gives @xmath72 . @xcite the band flatness can be expressed in terms of a figure of merit @xmath73 where @xmath74 and @xmath75 are the two gaps separating the narrow band of interest from the other orbitals and from the subband with opposite chern number and @xmath76 is the width of the narrow band . as has been pointed out , @xcite the lower subband can here become very flat , and as can be seen in fig . [ fig : gaps_kondo ] , the flatness can be further improved by going to larger crystal fields and reaches values @xmath77 . as these very flat bands can be achieved for large separation @xmath78 between the @xmath9 and @xmath37 states and as the band of interest then has almost purely @xmath9 character , it is natural to assume that one should be able to capture the most relevant processes with an effective @xmath9 model . ( this is in contrast to the situation starting from @xmath5 orbitals , where one finds intermediate @xmath78 to be optimal . @xcite in that case , the nearly flat bands can only be obtained if _ both _ orbitals contribute weight and one can not easily reduce the situation to a one - band system . ) the impact of the @xmath37 levels on the effective @xmath9 dispersion can be taken into account in second - order perturbation theory . this includes processes where a hole hops from the @xmath9 orbital at site @xmath51 to an @xmath37 state at @xmath79 and back again to an @xmath9 state at a third site @xmath80 , which may or may not be the same as @xmath51 . the denominator of these terms is the crystal - filed energy @xmath78 and the numerator is obtained from the products @xmath81 ( with @xmath82 designating @xmath9 and @xmath83 , @xmath84 the @xmath37 states ) . in order to evaluate the second - order hopping between sites @xmath51 and @xmath80 , these orbital hoppings have to be multiplied by the product of the berry phases @xmath85 and @xmath86 from eq . ( [ eq : hopp_chiral ] ) for all paths connecting @xmath51 and @xmath80 via one intermediate site @xmath87 . due to destructive interference , processes connecting nn and next - nearest neighbor ( nnn ) sites cancel while effective third - neighbor hopping , where there is only one path , remains . since third - neighbor spins in the chiral phase are always parallel , the total berry phase of this process is 1 in all directions , however , the hopping via a spin of different orientation in the middle reduces the hopping amplitude by @xmath88 , leading to @xmath89 a third - neighbor hopping @xmath90 turns out to have almost the same dispersion as the chiral subbands and can consequently almost cancel it in one subband . as its strength can be tuned by @xmath91 and @xmath78 , very flat subbands can be achieved , see fig . [ fig : gaps_kondo ] . , see eq . ( [ eq : merit ] ) , for finite hunds - rule coupling @xmath92 and @xmath93 . the bands designated as `` upper '' and `` lower '' refer to the two subbands of the @xmath9 states with spin parallel to the localized spin , which are separated by the gap opening in the spin - chiral phase , see fig . [ fig : bands_kondo].[fig : gapps_kondo ] ] while the previous section has illustrated how one can understand the occurrence of nearly flat bands in a three - orbital double - exchange model , i.e. , for infinite hund s rule coupling to some localized spins , this section will discuss finite hund s rule . figure [ fig : gapps_kondo ] shows the figure of merit for the band flatness eq . ( [ eq : merit ] ) for a few sets of hopping parameters and for @xmath94 depending on hund s rule coupling @xmath53 to the localized spin , see eq . ( [ eq : kondo_loc ] ) . as can be seen in fig . [ fig : gapps_kondo ] , the _ upper _ subband of the @xmath9 sector can now become nearly flat . ( for @xmath95 , one can of course still find flat lower subbands , as discussed above . ) the flatness of the upper subband can be explained by similar effective longer - range hoppings in second - order perturbation theory , this time also taking into account intermediate states with an electron in the upper kondo band , i.e. , with antiparallel spin . these additional terms can go either via the @xmath9 or via the @xmath37 orbitals and involve combined berry phases of the form @xmath96 . again , one has to sum over all possible intermediate sites @xmath79 and finds @xmath97 the nn , nnn and third - neighbor hoppings are here denoted by @xmath98 , @xmath99 , and @xmath100 . @xmath101 , @xmath102 , and @xmath103 give the excitation energies of the intermediate states with ( i ) a hole in the @xmath37 states with spin parallel , ( ii ) an electron in the @xmath9 states with spin anti - parallel and ( iii ) an electron in an @xmath37 state with spin anti - parallel . like the bare nn hopping , these effective hopping - matrix elements acquire an additional berry phase @xmath104 in the hamiltonian , which only depends on the relative orientation of spins on the initial and final sites . nnn hopping @xmath99 via the upper kondo band does not drop out , and nn hopping becomes renormalized . the flat chiral subbands that have been observed in a three - orbital @xmath0 hubbard model on the triangular lattice @xcite arise in situations similar to the finite-@xmath53 scenario , and a mapping to the kondo - lattice model was shortly mentioned in the supplemental material of ref . . the key point of the mapping is the observation that large crystal - field splitting @xmath78 , see ( [ eq : jt_xyz ] ) , leads to an orbital - selective mott insulator , where the @xmath37 levels are half filled and far from the fermi level , while the states near the fermi level have almost only @xmath9 character . the orbital degree of freedom is consequently quenched , because orbital occupations are already fully determined . a charge degree of freedom remains , as the @xmath9 orbital contains one electron per two sites . charge fluctuations of the half - filled @xmath5 levels , however , are suppressed due to the large mott gap between their occupied and empty states . they can thus be described as a spin degrees of freedom , and the situation is further simplified , because they form a total spin with @xmath105 due to hund s rule . the @xmath9 electron is likewise coupled via fm hund s rule to this spin . this situation mobile carriers coupled via fm hund s rule to localized spin degrees of freedom is captured by a fm kondo - lattice model . these considerations can be cast in a more formal setting starting from the mean - field decoupling @xcite of the onsite coulomb repulsion . the full interaction for equivalent @xmath0 orbitals reads @xmath106 where @xmath107 and @xmath108 holds in the case of equivalent @xmath0 orbitals . we now concentrate on the regime of interest , where the @xmath9 orbital is separated from the @xmath5 doublet by a sizable @xmath109 and does not have to be equivalent . as long as intraorbital interaction ( [ eq : intra ] ) and hund s rule coupling ( [ eq : hund ] ) dominate over interorbital interaction ( [ eq : inter ] ) and crystal - field splitting ( [ eq : jt_xyz ] ) , doubly occupied orbitals will be suppressed and the last term ( [ eq : pair ] ) will consequently not be important . moreover , there is no reason for spins in different orbitals , but on the same site , to point in different directions , as the only interactions between spins are fm , i.e. , we can use the same local quantization axis for all orbitals . in the case of doubly occupied orbitals , one spin can be seen as lying antiparallel and as introduced in sec . [ sec : berry ] , `` @xmath55 '' ( `` @xmath56 '' ) denotes a spin parallel ( antiparallel ) to the local quantization axis . the corresponding mean - field decoupling is @xmath110 where @xmath71 is a constant @xcite and @xmath111 . due to the last term ( [ eq : hundmf ] ) , an electron in orbital @xmath31 feels a fm coupling to a `` classical localized spin '' with length @xmath112 that points along the local quantization axis . the axis can be parametrized by angles @xmath113 and @xmath114 , which establishes the relation to sec . [ sec : berry ] . the first term ( [ eq : umf ] ) suppresses doubly occupied orbitals for all three orbitals . in the orbital - selective mott - insulator , @xmath115 for the @xmath37 states . the second term ( [ eq : upmf ] ) thus mainly enhances the effect of @xmath116 . the last term ( [ eq : hundmf ] ) becomes equivalent to the kondo term , eq . : @xmath117 finally , we note that the fact that onsite interactions @xmath118 and @xmath119 are only large , but _ not _ infinite , is important for the spin - chiral ground state : for very large @xmath118 and @xmath119 , the ground state becomes a ferromagnet . @xcite this can be related to the fact that the kondo - lattice model requires either finite @xmath53 @xcite or additional afm inter - site superexchange @xcite to support a spin - chiral instead of a fm state . at finite onsite interactions , virtual excitations with doubly occupied @xmath37 orbitals are possible and lead to second - order processes that are similar to the effective longer - range hoppings discussed above . in such a process , an @xmath37 electron hops into an occupied @xmath37 state at a nn site , creating a ( virtual ) intermediate state with energy @xmath120 , and hops back in the next step . such a process yields an energy gain @xmath121 and is only possible if the spins of the two involved electrons , which occupy the same orbital in the intermediate state , are opposite . the mechanism thus effectively provides the needed afm intersite superexchange and the spin - chiral state becomes stable for wide parameter regimes , including ranges supporting nearly flat upper chiral subbands . @xcite as we have discussed in the previous sec . [ sec : jfinite ] , the most realistic route to nearly flat bands with nontrivial topology on the triangular lattice arises at finite hubbard / kondo coupling , where effective second - neighbor hopping is generated in addition to nn and third - neighbor terms . however , we have found that all essential features of the band structure can be captured by using just nn and third - neighbor hoppings ( see figure in the supplemental material of ) . in the interest of simplicity , we consequently adopt here the kinetic energy @xmath122 where @xmath123 ( @xmath124 ) denote the unit vectors on the triangular lattice , and @xmath125 and @xmath126 are the nn and third - neighbor hopping , which can be related to eqs . ( [ eq : hopp_chiral ] ) and ( [ eq : t3 ] ) for the double - exchange scenario , to eqs . ( [ eq : t1full ] ) and ( [ eq : t3full ] ) for finite onsite interactions , or which can be taken as fit parameters . pauli matrices @xmath127 and unit matrix @xmath128 refer to the two sites of the electronic unit cell in the chiral state . @xcite the unit cell and the topologically non - trivial bands are due to the symmetry breaking involved in the underlying magnetic order . the dispersion of eq . is @xmath129 from now on , we use the effective nn hopping @xmath125 as unit of energy ; the band flatness can then be tuned by varying the ratio @xmath130 . + the longer range hopping @xmath126 determines the flatness of the bands of @xmath131 , which can be expressed by the figure of merit @xmath132 , see eq . ( [ eq : merit ] ) . figure [ fig : disp ] shows @xmath132 depending on @xmath126 , and one sees that ratios @xmath133 can be reached for @xmath134 . such flatness ratios can reasonably be achieved in the low - energy bands of a strongly correlated @xmath0 system on a triangular lattice . @xcite changing the sign of @xmath126 simply mirrors the dispersion vertically , i.e. , it is then the upper band that becomes nearly flat . when going away from maximal @xmath132 , the bands for smaller and larger @xmath126 differ qualitatively ; for @xmath135 , the fermi surface ( fs ) at some fillings is almost perfectly nested . we are going to discuss the impact of these differences in sec . [ sec : phdiag ] . the nn coulomb interaction @xmath136 is added to the kinetic term , giving the total hamiltonian @xmath137 . while the spin - chiral phase providing the nontrivial topology can only be expected to remain stable for not - too - large @xmath2 and doping of the flat band , we are going to study a variety of filling and interaction ranges to obtain a comprehensive picture of the model on finite - size systems . the interacting hamiltonian can be diagonalized exactly for small systems , using the lanczos method . @xcite unless otherwise noted , the results presented here are on a @xmath138 unit - cell torus ( i.e. , @xmath139 real - space sites ) . we do not project onto the flat subband , see fig . [ fig : disp](a ) , but instead model the whole system of both subbands with @xmath72 . while this approach increases the hilbert space and thus restricts system size , it has the advantage that competition with phases mixing the two subbands is included automatically . we use several observables to detect fci states and to distinguish them from other phases ; first , we discuss conclusions to be drawn from the eigenvalues , and later also include information obtained from the eigenstates , namely the many - body chern number as a topological invariant and the charge - structure factor indicating formation of a ( conventional ) charge - density wave . even though the results of our investigation do not allow for conclusive finite - size scaling , we have verified that the findings presented here are consistent for smaller as well as somewhat larger systems . furthermore , the main features of the fci states found are consistent for all aspect ratios yielding the same system size , apart from those that reduce the system to a one - dimensional chain . before passing , we note that we have obtained similar results to the ones presented here for several other filling fractions @xmath140 , with @xmath141 . conclusive evidence for 4/5 , 5/7 , 6/7 states could not be obtained , in agreement with reported results for another model . @xcite . in all panels @xmath142 . ] fci states manifest themselves in features of the obtained eigenvalue spectra . traditional fqh ground states are @xmath143-fold quasi - degenerate on a finite torus at filling @xmath140 . @xcite the same degeneracy is expected to occur for fci states as well , at least within certain well - defined limits . @xcite quasi - degenerate fqh ground - state eigenvalues must also exhibit spectral flow , leading each of them into another upon insertion of a flux quantum . @xcite magnetic fluxes can be modelled by introducing phase factors to the hopping from site @xmath144 to site @xmath145 , thus leading to the transformation @xmath146 $ ] . in this manner , an electron hopping around the lattice in the @xmath147 direction picks up a phase @xmath148 . . in all panels @xmath149 . ] the features described above are illustrated , as a function of @xmath150 , in fig . [ fig : evflux_v ] for @xmath3 . in this case , the three quasi - degenerate ground states emerge from the continuum as the interaction strength is increased . insertion of one flux quantum indeed leads from one ground - state eigenvalue to another . it can also be seen that spectral flow is a general property of the eigenvalue spectrum , and we have found that it also occurs even at @xmath151 . the @xmath4 ground states behave in a similar fashion . it can be seen that the fci states remain gapped for a range of both @xmath132 and @xmath2 . in order to define a physically meaningful phase boundary , however , one has to make sure that the ground states remain gapped upon flux insertion , not only in their own momentum sector , but also across momentum sectors . the reason is that the presence of impurities would mix the momentum sectors and therefore close the gap if the quasi - degenerate fci - state levels cross excited - state levels . as seen in fig . [ fig : evflux_v ] , interaction mostly changes the energy differences between groups of three quasi - degenerate eigenstates , which flow into each other upon flux insertion , and has less impact on the energy splitting within each manifold . when tuning the transition via increasing the band dispersion , the situation is somewhat different , see fig . [ fig : evflux_t3 ] . for @xmath152 , the fci ground state is destroyed by increasing the split between quasi - degenerate levels , permitting a higher - energy states to mix with the fci - state manifold for some flux values . the splitting between ground states in the traditional fqhe case is due to quasiparticle - quasihole excitations propagating around the torus and therefore depends mainly on system size , becoming zero in the thermodynamic limit . @xcite in the finite tight - binding lattice systems discussed here , the splitting is affected significantly by the residual kinetic energy of the partially filled band , as can be seen in fig . [ fig : evflux_t3 ] . since the dispersion is present also in the thermodynamic limit , it is not clear whether the splitting survives to the thermodynamic limit or not in this system . despite level rearrangements or increasing spread between eigenvalues , fci states remain topologically conjugate upon varying the band flatness or the interaction strength . a factor that can split this conjugacy is disorder , as will be shown in the sec . [ sec : disorder ] . at the same time , ground - state levels that remain gapped upon flux insertion along one direction may cross excited - state levels upon flux insertion along another direction . a more careful analysis , based both on eigenenergy as well as on eigenstate properties , is necessary to uniquely determine a gapped ground state for a given parameter set . this is the purpose of sec . [ sec : chern ] . before investigating eigenstates , however , the next section is going to discuss another aspect of eigenenergies , fractional statistics . -unit cell system , the filling fraction is @xmath153 . there are seven quasi - degenerate ground states ( marked in red ) , which are slightly separated from the rest of the states in the low - energy sector and exhibit spectral flow ( shown in different colors ) . their chern number is @xmath154 within the numerical error margin . ] the concept of generalized pauli principles , @xcite according to which states with clustered anyonic quasiparticles are energetically penalized , has served as a tool to indicate fractional statistics in fqh states . @xcite as was first demonstrated heuristically @xcite and later on substantiated theoretically , @xcite the same logic holds for fci states . for the @xmath3 fci state , the number of quasihole - state levels in a well separated , low - energy fock space is equal to the number of ( 1,3)-admissible partitions of the momentum sectors on the torus . the individual states per momentum sector can also be accounted for , due to an emergent translational symmetry ( see ref . for details ) . this symmetry is a characteristic feature of all the fci states we have found in the triangular lattice model . in this section , we demonstrate that quasihole states of the triangular - lattice model obey the same state - counting arguments . quasiholes can be introduced in a fci state by either removing electrons or increasing the system size . an example of both cases is presented in fig . [ fig : fractional ] . the counting rule is verified in both cases . the counting rule can also be easily generalized to all laughlin filling fractions @xmath155 and we have verified it for @xmath156 in this model . apart from the agreement to the counting rule , one can also notice the emergence of a `` daughter '' fci state within the gapped low - energy sector , as expected in the hierarchy picture . even though in the present context this is mainly a peculiarity of the small system sizes , it can nevertheless be viewed as a simple example of the formation of a hierarchical fci state among the quasiparticle states of a `` parent '' fci state . taking this observation one step further , it can be seen that the spectrum looks qualitatively different under flux insertion in the two cases presented in fig . [ fig : fractional ] , namely , the eigenvalue spectrum in the case where the filling is @xmath157 looks more like that of the non - interacting system with levels crossing upon flux insertion , whereas at @xmath153 levels come in small groups , which remain separated under flux insertion ( compare to fig . [ fig : evflux_v ] ) . this observation can be compared to the composite fermion theory of the fqhe , @xcite according to which the state of a fqh system at filling fractions with even denominators can be effectively described by free fermions with flux tubes attached to them . this composite - fermion view of the fci hierarchy of states has also been supported by recent numerical calculations . @xcite these two different facets of the same model can serve as an example of the compatibility between the hierarchy and composite - fermion pictures of the fqhe in an unconventional fqh system . the most unambiguous characteristic of a fci state is arguably its hall conductivity . the ground - state hall conductivity of a fci at filling @xmath140 should be exactly @xmath158 . occasionally , this quantity is referred to as `` fractional chern number '' , to highlight the connection to the topological aspects of fci states . the hall conductivity @xmath50 of a gapped , degenerate state , measured in units of @xmath159 in the following , can be efficiently calculated using the kubo formula : @xcite @xmath160 where @xmath161 is the number of unit cells , @xmath162 are the @xmath143-fold degenerate many - body ground states and @xmath163 are higher - energy eigenstates . the corresponding eigenenergies are @xmath164 . the @xmath148 dependence of the hamiltonian comes from magnetic fluxes going through each of the handles of the torus , as discussed in sec . [ sec : eff_model ] . , ( d - f ) their standard deviations @xmath165 from the exact average and relative deviations of hall conductivities @xmath166 as a function of grid size for the state @xmath167 , when it is in the excited - state quasi - continuum ( top ) and in the ground - state manifold ( middle and bottom ) . the integrations in eq . have been approximated by simple riemann sums . @xmath168 is the number of @xmath148 points taken in the range @xmath169 in each direction . ] the integrand in eq . is proportional to the berry curvature @xmath170 for each of the states in the degenerate ground state . for fci states , berry curvatures are periodic functions varying with @xmath148 . in fig . [ fig : berry ] , the many - body berry curvature for a specific state , which develops into one of the fci ground states , is shown . it is seen that its smoothness varies as this state emerges from the excited state continuum . the period extends over @xmath143 flux quanta in one direction and one flux quantum in the other , remains unchanged for all values of @xmath2 and it is the same for all states in the ground - state manifold . the berry curvature of the other two ground states is the same function , but translated by one and two flux quanta in the @xmath171 direction respectively . even when the many - body berry curvature of a fci state is strongly varying , it is often centered around the value corresponding to the filling fraction and , since it is periodic , the hall conductivity obtained by integration is very close to the quantized value . this occurs even if the fci state is not a ground state . in the example presented in fig . [ fig : berry ] , the hall conductivity is in all three cases equal to @xmath3 within the numerical accuracy , which we will now discuss . ) , for each of the three quasi - degenerate fci states as a function of ( a ) the interaction strength @xmath150 and ( b ) the third - neighbor hopping @xmath130 ( right ) . not the different scales on the vertical axes . ] the hall conductivities we have calculated are very close to the expected exact values in the fci regime , despite the fact that the integrations involved are performed numerically . the relative deviation with respect to the exact value for the case of @xmath3 is presented in fig . [ fig : berry ] . this example illustrates that the error in the hall conductivity due to the finite size of the system should be smaller than 1% , at least for the kubo formula approach , even for small systems . in the case of clearly gapped fci ground states , the accuracy is even better , due to the integrand being very smooth . in all cases , we have used simple riemann summation to evaluate integrals , in order to obtain upper bounds for the numerical errors . other methods , like the simpson rule , would converge for smaller grid sizes . it should be noted that when the hall conductivity of an individual state within the degenerate ground state is evaluated , the integration should be extended over the whole period of the berry curvature . this is no longer necessary when calculating the average hall conductivity and the integration range can then be restricted to @xmath172 . the effects of interaction and band dispersion can also be traced in the behavior of the hall conductivity . this is shown in fig . [ fig : chern ] . outside the fci regime , where low - energy levels cross upon flux insertion , the hall conductivity oscillates . it converges to @xmath158 at about the value of @xmath150 for which the fci states separate from the excited - state spectrum , forming a three - fold degenerate ground state . on the other hand , the hall conductivity of these states remains close to 1/3 with deviation from the flat - band limit , and only changes smoothly with @xmath130 , converging to the expected value close to the phase boundary discussed in sec . [ sec : phdiag ] . fqh states must be robust against disorder , as long as the energy scale of the disorder is smaller than the gap . @xcite the magnetic texture in the three - orbital model presented is generated spontaneously by itinerant electrons , so an expected source of disorder are local inhomogeneities in the chiral spin pattern . a simplistic approach to simulate this effect is to vary the flux picked up by an electron hopping along one selected bond of the finite cluster . this is done by varying the phase @xmath173 in the phase factor in front of the corresponding hopping . the effect of such a variation on hall conductivities at @xmath3 is shown in fig . [ fig : disorder ] . . ( c , d ) standard deviation @xmath174 of total berry curvature @xmath175 , defined by adding up the many - body berry curvatures @xmath170 of the three lowest - energy states for each value of @xmath148 , and relative deviation @xmath176 of hall conductivity as a function of grid size . @xmath168 is the number of @xmath148 points taken in the range @xmath177 in each direction . the results shown in the right panels are for @xmath178 , but are qualitatively the same for all other values of @xmath173 . ] as long as the energy scale of the disorder is small enough , the ground states remain separated by a gap from the rest of the spectrum and their average hall conductivity remains constant . the hall conductivities of the individual quasi - degenerate ground states are not smooth functions of magnetic disorder . in particular , as soon as the impurity is switched on , the hall conductivity of each individual state jumps to an integer value . also , berry curvatures are no longer smooth functions of @xmath148 . nevertheless , the hall conductivities are still numerically well defined , as demonstrated in the right panels of fig . [ fig : disorder ] , and their average , which is the proper observable quantity in the thermodynamic limit , remains constant as the disorder is varied . this invariance of the hall conductivity directly demonstrates the topological robustness of fci states . again in analogy to fqh states,@xcite disorder introduces further splitting , apart from the one due to dispersion discussed in sec . [ sec : spectrum ] , between fci ground - state eigenvalues , which no longer exchange places under flux insertion . this splitting is expected to be a finite - size effect and should disappear in the thermodynamic limit . a conclusive proof of this statement is however beyond the scope of this work . despite the similarities , the impact of disorder in general on the energy - scale balance necessary for fci states is qualitatively different than the situation in traditional fqh states . for example , disorder may have a different effect depending on the interaction strength and range , and may lead to another , possibly topologically trivial , state , as has been shown in a recent study on the checkerboard lattice involving chemical - potential disorder . @xcite early on , it was pointed out that an important prerequisite for the emergence of a fci state is the balance between three energy scales , namely the width of the topologically non - trivial band , the gap(s ) separating it from other bands and the coulomb interaction strength . @xcite if the ratio between the first two , as expressed by the figure of merit eq . ( [ eq : merit ] ) , is large , then the interaction can become strong enough to induce fci states , but remain small compared to the band gaps , which avoids mixing in wave function of different topological character . many investigations have been focused on models in the perfectly flat band limit ( exactly as in a landau level ) , supplemented by various types of interaction . rather recently , it was reported that the band flatness alone is not in fact a reliable indicator for the stability of fci states . @xcite + + + in this section , we discuss how a finite dispersion influences the stability of fci states in a triangular - lattice model , where we focus on band filling and fs nesting . the first presents a rather obvious difference to lls , as has recently also been mentioned in ref . : while a ll is expected to be particle - hole symmetric , the nearly flat subband of a lattice model is not . we are going to discuss filling fractions @xmath179 and @xmath4 that turn out to clearly exemplify this difference . the latter case , which was also shortly discussed in ref . , corresponds to @xmath180 filling of the original triangular lattice , where nn coulomb interaction can stabilize a cdw [ with particles at second - neighbor sites , see fig . [ fig : tri_cdw ] ] , while no such cdw is possible at @xmath3 . the competition between the fci and the cdw is in turn strongly affected by fs nesting , and we are going to see that the fci is far more stable for more dispersive bands without nesting than for flatter bands with a nested fermi surface . however , we are also going to see that very flat bands allow for fci states at the lowest interaction strengths , even though they are in our case very well nested . and @xmath150 , for ( a ) @xmath3 and ( b ) @xmath4 , shown in colorcode . the dashed lines indicate the phase boundary , defined by the condition that the gap remains open for all values of inserted magnetic flux . the flatness ratio @xmath132 of the flat band of @xmath131 ( bottom scale ) is adjusted by varying @xmath130 ( top scale ) . the maximum value of the flatness ratio ( @xmath181 ) is marked by the dotted lines . ] as mentioned in sec . [ sec : eff_model ] , varying @xmath130 allows us to tune the band flatness and to moreover switch between regimes with and without fs nesting . examples are shown in fig . [ fig : fbz_fs ] , where the dispersion as well as the fermi surface corresponding to @xmath4 are shown for some values of @xmath130 . both for near - optimal @xmath182 ( @xmath183 ) and for smaller @xmath184 ( @xmath185 ) , the fs contains hexagons with almost perfectly nested segments . a difference between the two cases is on one hand the flatness ratio , but on the other hand , the flatter bands also have an additional circular fs around the @xmath186 point . for @xmath187 , only the circular fs remains ( see the example with @xmath188 and @xmath189 ) and there is thus no longer good nesting . figure [ fig : phdiag ] shows the region where fci states are stable on a 24-site ( @xmath138 unit - cell ) system , as a function of coulomb interaction strength @xmath2 and the flatness ratio @xmath132 , which is in turn controlled by third - neighbor hopping , see fig . [ fig : disp ] . the two panels are for fillings @xmath3 and @xmath4 . in both cases , the ground - state manifold in the fci state is expected to be three - fold degenerate . the colorcode indicates the gap between the three ground states and the fourth lowest eigenstate in the absence of applied magnetic flux . in order to define a physically meaningful phase boundary , however , one has to make sure that the ground states remain gapped upon flux insertion , not only in their own momentum sector , but also across momentum sectors , see sec . [ sec : spectrum ] . the solid lines in fig . [ fig : phdiag ] indicate the phase boundary determined by taking these considerations into account . we have verified that the phase diagrams obtained from a 30-site torus are qualitatively the same as the ones shown here . the stability range of the fci differs for @xmath4 and system sizes commensurate with the cdw ( @xmath190 and @xmath191 sites ) , but the corresponding phase - diagram area is still finite . and @xmath150 at filling fractions ( a , b ) @xmath3 and ( c - f ) @xmath4 . the sharp peaks indicate the formation of a charge - ordered state . in the fci regime , the static structure factor of all three quasi - degenerate ground states is identical.[fig : sf ] ] for both fillings , the system is in a metallic state at small @xmath150 , while finite values of @xmath150 can lead to fci ground states for an extended range of the band flatness . in the case of @xmath3 , the fci state can be induced for any band dispersion by making @xmath2 large enough . the fci persists even for @xmath150 considerably larger than the band gap . this has also been observed for another fci model on the checkerboard lattice , @xcite and appears to indicate that the fermions are dilute enough to occupy mostly one subband regardless of @xmath150 , so that a mixing of the topological character of the two bands does not occur . in contrast , the @xmath4 fci states survive only for moderate values of @xmath150 . larger interaction strengths lead to a different ground state , which is , as we will argue below , a cdw . to determine the type of order in each of the regions in fig . [ fig : phdiag ] , we calculate key properties of the corresponding ground states . the fci states , despite the hints from the energy spectrum , have to be identified by their topological invariant , see sec . [ sec : chern ] . to find out whether there is tendency towards charge order in the rest of the phase diagram , we calculate the static ( charge-)structure factor ( ssf ) , defined as : @xmath192 where @xmath193 is the number of sites , @xmath194 is the electron - number operator acting on the site at position @xmath195 , @xmath196 is the average electron number and @xmath197 stands for the many - body ground state . charge - density modulations are marked by sharp features in @xmath198 at certain wave vectors . liquid states , on the other hand , should be featureless in comparison to charge - modulated states . even though the accessible system sizes are not large enough to exemplify the featurelessness of liquid states , a qualitative difference between liquid and charge - modulated states can be seen . examples are shown in fig . [ fig : sf ] . at @xmath3 and within the fci regime , @xmath199 remains almost unchanged upon variation of model parameters . the same holds , although less markedly , for the metallic , fermi - liquid - like state at small interaction strengths and small ( or very large ) @xmath130 . despite the fact that the shape of @xmath198 is distinct for the two liquid states , see figs . [ fig : sf](a ) and ( b ) , the differences are subtle . in order to distinguish between the fci and metallic states , one would thus rather use the topological invariant of their ground states , see sec . [ sec : chern ] , or the criterion that the quasi - degenerate ground states remain separated from higher states for all fluxes , as used for fig . [ fig : phdiag ] , see above . however , the charge - structure factor is very valuable in distinguishing the weakly correlated fermi liquid from a cdw driven by coulomb repulsion . at @xmath4 and @xmath200 , away from the fci regime , two peaks appear in @xmath198 and increase continuously upon increasing @xmath150 , see see fig . [ fig : sf](d ) . when choosing @xmath142 near the maximal flatness [ see fig . [ fig : sf](c ) ] or @xmath201 with bad fs nesting [ see fig . [ fig : sf](f ) ] , the peaks only begin to grow for large @xmath2 , where ground - state behavior upon flux insertion changes : a full gap is only obtained for lattice sizes commensurate with the ordering pattern , like the @xmath202 and @xmath203 lattices , but not on more general lattices . this indicates that the fci state breaks down and is replaced by a cdw . the peaks observed in @xmath198 for @xmath4 grow with @xmath2 , supporting their relation to a cdw . their wave vectors correspond to a state where particles sit at nnn sites on the triangular lattice , i.e. , to the regular charge pattern compatible with a filling of @xmath204 of the triangular lattice . as can be seen in fig . [ fig : tri_cdw ] , this pattern avoids any penalty due to nn coulomb interaction @xmath2 and large enough @xmath205 will thus eventually induce such a charge distribution . the phase diagrams in fig . [ fig : phdiag ] clearly show that the band flatness parametrized by @xmath132 is itself not a reliable indicator for the stability of fci states , as has also been pointed out recently in a different case . @xcite the energy gap for @xmath3 is rather symmetric with respect to the highest figure of merit , but the phase boundary determined by requiring a full gap over all momentum sectors and for all fluxes shows that the fci states are somewhat more stable for larger @xmath130 . in the case of @xmath4 , the asymmetry is far more striking ; as one can see in fig . [ fig : phdiag](b ) , fci states require rather flat bands with @xmath206 for @xmath207 , but extend to a band with a width comparable to the gap separating it from its counterpart for @xmath208 . having rather flat bands indeed makes it easier for small @xmath150 to induce fci states both at @xmath3 and @xmath4 , even though the optimal @xmath132 is still not quite the largest , at least for our system sizes . as soon as the bands acquire some dispersion , however , features beyond band flatness , in our case fs nesting , can strongly influence the stability of fci states by favoring competing states , in our case a cdw . another key feature , which extends previous results , @xcite is that in the case of competition between fci and other phases , perfect band flatness is not necessarily the ideal condition for the stability of fci ground states . before passing , a few more comments on the phase diagram have to be made . ideally , a ground - state property would be used to determine the phase boundaries . one such property , which is sensitive to phase transitions , is the ground - state fidelity , defined as a measure of the overlap @xmath209 , where @xmath210 is the ground - state wave function and @xmath30 is a control parameter varied in small steps @xmath211 . phase transitions are then marked by a divergence in the fidelity at the transition point . having calculated the ground - state fidelity upon varying interaction strength and band flatness for the cases presented here , we find that such divergences occur only at points where ground - state level crossings also occur , so the fidelity does not provide any extra information compared to the eigenvalue spectra . furthermore , it has recently been shown that the fidelity on finite systems can fail to register topological phase transitions , e.g. , between a fqh - like state and a fermi liquid . based on an earlier investigation , @xcite we have illustrated extensively how fci states can emerge in a strongly correlated multiorbital model and have shown how an effective spinless one - orbital model with nearly flat bands and non - zero chern number @xmath1 arises as the the low - energy limit of kondo - lattice and hubbard models for @xmath0 orbitals on a triangular lattice . nn coulomb interaction @xmath2 then stabilizes states with all the characteristics of fci states : the lowest eigenvalues have a near degeneracy corresponding to the denominator of the filling fraction and show spectral flow without closing of the energy gap . we have , moreover , demonstrated that the states of the ground - state manifold of this phase have a non - zero hall conductivity , which is shown to be precisely quantized and equal to the filling fraction , in units of @xmath159 . the exact quantization of the hall conductivity holds also when fci states are not ground states , and survives when the many - body berry curvature is not a particularly flat function . deviating from the exact filling fraction @xmath3 by either removing electrons or increasing the system size , we found indications of fractional quasihole statistics in the eigenvalue spectra , by application of the state - counting rule elaborated in ref . . the eigenvalue spectra contain features that remind one of both the hierarchy and the composite fermion pictures of fqh states , therefore pushing the correspondence between fci and fqh states one step further . since in the present context the magnetic texture is generated by strongly correlated itinerant electrons , imperfections in this texture are to be expected . we therefore investigate the impact of a localized impurity that mixes states across momentum sectors . disorder causes splitting of the energy levels corresponding to the degenerate fci ground states , thus making these states inequivalent in finite systems . however , the gap between these states and the excited - state spectrum remains open and their average hall conductivity remains accurately fixed to the value of the filling fraction , even though the hall conductivities of the individual states deviate from that value . this behavior illustrates the topological protection of the fci states . finally , we address the robustness of fci states with respect to band flatness , interaction strength , filling and fs nesting . very flat bands indeed turn out to require the weakest interaction to produce an fci state , corroborating the figure of merit @xmath132 [ see eq . ( [ eq : merit ] ) ] as a good first step in assessing candidate systems . as soon as one deviates from this limit , however , additional properties of the system become crucial : in contrast to a perfectly flat ll , strong particle - hole asymmetry can arise on a lattice , because some fillings allow cdws more easily than others . in addition , a cdw is favored by fs nesting and this in turn reduces the stability range of the fci states . nevertheless , it turns out that fci states can be obtained if either of the following conditions is met : ( i ) very flat bands , even near nesting and at a filling favorable to a cdw , and ( ii ) even rather dispersive bands , if the filling fraction or the absence of fs nesting are unfavorable to a cdw . this flexibility makes fci states appear more realistic . this work was supported by the deutsche forschungsgemeinschaft under the emmy - noether program ( s.k . and m.d . ) and the interphase program of the dutch science foundation nwo / fom ( jv ) . we thank j. van den brink for helpful discussions and b. a. bernevig for insightful comments . 58ifxundefined [ 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/physrevb.27.3383 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.51.605 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.52.1583 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.48.1559 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.49.957 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.67.937 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.80.1083 [ * * , ( ) ] link:\doibase 10.1103/physrevb.14.2239 [ * * , ( ) ] link:\doibase 10.1088/0953 - 8984/3/23/012 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.61.2015 [ * * , ( ) ] link:\doibase 10.1103/physrevb.74.085308 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.146802 [ * * , ( ) ] link:\doibase 10.1103/physrevb.62.6065 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.156402 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.236802 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.236803 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.236804 [ * * , ( ) ] link:\doibase 10.1038/ncomms1380 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.146803 [ * * , ( ) ] link:\doibase 10.1103/physrevx.1.021014 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.075116 [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevlett.108.126405 [ * * , ( ) ] @noop @noop @noop @noop http://link.aps.org/doi/10.1103/physrevb.86.201101 [ * * , ( ) ] arxiv:1204.1697v1 ( 2012 ) http://link.aps.org/doi/10.1103/physrevlett.109.186805 [ * * , ( ) ] @noop http://link.aps.org/doi/10.1103/physrevb.86.205125 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.075128 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.126803 [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevb.86.085129 [ * * , ( ) ] @noop @noop @noop link:\doibase 10.1103/physrevb.85.241308 [ * * , ( ) ] link:\doibase 10.1140/epjb / e2011 - 20857 - 6 [ * * , ( ) ] link:\doibase 10.1038/ncomms1602 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.266801 [ * * , ( ) ] http://arxiv.org/abs/1106.4439 [ * * , ( ) ] \doibase info : doi/10.1143/jpsj.79.083711 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.216405 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.266405 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.78.1323 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.91.257003 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1103/physrev.94.1498 [ * * , ( ) ] @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevb.41.9377 [ * * , ( ) ] link:\doibase 10.1103/physrevb.30.1097 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.68.3460 [ * * , ( ) ] link:\doibase 10.1103/physrevb.49.2947 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.63.199 [ * * , ( ) ] link:\doibase 10.1103/physrevb.31.3372 [ * * , ( ) ] link:\doibase 10.1103/revmodphys.82.1959 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.90.256802 [ * * , ( ) ] http://link.aps.org/doi/10.1103/physrevb.85.205124 [ * * , ( ) ]

we introduce a green s function method for handling radiative effects on false vacuum decay . in addition to the usual thin - wall approximation , we achieve further simplification by treating the bubble wall in the planar limit . as an application , we take the theory , extended with additional heavier scalars , wherein we calculate analytically both the functional determinant of the quadratic fluctuations about the classical soliton configuration and the first correction to the soliton configuration itself .

1501.07466

the association of the still recently - discovered @xmath2 scalar particle @xcite with the higgs boson of the standard model ( sm ) places the stability of the electroweak vacuum under question @xcite . this instability , arising at an energy scale around @xmath3 @xcite , results from the renormalization - group ( rg ) running of the higgs self - coupling , whose value is driven negative by contributions dominated by top - quark loops . state - of - the - art calculations suggest that the electroweak vacuum is metastable , having a lifetime longer than the present age of the universe and lying at the edge of the stable region @xcite , where seemingly - small corrections may have a material impact upon predictions . the uncertainty in such predictions remains , at present , dominated by that of the top - quark pole mass @xcite . even so , it has been suggested @xcite that , regardless of any improved precision in the experimental determination of the latter , the presence of planck - scale operators may weaken the claim of metastability . nevertheless , having , as yet , no experimental evidence of additional stabilizing physics between the electroweak and planck scales , it is provident to consider approaches to the calculation of tunneling rates that can consistently account for radiative corrections . the degree of vacuum metastability provides a strong criterion for the phenomenological viability of extensions to the sm . for example , supersymmetric scenarios can be ruled out if the electroweak symmetry - breaking vacuum decays into a color - breaking one in a timescale shorter than the age of the universe @xcite . in addition , transitions between vacua can also occur at finite temperature @xcite . in the context of early - universe cosmology , this is of interest because the corresponding first - order phase transitions may leave behind relic gravitational waves @xcite . moreover , such phase transitions may turn out to be pivotal for generating the cosmic matter - antimatter asymmetry @xcite . as a consequence of these applications and the wide range of phenomenological models , there are now routine methods for computing transition rates at both vanishing and finite temperature @xcite . vacuum transitions in scalar theories can be described in the following way @xcite . in the event that there are two non - degenerate vacua , an initially homogeneous system lying in the false vacuum will spontaneously nucleate bubbles of true vacuum , leading to the production of domain walls or `` kinks . '' the latter are the topological solitons that interpolate between regions of true and false vacuum . the study of these `` solitary wave '' solutions to non - linear equations of motion ( see e.g. ref . @xcite ) has a long history @xcite , and archetypal examples of such field configurations arise in the sine - gordon model @xcite and the @xmath4 theory with tachyonic mass @xmath5 . the semi - classical @xcite and quantum @xcite descriptions of false vacuum decay in the latter theory were presented in the seminal works by coleman and callan ( see also ref . early expansion on these works included induced vacuum decay @xcite and the incorporation of gravity @xcite . in order to decide whether a vacuum configuration is unstable , i.e. whether there exists a lowest - lying true vacuum , it is often necessary to account for the impact of radiative corrections . this is of particular relevance when the appearance or disappearance of minima is entirely a radiative effect @xcite , such as occurs for the coleman - weinberg ( cw ) mechanism of spontaneous symmetry breaking @xcite or in symmetry restoration at finite temperature @xcite . in phenomenological studies , this is commonly done by calculating the tunneling rate from the effective potential @xcite of a _ homogeneous _ field configuration @xcite , which is subsequently promoted to a space - time - dependent configuration . this practice is problematic for two reasons . firstly , the temporal and spatial inhomogeneity of the solitonic background is not fully taken into account @xcite . secondly , in the presence of tachyonic instabilities , e.g. when there are non - convex regions in the tree - level potential , the perturbatively - calculated effective potential receives a seemingly - pathological imaginary part . the latter has been shown @xcite to have a physical interpretation as a decay rate for an initially homogeneous field configuration ( see also ref . however , this subtlety can be circumvented by using constructions such as the coarse - grained effective action @xcite or , as is often employed in lattice simulations , the constraint effective potential @xcite ( cf . in addition , within the context of the sm , the standard rg improvement of the effective potential has recently been questioned @xcite . for the above reasons , it is reasonable to conclude that the use of the effective potential to calculate transition rates is neither satisfactory nor justifiable and it is desirable to consider alternative methods for determining quantum corrections , which can be applied to a wide range of models that feature vacuum decay . the first quantum corrections @xcite to the tunneling rate are those arising from the functional determinant over the quadratic fluctuations about the classical soliton configuration . in the case of one - dimensional operators , these determinants may be calculated using the gelfand - yaglom theorem @xcite , which may be generalized to higher dimensions in the case of radially - symmetric operators @xcite . general numerical techniques may then be obtained @xcite for calculating tunneling rates beyond the so - called thin - wall approximation , in which the width of the bubble wall is much smaller than its radius . these approaches have also been applied to radially - separable yang - mills backgrounds @xcite and scenarios in curved spacetime @xcite . alternatively , as we will employ , the functional determinant may be calculated by means of the so - called heat kernel method ( see e.g. refs . @xcite ) , based upon the schwinger proper - time representation and zeta function regularization @xcite . previously , this approach has been used to derive approximate analytic results for the one - loop fluctuation determinant beyond the thin - wall approximation @xcite , as well as one - loop corrections to sphaleron rates @xcite . the latter have also been calculated by direct integration of the green s function @xcite . recently , it has been shown that properties of topological solitons may be studied non - perturbatively using monte carlo and lattice simulations by considering correlation functions directly @xcite . other authors have proposed methods for calculating quantum corrections based upon functional renormalization techniques @xcite . in this article , we derive an _ result for the green s function of the @xmath6 theory in the background of the classical kink solution . within the thin- and planar - wall approximations , we illustrate that this green s function may be used to determine _ analytically _ the leading quantum corrections to both the semi - classical bounce action and the kink solution itself , thereby allowing us to compute the tunneling rate at the two - loop level , while isolating its diagrammatic interpretation . the latter calculation is performed within the context of a toy model extended with an additional @xmath1 heavier scalars , where the parametric dependence on @xmath1 allows the identification of a concrete example in which the calculated two - loop corrections dominate over the neglected higher - loop corrections . we illustrate that the problem of calculating these radiative corrections may be reduced to one of solving one - dimensional ordinary differential equations and integrals . thus , we anticipate that this methodical development may have numerical applications in the study of the decay rates of radiatively - generated metastable vacua , such as occur in the massless cw model @xcite or the higgs potential of the sm . similar methods based upon green s function techniques have been used previously to determine self - consistent bounce solutions _ numerically _ in the hartree approximation of the pure @xmath0 theory in both two and four dimensions @xcite . the remainder of this article is organized as follows . in sec . [ sec : classb ] , we review the calculation , _ la _ coleman and callan @xcite , of the classical `` bounce '' configuration , describing the semi - classical tunneling rate between two quasi - degenerate vacua and its first quantum corrections . in sec . [ sec : greens ] , we outline a green s function method for the evaluation of the functional determinant over the quantum fluctuations about the classical bounce , making comparison with existing calculations . subsequently , in sec . [ sec : corrb ] , we illustrate that this green s function method may be used to calculate analytically and self - consistently the first quantum corrections to the bounce itself . in sec . [ sec : conc ] , we conclude our discussions and highlight potential applications and future directions . finally , a number of mathematical appendices are included , outlining the technical details of the calculations summarized in secs . [ sec : greens ] and [ sec : corrb ] . we consider a real scalar field @xmath7 , with four - dimensional euclidean lagrangian @xmath8 and classical potential @xmath9 the mass squared is @xmath10 , @xmath11 is of mass dimension one , @xmath12 is dimensionless , @xmath13 is a constant and @xmath14 denotes the derivative with respect to the euclidean spacetime coordinate @xmath15 . throughout , we omit spacetime and field arguments for notational convenience when no ambiguity results . ( left panel ) and the inverted potential @xmath16 ( right panel ) . the arrow ( right panel ) indicates the trajectory of the `` bounce '' in imaginary time @xmath17.[fig : potential],title="fig : " ] ( left panel ) and the inverted potential @xmath16 ( right panel ) . the arrow ( right panel ) indicates the trajectory of the `` bounce '' in imaginary time @xmath17.[fig : potential],title="fig : " ] the classical potential in eq . has non - degenerate minima at @xmath18 as depicted in fig . [ fig : potential ] ( left panel ) , where we have defined @xmath19 and @xmath20 . the separation of the minima @xmath21 and the difference in their energy densities @xmath22 may be written in terms of the parameters @xmath23 where the approximations are valid in the limit @xmath24 , i.e. @xmath25 . for @xmath26 , @xmath27 and the minima at @xmath28 become degenerate , as we would expect . finally , the constant @xmath13 is chosen so that the potential vanishes in the false vacuum at @xmath29 , requiring @xmath30 and thus giving the barrier height to be @xmath31 and @xmath32 . the semi - classical probability for tunneling between the false ( @xmath33 ) and true ( @xmath34 ) vacua and its first quantum corrections were described in the seminal works by coleman and callan @xcite , which we now review . the classical equation of motion @xmath35 where @xmath36 denotes the derivative with respect to the field @xmath37 , is analogous to that of a particle moving in a potential @xmath38 . the boundary conditions of the `` bounce '' are @xmath39 and @xmath40 , where @xmath41 denotes the derivative with respect to @xmath42 . these correspond to a particle initially at @xmath43 rolling through the valley in @xmath38 , reaching a turning point close to @xmath44 , before rolling back to @xmath45 , see fig . [ fig : potential ] ( right panel ) . finally , in order to ensure that the action of the bounce is finite , we require @xmath46 . translating to four - dimensional hyperspherical coordinates , eq . takes the form @xmath47 where @xmath48 . the boundary conditions become @xmath49 and @xmath50 , where the latter ensures that the solution is well defined at the origin . thus , the bounce corresponds to a four - dimensional bubble of some radius @xmath51 , which separates the false vacuum ( @xmath33 ) outside from the true vacuum inside ( @xmath34 ) . analytically continuing to minkowski spacetime ( @xmath52 ) , the @xmath53 symmetry of the bounce becomes an @xmath54 symmetry , with the bubble expanding along the hyperbolic trajectory @xmath55 . the bounce action is @xmath56 \ ; , \ ] ] which can be written as @xmath57 , where @xmath58 are the contributions from the surface tension of the bubble and the energy of the true vacuum , respectively . in writing eq . , we have used the fact that for @xmath59 , i.e. for @xmath60 , we may show that the bounce @xmath37 satisfies the virial theorem @xmath61 i.e. it is the configuration of zero total energy density . notice that there is no contribution to the bounce action [ eq . ] from the exterior of the bubble , since the choice of the potential eq . @xmath13 , ensures that the false vacuum has zero energy density . in the thin - wall approximation , we may safely neglect the damping term in eq . and the contribution from the cubic self - interaction @xmath62 , as will be the case for the remainder of this article . we then obtain the well - known kink solution @xcite @xmath63 \ ; , \ ] ] with @xmath64 . the radius @xmath51 of the bubble is then obtained by extremizing the bounce action eq . , that is by minimizing the energy difference between the surface tension of the bubble and the true vacuum . this gives @xmath65 by considering the invariance of the bounce action in eq . under general coordinate transformations , i.e. @xmath66 , we may show that @xmath67 this is to say that @xmath68 in which case we find @xmath69 the decay rate of the false vacuum , i.e. the probability per unit time for the nucleation of a bubble of true vacuum , has the generic form @xcite @xmath70 here , @xmath71 is the three - volume within which the bounce may occur , arising from integrating over the center of the bounce , and @xmath72 contains the quantum corrections to the classical bounce action @xmath73 that are the subject of the remainder of this article . the tunneling probability in eq . may be obtained from the path integral @xmath74 \ = \ \int \ ! [ { \mathrm{d}}\phi ] \ ; e^{- \ , s[\phi ] / \hbar } \ ; , \ ] ] via @xmath75 | / t \ ; , \ ] ] where @xmath76 is the euclidean time of the bounce . in order to evaluate the functional integral over @xmath77 , we first expand around the classical bounce @xmath37 , whose equation of motion is obtained from @xmath78}{\delta \phi } \bigg|_{\phi \ , = \ , \varphi } \ = \ 0 \ ; .\ ] ] writing @xmath79 , where the factor of @xmath80 is written explicitly for bookkeeping purposes , we find @xmath81 \ = \ s[\varphi ] + \ , \frac{\hbar}{2 } \ ! \int \ ! { \mathrm{d}}^4 x \ ; \hat{\phi}(x ) \ , g^{-1}(\varphi;x ) \ , \hat{\phi}(x ) + \mathcal{o}(\hbar^{3/2 } ) \ ; , \ ] ] where @xmath82 \equiv b$ ] is the classical bounce action and @xmath83}{\delta \phi^2(x)}\bigg|_{\phi \ , = \ , \varphi } \ = \ - \ , \delta^{\ ! ( 4 ) } + u''(\varphi ; x ) \ ; , \ ] ] in which @xmath84 is the four - dimensional laplacian . before proceeding to perform the functional integration over the quadratic fluctuations about the bounce , we must consider the spectrum of the operator @xmath85 , which is not positive definite . by differentiating the equation of motion eq . with respect to @xmath86 and comparing with the eigenvalue equation @xmath87 it is straightforward to show that there exist four zero eigenmodes @xmath88 , transforming as a vector of @xmath89 and resulting from the translational invariance of the bounce . the normalization @xmath90 follows from eq . , since @xmath91 thus , @xmath92 . differentiating eq . with respect to @xmath93 and subsequently setting @xmath94 in those terms originating from the damping term , we can show that there also exists a discrete eigenmode @xmath95 . this eigenmode transforms as a scalar of @xmath89 , corresponding to dilatations of the classical bounce solution , and has the negative eigenvalue @xmath96 it is this lowest mode that is responsible for the path integral in eq . obtaining the non - zero imaginary part in eq . @xcite . alternatively , we may solve the eigenvalue problem directly in hyperspherical coordinates ( see appendix [ app : greens ] ) , by making the substitution @xmath97 . neglecting the damping term and setting @xmath98 in the centrifugal potential , we obtain the eigenspectrum @xmath99 the radial parts of the eigenfunctions are the associated legendre polynomials of the first kind and of order 2 , i.e. @xmath100 . thus , demanding normalizability , the quantum number @xmath101 is restricted to the set @xmath102 . from eq . , we see that the negative mode corresponds to @xmath103 ( @xmath104 , @xmath105 ) and the zero modes correspond to @xmath106 ( @xmath107 , @xmath108 ) , having degeneracy @xmath109 . the lowest two positive - definite eigenvalues are @xmath110 ( @xmath111 , @xmath112 ) and @xmath113 ( @xmath114 , @xmath115 ) . thus , for @xmath51 large , the `` continuum '' of positive - definite modes begins at @xmath116 , cf . @xcite . in order to perform the functional integral over the five negative - semi - definite discrete modes , we expand @xmath117 , where @xmath118 comprises the continuum of positive - definite eigenmodes . the functional measure then becomes @xmath119 \ = \ [ { \mathrm{d}}\phi_+ ] \prod_{i = 0}^4 ( 2 \pi \hbar)^{- 1/2 } { \mathrm{d}}a_i \ ; .\ ] ] the functional integral over the four zero eigenmodes ( @xmath120 ) is traded for an integral over the collective coordinates of the bounce @xcite ( see appendix [ app : zeromode ] ) and yields a factor @xmath121 the integral over the negative eigenmode ( @xmath122 ) may be performed using the method of steepest descent , giving an overall factor of latexmath:[$- i unphysical @xcite and depends on the choice of analytic continuation , thereby justifying the modulus sign in eq . . finally , the gaussian integral over the continuum of positive eigenmodes @xmath118 may be performed in the usual manner and we obtain @xmath124 \ = \ e^{- b / \hbar } \left| \frac{\lambda_0 \ , \mathrm{det}^{(5 ) } \ , g^{-1}(\varphi ) } { \frac{1}{4 } ( vt)^2 \big(\frac{b}{2 \pi \hbar}\big)^{\!4 } ( 4 \gamma^2)^5 \ , \mathrm{det}^{(5 ) } \ , g^{-1}(v ) } \right|^{-\tfrac{1}{2 } } \ ; , \ ] ] in which @xmath125 , cf . ref . @xcite , denotes the determinant calculated only over the continuum of positive - definite eigenmodes , i.e. omitting the zero and negative eigenmodes , whose contributions are included explicitly . in addition , we have normalized the determinant to that of the operator @xmath126 , evaluated in the false vacuum . substituting eq . into eq . , we find the tunneling rate per unit volume @xmath127 \ ; , \end{aligned}\ ] ] where @xmath128 contains the one - loop corrections from the quadratic fluctuations around the classical bounce . here , @xmath129 indicates that we are to trace over only the positive - definite eigenmodes . in this section , we outline the derivation of the green s function of the operator in eq . . the technical details are included for completeness in appendix [ app : greens ] . subsequently , we use this green s function to evaluate the functional determinant in eq . and obtain the correction from quadratic fluctuations . in addition , we calculate analytically the tadpole contribution to the effective equation of motion and point out that this may be used to calculate the first quantum corrections to the bounce . we have the inhomogeneous klein - gordon equation @xmath130 where @xmath131 is the four - dimensional dirac delta function . working in hyperspherical coordinates and writing @xmath132 , where @xmath133 are four - dimensional unit vectors , the green s function may be expanded as @xmath134 where @xmath135 and @xmath136 are the chebyshev polynomials of the second kind ( see appendix [ app : greens ] ) . the radial functions @xmath137 satisfy the inhomogeneous equation @xmath138 g_j(\varphi ; r , r')\ = \ \frac{\delta(r - r')}{r'^3 } \ ; .\end{aligned}\ ] ] for the thin wall , we safely neglect the damping term and approximate the centrifugal term by @xmath139 . for self - consistency of this approximation , we also replace the discontinuity on the rhs of eq . with @xmath140 . for generality of notation in what follows , it is then convenient to define @xmath141 being a function only of the normalized bounce @xmath142\ ] ] and the parameter @xmath143 the full green s function may then be written @xmath144 with the above approximations , the lhs of eq . is of pschl - teller form @xcite , having general solutions that may be expressed in terms of the associated legendre functions ( see appendix [ app : greens ] ) . we are then able to find the full analytic solution @xmath145 \ ; , \end{aligned}\ ] ] where @xmath146 is the generalized unit - step function . taking the coincidence limit @xmath147 , @xmath148 , the local contribution to the green s function @xmath149 in eq . takes the form @xmath150 where @xmath151 \ ; .\end{aligned}\ ] ] in eq . , the summation over @xmath152 corresponds to the contributions from the two towers of positive - definite eigenmodes of the operator @xmath153 , see eq . . for @xmath51 large , we may approximate the summation over @xmath154 by an integral over a continuous variable @xmath155 ( see appendix [ app : greens ] ) . in which case , we obtain @xmath156 with @xmath157 ] the continuum limit described above is entirely equivalent to the so - called planar - wall approximation . therein , for @xmath51 large , we align a set of coordinates @xmath158 with the bubble wall , as shown in fig . [ fig : planar ] . we may then fourier - transform with respect to the coordinates @xmath159 that lie within the three - dimensional wall , introducing a three - momentum @xmath160 , i.e. @xmath161 where we have let @xmath162 for notational convenience . the three - momentum - dependent green s function @xmath163 satisfies the inhomogeneous klein - gordon equation @xmath164 we may then show straightforwardly that @xmath165 where @xmath166 is as defined in eq . with @xmath167 given by eq . this planar - wall approximation is employed for the remainder of this article . before making use of the green s function calculated in the preceding section , we first derive the equation of motion for the quantum - corrected bounce . this calculation was first suggested by goldstone and jackiw @xcite and , in the following sections , we will illustrate that , within the thin- and planar - wall approximations , it may be completed analytically . the one - particle irreducible ( 1pi ) effective action @xcite is given by the legendre transform @xmath168 \ = \ - \ , \hbar \ , \ln z[j ] + \int \ ! { \mathrm{d}}^4 x \ ; j(x ) \phi(x ) \ ; , \ ] ] where @xmath169}{\delta j(x)}\ ] ] is a functional of the source @xmath170}{\delta \phi(x)}$ ] and @xmath171 \ , = \int \ ! [ { \mathrm{d}}\phi ] \ ; \exp \ ! \bigg [ - \frac{1}{\hbar } \bigg ( s[\phi ] - \ ! \int \ ! { \mathrm{d}}^4 x \ ; j(x ) \phi(x ) \bigg ) \bigg ] \ ; .\ ] ] in order to obtain the quantum corrections to the bounce @xmath37 , we wish to evaluate the functional integral in eq . by expanding around the configuration @xmath172 , which is the solution to the quantum equation of motion @xmath173}{\delta \phi } \bigg|_{\phi = \varphi^{(1 ) } } \ = \ 0 \ ; .\ ] ] here , the superscript `` @xmath174 '' indicates that @xmath172 contains the first quantum corrections to @xmath37 . it follows from eq . that @xmath172 can not extremize the classical action in the absence of the source @xmath175 , i.e. @xmath176}{\delta \phi(x)}\bigg|_{\phi \ , = \ , \varphi^{(1 ) } } \ = \ \neq \ 0 \ ; .\ ] ] writing @xmath177 , where the factor of @xmath80 is again written explicitly for bookkeeping , we proceed as in sec . [ sec : classb ] , expanding @xmath178 \ = \ s[\varphi^{(1 ) } ] + \hbar^{1/2 } \ ! \int \ ! { \mathrm{d}}^4 x \ ; j(x ) \ , \hat{\phi}^{(1)}(x ) \nonumber \\ & \quad + \frac{\hbar}{2 } \int \ ! { \mathrm{d}}^4 x \ ; \hat{\phi}^{(1)}(x ) \ , g^{-1}(\varphi^{(1 ) } ; x ) \ , \hat{\phi}^{(1)}(x ) + \cdots \ ; , \end{aligned}\ ] ] where @xmath179}{\delta \phi^2(x)}\bigg|_{\phi \ , = \ , \varphi^{(1 ) } } \ = \ - \ , \delta^{(4 ) } + u''(\varphi^{(1 ) } ; x ) \ ; .\ ] ] we now write @xmath180 and expand about the quadratic fluctuations evaluated around the classical bounce @xmath37 . thus , in performing the functional integral , we consider the same spectrum of negative and zero eigenmodes as in sec . [ sec : classb ] . finally , by expanding the effective action @xmath181 $ ] in eq . around @xmath182 ( see ref . @xcite ) , we obtain @xmath183 \ = \ s[\varphi^{(1 ) } ] + \frac{i \pi \hbar}{2}+\hbar^2b^{(2)}{}'[\varphi ] \nonumber \\ & \qquad + \frac{\hbar}{2 } \ln \left| \frac{\lambda_0 \ , \mathrm{det}^{(5 ) } \ , g^{-1}(\varphi ) } { \frac{1}{4 } ( vt)^2 \big ( \frac{b}{2 \pi \hbar } \big)^4 ( 4 \gamma^2)^5 \mathrm{det}^{(5 ) } \ , g^{-1}(v ) } \right| + \cdots \;,\end{aligned}\ ] ] where @xmath184\ & = \ \frac{1}{2}\int\!{\mathrm{d}}^4 x\;\delta\varphi(x)\nonumber\\&\quad\times \frac{\delta}{\delta \varphi^{(1)}(x)}\,\ln\frac{\mathrm{det}^{(5)}g^{-1}(\varphi^{(1)})}{\mathrm{det}^{(5)}g^{-1}(v)}\bigg|_{\varphi^{(1)}=\varphi}\;.\end{aligned}\ ] ] functionally differentiating eq . with respect to @xmath172 , we obtain the equation of motion for the corrected bounce @xmath185 where @xmath186 containing the tadpole contribution @xmath187 comparing the functional derivative of eq . with eqs . and , we see that this evaluation of the effective action is self - consistent so long as the source @xmath188 which is , as expected , non - vanishing . we may show that the correction to the classical bounce @xmath189 satisfies the equation of motion @xmath190 the corrected bounce action @xmath191 $ ] contains contributions at order @xmath192 . specifically , @xmath193 \ = \ s[\varphi ] \nonumber \\ & \quad + \ , \frac{\hbar^2}{2 } \ ! \int \ ! { \mathrm{d}}^4 x \ ; \delta \varphi(x ) \ , g^{-1}(\varphi ; x ) \ , \delta \varphi(x ) + \mathcal{o}(\hbar^{3 } ) \ ; , \end{aligned}\ ] ] where we have used eqs . and . thus , using eq . , we may write @xmath194 \ = \ b + \hbar^2 b^{(2 ) } \ ; , \ ] ] where @xmath195 hence , we obtain the tunneling rate per unit volume @xmath196/\hbar } | / ( vt ) \nonumber\\ & = \ \bigg ( \frac{b}{2 \pi\hbar } \bigg)^{\ ! 2 } ( 2 \gamma)^5 | \lambda_0 |^{- \frac{1}{2}}\nonumber \\ & \quad \times \exp \ ! \bigg [ - \frac{1}{\hbar } \bigg ( b + \hbar b^{(1 ) } + \hbar^2 b^{(2)}+\hbar^2b^{(2 ) } { } ' \bigg ) \bigg ] \ , , \end{aligned}\ ] ] where @xmath73 is the classical bounce action ; @xmath197 , given in eq . , contains the corrections from quadratic fluctations about the classical bounce ; and @xmath198 , given in eq . , contains the contribution arising from the quantum corrections to the bounce itself . we note that @xmath199 such that the @xmath200 corrections to the quadratic fluctuations flip the sign of the contribution to the bounce action from the @xmath200 corrections to the bounce itself . we will now proceed to calculate explicitly the tadpole contribution appearing in eq . . introducing an ultraviolet cutoff @xmath201 , the @xmath202 integral can be performed in eq . , and we obtain @xmath203 \ ; .\end{aligned}\ ] ] we choose to define the physical mass and coupling in the homogeneous non - solitonic background . the renormalization conditions are then as follows : [ eq : renormconds ] @xmath204 where @xmath205 is the cw effective potential @xcite . the resulting mass and coupling counterterms are @xmath206 we then arrive at the renormalized tadpole correction @xmath207 \ ; .\end{aligned}\ ] ] we may calculate the traces appearing in the exponent of eq . , which arise from the functional determinant of the operator @xmath208 in eq . , by using the heat kernel method ( see e.g. ref . specifically , the trace may be written in the form @xmath209 the heat kernel @xmath210 is the solution to the heat - flow equation @xmath211 and satisfies the condition @xmath212 . it is convenient to work in terms of the laplace transform of the heat kernel @xmath213 which is the solution to @xmath214 in the planar - wall approximation , we take @xmath215 where @xmath216 satisfies @xmath217 comparing eq . with eq . , we see that @xmath218 is nothing other than the green s function @xmath219 in eq . with the replacement @xmath220 in @xmath167 , see eq . thus , we may write @xmath221(\tau ) \ ; , \end{aligned}\ ] ] where we have defined @xmath222 and @xmath223(\tau ) \ = \ \int_{\mathcal{c } } \ ! \frac{{\mathrm{d}}s}{2\pi i } \ ; e^{-s\tau } \ , \widetilde{g}(u , m)\ ] ] is the inverse laplace transform with respect to @xmath224 , with @xmath225 indicating the bromwich contour . we may perform the integrals in eq . analytically , proceeding in order from right to left and beginning with the inverse laplace transform . we then obtain the unrenormalized correction to the bounce action @xmath226 the technical details of the relevant integrations are included in appendix [ app : greens ] . adding the counterterm @xmath227 we obtain the final renormalized result @xmath228 in appendix [ app : greens ] , we reproduce this result by the method presented in ref . @xcite . ] repeating the analysis presented therein , as outlined in appendix [ app : greens ] , we find a result in agreement with eq . reported here , suggesting a numerical error in the factor of 50 above . ] we now discuss an example of the role played by loop corrections to the bounce itself . within the perturbation expansion , one should expect that these lead to second - order corrections to the classical action of the soliton simply because the latter is evaluated for a stationary path . there are , however , important situations , in which all one - loop contributions must be resummed in order to capture the leading quantum corrections to the action . examples include situations where the symmetry - breaking minima of the potential emerge radiatively through the cw mechanism @xcite . in the absence of a soliton , this implies that the classical solution , i.e. the homogeneous expectation value of the field , has to be found consistently by minimizing the one - loop effective potential as a _ function _ of the field expectation value itself . analogously , in order to find the decay rate of the false vacuum , the bounce must be computed consistently from the one - loop effective action , which is a _ functional _ of the bounce itself . the methods presented in this article reduce the problem of tunneling in radiatively - generated potentials to one - dimensional ordinary differential equations and integrals . it is anticipated that it should be possible to derive numerical solutions in future work . for the purpose of illustration , however , we remain herein on the ground of analytic and perturbative approximations . in order to enhance the corrections to the bounce compared to other quantum effects that appear at second order in perturbation theory , we extend the model in eq . with @xmath1 copies of an additional scalar field @xmath229 by adding to the lagrangian the terms @xmath230 here , we have chosen the coupling @xmath12 to be identical to the self - coupling of @xmath77 for the sake of simplicity in the green s function of the @xmath229 fields . since @xmath231 , the additional scalars do not impact upon the classical bounce in sec . [ sec : classb ] or the discussion of the green s function in sec . [ sec : greens ] . the klein - gordon equation for @xmath232 takes the form @xmath233 s(\varphi ; x , x ' ) \ = \ \delta^{(4)}(x - x ' ) \ ; .\ ] ] comparing with that of @xmath77 in eq . , we see that the green s function @xmath234 may be obtained straightforwardly from @xmath235 in eq . by making the replacement @xmath236 the renormalized tadpole contribution from each @xmath232 field , integrated over the three - momentum @xmath160 , is given by @xmath237 \ ; , \ ] ] where we have assumed @xmath238 for simplicity . the full form of @xmath239 and the relevant counterterms are provided in appendix [ app : scalc ] . the renormalized correction to the classical bounce @xmath189 is governed by the equation of motion @xmath240 \delta \varphi \ = \ \big ( \pi^r(u ) + n \sigma^r(u ) \big ) \ , \varphi \ ; , \end{aligned}\ ] ] cf . we obtain the solution by making use of the green s function @xmath241 , writing @xmath242 where we have used eq . in order to substitute @xmath37 . as a function of @xmath243 for @xmath244 ( solid ) , @xmath245 ( dashed ) , @xmath246 ( dash - dotted ) and @xmath247 ( dotted).[fig : deltaphi ] ] as a function of @xmath243 for @xmath244 ( solid ) , @xmath245 ( dashed ) , @xmath246 ( dash - dotted ) and @xmath247 ( dotted ) . we see clearly that the impact of the tadpole correction is to broaden the bubble wall.[fig : correctedbounce],title="fig : " ] + as a function of @xmath243 for @xmath244 ( solid ) , @xmath245 ( dashed ) , @xmath246 ( dash - dotted ) and @xmath247 ( dotted ) . we see clearly that the impact of the tadpole correction is to broaden the bubble wall.[fig : correctedbounce],title="fig : " ] we note at this point that @xmath235 is singular as @xmath248 ( or , equivalently , @xmath249 ) . nonetheless , the integral in eq . remains finite , since @xmath166 is multiplied with an odd function , whereas the singularity resides in its even part . it is therefore useful to define @xmath250 within the domain @xmath251 , this function can be expressed as @xmath252 + ( u \leftrightarrow u ' ) \ ; .\end{aligned}\ ] ] defining in addition @xmath253 \ ; , \\ p_1(u ) \ & = \ \gamma \ ! \int_{-1}^{1 } \ ! { \mathrm{d}}u ' \ ; u ' g^{\rm odd}(u , u ' ) \nonumber \\ & = \ \frac{1 - u^2}{8 } \ln \frac{1 + u}{1 - u } \ ; , \\ p_2(u ) \ & = \ \gamma \ ! \int_{-1}^1 \ ! { \mathrm{d}}u ' \ ; u'^3 g^{\rm odd}(u , u ' ) \nonumber \\ & = \ \frac{1 - u^2}{8 } \bigg [ \ln \frac{1 + u}{1 - u } - \frac{4}{3 } \ , u \bigg ] \ ; , \end{aligned}\ ] ] we find the result @xmath254 \ ; .\end{aligned}\ ] ] in fig . [ fig : deltaphi ] , we plot @xmath255 as a function of @xmath243 for a range of values of @xmath256 . we see from fig . [ fig : correctedbounce ] , which plots the corrected bounce @xmath257 for the same range , that the impact of this correction is to lower the height and broaden the width of the bubble wall . we note that this behaviour is in qualitative agreement with the results of the self - consistent numerical analysis in ref . @xcite , presented there for the pure @xmath0 theory in @xmath258 dimensions . diagrammatic representation of various contributions to the effective action : ( a ) is the one - loop term @xmath259 , ( b ) is the @xmath260 contribution to @xmath198 , and ( c ) and ( d ) are @xmath261 terms . solid lines represent the propagator @xmath262 , dotted lines @xmath263 . crosses denote insertions of the bounce @xmath37.,title="fig:",width=283 ] + .4 cm diagrammatic representation of various contributions to the effective action : ( a ) is the one - loop term @xmath259 , ( b ) is the @xmath260 contribution to @xmath198 , and ( c ) and ( d ) are @xmath261 terms . solid lines represent the propagator @xmath262 , dotted lines @xmath263 . crosses denote insertions of the bounce @xmath37.,title="fig:",width=181 ] substituting eq . into eq . , we find the correction to the bounce action @xmath264 \ ; .\end{aligned}\ ] ] in order to obtain a finite result for eq . , we have added to @xmath13 the correction @xmath265 ensuring that the potential continues to vanish in the false vacuum . the corrections appearing in eq . should be compared to the renormalized logarithm of the determinants of the klein - gordon operators of the @xmath229 fields in the background given by @xmath37 , which are given by @xmath266 n \ ; .\end{aligned}\ ] ] in comparison , the leading term in eq . is suppressed by a factor @xmath267 . the one - loop corrections @xmath197 and @xmath268 are both negative , thereby increasing the tunneling rate . it is interesting to note that , although the contribution @xmath198 to the tunneling action from the corrections to the bounce itself is positive , the net contribution of @xmath269 is still negative , again increasing the tunneling rate . in fig . [ fig : feyn : radcor ] , we present a diagrammatic representation of the corrections to the bounce action . it is also useful in order to see that there appear no contributions of @xmath270 relative to the bounce action @xmath73 in addition to those from @xmath198 . in order to avoid proliferation , we only show the leading contributions in @xmath271 for a given type of diagram . at one - loop order , there is the vacuum bubble in terms of the propagator @xmath272 of the @xmath229 fields , fig . [ fig : feyn : radcor](a ) , which gives the contribution @xmath273 relative to @xmath73 from @xmath274 in eq . . on substituting @xmath275 in the form of eq . into the action [ eq . ] , we see that the diagram corresponding to the @xmath276 term in @xmath277 is given by fig . [ fig : feyn : radcor](b ) , where , when counting the powers of @xmath12 , one should note that each explicit insertion of @xmath37 contributes a factor of @xmath278 . finally , at two - loop order , there are the diagrams figs . [ fig : feyn : radcor](c ) and ( d ) , which we do not compute but yield contributions of @xmath279 relative to @xmath73 . these contributions are therefore suppressed by a relative factor of @xmath271 relative to the @xmath280 in @xmath277 , as is familiar from the standard approximation scheme known as the @xmath271 expansion @xcite . we should remark that these arguments do not hold , of course , for the contribution to @xmath198 from the @xmath77 tadpole , which we include here for completeness . the latter is formally the same order as other two - loop diagrams , involving only @xmath77 , that are not captured in the 1pi approximation employed here . this observation is true also of the hartree approximation for the pure @xmath0 theory analyzed numerically in refs . nevertheless , these additional two - loop diagrams remain subdominant compared to the @xmath281 and @xmath280 contributions from the @xmath229 tadpole in eq . . finally , we note that approximating @xmath275 as a small perturbation to @xmath37 , using eq . , requires for consistency that @xmath282 , such that within the range of validity of present approximations , we can not obtain @xmath283 . nevertheless for large @xmath1 , @xmath269 can be the dominant two - loop contribution to the effective action . within the context of @xmath6 theory , we have described a green s function method for handling radiative effects on false vacuum decay . by this means and employing the thin- and planar - wall approximations , we have been able to calculate analytically and in a straightforward manner both the functional determinant of the quadratic fluctuations about the classical soliton configuration and the first correction to the configuration itself . this green s function method is well suited to numerical evaluation and , as a consequence , should be applicable to potentials of more general form . as such , we anticipate that it may be of particular use when the non - degeneracy of minima is purely radiatively generated . examples of the latter include the spontaneous symmetry breaking of the massless cw model @xcite or the instability of the electroweak vacuum . other applications might include the calculation of corrections to inflationary potentials in the time - dependent inflaton background , for instance in inflection - point or @xmath72-term inflation @xcite , which exploit the flat directions and saddle points of the mssm potential . furthermore , the use of green s functions naturally admits the introduction of finite - temperature effects or extension to non - trivial background spacetimes . green s functions have proved to be central objects within perturbative calculations throughout quantum field theory and it is therefore unsurprising that we find these suitable to treat solitons in @xmath6 theory as well . we take this as an encouragement that further theoretically- and phenomenologically - interesting systematic results on false vacuum decay may be within reach . the authors would like to thank jrgen baacke , daniel litim and holger gies for helpful correspondence and discussions . the work of p.m. is supported by a university foundation fellowship ( tuff ) from the technische universitt mnchen . the work of b.g . is supported by the gottfried wilhelm leibniz programme of the deutsche forschungsgemeinschaft ( dfg ) . both authors acknowledge support from the dfg cluster of excellence origin and structure of the universe . in order to perform the functional integration over the zero modes , we insert four copies of unity in faddeev - popov form @xcite : @xmath284 here , @xmath285 is not summed over and @xmath286 where we recall that @xmath287 it follows that @xmath288 ignoring terms that are formally @xmath289 . thus , @xmath290 we then have @xmath291 in this appendix , we include the technical details of the calculations outlined in secs . [ sec : greens ] and [ sec : corrb ] . all functional identities used in what follows may be found in ref . @xcite . in @xmath292 dimensions , the green s function satisfies the inhomogeneous klein - gordon equation @xmath293 where @xmath294 is the dirac delta function and @xmath295 is the laplacian . given the @xmath296 invariance of the bounce @xmath37 , it is convenient to work in hyperspherical coordinates , in which case the laplacian takes the form @xmath297 where @xmath298 is the laplace - beltrami operator on the @xmath299 sphere . we proceed by performing a partial - wave decomposition of the green s function : @xmath300 where @xmath301 , @xmath302 and @xmath303 are the hyperspherical harmonics ( see e.g. ref . @xcite ) , satisfying the eigenvalue equation @xmath304 with @xmath305 . the hyperradial function @xmath306 satisfies @xmath307 g_j(\varphi ; r , r ' ) \nonumber \\ & \qquad = \ r'^{1 - d } \delta(r - r ' ) \ ; .\end{aligned}\ ] ] since , for each @xmath154 , the @xmath308 modes are degenerate , we may use the sum rule @xcite @xmath309 where @xmath310 , @xmath311 , @xmath312 is the pochhammer symbol and the @xmath313 are the jacobi polynomials . for @xmath314 , @xmath315 , @xmath316 , and we have @xmath317 hence , @xmath318 , as we would expect . for @xmath319 , @xmath320 , and we have @xmath321 where @xmath322 is the chebyshev polynomial of the first kind . we then obtain @xmath323 where we have used the trigonometric form @xmath324 . for @xmath325 , @xmath326 , and @xmath327 where @xmath328 are the legendre polynomials . thus , we obtain the familiar three - dimensional expansion @xmath329 finally , for @xmath330 , @xmath331 , and @xmath332 where @xmath136 are the chebyshev polynomials of the second kind . hence , we find @xmath333 as appearing in eq . . in the coincident limit @xmath334 , @xmath335 , and we have @xmath336 alternatively , in @xmath292 dimensions , we may use @xmath337 in eq . , giving @xmath338 completing the square in the centrifugal potential in eq . , we make the following approximation for large @xmath51 : @xmath339 where @xmath340 , as before . we may then promote @xmath341 to a continuous variable @xmath202 , obtaining @xmath342 where we have used the general notation employed in sec . [ sec : greens ] , see eq . , with @xmath167 given by eq . we note that for @xmath343 , we have dropped terms @xmath344 and higher within the integrand . for large @xmath51 , we neglect the damping term in the radial equation [ eq . ] and set @xmath98 in the centrifugal potential and discontinuity , giving @xmath345 g_j(\varphi ; r , r ' ) \ = \ \frac{\delta(r - r')}{r^{d - 1 } } \ ; .\ ] ] since the solution depends only on the normalized bounce @xmath346 $ ] , it is convenient to define @xmath347 cf . [ sec : greens ] . equation may then be recast in the form @xmath348 g(u , u ' , m ) \nonumber \\ & \qquad \qquad = \ - \ , \gamma^{-1 } \delta(u - u ' ) \ ; , \end{aligned}\ ] ] where @xmath349 splitting around the discontinuity at @xmath350 , we decompose @xmath351 where @xmath352 are the solutions to the homogeneous equation @xmath353 g^{\gtrless}(u , u ' , m ) \ = \ 0 \ ; .\ ] ] the latter is the associated legendre differential equation and we obtain the general solutions @xmath354 where @xmath355 and @xmath356 are the associated legendre functions of the first and second kind , respectively . matching around the delta function in the inhomogeneous equation , we require @xmath357 thus , we find @xmath358 } \ ; , \\ b^ > - b^ < \ = \ \frac{1}{\gamma \big ( 1 - u'^2 \big ) } \ , \frac{p_2^m(u')}{w[q_2^m(u ' ) , p_2^m(u ' ) ] } \ ; , \end{aligned}\ ] ] where @xmath359 $ ] is the wronskian , having the explicit form @xmath360 \ = \ \frac{(n - m + 1)_{2m}}{1 - u'^2 } \ ; , \ ] ] with the pochhammer symbol defined in eq . . we also require the boundary condition that @xmath235 go to zero as @xmath361 , giving @xmath362 we may now solve for the remaining non - zero coefficients and obtain @xmath363 with @xmath364 . here , we have used the identity @xmath365 finally , we employ the representation @xmath366 of the associated legendre function of the first kind in terms of the jacobi polynomials . for @xmath107 , the polynomial expansion of the latter terminates , and we have @xmath367 \ ; .\end{aligned}\ ] ] after some algebraic simplification , we then arrive at the final analytic solution , as presented in eq . of sec . [ sec : greens ] . the normalized heat kernel @xmath368 , see sec . [ sec : greens ] , is given in terms of the inverse laplace transform @xmath369(\tau ) \ ; , \ ] ] where @xmath370 with @xmath371 the inverse laplace transform may be performed by using the shift , scaling and division properties @xmath372(\tau ) \ = \ e^{- b\tau } f(\tau ) \ ; , \\ \mathcal{l}_{s}^{-1}[f(a s)](\tau ) \ \frac{1}{a } \ , f(\tau / a ) \ ; , \\ \mathcal{l}_{s}^{-1}[s^{-1 } f(s)](\tau ) \ = \ \int_{0}^{\tau } \ ! { \mathrm{d}}\tau ' \ ; f(\tau ' ) \ ; , \end{gathered}\ ] ] where @xmath373(\tau)$ ] , as well as the elementary transformation @xmath374(\tau ) \ , = \ , \frac{\tau^{z - 1}}{\gamma(z ) } \ ; , \qquad \mathrm{re } \ , z \ > \ 0 \ ; .\ ] ] we find @xmath375(\tau ) = \frac{\gamma^2}{n } \ , e^{[\gamma^2(n^2 - 4 ) - k^2]\tau } \mathrm{erf}(n\gamma\sqrt{\tau } ) \ ; , \ ] ] where @xmath376 is the error function . hence , we have @xmath377 generalizing to @xmath292 dimensions , using the continuum limit in eq . , the correction to the bounce action arising from the functional determinant is therefore @xmath378 where @xmath379 is the solid angle subtended by the @xmath380 dimensional hypersphere . the integral over @xmath93 is dominated by @xmath60 , such that ( for @xmath152 ) @xmath381 we are then left with @xmath382 cf . the form presented in ref . @xcite . we may now proceed in one of two ways : ( i ) performing the @xmath17 integration first , we must regularize the @xmath202 integral , for instance by introducing an ultraviolet cutoff @xmath201 ; or ( ii ) performing the @xmath202 integral first , we must instead regularize the @xmath17 integral . the latter is the approach presented in ref . @xcite , which we reproduce in what follows for comparison . \(i ) performing the @xmath17 integral first gives @xmath383 subsequently , performing the @xmath202 integral for @xmath330 , we obtain the result in eq . . \(ii ) instead , performing instead the @xmath202 integral first , we obtain @xmath384 which is regularized by introducing a large mass @xmath385 as follows : @xmath386 we may proceed by using the series representation of the error function @xmath387 where @xmath388 denotes the double factorial . the @xmath17 integral may now be performed , and we obtain @xmath389 considering the derivative with respect to @xmath390 , we have @xmath391 \ ; , \end{aligned}\ ] ] where @xmath392 is the digamma function . in order to take the limit @xmath393 safely , we must take note of the poles occurring in @xmath394 and @xmath392 for non - positive integers . such poles occur in even dimensions for @xmath395 . after treating the limit @xmath393 , we find for @xmath292 odd ( including @xmath396 ) @xmath397 on the other hand , for @xmath292 even , we find @xmath398 \ ; , \end{aligned}\ ] ] where @xmath399 are the harmonic numbers , which we have supplemented with @xmath400 for notational simplicity . for @xmath401 , we then obtain @xmath402 \ ; .\end{aligned}\ ] ] lastly , performing the summations , we arrive at the result @xmath403 defining the counterterms in the proper - time representation , see ref . @xcite , and fixing the renormalization conditions as in eq . , we find the counterterms @xmath404 giving @xmath405 adding these to eq . , we obtain agreement with eq . . in this final appendix , we highlight the main technical details of the derivation of the green s function and corrections to the bounce from the @xmath229 fields . proceeding as for the isolated @xmath37 case , see sec . [ sec : greens ] , the renormalization is fixed using the cw effective potential @xcite , evaluated in a homogeneous false vacuum . the renormalization conditions are then @xmath412 \ ; , \\ & \delta \lambda_{\varphi } \ = \ - \ , \frac{3 \lambda^2}{32 \pi^2 } \bigg [ \ln \frac{\gamma^2}{\lambda^2 } + 5(n + 1 ) \nonumber \\ & \ + n \ln \frac{6\gamma^2 + m_{\chi}^2}{4 \lambda^2 } - 3 n \bigg ( \frac{m_{\chi}^2 + 2\gamma^2 } { m^2_{\chi } + 6\gamma^2 } \bigg)^{\ ! \ ! 2 } \bigg ] \ ; , \\ & \delta \lambda_{\varphi\chi } \ = \ - \ , \frac{\lambda^2}{32 \pi^2 } \bigg ( \ln \frac{\gamma^2}{\lambda^2}+4\ln\frac{6\gamma^2+m_{\chi}^2}{4\lambda^2}\nonumber\\ & \quad + \frac{136\gamma^2}{m_{\chi}^2 + 2\gamma^2}\ln\frac{6\gamma^2+m_{\chi}^2}{4\gamma^2 } + 9\,\frac{m_{\chi}^2 - 2\gamma^2}{m_{\chi}^2 + 6\gamma^2 } \bigg ) \ ; , \displaybreak\\ & \delta\lambda_{\chi}\ = \ -\,\frac{\lambda^2}{32\pi^2}\bigg(\frac{1}{2}\ln\frac{\gamma^2}{\lambda^2}+\frac{\big(m_{\chi}^2 - 10\gamma^2\big)^2 + 432\gamma^4}{\big(m_{\chi}^2 + 2\gamma^2\big)^{2}}\nonumber\\&\quad + 24\gamma^2\frac{\big(m_{\chi}^2 - 2\gamma^2\big)^2 - 112\gamma^4}{\big(m_{\chi}^2 + 2\gamma^2\big)^3}\ln\frac{6\gamma^2+m_{\chi}^2}{4\gamma^2}\bigg)\;. \end{aligned}\ ] ] proceeding as for @xmath37 , we find the unrenormalized tadpole contribution of the @xmath229 fields @xmath413 \ ; .\end{aligned}\ ] ] after adding the counterterms , we obtain @xmath414 \ ; .\end{aligned}\ ] ] we note that the expression in eq . agrees with the renormalized tadpole contribution from @xmath77 in eq . for @xmath415 , as we would expect . assuming @xmath416 , we may expand eq . to leading order in @xmath417 , giving eq . . the one - loop correction to the bounce action from the determinant over the quadratic fluctuations in the @xmath229 fields is given by @xmath418(\tau ) \ ; , \end{aligned}\ ] ] where @xmath419 is obtained from eqs . and with @xmath420 continuing as in sec . [ sec : greens ] , we find @xmath421 \ ; .\end{aligned}\ ] ] adding the counterterm @xmath422 \ ; , \end{aligned}\ ] ] obtained in analogy with eq . , we find @xmath423 \ ; .\end{aligned}\ ] ] the result in eq . reduces to that found in eq . for @xmath424 and @xmath425 , as we would expect . instead , taking @xmath426 , we obtain the expression in eq . .

we consider the class of nonlinear optimal control problems ( ocp ) with polynomial data , i.e. , the differential equation , state and control constraints and cost are all described by polynomials , and more generally for ocps with smooth data . in addition , state constraints as well as state and/or action constraints are allowed . we provide a simple hierarchy of lmi ( linear matrix inequality)-relaxations whose optimal values form a nondecreasing sequence of lower bounds on the optimal value . under some convexity assumptions , the sequence converges to the optimal value of the ocp . preliminary results show that good approximations are obtained with few moments .

math0703377

solving a general nonlinear optimal control problem ( ocp ) is a difficult challenge , despite powerful theoretical tools are available , e.g. the maximum principle and hamilton - jacobi - bellman ( hjb ) optimality equation . the problem is even more difficult in the presence of state and/or control constraints . state constraints are particularly difficult to handle , and the interested reader is referred to capuzzo - dolcetta and lions @xcite and soner @xcite for a detailed account of hjb theory in the case of state constraints . there exist many numerical methods to compute the solution of a given optimal control problem ; for instance , _ multiple shooting _ techniques which solve two - point boundary value problems as described , e.g. , in @xcite , or _ direct methods _ , as , e.g. , in @xcite , which use , among others , descent or gradient - like algorithms . to deal with optimal control problems with state constraints , some adapted versions of the maximum principle have been developed ( see @xcite , and see @xcite for a survey of this theory ) , but happen to be very hard to implement in general . on the other hand , the ocp can be written as an infinite - dimensional linear program ( lp ) over two spaces of measures . this is called the _ weak _ formulation of the ocp in vinter @xcite ( stated in the more general context of differential inclusions ) . the two unknown measures are the state - action _ occupation measure _ ( o.m . ) _ up to _ the final time @xmath0 , and the state o.m . _ at _ time @xmath0 . the optimal value of the resulting lp always provides a lower bound on the optimal value of the ocp , and under some convexity assumptions , both values coincide ; see vinter @xcite and hernandez - hernandez et al . @xcite as well . the dual of the original infinite dimensional lp has an interpretation in terms of `` subsolutions '' of related hjb - like optimality conditions , as for the unconstrained case . the only difference with the unconstrained case is the underlying function space involved , which directly incorporate the state constraints . namely , the functions are only defined on the state constraint set . an interesting feature of this lp approach with o.m.s is that state constraints , as well as state and/or action constraints are all easy to handle ; indeed they simply translate into constraints on the supports of the unknown o.m.s . it thus provides an alternative to the use of maximum principles with state constraints . however , although this lp approach is valid for any ocp , solving the corresponding ( infinite - dimensional ) lp is difficult in general ; however , general lp approximation schemes based on grids have been proposed in e.g. hernandez and lasserre @xcite . this lp approach has also been used in the context of discrete - time markov control processes , and is dual to bellman s optimality principle . for more details the interested reader is referred to borkar @xcite , hernandez - lerma and lasserre @xcite and many references therein . for some continuous - time stochastic control problems ( e.g. , modeled by diffusions ) and optimal stopping problems , the lp approach has also been used with success to prove existence of stationary optimal policies ; see for instance cho and stockbridge @xcite , helmes and stockbridge @xcite , helmes et al . @xcite , kurtz and stockbridge @xcite , and also fleming and vermes @xcite . in some of these works , the moment approach is also used to approximate the resulting infinite - dimensional lp . * contribution . * in this paper , we consider the particular class of nonlinear ocp s with state and/or control constraints , for which all data describing the problem ( dynamics , state and control constraints ) are _ polynomials_. the approach also extends to the case of problems with _ smooth _ data and compact sets , because polynomials are dense in the space of functions considered ; this point of view is detailed in [ smooth ] . in this restricted polynomial framework , the lp approach has interesting additional features that can be exploited for effective numerical computation . indeed , what makes this lp approach attractive is that for the class of ocps considered : @xmath1 only the _ moments _ of the o.m.s appear in the lp formulation , so that we already end up with countably many variables , a significant progress . @xmath1 constraints on the support of the o.m.s translate easily into either lp or sdp ( semi definite programming ) _ necessary _ constraints on their moments . even more , for ( semi - algebraic ) compact supports , relatively recent powerful results from real algebraic geometry make these constraints also _ sufficient_. @xmath1 when truncating to finitely many moments , the resulting lp or sdp s are solvable and their optimal values form a monotone nondecreasing sequence ( indexed by the number of moments considered ) of lower bounds on the optimal value of the lp ( and thus of the ocp ) . therefore , based on the above observations , we propose an approximation of the optimal value of the ocp via solving a hierarchy of sdps ( or linear matrix inequalities , lmis)that provides a monotone nondecreasing sequence of lower bounds on the optimal value of the weak lp formulation of the ocp . in adddition , under some compactness assumption of the state and control constraint sets , the sequence of lower bounds is shown to converge to the optimal value of the lp , and thus the optimal value of the ocp when the former and latter are equal . as such , it could be seen as a complement to the above shooting or direct methods , and when the sequence of lower bounds converges to the optimal value , a test of their efficiency . finally this approach can also be used to provide a _ certificate _ of unfeasibility . indeed , if in the hierarchy of lmi - relaxations of the minimum time ocp , one is infeasible then the ocp itself is infeasible . it turns out that sometimes this certificate is provided at an early stage in the hierarchy , i.e. with very few moments . this is illustrated on two simple examples . in a pioneering paper , dawson @xcite had suggested the use of _ moments _ in the lp approach with o.m.s , but results on the @xmath2-moment problem by schmdgen @xcite and putinar @xcite were either not available at that time . later , helmes and stockbridge @xcite and helmes , rhl and stockbridge @xcite have used lp moment conditions for computing some exit time moments in some diffusion model , whereas for the same models , lasserre and prieto - rumeau @xcite have shown that sdp moment conditions are superior in terms of precision and number of moments to consider ; in @xcite , they have extended the moment approach for options pricing problems in some mathematical finance models . more recently , lasserre , prieur and henrion @xcite have used the o.m . approach for minimum time ocp without state constraint . preliminary experimental results on brockett s integrator example , and the double integrator show fast convergence with few moments . let @xmath3 and @xmath4 be nonzero integers . consider on @xmath5 the control system @xmath6 where @xmath7 is smooth , and where the controls are bounded measurable functions , defined on intervals @xmath8 $ ] of @xmath9 , and taking their values in a _ subset @xmath10 of @xmath11 . let @xmath12 , and let @xmath13 and @xmath2 be _ compact _ subsets of @xmath5 . for @xmath14 , a control @xmath15 admissible _ on @xmath16 $ ] whenever the solution @xmath17 of ( [ system ] ) , such that @xmath18 , is well defined on @xmath16 $ ] , and satisfies @xmath19,\ ] ] and @xmath20 denote by @xmath21 the set of _ admissible controls _ on @xmath16 $ ] . for @xmath22 , the _ cost _ of the associated trajectory @xmath17 is defined by @xmath23 where @xmath24 and @xmath25 are smooth functions . consider the _ optimal control problem _ of determining a trajectory solution of ( [ system ] , starting from @xmath18 , satisfying the state and control constraints ( [ cont1 ] ) , the terminal constraint ( [ cont2 ] ) , and minimizing the cost ( [ cost ] ) . the final time @xmath0 may be fixed or not . if the final time @xmath0 is fixed , we set @xmath26 and if @xmath0 is free , we set @xmath27 note that a particular ocp is the minimal time problem from @xmath28 to @xmath2 , by letting @xmath29 , @xmath30 . in this particular case , the minimal time is the first _ hitting time _ of the set @xmath2 . it is possible to associate a stochastic or deterministic ocp with an abstract infinite dimensional linear programming ( lp ) problem @xmath31 together with its dual @xmath32 ( see for instance hernndez - lerma and lasserre @xcite for discrete - time markov control problems , and vinter @xcite , hernandez et al . @xcite for deterministic optimal control problems , as well as many references therein ) . we next describe this lp approach in the present context . for a topological space @xmath33 , let @xmath34 be the banach space of finite signed borel measures on @xmath33 , equipped with the norm of total variation , and denote by @xmath35 its positive cone , that is , the space of finite measures on @xmath33 . let @xmath36 be the banach space of bounded continuous functions on @xmath33 , equipped with the sup - norm . notice that when @xmath33 is compact hausdorff , then @xmath37 , i.e. , @xmath34 is the topological dual of @xmath36 . let @xmath38\times \x$ ] , @xmath39 , and let @xmath40 be the banach space of functions @xmath41 that are continuously differentiable . for ease of exposition we use the same notation @xmath42 ( resp . @xmath43 ) for a polynomial @xmath44 $ ] ( resp . @xmath45 $ ] ) and its restriction to the compact set @xmath46 ( resp . @xmath2 ) . next , with @xmath47 , let @xmath48 be the mapping @xmath49 notice that @xmath50 for every @xmath51 , because both @xmath13 and @xmath10 are compact , and @xmath52 is understood as its restriction to @xmath53 . next , let @xmath54 be the linear mapping @xmath55 where @xmath56 , for all @xmath57 . obviously , @xmath58 is continuous with respect to the strong topologies of @xmath40 and @xmath59 . both @xmath60 and @xmath61 form a _ dual pair _ of vector spaces , with duality brackets @xmath62 and @xmath63 . let @xmath64 be the adjoint mapping of @xmath58 , defined by @xmath65 for all @xmath66 . [ continuity ] * the mapping @xmath67 is continuous with respect to the weak topologies @xmath68 , and @xmath69 . * since the mapping @xmath58 is continuous in the strong sense , it is also continuous with respect to the weak topologies @xmath70 and @xmath71 . * in the case of a _ free _ terminal time @xmath72 , it suffices to redefine @xmath73\times \k)$ ] by @xmath74 . the operator @xmath75\times \k)\to c_1(\sigma)^*$ ] is still defined by ( [ map * ] ) , for all @xmath76\times\k)\times c_1(\si)$ ] . + for time - homogeneous free terminal time problems , one only needs functions @xmath77 of @xmath78 only , and so @xmath79 and @xmath80 . let @xmath14 , and let @xmath81 be a control such that the solution of ( [ system ] ) , with @xmath18 , is well defined on @xmath16 $ ] . define the probability measure @xmath82 on @xmath5 , and the measure @xmath83 on @xmath16\times\rr^n\times \rr^m$ ] , by @xmath84,\quad d\in\mathcal{b}_n,\\ \label{mu } \mu^{\mathbf{u}}(a\times b\times c)&:=&\int_{[0,t]\cap a}\,\i_{b\times c}\,[(x(t),u(t))]\,dt,\end{aligned}\ ] ] for all _ rectangles _ @xmath85 , with @xmath86 , and where @xmath87 ( resp . @xmath88 ) denotes the usual borel @xmath89-algebra associated with @xmath5 ( resp . @xmath11 ) , and @xmath90 the borel @xmath89-algebra of @xmath16 $ ] , and @xmath91 the indicator function of the set @xmath92 . the measure @xmath93 is called the _ occupation measure _ of the state - action ( deterministic ) process @xmath94 _ up to _ time @xmath0 , whereas @xmath95 is the occupation measure of the state @xmath96 _ at _ time @xmath0 . if the control @xmath97 is admissible on @xmath16 $ ] , i.e. , if the trajectory @xmath17 satisfies the constraints ( [ cont1 ] ) and ( [ cont2 ] ) , then @xmath82 is a probability measure supported on @xmath2 ( i.e. @xmath98 ) , and @xmath83 is supported on @xmath16\times \x\times \u$ ] ( i.e. @xmath99 ) . in particular , @xmath100 . conversely , if the support of @xmath93 is contained in @xmath101\times \x\times \u$ ] and if @xmath102 , then @xmath103 for almost every @xmath104 $ ] . indeed , with ( [ mu ] ) , @xmath105\,ds\ ] ] @xmath106\,=\,1\quad \mbox{a.e . in } [ 0,t],\ ] ] and hence @xmath103 , for almost every @xmath107 $ ] . if moreover the support of @xmath95 is contained in @xmath2 , then @xmath108 . therefore , @xmath97 is an admissible control on @xmath16 $ ] . then , observe that the optimization criterion ( [ crit1 ] ) now writes @xmath109 and one infers from ( [ system ] ) , ( [ cont1 ] ) and ( [ cont2 ] ) that @xmath110 for every @xmath111 ( where @xmath112 for every @xmath113 ) , or equivalently , in view of ( [ map ] ) and ( [ map * ] ) , @xmath114 this in turn implies that @xmath115 therefore , consider the infinite - dimensional linear program @xmath31 @xmath116 ( where @xmath117 ) . denote by @xmath118 its optimal value and @xmath119 is the infimum is attained , in which case @xmath31 is said to be _ solvable_. the problem @xmath31 is said _ feasible _ if there exists @xmath120 such that @xmath121 . note that @xmath31 is feasible whenever there exists an admissible control . the linear program @xmath31 is a rephrasing of the ocp ( [ system])([crit1 ] ) in terms of the _ occupation measures _ of its trajectories @xmath94 . its dual lp reads @xmath122 where @xmath123 denote by @xmath124 its optimal value and @xmath125 is the supremum is attained , i.e. if @xmath32 is solvable . discrete - time stochastic analogues of the linear programs @xmath31 and @xmath32 are also described in hernndez - lerma and lasserre @xcite for discrete time markov control problems . similarly see cho and stockbridge @xcite , kurtz and stockbridge @xcite , and helmes and stcokbridge @xcite for some continuous - time stochastic problems . [ thlp ] if @xmath31 is feasible , then : * @xmath31 is solvable , i.e. , @xmath126 . * there is no duality gap , i.e. , @xmath127 . * if moreover , for every @xmath128 , the set @xmath129 is convex , and the function @xmath130 is convex , then the ocp ( [ system])([crit1 ] ) has an optimal solution and @xmath131 for a proof see [ sec - thlp ] . theorem [ thlp](iii ) is due to vinter @xcite . the linear program @xmath31 is infinite dimensional , and thus , not tractable as it stands . therefore , we next present a relaxation scheme that provides a sequence of semidefinite programming , or linear matrix inequality relaxations ( in short , lmi - relaxations ) @xmath132 , each with _ finitely many _ constraints and variables . let @xmath133=[x_1,\ldots x_n]$ ] ( resp . @xmath134=\rr[t , x_1,\ldots x_n , u_1,\ldots , u_m]$ ] ) denote the set of polynomial functions of the variable @xmath78 ( resp . , of the variables @xmath135 ) . assume that @xmath13 and @xmath2 ( resp . , @xmath10 ) are compact semi - algebraic subsets of @xmath5 ( resp . of @xmath11 ) , of the form @xmath136 for some finite index sets @xmath137 , @xmath138 and @xmath139 , where @xmath140 , @xmath141 and @xmath142 are polynomial functions . define @xmath143 to highlight the main ideas , in this section we assume that @xmath52 , @xmath43 and @xmath144 are polynomial functions , that is , @xmath145 $ ] , @xmath146 $ ] , and @xmath147 is polynomial , i.e. , every component of @xmath52 satisfies @xmath148 $ ] , for @xmath149 . observe the following important facts . the restriction of @xmath150 $ ] to @xmath46 belongs to @xmath40 . therefore , @xmath151,\ ] ] because @xmath46 being compact , polynomial functions are _ dense _ in @xmath40 for the sup - norm . indeed , on a compact set , simultaneously _ approximate a function and its ( continuous ) partial derivatives by a polynomial and its derivatives , uniformly ( see @xcite pp . 65 - 66 ) . hence , the linear program @xmath31 can be written @xmath152 , \end{array}\right.\ ] ] or , equivalently , by linearity , @xmath153 the constraints of @xmath31 , @xmath154 define countably many _ linear _ equality constraints linking the _ moments _ of @xmath155 and @xmath156 , because if @xmath42 is polynomial then so are @xmath157 and @xmath158 , for every @xmath159 , and @xmath160 . and so , @xmath161 is polynomial . the functions @xmath162 being also polynomial , the cost @xmath163 of the ocp ( [ system])([crit1 ] ) is also a linear combination of the moments of @xmath155 and @xmath156 . therefore , the linear program @xmath31 in ( [ equivp ] ) can be formulated as a lp with countably many variables ( the moments of @xmath155 and @xmath156 ) , and countably many linear equality constraints . however , it remains to express the fact that the variables should be moments of some measures @xmath155 and @xmath156 , with support contained in @xmath53 and @xmath2 respectively . at this stage , one will make some ( weak ) additional assumptions on the polynomials that define the compact semi - algebraic sets @xmath164 and @xmath10 . under such assumptions , one may then invoke recent results of real algebraic geometry on the representation of polynomials positive on a compact set , and get necessary and sufficient conditions on the variables of @xmath31 to be indeed moments of two measures @xmath155 and @xmath156 , with appropriate support . we will use putinar s positivstellensatz @xcite described in the next section , which yields sdp constraints on the variables . one might also use other representation results like e.g. krivine @xcite and vasilescu @xcite , and obtain _ linear _ constraints on the variables ( as opposed to sdp constraints ) . this is the approach taken in e.g. helmes et al . however , a comparison of the use of lp - constraints versus sdp constraints on a related problem @xcite has dictated our choice of the former . finally , if @xmath42 in ( [ c ] ) runs only over all monomials of degree less than @xmath165 , one then obtains a corresponding relaxation @xmath166 of @xmath31 , which is now a finite - dimensional sdp that one may solve with public software packages . at last , one lets @xmath167 . for a multi - index @xmath168 , and for @xmath169 , denote @xmath170 . consider the canonical basis @xmath171 ( resp . , @xmath172 ) of @xmath133 $ ] ( resp . , of @xmath134 $ ] ) . given two sequences @xmath173 and @xmath174 of real numbers , define the linear functional @xmath175\to\rr$ ] by @xmath176 and similarly , define the linear functional @xmath177\to\rr$ ] by @xmath178 where @xmath179 . note that , for a given measure @xmath156 ( resp . , @xmath155 ) on @xmath180 ( resp . , on @xmath181 ) , there holds , for every @xmath146 $ ] ( resp . , for every @xmath182)$ ] , @xmath183 where the real numbers @xmath184 are the moments of the measure @xmath156 ( resp . , @xmath185 , where @xmath186 is the sequence of moments of the measure @xmath155 ) . for a given sequence @xmath174 of real numbers , the _ moment _ matrix @xmath187 of order @xmath165 associated with @xmath186 , has its rows and columns indexed in the canonical basis @xmath188 , and is defined by @xmath189 where @xmath190 . the moment matrix @xmath191 of order @xmath165 associated with a given sequence @xmath173 , has its rows and columns indexed in the canonical basis @xmath192 , and is defined in a similar fashion . note that , if @xmath186 has a representing measure @xmath155 , i.e. , if @xmath186 is the sequence of moments of the measure @xmath155 on @xmath181 , then @xmath193 , for every @xmath145 $ ] , and if @xmath194 denotes the vector of coefficients of @xmath145 $ ] of degree less than @xmath165 , then @xmath195 this implies that @xmath187 is symmetric nonnegative ( denoted @xmath196 ) , for every @xmath165 . the same holds for @xmath191 . conversely , not every sequence @xmath197 that satisfies @xmath198 for every @xmath165 , has a representing measure . however , several sufficient conditions exist , and in particular the following one , due to berg @xcite . [ prop1 ] if @xmath199 satisfies @xmath200 for every @xmath201 , and @xmath198 for every integer @xmath165 , then @xmath197 has a representing measure on @xmath5 , with support contained in the unit ball @xmath202^{n}$ ] . we next present another sufficient condition which is crucial in the proof of our main result . for a given polynomial @xmath203 $ ] , written @xmath204 define the _ localizing _ matrix @xmath205 associated with @xmath206 , and with rows and columns also indexed in the canonical basis of @xmath134 $ ] , by @xmath207 the localizing matrix @xmath208 associated with a given sequence @xmath173 is defined similarly . note that , if @xmath186 has a representing measure @xmath155 on @xmath181 with support contained in the level set @xmath209 , and if @xmath145 $ ] has degree less than @xmath165 , then @xmath210 hence , @xmath211 , for every @xmath165 . let @xmath212 $ ] be the convex cone generated in @xmath133 $ ] by all squares of polynomials , and let @xmath213 be the compact basic semi - algebraic set defined by @xmath214 for some family @xmath215 $ ] . [ def1 ] the set @xmath213 defined by ( [ setomega ] ) satisfies putinar s condition if there exists @xmath216 $ ] such that @xmath217 for some family @xmath218 , and the level set @xmath219 is compact . putinar s condition is satisfied if e.g. the level set @xmath220 is compact for some @xmath159 , or if all the @xmath221 s are linear , in which case @xmath222 is a polytope . in addition , if one knows some @xmath223 such that @xmath224 whenever @xmath225 , then it suffices to add the redundant quadratic constraint @xmath226 in the definition ( [ setomega ] ) of @xmath222 , and putinar s condition is satisfied ( take @xmath227 ) . [ putinarthm ] assume that the set @xmath222 defined by ( [ setomega ] ) satisfies putinar s condition . * if @xmath228 $ ] and @xmath229 on @xmath222 , then @xmath230 for some family @xmath231 . * let @xmath199 be a sequence of real numbers . if @xmath232 then @xmath197 has a representing measure with support contained in @xmath222 . consider the linear program @xmath31 defined by ( [ equivp ] ) . since the semi - algebraic sets @xmath164 and @xmath10 defined respectively by ( [ setx ] ) , ( [ setk ] ) and ( [ setu ] ) are compact , with no loss of generality , we assume ( up to a scaling of the variables @xmath233 and @xmath234 ) that @xmath235 , @xmath236^n$ ] and @xmath237^m$ ] . next , given a sequence @xmath238 indexed in the basis of @xmath134 $ ] denote @xmath239 , @xmath240 and @xmath241 its marginals with respect to the variables @xmath234 , @xmath78 and @xmath15 , respectively . these sequences are indexed in the canonical basis of @xmath242 $ ] , @xmath133 $ ] and @xmath243 $ ] repectively . for instance , writing @xmath244 , @xmath245 let @xmath246 be an integer such that @xmath247 , where @xmath248 is defined by ( [ defdegxku ] ) . for every @xmath249 , consider the lmi - relaxation @xmath250 whose optimal value is denoted by @xmath251 . * ocp with free terminal time . * for the ocp ( [ crit1libre ] ) , i.e. , with _ free _ terminal time @xmath72 , we need adapt the notation because @xmath0 is now a variable . as already mentioned in remark [ continuity](iii ) , the measure @xmath156 in the infinite dimensional linear program @xmath31 defined in ( [ p ] ) , is now supported in @xmath252\times \k$ ] ( and @xmath253\times \k$ ] after re - scaling ) instead of @xmath2 previously . hence , the sequence @xmath197 associated with @xmath156 is now indexed in the basis @xmath254 of @xmath150 $ ] instead of @xmath192 previously . therefore , given @xmath255 indexed in that basis , let @xmath256 and @xmath257 be the subsequences of @xmath197 defined by : @xmath258 then again ( after rescaling ) , the lmi - relaxation @xmath166 now reads @xmath259 the particular case of minimal time problem is obtained with @xmath29 , @xmath30 . for _ time - homogeneous _ problems , i.e. , when @xmath43 and @xmath52 do not depend on @xmath234 , one may take @xmath155 ( resp . @xmath156 ) supported on @xmath260 ( resp . @xmath2 ) , which simplifies the associated lmi - relaxation ( [ sdpqrt ] ) . the main result is the following . [ main ] let @xmath261^n$ ] , and @xmath262^m$ ] be compact basic semi - algebraic - sets respectively defined by ( [ setx ] ) , ( [ setk ] ) and ( [ setu ] ) . assume that @xmath164 and @xmath10 satisfy putinar s condition ( see definition ( [ def1 ] ) ) , and let @xmath166 be the lmi - relaxation defined in ( [ sdpq ] ) . then , * @xmath263 as @xmath167 ; * if moreover , for every @xmath128 , the set @xmath129 is convex , and the function @xmath264 is convex , then @xmath265 , as @xmath167 . the proof of this result is postponed to the appendix in section [ proofmain ] . we describe the dual of the lmi - relaxation @xmath166 which is also a semidefinite program , denoted @xmath266 , and relate @xmath266 with the dual @xmath32 of @xmath31 , defined in ( [ pstar ] ) . + let @xmath267 be the cardinal of the set @xmath268 , and given @xmath269 , let @xmath270 $ ] be the polynomial @xmath271 consider the semidefinite program : @xmath272,\,q^u_k\in\rr[u],\,q^x_j\in\rr[x],\,l_j\in\rr[x]\\ \\ \{q_0,q_j^x , q_k^u , l_0,l_j\}\mbox { s.o.s . ( sums of squares polynomials ) } , \quad\mbox{and}\\ { \rm deg}\,l_j\theta_j,\ : { \rm deg}\,q_j^xv_j,\ : { \rm deg}\,q_k^uw_k,\ : { \rm deg}\,q_0\leq\,2r-2;\:{\rm deg}\,l_0\,\leq\,2r . \end{array}\right .. \ ] ] the lmi @xmath266 is a reinforcement of @xmath32 in the following sense : * the unknown function @xmath51 is now replaced with a polynomial @xmath270 $ ] of degree less than @xmath273 ; * the constraint @xmath274 for @xmath275 , is now replaced with the constraint @xmath276 on @xmath53 and the polynomial @xmath277 which is nonnegative on @xmath53 , has putinar s representation @xmath278 ; * the constraint @xmath279 for @xmath280 , is replaced with the constraint @xmath281 on @xmath2 , and the polynomial @xmath282 which is nonnegative on @xmath2 , has putinar s representation @xmath283 . assume that @xmath266 is solvable . a natural question is to know whether or not we can use an optimal solution @xmath284 of @xmath266 to obtain some information on an optimal solution of @xmath31 . the most natural idea is to look for the zero set in @xmath53 of the polynomial @xmath285 indeed , under the assumptions of theorem [ main ] , if @xmath286 , then @xmath287 , and so , the polynomial @xmath270 $ ] seems to be a good candidate to approximate a nearly optimal solution @xmath51 of @xmath32 . next , as @xmath266 is an approximation of a weak formulation of the hjb optimality equation , one may hope that the zero set in @xmath53 of the polynomial @xmath288 provides some good information on the possible states @xmath289 and controls @xmath290 at time @xmath234 in an optimal solution of the ocp ( [ system])([crit1 ] ) . that is , fixing an arbitrary @xmath291 $ ] , one may solve the equation @xmath292 and look for solutions @xmath293 . all these issues deserve further investigation beyond the scope of the present paper . however , at least in the minimum time problem for the ( state and control constrained ) double integrator example considered in [ ex1 ] , we already have some numerical support for the above claims . for minimum time ocps , i.e. , with free terminal time @xmath0 and instantaneous cost @xmath29 , and @xmath30 , the lmi - relaxations @xmath166 defined in ( [ sdpqrt ] ) may provide _ certificates _ of non controllability . indeed , if for a given initial state @xmath294 , some lmi - relaxation @xmath166 in the hierarchy has _ no _ feasible solution , then the initial state @xmath28 _ can not _ be steered to the origin in finite time . in other words , @xmath295 provides a certificate of uncontrollability of the initial state @xmath28 . it turns out that sometimes such certificates can be provided at cheap cost , i.e. , with lmi - relaxations of low order @xmath165 . this is illustrated on the zermelo problem in [ controllability ] . moreover , one may also consider controllability in given finite time @xmath0 , that is , consider the lmi - relaxations as defined in ( [ sdpq ] ) with @xmath0 fixed , and @xmath30 , @xmath29 . again , if for a given initial state @xmath294 , the lmi - relaxation @xmath166 has no feasible solution , the initial state @xmath28 can not be steered to the origin in less than @xmath0 units of time . and so , @xmath295 also provides a certificate of uncontrollability of the initial state @xmath28 . in the previous section , we assumed , to highlight the main ideas , that @xmath52 , @xmath43 and @xmath144 were polynomials . in this section , we generalize theorem [ main ] , and simply assume that @xmath52 , @xmath43 and @xmath144 are smooth . consider the linear program @xmath31 defined in the previous section @xmath296 . \end{array}\right.\ ] ] since the sets @xmath13 , @xmath2 and @xmath10 , defined previously , are compact , it follows from @xcite ( see also @xcite ) that @xmath52 ( resp . @xmath43 , resp . @xmath144 ) is the limit in @xmath297 ( resp . @xmath297 , resp . @xmath298 ) of a sequence of polynomials @xmath299 ( resp . @xmath300 , resp . @xmath301 ) of degree @xmath302 , as @xmath303 . hence , for every integer @xmath302 , consider the linear program @xmath304 @xmath305 , \end{array}\right.\ ] ] the smooth analogue of @xmath31 , where the linear mapping @xmath306 is defined by @xmath307 and where @xmath308 is defined by @xmath309 for every integer @xmath310 , let @xmath311 denote the lmi - relaxation ( [ sdpq ] ) associated with the linear program @xmath304 . recall that from theorem [ main ] , if @xmath2 , @xmath13 and @xmath10 satisfy putinar s condition , then @xmath312 as @xmath313 ; the next result , generalizing theorem [ main ] , shows that it is possible to let @xmath302 tend to @xmath314 . for convenience , set @xmath315 [ mainbis ] let @xmath261^n$ ] , and @xmath262^m$ ] be compact semi - algebraic - sets respectively defined by ( [ setx ] ) , ( [ setk ] ) and ( [ setu ] ) . assume that @xmath164 and @xmath10 satisfy putinar s condition ( see definition ( [ def1 ] ) ) . then , * @xmath316 . * moreover if for every @xmath128 , the set @xmath129 is convex , and the function @xmath264 is convex , then @xmath317 . the proof of this result is in the appendix , section [ proofmainbis ] . from the numerical point of view , depending on the functions @xmath52 , @xmath43 , @xmath144 , the degree of the polynomials of the approximate ocp @xmath304 may be required to be large , and hence the hierarchy of lmi - relaxations @xmath318 in ( [ sdpq ] ) might not be efficiently implementable , at least in view of the performances of public sdp solvers available at present . the previous construction extends to smooth optimal control problems on riemannian manifolds , as follows . let @xmath223 and @xmath319 be smooth riemannian manifolds . consider on @xmath223 the control system ( [ system ] ) , where @xmath320 is smooth , and where the controls are bounded measurable functions , defined on intervals @xmath8 $ ] of @xmath9 , and taking their values in a _ subset @xmath10 of @xmath319 . let @xmath321 , and let @xmath13 and @xmath2 be compact subsets of @xmath223 . admissible controls are defined as previously . for an admissible control @xmath97 on @xmath16 $ ] , the cost of the associated trajectory @xmath17 is defined by ( [ cost ] ) , where where @xmath322 and @xmath323 are smooth functions . according to nash embedding theorem @xcite , there exist an integer @xmath3 ( resp . @xmath4 ) such that @xmath223 ( resp . @xmath319 ) is smoothly isometrically embedded in @xmath5 ( resp . @xmath11 ) . in this context , all previous results apply . this remark is important for the applicability of the method described in this article . indeed , many practical control problems ( in particular , in mechanics ) are expressed on manifolds , and since the optimal control problem investigated here is global , they can not be expressed in general as control systems in @xmath5 ( in a global chart ) . we consider here the minimal time ocp , that is , we aim to approximate the _ minimal time _ to steer a given initial condition to the origin . we have tested the above methodology on two test ocps , the double and brockett _ integrators _ , for which the associated optimal value @xmath324 can be calculated exactly . the numerical examples in this section were processed with our matlab package _ _ gloptipoly 3__henrion / software ] . consider the double integrator system in @xmath325 @xmath326 where @xmath327 is the state and the control @xmath328 , satisfies the constraint @xmath329 , for all @xmath330 . in addition , the state is constrained to satisfy @xmath331 , for all @xmath234 . in this time - homogeneous case , and with the notation of section [ general ] , we have @xmath332 , @xmath333 , and @xmath334 $ ] . the theorem obviously extends , up to scaling , to the case of arbitrary compact subsets @xmath335 and @xmath336^m$ ] . observe that @xmath13 is not compact and so the convergence result of theorem [ main ] may not hold . in fact , we may impose the additional constraint @xmath337 for some large @xmath223 ( and modify @xmath13 accordingly ) , because for initial states @xmath28 with @xmath338 relatively small with respect to @xmath223 , the optimal trajectory remains in @xmath13 . however , in the numerical experiments , we have not enforced an additional constraint . we have maintained the original constraint @xmath339 in the localizing constraint @xmath340 , with @xmath341 . for this very simple system , one is able to compute exactly the optimal minimum time . denoting @xmath342 the minimal time to reach the origin from @xmath327 , we have : if @xmath343 and @xmath344 then @xmath345 . if @xmath346 and @xmath344 then @xmath347 . if @xmath348 and @xmath344 then @xmath349 . table [ doubleinte - x0 ] displays the values of the initial state @xmath294 , and denoting @xmath350 the optimal value of the lmi - relaxation ( [ sdpqrt ] ) for the minimum time ocp ( [ inte ] ) with initial state @xmath28 , tables [ doubleinte-2 ] , [ doubleinte-3 ] , and [ doubleinte-5 ] display the numerical values of the ratii @xmath351 for @xmath352 and @xmath353 respectively . .double integrator : data initial state @xmath354 [ cols="^,^,^,^,^,^,^,^,^,^,^,^",options="header " , ] 0.4598 & 0.3964 & 0.3512 & 0.9817 & 0.9674 & 0.9634 & 0.9628 & 0.9608 & 0.9600 & 0.9596 & 0.9595 + 0.4534 & 0.3679 & 0.9653 & 0.9347 & 0.9355 & 0.9383 & 0.9385 & 0.9386 & 0.9413 & 0.9432 & 0.9445 + 0.4390 & 0.9722 & 0.8653 & 0.8457 & 0.8518 & 0.8639 & 0.8720 & 0.8848 & 0.8862 & 0.8983 & 0.9015 + 0.4301 & 0.7698 & 0.7057 & 0.7050 & 0.7286 & 0.7542 & 0.7752 & 0.7964 & 0.8085 & 0.8187 & 0.8351 + 0.4212 & 0.4919 & 0.5039 & 0.5422 & 0.5833 & 0.6230 & 0.6613 & 0.6870 & 0.7121 & 0.7329 & 0.7513 + 0.0000 & 0.2238 & 0.3165 & 0.3877 & 0.4476 & 0.5005 & 0.5460 & 0.5839 & 0.6158 & 0.6434 & 0.6671 + 0.4501 & 0.3536 & 0.3950 & 0.4403 & 0.4846 & 0.5276 & 0.5663 & 0.5934 & 0.6204 & 0.6474 & 0.6667 + 0.4878 & 0.4493 & 0.4699 & 0.5025 & 0.5342 & 0.5691 & 0.5981 & 0.6219 & 0.6446 & 0.6647 & 0.6824 + 0.5248 & 0.5142 & 0.5355 & 0.5591 & 0.5840 & 0.6124 & 0.6312 & 0.6544 & 0.6689 & 0.6869 & 0.7005 + 0.5629 & 0.5673 & 0.5842 & 0.6044 & 0.6296 & 0.6465 & 0.6674 & 0.6829 & 0.6906 & 0.7083 & 0.7204 + 0.6001 & 0.6099 & 0.6245 & 0.6470 & 0.6617 & 0.6792 & 0.6972 & 0.7028 & 0.7153 & 0.7279 & 0.7369 + 0.5418 & 0.4400 & 0.3630 & 0.9989 & 0.9987 & 0.9987 & 0.9985 & 0.9984 & 0.9983 & 0.9984 & 0.9984 + 0.5115 & 0.3864 & 0.9803 & 0.9648 & 0.9687 & 0.9726 & 0.9756 & 0.9778 & 0.9798 & 0.9815 & 0.9829 + 0.4848 & 0.9793 & 0.8877 & 0.8745 & 0.8847 & 0.8997 & 0.9110 & 0.9208 & 0.9277 & 0.9339 & 0.9385 + 0.4613 & 0.7899 & 0.7321 & 0.7401 & 0.7636 & 0.7915 & 0.8147 & 0.8339 & 0.8484 & 0.8605 & 0.8714 + 0.4359 & 0.5179 & 0.5361 & 0.5772 & 0.6205 & 0.6629 & 0.7013 & 0.7302 & 0.7540 & 0.7711 & 0.7891 + 0.0000 & 0.2458 & 0.3496 & 0.4273 & 0.4979 & 0.5571 & 0.5978 & 0.6409 & 0.6719 & 0.6925 & 0.7254 + 0.4556 & 0.3740 & 0.4242 & 0.4789 & 0.5253 & 0.5767 & 0.6166 & 0.6437 & 0.6807 & 0.6972 & 0.7342 + 0.4978 & 0.4709 & 0.5020 & 0.5393 & 0.5784 & 0.6179 & 0.6477 & 0.6776 & 0.6976 & 0.7192 & 0.7347 + 0.5396 & 0.5395 & 0.5638 & 0.5955 & 0.6314 & 0.6600 & 0.6856 & 0.7089 & 0.7269 & 0.7438 & 0.7555 + 0.5823 & 0.5946 & 0.6190 & 0.6453 & 0.6703 & 0.7019 & 0.7177 & 0.7382 & 0.7539 & 0.7662 & 0.7767 + 0.6255 & 0.6434 & 0.6656 & 0.6903 & 0.7193 & 0.7326 & 0.7543 & 0.7649 & 0.7776 & 0.7917 & 0.8012 + 0.7550 & 0.5539 & 0.3928 & 0.9995 & 0.9995 & 0.9995 & 0.9994 & 0.9992 & 0.9988 & 0.9985 & 0.9984 + 0.6799 & 0.4354 & 0.9828 & 0.9794 & 0.9896 & 0.9923 & 0.9917 & 0.9919 & 0.9923 & 0.9923 & 0.9938 + 0.6062 & 0.9805 & 0.9314 & 0.9462 & 0.9706 & 0.9836 & 0.9853 & 0.9847 & 0.9848 & 0.9862 & 0.9871 + 0.5368 & 0.8422 & 0.8550 & 0.8911 & 0.9394 & 0.9599 & 0.9684 & 0.9741 & 0.9727 & 0.9793 & 0.9776 + 0.4713 & 0.6417 & 0.7334 & 0.8186 & 0.8622 & 0.9154 & 0.9448 & 0.9501 & 0.9505 & 0.9665 & 0.9637 + 0.0000 & 0.4184 & 0.5962 & 0.7144 & 0.8053 & 0.8825 & 0.9044 & 0.9210 & 0.9320 & 0.9544 & 0.9534 + 0.4742 & 0.5068 & 0.6224 & 0.7239 & 0.7988 & 0.8726 & 0.8860 & 0.9097 & 0.9263 & 0.9475 & 0.9580 + 0.5410 & 0.6003 & 0.6988 & 0.7585 & 0.8236 & 0.8860 & 0.9128 & 0.9257 & 0.9358 & 0.9452 & 0.9528 + 0.6106 & 0.6826 & 0.7416 & 0.8125 & 0.8725 & 0.9241 & 0.9305 & 0.9375 & 0.9507 & 0.9567 & 0.9604 + 0.6864 & 0.7330 & 0.7979 & 0.8588 & 0.9183 & 0.9473 & 0.9481 & 0.9480 & 0.9559 & 0.9634 & 0.9733 + 0.7462 & 0.8032 & 0.8564 & 0.9138 & 0.9394 & 0.9610 & 0.9678 & 0.9678 & 0.9696 & 0.9755 & 0.9764 + in figures [ figinte2 ] , [ figinte3 ] , and [ figinte5 ] one displays the level sets of the ratii @xmath355 for @xmath352 and @xmath353 respectively . the closer to white the color , the closer to @xmath356 the ratio @xmath355 . finally , for this double integrator example we have plotted in figure [ figinte6 ] the level sets of the function @xmath357 where @xmath342 is the known optimal minimum time to steer the initial state @xmath78 to @xmath358 ; the problem being time - homogeneous , one may take @xmath359 $ ] instead of @xmath150 $ ] . for instance , one may verify that when the initial state is in the region where the approximation is good , then the whole optimal trajectory also lies in that region . consider the so - called _ brockett system _ in @xmath360 @xmath361 where @xmath362 , and the control @xmath363 satisfies the constraint @xmath364 in this case , we have @xmath365 , @xmath366 , and @xmath10 is the closed unit ball of @xmath325 , centered at the origin . note that set @xmath13 is not compact and so the convergence result of theorem [ main ] may not hold , see the discussion at the beginning of example [ ex1 ] . nevertheless , in the numerical examples , we have not enforced additional state constraints . let @xmath342 be the minimum time needed to steer an initial condition @xmath367 to the origin . we recall the following result of @xcite ( in fact given for equivalent ( reachability ) ocp of steering the origin to a given point @xmath78 ) . consider the minimum time ocp for the system ( [ brosys ] ) with control constraint ( [ cont ] ) . the minimum time @xmath342 needed to steer the origin to a point @xmath368 is given by @xmath369 where @xmath370 is the unique solution in @xmath371 of @xmath372 moreover , the function @xmath0 is continuous on @xmath360 , and is analytic outside the line @xmath373 . [ remensemblecourbes ] along the line @xmath373 , one has @xmath374 the singular set of the function @xmath0 , _ i.e. _ the set where @xmath0 is not @xmath375 , is the line @xmath373 in @xmath360 . more precisely , the gradients @xmath376 , @xmath377 , are discontinuous at every point @xmath378 , @xmath379 . for the interested reader , the level sets @xmath380 , with @xmath381 , are displayed , e.g. , in prieur and trlat @xcite . recall that the convergence result of theorem [ main ] is guaranteed for @xmath13 compact only . however , in the present case @xmath365 is not compact . one possibility is to take for @xmath13 a large ball of @xmath382 centered at the origin because for initial states @xmath28 with norm @xmath383 relatively small , the optimal trajectory remains in @xmath13 . however , in the numerical experiments presented below , we have chosen to maintain @xmath365 , that is , the lmi - relaxation @xmath166 does not include any localizing constraint @xmath384 on the moment sequence @xmath240 . recall that @xmath385 as @xmath165 increases , i.e. , the more moments we consider , the closer to the exact value we get . in table [ table_example2 ] we have displayed the optimal values @xmath251 for @xmath386 different values of the initial state @xmath18 , in fact , all @xmath386 combinations of @xmath387 , @xmath388 , and @xmath389 . so , the entry @xmath390 of table [ table_example2 ] for the second lmi - relaxation is @xmath391 for the initial condition @xmath392 . at some ( few ) places in the table , the @xmath393 indicates that the sdp solver encountered some numerical problems , which explains why one finds a lower bound @xmath394 slightly higher than @xmath395 , when practically equal to the exact value @xmath396 . 0.0000 & 0.9999 & 1.9999 & 2.9999 + 0.0140 & 1.0017 & 2.0010 & 3.0006 + 0.0243 & 1.0032 & 2.0017 & 3.0024 + 0.0295 & 1.0101 & 2.0034 & 3.0040 + + 0.0000 & 0.9998 & @xmath397 & @xmath398 + 0.2012 & 1.1199 & 2.0762 & 3.0453 + 0.3738 & 1.2003 & 2.1631 & 3.1304 + 0.4946 & 1.3467 & 2.2417 & 3.1943 + + 0.0000 & 0.9995 & @xmath399 & @xmath400 + 0.7665 & 1.3350 & 2.1563 & 3.0530 + 1.0826 & 1.7574 & 2.4172 & 3.2036 + 1.3804 & 2.0398 & 2.6797 & 3.4077 + + 0.0000 & 0.9992 & 1.9977 & 2.9952 + 1.2554 & 1.5925 & 2.1699 & 3.0478 + 1.9962 & 2.1871 & 2.5601 & 3.1977 + 2.7006 & 2.7390 & 2.9894 & 3.4254 + + 0.0000 & 1.0000 & 2.0000 & 3.0000 + 2.5066 & 1.7841 & 2.1735 & 3.0547 + 3.5449 & 2.6831 & 2.5819 & 3.2088 + 4.3416 & 3.4328 & 3.0708 & 3.4392 + notice that the upper triangular part ( i.e. , when both first coordinates @xmath401 of the initial condition are away from zero ) displays very good approximations with low order moments . in addition , the further the coordinates from zero , the best . for another set of initial conditions @xmath402 and @xmath403 table [ table_example3 ] displays the results obtained at the lmi - relaxation @xmath404 . 1.7979 & 2.3614 & 3.2004 + 2.3691 & 2.6780 & 3.3341 + 2.8875 & 3.0654 & 3.5337 + + 1.8257 & 2.3636 & 3.2091 + 2.5231 & 2.6856 & 3.3426 + 3.1895 & 3.1008 & 3.5456 + the regularity property of the minimal - time function seems to be an important topic of further investigation . consider the so - called zermelo problem in @xmath325 studied in bokanowski et al . @xcite @xmath405 with @xmath406 . the state @xmath78 is constrained to remain in the set @xmath407\times [ -2,2]\subset\rr^2 $ ] , and we also have the control constraints @xmath408 , as well as @xmath409 $ ] . the target @xmath2 to reach from an initial state @xmath28 is the ball @xmath410 with @xmath411 . with the change of variable @xmath412 , @xmath413 , and @xmath414 , this problem is formulated as a minimum time ocp with associated hierarchy of lmi - relaxations @xmath318 defined in ( [ sdpqrt ] ) . therefore , if an lmi - relaxation @xmath166 at some stage @xmath165 of the hierarchy is infeasible then the ocp itself is infeasible , i.e. , the initial state @xmath28 can not be steered to the target @xmath2 while the trajectory remains in @xmath13 . figures [ zermelo1 ] and [ zermelo2 ] display the uncontrollable initial states @xmath294 found with the infeasibility of the lmi - relaxations @xmath415 and @xmath416 respectively . with @xmath415 , i.e. by using only second moments , we already have a very good approximation of the controllable set displayed in @xcite , and @xmath416 brings only a small additional set of uncontrollable states . we first prove item ( i ) . consider the linear program @xmath31 defined in ( [ p ] ) , assumed to be feasible . from the constraint @xmath417 , one has @xmath418 in particular , taking @xmath419 and @xmath420 , it follows that @xmath421 and @xmath422 . hence , for every @xmath423 satisfying @xmath417 , the pair @xmath424 belongs to the unit ball @xmath425 of @xmath426 . from banach - alaoglu theorem , @xmath425 is compact for the weak @xmath427 topology , and even sequentially compact because @xmath425 is metrizable ( see e.g. hernndez - lerma and lasserre ( * ? ? ? * lemma 1.3.2 ) ) . since @xmath67 is continuous ( see remark [ continuity ] ) , it follows that the set of @xmath423 satisfying @xmath417 is a closed subset of @xmath428 , and thus is compact . moreover , since the linear program @xmath31 is feasible , this set is nonempty . finally , since the linear functional to be minimized is continuous , @xmath31 is solvable . we next prove item ( ii ) . consider the set @xmath429 to prove that @xmath430 is closed , consider a sequence @xmath431 of @xmath432 such that @xmath433 for some @xmath434 . it means that @xmath435 , and @xmath436 . in particular , taking @xmath437 and @xmath438 , there must hold @xmath439 hence , there exist @xmath440 and a ball @xmath441 of @xmath442 , such that @xmath443 for every @xmath444 . since @xmath441 is compact , up to a subsequence @xmath445 converges weakly to some @xmath446 . this fact , combined with ( [ net ] ) and the continuity of @xmath67 , yields @xmath447 , and @xmath448 . therefore , the set @xmath430 is closed . + from anderson and nash ( * ? ? ? * theorems 3.10 and 3.22 ) , it follows that there is no duality gap between @xmath31 and @xmath32 , and hence , with ( i ) , @xmath127 . item ( iii ) follows from vinter ( * ? ? ? * theorems 2.1 and 2.3 ) , applied to the mappings @xmath449 for @xmath450 . @xmath451 first of all , observe that @xmath166 has a feasible solution . indeed , it suffices to consider the sequences @xmath197 and @xmath186 consisting of the moments ( up to order @xmath273 ) of the occupations measures @xmath452 and @xmath453 associated with an admissible control @xmath454 of the ocp ( [ system])-([crit1 ] ) . next , observe that , for every couple @xmath455 satisfying all constraints of @xmath166 , there must holds @xmath456 and @xmath457 . indeed , it suffices to choose @xmath419 and @xmath458 ( or @xmath234 ) in the constraint @xmath459 we next prove that , for @xmath165 sufficiently large , one has @xmath460 , @xmath461 , @xmath462 , for every @xmath159 , and @xmath463 . we only provide the details of the proof for @xmath240 , the arguments being similar for @xmath241 , @xmath239 and @xmath197 . let @xmath212 $ ] be the space of sums of squares ( s.o.s . ) polynomials , and let @xmath464 $ ] be the _ quadratic modulus _ generated by the polynomials @xmath465 $ ] that define @xmath13 , i.e. , @xmath466\quad\vert\quad \sigma\,=\,\sigma_0+\sum_{j\in j}\sigma_j\,v_j\quad \mbox{with } \sigma_j\in\sigma^2,\quad \forall\,j\in \{0\}\cup j\}.\ ] ] in addition , let @xmath467 be the set of elements @xmath89 of @xmath468 which have a representation @xmath469 for some s.o.s . family @xmath470 with @xmath471 and @xmath472 for every @xmath473 . let @xmath474 be fixed . since @xmath475^n$ ] , there holds @xmath476 on @xmath13 , for every @xmath201 with @xmath477 . therefore , since @xmath13 satisfies putinar condition ( see definition [ def1 ] ) , the polynomial @xmath478 belongs to @xmath468 ( see putinar @xcite ) . moreover , there exists @xmath479 such that @xmath480 for every @xmath477 . of course , @xmath481 for every @xmath477 , whenever @xmath482 . for every feasible solution @xmath186 of @xmath483 one has @xmath484 this follows from @xmath457 , @xmath485 and @xmath486 , which implies @xmath487 with similar arguments , one redefines @xmath479 so that , for every couple @xmath455 satisfying the contraints of @xmath483 , one has @xmath488 from this , and with @xmath489 , we immediately deduce that @xmath490 whenever @xmath491 . in particular , @xmath492 , which proves that @xmath493 , and so @xmath494 for @xmath165 sufficiently large . + let @xmath495 ( by theorem [ thlp ] ) , let @xmath496 , and let @xmath497 be a nearly optimal solution of @xmath166 with value @xmath498 complete the finite vectors @xmath499 and @xmath500 with zeros to make them infinite sequences . since for arbitrary @xmath501 one has @xmath502 whenever @xmath503 , provided @xmath165 is sufficiently large , by a standard diagonal argument , there exists a subsequence @xmath504 and two infinite sequences @xmath197 and @xmath186 , with @xmath505 and @xmath506 , for all @xmath507 , and such that @xmath508 next , let @xmath165 be fixed arbitrarily . observe that @xmath509 implies @xmath510 whenever @xmath511 , and similarly , @xmath512 . therefore , from ( [ convergence ] ) and @xmath510 , we deduce that @xmath198 , and similarly @xmath196 . since this holds for arbitrary @xmath165 , and @xmath513 for all @xmath507 , one infers from proposition [ prop1 ] that @xmath197 and @xmath186 are moment sequences of two measures @xmath156 and @xmath155 with support contained in @xmath202^n$ ] and @xmath253\times [ -1,1]^n\times [ -1,1]^m$ ] respectively . in addition , from the equalities @xmath514 and @xmath515 for every @xmath159 , it follows that @xmath156 and @xmath155 are probability measures on @xmath202^n$ ] , and @xmath253\times [ -1,1]^n\times [ -1,1]^m$ ] . next , let @xmath516 be fixed , arbitrarily . from @xmath517 and the convergence ( [ convergence ] ) , we obtain @xmath518 that is , @xmath519 . since @xmath516 is arbitrary , we have @xmath520,\ ] ] which implies that @xmath417 . let @xmath240 , @xmath241 and @xmath239 denote the moment vectors of the marginals of @xmath155 with respect to the variables @xmath521 , @xmath522 and @xmath523 , respectively , i.e. , @xmath524 and @xmath525 for every @xmath526 . with @xmath165 fixed arbitrarily , and using again ( [ convergence ] ) , we also have @xmath527 for every @xmath528 , and @xmath529 since @xmath13 , @xmath2 and @xmath10 satisfy putinar s condition ( see definition [ def1 ] ) , from theorem [ putinarthm ] ( putinar s positivstellensatz ) , @xmath197 is the moment sequence of a probability measure @xmath156 supported on @xmath530^n$ ] . similarly , @xmath240 is the moment sequence of a measure @xmath531 supported on @xmath475^n$ ] , @xmath241 is the moment sequence of a measure @xmath532 supported on @xmath262^m$ ] , and @xmath239 is the moment sequence of a measure @xmath533 supported on @xmath253 $ ] . since measures on compact sets are moment determinate , it follows that @xmath531 , @xmath532 , and @xmath533 are the marginals of @xmath155 with respect to @xmath78 , @xmath15 and @xmath234 respectively . therefore , @xmath155 has its support contained in @xmath53 , and from @xmath417 it follows that @xmath423 satisfies all constraints of the problem @xmath31 . moreover , one has @xmath534 hence , @xmath423 is an optimal solution of @xmath31 , and @xmath535 ( the sequence is monotone nondecreasing ) . item ( i ) is proved . item ( ii ) follows from theorem [ thlp ] ( iii ) . @xmath451 it suffices to prove that @xmath536 as @xmath537 . for every integer @xmath302 , @xmath538 is attained for a couple of measures @xmath539 . as in the proof of theorem [ thlp ] , the sequence @xmath540 is bounded in @xmath432 , and hence , up to a subsequence , it converges to an element @xmath423 of this space for the weak @xmath427 topology . on the one hand , by definition , @xmath541 for every @xmath302 . on the other , @xmath542 tends strongly to @xmath67 , and so @xmath543 moreover , since @xmath544 tends strongly to @xmath545 in @xmath546 , one has @xmath547 and so @xmath548 . we next prove that @xmath549 . since @xmath539 is an optimal solution of @xmath304 , @xmath550 hence , passing to the limit , @xmath551 and so , @xmath423 is an optimal solution of @xmath31 , i.e. , @xmath549 . the research of didier henrion was partly supported by project no . 102/06/0652 of the grant agency of the czech republic and research program no . msm6840770038 of the ministry of education of the czech republic . lasserre , c. prieur , d. henrion . nonlinear optimal control : numerical approximations via moments and lmi relaxations . 44th ieee conference on decision and control _ , sevilla , spain , december 2005 , pp .

in quantum systems with short - range interactions , causality imposes nontrivial constraints on low - energy scattering parameters . we investigate these causality constraints for systems where a long - range coulomb potential is present in addition to a short - range interaction . the main result is an upper bound for the coulomb - modified effective range parameter . we discuss the implications of this bound to the effective field theory for nuclear halo systems . in particular , we consider several examples of proton nucleus and nucleus nucleus scattering . for the bound - state regime , we find relations for the asymptotic normalization coefficients ( ancs ) of nuclear halo states . as an application of these relations , we extract the ancs of the excited and states ino fromc scattering data .

1210.8304

the constraints of causality for two - body scattering with finite - range interactions were first derived by wigner @xcite . the causality bound can be understood as a lower bound on the time delay between the incoming and outgoing wave packets , @xmath4 . when @xmath4 is negative , the outgoing wave packet departs earlier than for the non - interacting system . however , the incoming wave must first reach the interaction region before the outgoing wave can leave . in low - energy scattering this manifests itself as an upper bound on the effective range parameter . in ref . @xcite , phillips and cohen derived this bound for s - wave scattering with finite - range interactions . some constraints on nucleon nucleon scattering and the chiral two - pion exchange potential were considered in ref . @xcite , and relations between the scattering length and effective range have been explored for one - boson exchange potentials @xcite and van der waals potentials @xcite . in refs . @xcite the causality bounds for finite - range interactions were extended to an arbitrary number of spacetime dimensions and arbitrary angular momentum . the extension to systems with partial wave mixing was first studied in ref . @xcite . in this paper , we consider the causality constraints for the scattering of two charged particles with an arbitrary finite - range interaction . our analysis is the first study of causality bounds that takes into account the long - range coulomb force . the results presented here are relevant to studies of low - energy scattering of nuclei and nucleons using effective field theory ( eft ) . in particular , they are important for the application of effective field theory to nuclear halo systems ( cf . refs . @xcite ) . this so - called halo eft utilizes the separation of energy scales between the internal excitations of the core nucleus and soft halo physics . an effective lagrangian is built order by order from local interactions and used to describe low - energy phenomena such as shallow bound states , charge radii , near - threshold resonances , radiative capture reactions , and soft photodissociation . halo eft has been used to describe neutron alpha scattering @xcite and alpha alpha scattering @xcite . bound single - neutron halo systems such as @xmath5be @xcite and @xmath6li @xcite have been studied as well as various two - neutron halo systems @xcite . there is an important connection between causality bounds and the convergence of effective field theory calculations with increasing order @xcite . for local contact interactions , the range of the effective interaction is controlled by the momentum cutoff scale of the effective theory . in effective theories with non - perturbative renormalization , which typically occur in nuclear physics , exact cutoff - independence can generally not be achieved . there is a `` natural '' value of the cutoff at which all higher - order corrections scale as expected from dimensional analysis . if the cutoff is taken larger , `` new physics '' intervenes , the corrections scale unnaturally , and unitarity violations may occur . this is different from what one encounters in high - energy particle physics where the renormalization is typically perturbative and cutoff momenta can be taken arbitrarily large . for calculations using dimensional regularization , the renormalization scale plays a similar role in regulating ultraviolet physics . the term `` new physics '' , in this context , refers to details left out ( integrated out ) in the effective theory . in the case of halo eft , these details are the finite size of the core nucleus and its internal excitations as well as the exponential tail of the pion - exchange interaction . problems with convergence of the effective theory can occur if the cutoff scale is set higher than the scale of the new physics . however , it is useful to have a more quantitative measure of when problems may appear , and this is where the causality bound provides a useful diagnostic tool . for each scattering channel we use the physical scattering parameters to compute a quantity called the causal range , @xmath7 . it is the minimum range for finite - range interactions consistent with the requirements of causality and unitarity . for any fixed cutoff scale , the causality bound marks a branch cut of the effective theory when viewed as a function of physical scattering parameters @xcite . the coupling constants of the effective theory become complex when scattering parameters violating the causality bound are enforced . these branch cuts do not appear in perturbation theory ; however , a nearby branch point can spoil the absolute convergence of the perturbative expansion . our results can be viewed as a guide for improving the convergence of halo eft calculations . in particular , if the cutoff momentum used in a calculation is too high , then problems with convergence may appear in some observables . consequently , the causal range can be used to estimate the `` natural '' ultraviolet cutoff @xmath8 of the effective theory as @xmath9 . the natural cutoff is optimal in the sense that no known infrared physics is left out of the theory and that all corrections involving the ultraviolet cutoff scale naturally @xcite . increasing the cutoff beyond the natural value will not improve the accuracy of the calculation . the causality bounds also have an impact in the regime of bound states . for two - body halo states , or more generally whenever there is a shallow two - body bound state close to threshold , the same integral identity that yields the causality bound for the effective range can be used to derive a relation between the asymptotic normalization constant ( anc ) of the bound - state wave function , the binding momentum , and the effective range for the scattering of the two halo constituents . this relation can be shown to be equivalent to a result previously derived by sparenberg _ et al . _ its significance lies in the fact that the anc is an important input parameter for the calculation of near - threshold radiative capture and photodissociation reactions . the causality bounds also constrain the range of model potentials that are fitted to scattering data in order to extract ancs . the organization of this paper is as follows . we first review the general theory of scattering for two charged particles with additional short - range interactions . our analysis includes both attractive and repulsive coulomb forces . in the next section we derive the charged - particle causality bounds for arbitrary values of the orbital angular momentum . using the causality bounds , we extract and discuss the causal range for several nuclear scattering processes including proton proton , proton deuteron , proton@xmath10he , proton alpha , and alpha alpha scattering . we then elucidate the relation for the anc and extract the ancs of the excited @xmath0 and @xmath1 states in @xmath2o from @xmath3c scattering data as an application . we conclude with a summary of the main results and provide an outlook . in the appendices , we provide technical details of the derivation as well as numerical examples . in this section , we review some preliminaries that we will need later in order to derive the desired causality bounds . we consider a two - particle system with reduced mass @xmath11 interacting via a finite - range potential with range @xmath12 . we write the interaction as a real symmetric operator with kernel @xmath13 satisfying the finite - range condition , @xmath14 in particular , we assume that the interaction is energy - independent . after giving a detailed formal derivation of the causality bounds in what follows , we will come back to the question what the above assumptions mean for the application to ( halo ) eft calculations in sec . [ sec : results ] . in the absence of coulomb interactions the system is described by the radial schrdinger equation , @xmath15 where by @xmath16 we explicitly indicate a solution for center - of - mass momentum @xmath17 in the following . following the conventions of ref . @xcite , the normalization of @xmath18 is chosen such that for @xmath19 we have @xmath20 \ , , \label{eq : u - asympt}\ ] ] where @xmath21 and @xmath22 are the riccati - bessel functions and @xmath23 is the scattering phase shift . if the particles carry electromagnetic charges @xmath24 and @xmath25 , respectively , there is a coulomb potential in addition to the finite - range interaction . the radial schrdinger equation now reads @xmath26 with the coulomb parameter @xmath27 the normalization of @xmath28 is now chosen such that for @xmath19 we have @xmath29 \ , , \label{eq : w - asympt}\ ] ] where @xmath30 and @xmath31 are the regular and irregular coulomb wave functions @xcite , respectively , and @xmath32 is the phase shift of the full solution @xmath28 compared to the regular coulomb function @xmath30 @xcite . our normalization is the same as chosen in ref . @xcite , _ i.e. _ , for @xmath19 our solution @xmath33 coincides with the function @xmath34 defined in eq . ( 42 ) of that paper . ] the factor @xmath35 is given by @xmath36 @xmath37 and @xmath38 ^ 2 } \prod\limits_{s=1}^\ell ( s^2+\eta^2 ) \cdot c_{\eta,0}^2 \ , . \label{eq : c_eta}\ ] ] for the coulomb wave functions @xmath30 and @xmath31 we use the conventions introduced by yost , breit and wheeler in ref . explicitly , we write them as @xcite @xmath39 @xmath40 [ eq : fg ] where @xmath41 the functions @xmath42 and @xmath43 are whittaker functions , which can be expressed in terms of hypergeometric functions as @xmath44@xmath45 @xmath46 is kummer s function of the first kind , @xmath47 and @xmath48 a comprehensive discussion of the functions @xmath49 and @xmath50 can be found in ref . @xcite . for asymptotically large @xmath51 the coulomb wave functions behave as @xmath52 with the coulomb phase shift @xmath53 in the limit @xmath54 , their behavior is [ eq : fg - rho - zero ] @xmath55 for a system without coulomb force and only a short - range interaction , one has the well - known effective range expansion @xmath56 where @xmath57 and @xmath58 are the scattering and effective range parameters , respectively . for a system of charged particles we have instead the coulomb - modified effective range expansion @xmath59 where @xmath60 @xmath61 and @xmath62 is the logarithmic derivative of the gamma function , also called digamma function . for @xmath63 , the expansion simplifies to @xmath64 a derivation of eq . for the case of proton proton scattering can be found , for example , in ref . see also ref . @xcite for a detailed discussion . the analytic properties of the @xmath63 modified effective range function are investigated in ref . @xcite .- dependent function on the left hand side of bethe s expansion appears to differ from our @xmath65 by two times the euler - mascheroni constant @xmath66 . this apparent conflict can be resolved by noting that the @xmath67 in eq . ( 51 ) of ref . @xcite is _ not _ the function defined in eq . ( 47a ) of the same paper , but rather given by @xmath68 $ ] , where in this latter expression the @xmath69 from eq . ( 47a ) is meant . the limiting process then yields exactly the term @xmath70 . ] in ref . @xcite , boll and gesztesy derived a very general form of the coulomb - modified effective range expansion for an arbitrary number of spatial dimensions . specializing their result to the three - dimensional case , a version of eq . can be written as @xmath71 with rather than @xmath72 . ] @xmath73 the latter function can be rewritten using @xmath74 with @xmath75 as defined in eq . . the expressions given here reproduce eqs . and when one explicitly assumes that the momentum @xmath17 is real . in fact , one has to rewrite eq . in this manner in order to get an effective range function that is analytic in @xmath76 around threshold . the conventions for the coulomb - modified efffective range expansion for general @xmath77 that we use throughout this work , eqs . and , are the same as in ref . @xcite . we note , however , that a different convention is also used in the literature which differs from ours by an overall momentum - independent factor . the effective range expansion given in refs . @xcite uses the normalization @xmath78 = -\frac{1}{\tilde{a}_{\ell}^{c } } + \frac{1}{2}\tilde{r}_{\ell}^{c}\,p^{2}+\cdots \ , . \label{eq : ere - alternative}\ ] ] this choice of normalization has the advantage of having a more direct correspondence to the ordinary effective range expansion without coulomb effects . for @xmath63 , both conventions give the same expression . with the coulomb wave functions and coulomb - modified effective range expansions at our hands , we can now closely follow the derivation presented in ref . @xcite for scattering in the absence of coulomb interactions . we consider solutions of the radial schrdinger equation for two different momenta @xmath79 and @xmath80 . introducing the short - hand notation @xmath81 _ i.e. _ , suppressing the angular - momentum subscript @xmath77 for convenience , we get @xmath82 \label{eq : w_ba - w - int - pre}\end{gathered}\ ] ] by subtracting @xmath83 times the equation for @xmath84 from that for @xmath84 multiplied by @xmath83 , as it is done in ref . @xcite , and integrating from a small radius @xmath85 to @xmath86 . we assume that our interaction @xmath13 is such that it alone permits a solution that is sufficiently regular at the origin , _ i.e. _ , @xmath87 and @xmath88 stays finite as @xmath89 , where @xmath90 is a solution of eq . . as boundary condition for the solutions @xmath91 of the full radial schrdinger equation we can then demand as well that they vanish with finite derivative at the origin . if we only had the coulomb potential and no additional interaction , this is fulfilled by the regular coulomb function @xmath49 , _ cf . _ eq . . we can thus take the limit @xmath92 in eq . and get the relation @xmath93(r ) = ( p_b^2-p_a^2)\int_0^r{\mathrm{d}}r'\,w_a(r')w_b(r ' ) \ , , \label{eq : w_ba - w - int}\ ] ] where the wronskian @xmath94 $ ] is defined as @xmath93(r ) = w_b(r)w_a'(r ) - w_a(r)w_b'(r ) \,.\ ] ] following further the derivation presented in ref . @xcite , we re - express the solutions @xmath95 in terms of functions @xmath96 and @xmath97 such that @xmath98 for @xmath19 , with @xmath96 analytic in @xmath99 , @xmath100 and [ eq : g - exp]@xmath101 the @xmath97 contains a term which is non - analytic in @xmath99 and is proportional to @xmath96 . the remainder @xmath102 , however , is analytic in @xmath99 , @xmath103 explicit expressions for these functions can be obtained from the results of boll and gesztesy @xcite ; they are discussed further in appendix [ sec : coulombwf - bg ] . combining eqs . and , we find @xmath104 and @xmath105 [ eq : fg - fg ] we insert now the modified effective range expansion into the asymptotic coulomb wave function . note also that @xmath106 where the last step follows from the relations in appendix [ sec : coulombwf - bg ] . the term involving @xmath107 exactly drops out and we are left with @xmath108 thus it is possible to choose a normalization such that @xmath95 is analytic in @xmath76 . combining this with the expansions and , we arrive at @xmath109 + { \mathcal{o}}(p^4)\ , . \label{eq : w - asympt - ere}\ ] ] from here , we can proceed exactly as in ref . @xcite . for the wronskian of two solutions @xmath83 and @xmath84 for @xmath19 we find @xmath110(r ) = ( p_b^2-p_a^2)\bigg\{\frac12r_\ell^c w[f_0,g_0](r ) + \left(\frac1{a_\ell^c}\right)^{\!2 } w[f_2,f_0](r ) \\ - \frac1{a_\ell^c}\left[w[f_2,g_0](r ) - w[g_2,f_0](r)\right ] + w[g_2,g_0](r)\bigg\ } + { \mathcal{o}}(p_{a , b}^4 ) \ , . \label{eq : w_ba - w - exp}\end{gathered}\ ] ] note that in the @xmath111 we have also included terms of the form @xmath112 . we set @xmath113 in eq . and furthermore take the limit @xmath114 . using the expansion , we get @xmath115(r ) = b_\ell^c(r ) - 2\int_0^r{\mathrm{d}}r'\left[w_\ell^{(0)}(r')\right]^2 \label{eq : rc - bc - raw}\ ] ] for @xmath19 , with @xmath116 and @xmath117(r ) - \frac2{a_\ell^c}\left\{w[f_2,g_0](r ) + w[g_2,f_0](r)\right\ } + \frac2{\left(a_\ell^c\right)^2 } w[f_2,f_0](r ) \ , . \label{eq : bwronksians}\ ] ] written as a function of @xmath118 , the wronskian of the coulomb wave functions is @xmath119(\rho ) = -w[g_\ell^{(p)},f_\ell^{(p)}](\rho ) = -1 \,.\ ] ] since @xmath120 and @xmath121\equiv0 $ ] , we also have @xmath122(r ) = w[f,\tilde{g}](r ) = -1 \,.\ ] ] plugging in the expansions and , we see that @xmath123(r)=-1 $ ] for the leading - order functions , and @xmath124(r)=w[g_{2},f_{0}](r)$ ] for the terms at @xmath125 . inserting these relations into eq . , we get @xmath126 ^ 2 \ , , \label{eq : rc - bc}\ ] ] where @xmath127 has been simplified to @xmath117(r ) - \frac4{a_\ell^c}w[f_2,g_0](r ) + \frac2{\left(a_\ell^c\right)^2}w[f_2,f_0](r ) \ , . \label{eq : bc}\ ] ] since the integral in eq . is positive definite , the resulting causality bound is @xmath128 we now want to derive the explicit form of the function @xmath129 . to do this , we need expressions for the wronskians that appear in eq . . we can obtain them by first noting that @xmath96 and @xmath102 , being linear combinations of coulomb wave functions ( with @xmath17-dependent coefficients ) , are solutions of the coulomb schrdinger equation , @xmath130x(p , r ) = 0 \ , , \label{eq : sg - rad - coulomb}\ ] ] which , of course , corresponds to setting @xmath131 in eq . . here and in the following , @xmath132 stands for either @xmath133 or @xmath134 . inserting the expansion @xmath135 into eq . and comparing orders in @xmath99 , we find that @xmath136 x_0(r ) = 0 \ , , \label{eq : dg - x0}\ ] ] _ i.e. _ , @xmath137 is a solution of the zero - energy coulomb schrdinger equation , and @xmath136x_2(r ) = x_0(r ) \ , . \label{eq : dg - x2}\ ] ] from this we readily obtain the differential equations [ eq : dg - wfg]@xmath138(r ) & = \left[f_0(r)\right]^2 \ , , \\ \label{eq : dg - wg2g0 } \frac{{\mathrm{d}}}{{\mathrm{d}}r } w[g_2,g_0](r ) & = \left[g_0(r)\right]^2 \ , , \\ \label{eq : dg - wf2g0 } \frac{{\mathrm{d}}}{{\mathrm{d}}r } w[f_2,g_0](r ) & = f_0(r)g_0(r)\end{aligned}\ ] ] for the desired wronskians . put together , this yields a simple first - order differential equation for @xmath129 , @xmath139 from eqs . ( a.7 ) and ( a.8 ) in ref . @xcite we have the explicit expressions [ eq : f0-g0-rep]@xmath140 for @xmath141 , where @xmath142 and @xmath143 are modified bessel functions , and [ eq : f0-g0-attr]@xmath144 for @xmath145 , where @xmath146 and @xmath147 are the ordinary bessel functions . and @xmath148 are indeed solutions of . ] in the attractive case ( @xmath145 ) that involves the hankel function @xmath149 times @xmath150 instead of @xmath151 . with that , however , @xmath148 would not be real , which it should be . our @xmath148 as in eq . is taken from the results of lambert @xcite . ] using these expressions for @xmath152 and @xmath148 and eq . we can determine @xmath129 up to an integration constant . in order to fix this constant , we must work directly with the wronskians in eq . ( [ eq : bwronksians ] ) . before we do that , however , we first discuss the general form of @xmath129 . let us break apart the function as a sum of two functions , @xmath153 and @xmath154 , and a constant term @xmath155 , @xmath156 we take @xmath153 to be a function consisting entirely of a sum of terms that have a pole at @xmath157 , ranging from order @xmath158 to @xmath77 , @xmath159 by furthermore requiring the function @xmath154 to vanish at @xmath157 , the decomposition in eq . is unique . where exactly the contributions to the three terms in the decomposition come from can be inferred from the behavior of @xmath96 and @xmath97 at the origin . from eq . combined with eq . we find that @xmath160 this implies that every term in the expansion of @xmath96 is @xmath161 . therefore , @xmath162(r ) = 0\ ] ] for all @xmath77 , which means that this wronskian only yields contributions to the @xmath163 . furthermore , from eq . ( 17 ) in ref . @xcite we know that the irregular coulomb wave function has the asymptotic behavior @xmath164 with @xmath165 fulfilling @xmath166 . using eq . then yields @xmath167 we note that @xmath168 has exactly the same behavior near @xmath157 and can thus show that @xmath169 is subleading as @xmath89 , @xmath170 for @xmath171 . from this we infer that @xmath172(r ) = 0\ ] ] for all @xmath77 , so also from this wronskian we only get contributions to @xmath163 . both the singular @xmath173 and the constant @xmath174 , therefore , only come from the wronskian @xmath175(r)$ ] . for @xmath63 the situation is still simple because the above analysis also tells us that @xmath176(r ) = 0 { { \ \ } \text{for}{\ \ } } \ell=0 \,,\ ] ] _ i.e. _ , @xmath177 is given entirely by @xmath178 . with the knowledge that it vanishes at the origin , it is actually straightforward to give an explicit expression for @xmath179 in terms of antiderivatives of the right hand side of eq . , where one has to insert the @xmath180 and @xmath181 from eqs . and the result , obtained by integrating from @xmath182 to @xmath86 , is @xmath183{c } { \frac12}\\ 0,1,-2 \end{array}\right ) \\ & \hspace{2em } + 4\sqrt{\pi}\gamma r^{2}\ , g_{2,4}^{3,1}\left(4\gamma r\middle|\begin{array}[c]{c } -1,{\frac12}\\ -1,0,1,-2 \end{array}\right ) \label{eq : b0-rep } \end{split}\ ] ] for the repulsive case , and @xmath184{c } { \frac12},{-\frac12}\\ 0,1,-2,{-\frac12 } \end{array}\right ) \\ & \hspace{2em } + 2\pi^2\left[\frac{\gamma^2r^3}{3}\ , { _ { 1}f_{2}}\left(\frac{3}{2};2,4;4\gamma r\right ) -\frac{2\gamma r^2}{\sqrt\pi}\ , g^{3,1}_{3,5}\left(-4\gamma r\middle|\begin{array}[c]{c } -1,{\frac12},{-\frac12}\\ -1,0,1,-2,{-\frac12 } \end{array}\right)\right ] \label{eq : b0-attr } \end{split}\ ] ] for an attractive coulomb interaction . in the above equations , @xmath185 and @xmath186 denote the ( generalized ) hypergeometric and meijer @xmath187-functions , respectively . for general @xmath188 , @xmath189(r)$ ] is singular at @xmath157 and the analysis becomes more complicated . for practical purposes , one can simply use power - series expansions for the bessel functions that appear in the expressions for the zero - energy functions and integrate these term by term until a desired precision is reached . the only additional ingredients needed are the values for the constant terms @xmath174 because these are obviously not generated by the integration . in table [ tab : wg2g0-const ] we list these constants for @xmath190 . .constant term @xmath155 in eq . ( [ eq : xyz ] ) for @xmath190 . @xmath191 is the euler - mascheroni constant . the values are the same for repulsive ( @xmath141 ) and attractive ( @xmath145 ) coulomb potentials . [ cols="^,^,^,^",options="header " , ] finally , we note that sparenberg _ et al . _ @xcite have previously derived an anc relation equivalent to ours , but written in terms of the scattering length instead of the effective range parameter . the equivalence of the two relations up to the contributions of higher order shape parameters is shown in appendix [ sec : sparenberg - equiv ] . the @xmath192 nucleus has two excited states lying just below the @xmath193@xmath194 threshold , a @xmath0 at about @xmath195 and an even more shallow @xmath1 at only @xmath196 . the properties of these states play an important role in astrophysical helium burning processes @xcite . in the following , we calculate asymptotic normalization constants for both states under the assumption that they can be described in a @xmath197 halo picture . we use the recent data obtained by tischhauser _ et al . _ @xcite ( for the actual phase shifts see their ref . [ 32 ] ) in order to extract the coulomb - modified effective range for the @xmath193@xmath194 p- and d - wave channels . focusing first on the d - wave , we note that the combination of the strong coulomb repulsion and the @xmath198 centrifugal barrier makes the low - energy phase shifts very small over a wide energy range . moreover , there is a narrow resonance at a center - of - mass energy of about @xmath199 , which strongly constrains the energy region for a straightforward fit to the effective range function . from a simple fit to the data up to the narrow resonance one only obtains an effective range parameter with an uncertainty too large ( 50% ) for an extraction of the anc since the latter depends on @xmath200 very sensitively . to mitigate this problem , we use the position of the @xmath0 oxygen state as an additional input parameter for a self - consistent extraction of the effective range . in the following , we describe this procedure in more detail . at the position of the bound state , where @xmath201 , the coulomb - modified effective range expansion reduces to @xmath202 where now @xmath203 a straightforward calculation shows that for the prefactor we have @xmath204 furthermore , for the digamma function we have the asymptotic expansion ( eq . ( 6.3.18 ) in ref . @xcite ) @xmath205 for @xmath206 , where the @xmath207 are the bernoulli numbers , @xmath208 noting that the sum in eq . only starts at @xmath209 , we find that @xmath210 only starts at order @xmath211 , and inserting the precise relation into eq . , we arrive at @xmath212 neglecting the effective range contribution at leading order , we insert the binding momentum @xmath213 of the @xmath0 state to get a first approximation for the scattering length parameter . this is then used to constrain a subsequent fit to the phase shift data up to about @xmath214 , _ i.e. _ , just below the narrow resonance . we include a single @xmath215 shape parameter in the fit since some curvature is clearly necessary . the @xmath216 obtained from this is then used in eq . to get a better result for @xmath217 , which , in turn , is fed back into the fit . iterating this procedure a couple of times yields a well - converged self - consistent result for @xmath216 . after eight iterations , we find @xmath218 for the @xmath193@xmath194 d - wave . including a second shape parameter in the fitting procedure only changes this result within the given uncertainty , so we conclude that for the energy range we have been fitting , a single shape parameter really is sufficient to account for the curvature . inserting the fit result into eq . yields @xmath219 for the @xmath0 state in @xmath192 . including in eq . an @xmath220 term of the order of the shape parameter gives a consistent result for @xmath221 within the error given above . when we apply the same same procedure to the @xmath1 state just @xmath222 below the @xmath193@xmath194 threshold , we also see a nice convergence and obtain the results @xmath223 and @xmath224 the overall picture for the @xmath1 state is somewhat more complicated , however . on the one hand , the extraction of the effective range parameters is easier in this case because there is no narrow resonance limiting the fit range ( we have used the data up to @xmath225 ) . by doing a simple fit without the self - consistent iteration we get values for the effective range and the anc that are slightly smaller , but overlap with the results given above when the respective uncertainties are taken into account . on the other hand , allowing for a second shape parameter changes the value of the effective range quite dramatically to @xmath226 , which leads to an imaginary anc . this could indicate that the effective range parameters from the simple fit violate the causality bound . the fact that @xmath200 has to be such that eq . yields a real value for the anc can be interpreted as a weaker remnant of the original causality bound relation . alternatively , the cluster picture might not be applicable for the shallow @xmath1 state in @xmath192 . our result for the anc of the @xmath0 state is about a factor five smaller than the value @xmath227 obtained by brune _ et al . _ @xcite , while the value for @xmath228 is about a factor two smaller . other , more recent determinations @xcite have found even larger values . for the interpretation of this discrepancy , note that our calculations are predictions of the ancs based only on alpha carbon elastic scattering data and the assumption that the system can be approximately described in an effective two - body picture with a finite - range interaction . in the references mentioned above , the ancs are extracted from alpha carbon transfer measurements . note furthermore that a comparison of the experimental extractions in ref . @xcite exhibits quite some discrepancy ( factors of two up to roughly an order of magnitude ) also between the cited individual experimental determinations of the ancs . sparenberg _ et al . _ @xcite have carried out a similar analysis for the @xmath0 state based on their anc relation ( _ cf . _ ref . they subtracted the narrow resonance from the @xmath193@xmath194 d - wave phase shift data in order to extract a set of higher - order shape parameters and concluded that present day data are not sufficient to constrain the anc strongly . as discussed above , our approach of performing a self - consistent fit to the data below the resonance constrained by the separation energy and using the effective range instead of the scattering length as input in the anc relation improves the stability of the extraction . however , compared to other determinations our anc values are generally smaller by a factor two to five . we thus conclude that this issue requires further study . in this paper , we have investigated the constraints imposed by causality on the low - energy scattering parameters of charged particles interacting via a short - range interaction and a long - range coulomb potential . similar to the case of neutral particles without coulomb interaction @xcite , our considerations lead to a constraint on the maximum value of the coulomb - modified effective range . while conceptionally straightforward , the calculation of the wronskians of the wave functions required for the derivation of the bound is intricate . we have calculated them through term - by - term integration of the power series expansion of the zero - energy wave functions and additionally determining the integration constants that are not generated by this process . we define the causal range as the minimum value of the interaction range consistent with the causality bound . in effective field theories with contact interactions such as the halo eft , the natural momentum cutoff is of the order of the inverse of the causal range . if the natural cutoff is not known from other considerations , its size can be estimated from the causal range . if the momentum cutoff used in a calculation is too high , then problems with convergence of higher order corrections can appear . for example , the convergence pattern might be such that an improvement in higher orders of the eft can only be sustained through large cancellations between individual terms . such an unnatural pattern would be especially undesirable for the stability of numerical ( lattice ) calculations . our results can thus be viewed as a guide for improving the convergence pattern of eft calculations with contact interactions . in lattice simulations of halo eft , the lattice spacing should not be taken smaller than the causal range . we have analyzed the causal ranges for a variety of systems ranging from proton proton scattering to alpha alpha scattering . our results for causal ranges in different partial waves in these systems typically vary by factors of 2 - 3 . the precise values are quite sensitive to small uncertainties in the effective range parameters . in channels with a large negative effective range the causal range is very close to zero , which implies that causality provides almost no constraints on the range of the interaction in this case . thus the causal range provides a good order of magnitude estimate of the range of interaction , but drawing more quantitative conclusions about the structure of the underlying potentials is difficult . after an analytic continuation to the bound state regime , the integral relations for the causality bound can also be used to derive a model - independent expression for the anc for shallow bound states . if the state is a two - body halo state , the relation can be used to extract the anc from low - energy scattering parameters . up to higher order shape parameters , our relation is equivalent to the one previously derived by sparenberg _ et al . _ in ref . @xcite ( see appendix [ sec : sparenberg - equiv ] for details ) . one difference is that we express the anc in terms of the binding momentum and the effective range rather than the binding momentum and the scattering length . we find this form more suitable for the extraction of ancs from scattering data since the effective range is typically more precisely determined than the scattering length for shallow states . moreover , extracting the effective range in a self - consistent fit from the scattering data that reproduces the correct separation energy improves the stability of the extraction . finally , we illustrate our relation by extracting the ancs of the excited @xmath0 and @xmath1 states in @xmath2o from @xmath3c scattering data . compared to previous extractions @xcite , our values are generally smaller . whether this difference is physically significant requires further study . the application of our relation to other shallow cluster states and a benchmark against model calculations would also be very interesting . we thank renato higa and gautam rupak for useful discussions . we also thank the institute for nuclear theory at the university of washington for its hospitality and the department of energy for partial support during the completion of this work . s.k . furthermore thanks the north carolina state university , where a part of this research was worked out , for its hospitality . this research was supported in part by the dfg through sfb / tr 16 `` subnuclear structure of matter , '' the bmbf under contracts no . 06bn9006 , 06bn7008 , and by the us department of energy under contract no . de - fg02 - 03er41260 . s.k . was supported by the `` studienstiftung des deutschen volkes '' and by the bonn - cologne graduate school of physics and astronomy . to determine the constants @xmath155 introduced in sec . [ sec : wronskians ] , we consider the explicit form of @xmath102 . from the results of boll and gesztesy @xcite , we have by inserting @xmath229 , @xmath230 and assuming that the momentum @xmath17 is real and positive . ] @xmath231 r^n\right\ } \\ + { \mathrm{re}}\left\{\frac{r^{-\ell}}{2\ell+1}\,{\mathrm{e}}^{-{\mathrm{i}}pr } \cdot\sum\limits_{n=0}^{2\ell } d_{\ell , n}(p)\,r^n\right\ } \ , , \label{eq : g - tilde - series}\end{gathered}\ ] ] where @xmath232 @xmath233 \,,\ ] ] @xmath234 and @xmath235 with this result and the appropriate expression for @xmath181 from eq . or , one can use the following procedure to calculate the @xmath155 , _ i.e. _ , the terms of order @xmath236 in the wronskian @xmath175(r)$ ] . 1 . note that @xmath237(r ) = p^{2}\,w[g_{2},g_{0}](r ) + { \mathcal{o}}(p^{4})\ ] ] and calculate this wronskian using a truncated version ( including terms up to the order @xmath238 in @xmath86 is sufficient ) of @xmath102 as given in eq . . 2 . from the result , extract the terms that are of the order @xmath236 . 3 . from that expression then extract the terms that are of the order @xmath76 . they constitute the @xmath239-contributions in a series expansion of @xmath175(r)$ ] which can not be obtained from a term - by - term integration of @xmath240 . with a computer algebra software , this prescription is straightforward to implement . the results for @xmath241 are shown in table [ tab : wg2g0-const ] in sec . [ sec : wronskians ] . at this point we remark that there is a recent publication by seaton @xcite discussing coulomb wave functions that are explicitly analytic in the energy ( or the momentum squared ) . in principle , it is possible to use these results to get explicit expressions for @xmath242 and @xmath243 ( in addition to the already known zero - energy functions ) , and then simply calculate all the wronskians directly . however , the analytic irregular coulomb function defined by seaton is slightly different from our @xmath102 . more importantly , not all coefficients needed for the expansions in energy are given explicitly . finally , writing everything in terms of the wave functions of boll and gesztesy paves the way for a generalization of the results presented in the current paper to an arbitrary number of spatial dimensions . knowing now what the constant terms @xmath174 that come from @xmath175(r)$ ] should be , it is possible to write down explicit expressions for the causality bound functions @xmath244 . to do that we use that the antiderivative of the right hand side of eq . can be expressed in terms of ( generalized ) hypergeometric functions @xmath185 and meijer @xmath187-functions @xmath186 . the only additional point to be taken into account is that the antiderivative of @xmath240 in general includes a constant term that is different from the desired @xmath174 . hence , one has to determine this term and add another constant such that the sum is exactly equal to the @xmath174 given in table [ tab : wg2g0-const ] . for the antiderivative of @xmath240 that we are using in the following , up to a minus sign the correction term is just the @xmath245 introduced in sec . [ sec : anc - relations ] . , at least for the case of a repulsive coulomb interaction , where this integral is convergent . ] the @xmath63 result for a repulsive coulomb potential is given by eq . . for @xmath246 , we get @xmath247 \\ & \hspace{2em } - \frac{4r^2}{\sqrt\pi}\left(a_{1}^{c}\right)^{-1 } g^{2,2}_{2,4}\left(4\gamma r\middle|\begin{array}[c]{c } -1,\frac{1}{2}\\ 0,3,-3,-2 \end{array}\right ) \\ & \hspace{2em } - \frac{\sqrt{\pi}\gamma^{3}r^{2}}{9}\,g^{4,0}_{2,4}\left ( 4\gamma r\middle|\begin{array}[c]{c } -1,\frac{1}{2}\\ -3,-2,0,3 \end{array}\right ) + \frac{\gamma}{54 } \ , , \end{split } \label{eq : b1-rep}\ ] ] and for @xmath198 the result is @xmath248 \\ & \hspace{2em}- \frac{4r^2}{\sqrt\pi}\left(a_{2}^{c}\right)^{-1 } g^{2,2}_{2,4}\left(4\gamma r\middle|\begin{array}[c]{c } -1,\frac{1}{2}\\ 0,5,-5,-2 \end{array}\right ) \\ & \hspace{2em}- \frac{\sqrt{\pi}\gamma^{5}r^{2}}{3600}\,g^{4,0}_{2,4}\left ( 4\gamma r\middle|\begin{array}[c]{c } -1,\frac{1}{2}\\ -5,-2,0,5 \end{array}\right ) + \frac{\gamma^3}{21600 } \ , . \end{split } \label{eq : b2-rep}\ ] ] for an attractive coulomb potential , the @xmath63 result is given by eq . . the @xmath246 result reads @xmath247 \\ & \hspace{2em}+ 4\sqrt\pi r^2\left(a_{1}^{c}\right)^{-1 } g^{2,2}_{3,5}\left ( -4\gamma r\middle|\begin{array}[c]{c } -1,\frac12,{-\frac12}\\ 0,3,-3,-2,{-\frac12 } \end{array}\right ) \\ & \hspace{2em}+ \frac{\pi^{3/2}\gamma^3r^2}{18}\left[2\,g^{4,0}_{3,5}\left ( -4\gamma r\middle|\begin{array}[c]{c } { -\frac52},-1,{\frac12}\\ -3,-2,0,3,{-\frac52 } \end{array}\right)\right . \\ & \hspace{9em}\left.- g^{1,2}_{2,4}\left ( -4\gamma r\middle|\begin{array}[c]{c } -1,{\frac12}\\ 3,-3,-2,0 \end{array}\right)\right ] + \frac{\gamma}{54 } \ , , \end{split } \label{eq : b1-attr}\ ] ] and for @xmath198 one finds @xmath249 \\ & \hspace{2em}+ 4\sqrt\pi r^2\left(a_{2}^{c}\right)^{-1}g^{2,2}_{3,5}\left ( -4\gamma r\middle|\begin{array}[c]{c } -1,{\frac12},{-\frac12}\\ 0,5,-5,-2,{-\frac12 } \end{array}\right ) \\ & \hspace{2em } + \frac{\pi^{3/2}\gamma^{5}r^{2}}{7200}\left[2\ , g^{4,0}_{3,5}\left(-4\gamma r\middle|\begin{array}[c]{c } { -\frac92},-1,{\frac12}\\ -5,-2,0,5,{-\frac92 } \end{array}\right)\right . \\ & \hspace{9em}\left.- g^{1,2}_{2,4}\left ( -4\gamma r\middle|\begin{array}[c]{c } -1,{\frac12}\\ 5,-5,-2,0 \end{array}\right)\right ] + \frac{\gamma^3}{21600 } \ , . \end{split } \label{eq : b2-attr}\ ] ] boll and gesztesy define the coulomb wave functions @xmath250 and @xmath251 [ eq : fg - n ] where , in our case , @xmath229 and @xmath252 , @xmath253 , @xmath254 are as defined in eq . . using the formula . ] @xmath255 eq . ( 5.1 ) in ref . @xcite , one finds that @xmath256 and @xmath257 \,.\ ] ] [ eq : fg - n - fg ] it is shown in ref . @xcite that @xmath258 is analytic in @xmath99 . furthermore , from eqs . ( 3.16 ) , ( 3.17 ) and ( 4.1 ) in that paper it follows that @xmath259 where @xmath260 is the function defined in eq . , and where @xmath261 is analytic in @xmath99 . the functions @xmath96 and @xmath102 that we introduce in sec . [ sec : rewritewf ] are exactly the analytic wave functions defined above , _ i.e. _ , @xmath262 it should be noted that lambert @xcite already defines coulomb wave functions that are analytic in @xmath99 , but only for the most common case of three spatial dimensions . boll and gesztesy extend lambert s results to an arbitrary number of dimensions @xmath263 . more importantly , at least for our application , in eq . ( 4.3 ) of ref . @xcite , boll and gesztesy give an explicit expression for @xmath261 . since there are two typos in their original equation , in the first line should be @xmath264 and the @xmath265 in the last line should be @xmath266 . ] we quote the whole expression for completeness . slightly altering the notation to match our conventions , we have @xmath267(2p)^{n-3 } \left|\gamma\left(\frac{n-1}2+\frac{{\mathrm{i}}\gamma}{2p}\right)\right|^2 \left|\gamma\left(1+\frac{{\mathrm{i}}\gamma}{2p}\right)\right|^{-2 } \gamma\log\left(|\gamma|r\right)\cdot f_n^{(0)}(p , r ) \\ - \gamma\,\mathrm{re}\bigg\{\left[\gamma(n-1)^{-2}\right](2p)^{n-3 } \left|\gamma\left(\frac{n-1}2+\frac{{\mathrm{i}}\gamma}{2p}\right)\right|^2 \left|\gamma\left(1+\frac{{\mathrm{i}}\gamma}{2p}\right)\right|^{-2}r^{(n-1)/2 } \,{\mathrm{e}}^{-{\mathrm{i}}pr } \\ \times\sum\limits_{k=0}^\infty\frac{\gamma\left(\frac{n-1}2+k -\frac{{\mathrm{i}}\gamma}{2p}\right)\gamma(n-1)(2{\mathrm{i}}pr)^k } { \gamma\left(\frac{n-1}2-\frac{{\mathrm{i}}\gamma}{2p}\right ) \gamma(n-1+k)\gamma(k+1)}\left[\psi(k+1)+\psi(k+n+1)\right]\bigg\ } \\ + \gamma\,\mathrm{re}\bigg\{\left[\gamma(n-1)^{-2}\right](2p)^{n-3 } \left|\gamma\left(\frac{n-1}2+\frac{{\mathrm{i}}\gamma}{2p}\right)\right|^2 \left|\gamma\left(1+\frac{{\mathrm{i}}\gamma}{2p}\right)\right|^{-2}r^{(n-1)/2 } \,{\mathrm{e}}^{-{\mathrm{i}}pr } \\ \times\sum\limits_{k=0}^\infty\frac{\gamma\left(\frac{n-1}2+k -\frac{{\mathrm{i}}\gamma}{2p}\right)\gamma(n-1)(2{\mathrm{i}}pr)^k } { \gamma\left(\frac{n-1}2-\frac{{\mathrm{i}}\gamma}{2p}\right)\gamma(n-1+k)\gamma(k+1 ) } \sum\limits_{s=1}^{\frac{n-1}2+k-1}\left(s-\frac{{\mathrm{i}}\gamma}{2p}\right)^{-1 } \bigg\ } \\ + \mathrm{re}\bigg\{(n-2)^{-1}(2{\mathrm{i}}p)^{n-2}r^{(n-1)/2}\,{\mathrm{e}}^{-{\mathrm{i}}pr } \sum\limits_{q=0}^{n-3}\frac{\gamma\left(\frac{3-n}2+q-\frac{{\mathrm{i}}\gamma } { 2p}\right)\gamma(3-n)(2{\mathrm{i}}pr)^{q+2-n } } { \gamma\left(\frac{3-n}2-\frac{{\mathrm{i}}\gamma}{2p}\right)\gamma(q+3-n)\gamma(q+1 ) } \bigg\ } \ , , \label{eq : g - tilde - raw}\end{gathered}\ ] ] valid for any odd @xmath268 and where the sum over @xmath269 in the fourth line is defined to give zero for @xmath270 and @xmath271 . in order to check our relations and to get a better understanding of the values for the causal range , we consider some explicit examples numerically . by cutting off the singular parts of the potential ( _ i.e. _ , the coulomb potential and the angular momentum term for @xmath272 ) at very small distances , it is a simple task to numerically solve the radial schrdinger equation in configuration space . from the radial wave functions one can extract the coulomb - modified phase shifts by looking for a zero at some large ( _ i.e. _ , much larger than the range @xmath12 of the short - range potential ) distance , @xmath273 and then calculating @xmath274 for the simplest case of a local step potential , @xmath275 one can of course also obtain the phase shift directly by matching the wave functions at @xmath276 . the effective range parameters are then obtained by repeating the calculation for several ( small ) momenta and fitting eq . to the results . in order to test eq . directly one needs the wave function to calculate the integral @xmath277 ^ 2 = \lim_{p\to0}\int_0^r{\mathrm{d}}r'\left[w_\ell^{(p)}(r')\right]^2 \,.\ ] ] even if we do not actually take the limit @xmath278 but rather just insert some small @xmath279 ( in units of an arbitrary inverse length scale ) , we find that the relation @xmath280 ^ 2 \ , , \label{eq : rc - bc - approx}\ ] ] is typically fulfilled to better than one percent accuracy for the simple step potential defined above . for illustration , in the following we choose units where the radial distance is measured in @xmath281 . the potential range is set to @xmath282 and its strength is measured in @xmath283 . , where @xmath284 is used for the unit conversion . ] furthermore , the reduced mass and coulomb parameter are set to the values for the proton proton system , _ i.e. _ , @xmath285 and @xmath286 . .,scaledwidth=62.0% ] . the thin lines show the corresponding scattering lengths.,scaledwidth=62.0% ] in figs . [ fig : rc - step - rep - gpp ] and [ fig : rc - step - attr - gpp ] we show the results ( for @xmath190 ) for both repulsive and attractive step potentials . quite interestingly , the @xmath63 causal range for the repulsive potential stays at zero ( meaning that one could reproduce the underlying values of the scattering length and the effective range even with a contact interaction ) until a potential strength of about @xmath287 . for higher partial waves the causal range takes a nonzero value for much weaker potentials , but the rise is less steep . in general , it is remarkable that the causal range is typically considerably smaller than the actual potential range ( @xmath288 ) . for attractive potentials the causal range grows much faster as the potential strength ( now negative ) increases . in contrast to what one might expect , no special features are seen in the causal ranges as the potential becomes strong enough to support a new bound state close to threshold , _ i.e. _ , when there is a pole in the scattering length parameter . to conclude this section , we show the general dependence of the causal range on both the scattering length and the effective range , which has the advantage of not depending on a certain model potential . for illustration , we again measure distances in @xmath281 and set the coulomb parameter to the value of the proton proton system . in figs . [ fig:3drc - gpp-0 ] and [ fig:3drc - gpp-1 ] we show the results for @xmath63 and @xmath246 . for negative effective range @xmath289 , the causal range stays essentially zero . for positive effective range , it increases as the absolute value of @xmath290 becomes larger . if one gradually turns off the coulomb interaction by letting @xmath291 , the @xmath246 plot stays almost unchanged , whereas the @xmath63 result remains qualitatively the same , but with a much steeper rise in the quadrant where @xmath292 and @xmath293 . and @xmath63 in dependence of @xmath294 and @xmath295 , both measured in @xmath281.,scaledwidth=80.0% ] and @xmath246 in dependence of @xmath296 and @xmath297 , measured in @xmath298 and @xmath299 , respectively.,scaledwidth=80.0% ] in this appendix , we demonstrate the equivalence of our anc relation to the one derived by sparenberg _ et al . _ @xcite . their anc relation , translated to our notational convention , reads @xmath300 the tilde on the @xmath301 is there to indicate that sparenberg _ et al . _ use the convention for the coulomb - modified effective range expansion as in refs . @xcite , which , as noted at the end of sec . [ sec : ere ] , differs from the one used in this paper by an overall factor @xmath302^{2}$ ] . combining equations in ref . @xcite and again matching to our notation , one finds that @xmath303 = -\frac1{\tilde{a}^c_\ell}+\frac12\tilde{r}^c_\ell\,p^2 + \cdots \,.\ ] ] note that this is just eq . with the prefactors combined . more explicitly , we have @xmath304 the final step of the derivation by sparenberg _ et al . _ eliminates the effective range in favor of the scattering length . without invoking this final step their relation reads @xmath305 where we are now using our convention for the effective range expansion . with the definition of @xmath306 from eq . and the values for @xmath307 from table [ tab : cg2g0 ] one sees that at least for @xmath241 this is exactly equivalent to our eq . with the @xmath308 set to zero . in order to prove the equivalence for arbitrary @xmath77 one would need a general expression for @xmath307 . this , in turn , requires knowledge of the constant terms in @xmath189(r)$ ] for arbitrary @xmath77 . it would thus probably be more interesting to turn the argument around , _ i.e. _ , take the equivalence for granted and derive from it a general expression for the constant terms in the wronskians . the only additional ingredient one would need for this procedure is a general series expansion for the meijer @xmath187-functions that arise from the integral of @xmath309 . e. p. wigner , phys . * 98 * ( 1955 ) 145 . d. r. phillips and t. d. cohen , phys . b * 390 * ( 1997 ) 7 [ nucl - th/9607048 ] . m. pavon valderrama and e. ruiz arriola , phys . c * 74 * ( 2006 ) 054001 [ nucl - th/0506047 ] . a. calle cordon and e. ruiz arriola , phys . c * 81 * ( 2010 ) 044002 [ arxiv:0905.4933 [ nucl - th ] ] . a. c. cordon and e. ruiz arriola phys . rev . a * 81 * ( 2010 ) 044701 [ arxiv:0912.1714 [ cond-mat.other ] ] . w . hammer and d. lee , b * 681 * ( 2009 ) 500 [ arxiv:0907.1763 [ nucl - th ] ] . hammer and d. lee , annals phys . * 325 * ( 2010 ) 2212 [ arxiv:1002.4603 [ nucl - th ] ] . s. elhatisari and d. lee , eur . j. a * 48 * ( 2012 ) 110 [ arxiv:1206.1207 [ nucl - th ] ] . m. v. zhukov , b. v. danilin , d. v. fedorov , j. m. bang , i. j. thompson , and j. s. vaagen , phys . * 231 * ( 1993 ) 151 . k. riisager , rev . phys . * 66 * ( 1994 ) 1105 . c. a. bertulani , h .- w . hammer , and u. van kolck , nucl . a * 712 * ( 2002 ) 37 [ nucl - th/0205063 ] . f. bedaque , h .- w . hammer , and u. van kolck , phys . lett . b * 569 * ( 2003 ) 159 [ nucl - th/0304007 ] . r. higa , h .- w . hammer , and u. van kolck , nucl . a * 809 * ( 2008 ) 171 [ arxiv:0802.3426 [ nucl - th ] ] . .- w . hammer and d. r. phillips , nucl . phys . a * 865 * ( 2011 ) 17 [ arxiv:1103.1087 [ nucl - th ] ] . g. rupak and r. higa , phys . * 106 * ( 2011 ) 222501 [ arxiv:1101.0207 [ nucl - th ] ] . l. fernando , r. higa , and g. rupak , arxiv:1109.1876 [ nucl - th ] . d. l. canham and h .- w . hammer , eur . j. a * 37 * ( 2008 ) 367 [ arxiv:0807.3258 [ nucl - th ] ] . d. l. canham and h .- w . hammer , nucl . a * 836 * ( 2010 ) 275 [ arxiv:0911.3238 [ nucl - th ] ] . j. rotureau and u. van kolck , arxiv:1201.3351 [ nucl - th ] . g. p. lepage , hep - ph/0506330 . g. p. lepage , nucl - th/9706029 . sparenberg , p. capel , and d. baye , phys . c * 81 * ( 2010 ) 011601 [ arxiv:0907.5166 [ nucl - th ] ] . f. l. yost , j. a. wheeler , and g. breit , phys . * 49 * ( 1936 ) 174 . s. a. afzal , a. a. z. ahmad , and s. ali , rev . mod . phys . * 41 * ( 1969 ) 247 . k. a. scaldeferri , d. r. phillips , c. w. kao and t. d. cohen , phys . c * 56 * ( 1997 ) 679 [ nucl - th/9610049 ] . v. orlov and yu . p. orevkov , phys . * 69 * ( 2006 ) 828 . s. knig , d. lee and h .- w . hammer , annals phys . * 327 * ( 2012 ) 1450 [ arxiv:1109.4577 [ hep - lat ] ] . c. r. brune , w. h. geist , r. w. kavanagh , and a. k. d. veal , phys . * 83 * ( 1999 ) 4025 . g. wallerstein _ et al . . mod . * 69 * ( 1997 ) 995 . l. buchmann , r. e. azuma , c. a. barnes , j. humblet , and k. langanke , phys . c * 54 * ( 1996 ) 393 . p. tischhauser _ , phys . c * 79 * ( 2009 ) 055803 . s. adhikari and c. basu , phys . b * 704 * ( 2011 ) 308 . a. belhout _ et al . _ , nucl . phys . a * 793 * ( 2007 ) 178 .

the concentration of empirical measures is studied for dependent data , whose joint distribution satisfies poincar - type or logarithmic sobolev inequalities . the general concentration results are then applied to spectral empirical distribution functions associated with high - dimensional random matrices .

1011.6165

let @xmath0 be a random vector in @xmath1 with distribution @xmath2 . we study rates of approximation of the average marginal distribution function @xmath3 by the empirical distribution function @xmath4 we shall measure the distance between @xmath5 and @xmath6 by means of the ( uniform ) kolmogorov metric @xmath7 , as well as by means of the @xmath8-metric @xmath9 the latter , also called the kantorovich rubinstein distance , may be interpreted as the minimal cost needed to transport the empirical measure @xmath6 to @xmath5 with cost function @xmath10 ( the price paid to transport the point @xmath11 to the point @xmath12 ) . the classical example is the case where all @xmath13 s are independent and identically distributed ( i.i.d . ) , that is , when @xmath2 represents a product measure on @xmath1 with equal marginals , say , @xmath5 . if it has no atoms , the distributions of the random variables @xmath14 are weakly convergent to the kolmogorov law . moreover , by the dvoretzky kiefer wolfowitz theorem , the r.v.s @xmath15 are uniformly sub - gaussian and , in particular , @xmath16 up to a universal factor @xmath17 ( @xcite ; cf . @xcite for history and sharp bounds ) . this result , together with the related invariance principle , has a number of extensions to the case of dependent observations , mainly in terms of mixing conditions imposed on a stationary process ; see , for example , @xcite . on the other hand , the observations @xmath18 may also be generated by non - trivial functions of independent random variables . of particular importance are random symmetric matrices @xmath19 , @xmath20 , with i.i.d . entries above and on the diagonal . arranging their eigenvalues @xmath21 in increasing order , we arrive at the spectral empirical measures @xmath6 . in this case , the mean @xmath22 also depends on @xmath23 and converges to the semicircle law under appropriate moment assumptions on @xmath24 ( cf . , for example , @xcite ) . the example of matrices strongly motivates the study of deviations of @xmath6 from the mean @xmath5 under general analytical hypotheses on the joint distribution of the observations , such as poincar or logarithmic sobolev inequalities . a probability measure @xmath2 on @xmath1 is said to satisfy a _ poincar - type _ or _ spectral gap inequality with constant _ @xmath25 ( @xmath26 ) if , for any bounded smooth function @xmath27 on @xmath1 with gradient @xmath28 , @xmath29 in this case , we write @xmath30 for short . similarly , @xmath2 satisfies a logarithmic sobolev inequality with constant @xmath25 and we write @xmath31 if , for all bounded smooth @xmath27 , @xmath32 here , as usual , @xmath33 stands for the variance of @xmath27 and @xmath34 denotes the entropy of @xmath35 under the measure @xmath2 . it is well known that lsi(@xmath25 ) implies pi(@xmath25 ) . these hypotheses are crucial in the study of concentration of the spectral empirical distributions , especially of the linear functionals @xmath36 with individual smooth @xmath37 on the line ; see , for example , the results by guionnet and zeitouni @xcite , chatterjee and bose @xcite , davidson and szarek @xcite and ledoux @xcite . a remarkable feature of this approach to spectral analysis is that no specific knowledge about the non - explicit mapping from a random matrix to its spectral empirical measure is required . instead , one may use general lipschitz properties only , which are satisfied by this mapping . as for the general ( not necessarily matrix ) scheme , we shall only require the hypotheses ( [ for1.1 ] ) or ( [ for1.2 ] ) . in particular , we derive the following from ( [ for1.1 ] ) . [ teo1.1 ] under @xmath30 on @xmath1 @xmath38 , @xmath39 where @xmath40 and @xmath17 is an absolute constant . note that the poincar - type inequality ( [ for1.1 ] ) is invariant under shifts of the measure @xmath2 , while the left - hand side of ( [ for1.3 ] ) is not . this is why the bound on the right - hand side of ( [ for1.3 ] ) should also depend on the means of the observations . in terms of the ordered statistics @xmath41 of the random vector @xmath0 , there is a general two - sided estimate for the mean of the kantorovich rubinstein distance : @xmath42 ( see remarks at the end of section [ sec4 ] ) . hence , under the conditions of theorem [ teo1.1 ] , one may control the local fluctuations of @xmath43 ( on average ) , which typically deviate from their mean by not more than @xmath44 . under a stronger hypothesis , such as ( [ for1.2 ] ) , one can obtain more information about the fluctuations of @xmath45 for individual points @xmath11 and thus get some control of the kolmogorov distance . similarly to the bound ( [ for1.3 ] ) , such fluctuations will , on average , be shown to be at most @xmath46 in the sense that @xmath47 , where @xmath48 is the lipschitz seminorm of @xmath5 ( see proposition [ pro6.3 ] ) . as for the kolmogorov distance , we prove the following theorem . [ teo1.2 ] assume that @xmath5 has a density , bounded by a number @xmath48 . under @xmath31 , for any @xmath49 , @xmath50 in particular , @xmath51 where @xmath52 and @xmath17 are positive absolute constants . in both cases , the stated bounds are of order @xmath53 up to a @xmath54 term with respect to the dimension @xmath23 . thus , they are not as sharp as in the classical i.i.d . case . indeed , our assumptions are much weaker and may naturally lead to weaker conclusions . let us look at two examples illustrating the bounds obtained in the cases that essentially differ from the i.i.d . case . [ ex1 ] let @xmath13 be independent and uniformly distributed in the intervals @xmath55 , @xmath56 . their joint distribution is a product measure , satisfying ( [ for1.1 ] ) and ( [ for1.2 ] ) with some absolute constant @xmath57 . clearly , @xmath5 is the uniform distribution in @xmath58 so @xmath59 and @xmath60 is of order @xmath61 . as is easy to see , @xmath62 is also of order @xmath61 , that is , the bound ( [ for1.6 ] ) is sharp up to a @xmath63 term . also , since @xmath64 is of order @xmath23 , both sides of ( [ for1.3 ] ) are of order 1 . in particular , this shows that the quantity @xmath64 can not be removed from ( [ for1.3 ] ) . [ ex2 ] let all @xmath65 , where @xmath66 is uniformly distributed in @xmath67 $ ] . note that all random variables are identically distributed with @xmath68 . the joint distribution @xmath2 represents a uniform distribution on the main diagonal of the cube @xmath67^n$ ] , so it satisfies ( [ for1.1 ] ) and ( [ for1.2 ] ) with @xmath69 , where @xmath52 is absolute . in this case , @xmath5 is a uniform distribution on @xmath67 $ ] , so @xmath70 and @xmath60 is of order 1 . hence , both sides of ( [ for1.6 ] ) are of order 1 . next , we restrict the above statements to the empirical spectral measures @xmath6 of the @xmath23 eigenvalues @xmath21 of a random symmetric matrix @xmath19 , @xmath71 , with independent entries above and on the diagonal ( @xmath72 ) . assume that @xmath73 and @xmath74 so that the means @xmath22 converge to the semicircle law @xmath75 with mean zero and variance one . the boundedness of moments of @xmath24 of any order will be guaranteed by ( [ for1.1 ] ) . [ teo1.3 ] if the distributions of the @xmath24 s satisfy the poincar - type inequality @xmath30 on the real line , then @xmath76 where @xmath17 is an absolute constant . moreover , under @xmath31 , @xmath77 by the convexity of the distance , we always have @xmath78 . in some random matrix models , the kolmogorov distance @xmath79 is known to tend to zero at rate at most @xmath80 . for instance , it is true when the distributions of the @xmath81 s have a non - trivial gaussian component ( see @xcite ) . hence , if , additionally , @xmath31 is satisfied , then we get that for any @xmath82 , @xmath83 it is unknown whether this bound is optimal . note , however , that in the case of gaussian @xmath81 , the distance @xmath79 is known to be of order @xmath84 @xcite . therefore , @xmath85 which is a slight improvement of a bound obtained in @xcite . in fact , as was recently shown in @xcite , we have @xmath86 in the presence of the @xmath30-hypothesis . hence , the bound ( [ for1.9 ] ) always holds under @xmath31 with constants depending only on @xmath57 . it seems natural to try to relax the @xmath31-hypothesis in ( [ for1.8 ] ) and ( [ for1.9 ] ) to @xmath30 . in this context , let us mention a result of chatterjee and bose @xcite , who used fourier transforms to derive from @xmath30 a similar bound , @xmath87 as for ( [ for1.7 ] ) , let us return to the two - sided bound ( [ for1.4 ] ) which holds with @xmath88 by the convention that the eigenvalues are listed in increasing order . the asymptotic behavior of distributions of @xmath13 with fixed or varying indices has been studied by many authors , especially in the standard gaussian case . in particular , if @xmath89 is fixed , while @xmath23 grows , @xmath90 converges in distribution to ( a variant of ) the tracy widom law so the @xmath91 are of order @xmath92 . this property still holds when @xmath24 are symmetric and have sub - gaussian tails ; see @xcite and @xcite for the history and related results . although this rate is consistent with the bound ( [ for1.7 ] ) , the main contribution in the normalized sum ( [ for1.4 ] ) is due to the intermediate terms ( in the bulk ) and their rate might be different . it was shown by gustavsson @xcite for the gue model that if @xmath93 , then @xmath13 is asymptotically normal with variance of order @xmath94 . hence , it is not surprising that @xmath95 , see @xcite . the paper is organized as follows . in section [ sec2 ] , we collect a few direct applications of the poincar - type inequality to linear functionals of empirical measures . they are used in section [ sec3 ] to complete the proof of theorem [ teo1.1 ] . in the next section , we discuss deviations of @xmath96 from its mean . in section [ sec5 ] , we turn to logarithmic sobolev inequalities . here , we shall adapt infimum - convolution operators to empirical measures and apply a result of @xcite on the relationship between infimum - convolution and log - sobolev inequalities . in section [ sec6 ] , we illustrate this approach in the problem of dispersion of the values of the empirical distribution functions at a fixed point . in section [ sec7 ] , we derive bounds on the uniform distance similar to ( [ for1.5 ] ) and ( [ for1.6 ] ) and give a somewhat more general form of theorem [ teo1.2 ] . in section [ sec8 ] , we apply the previous results to high - dimensional random matrices to prove theorem [ teo1.3 ] and obtain some refinements . finally , since the hypotheses ( [ for1.1 ] ) and ( [ for1.2 ] ) play a crucial role in this investigation , we collect in the appendix a few results on sufficient conditions for a measure to satisfy pi and lsi . we assume that the random variables @xmath18 have a joint distribution @xmath2 on @xmath1 , satisfying the poincar - type inequality ( [ for1.1 ] ) . for a bounded smooth function @xmath37 on the real line , we apply it to @xmath97 where @xmath6 is the empirical measure , defined for the ` observations ' @xmath98 . since @xmath99 we obtain an integro - differential inequality , which may viewed as an empirical poincar - type inequality for the measure @xmath2 . [ pro2.1 ] under @xmath30 , for any smooth @xmath5-integrable function @xmath37 on @xmath100 such that @xmath101 belongs to @xmath102 , we have @xmath103 recall that @xmath22 denotes the mean of the empirical measures . the inequality continues to hold for all locally lipschitz functions with the modulus of the derivative , understood in the generalized sense , that is , @xmath104 . as long as @xmath105 is finite , @xmath106 is also finite and ( [ for2.3 ] ) holds . the latter may be extended to all @xmath107-spaces by applying the following general lemma . [ lem2.2 ] under @xmath30 , any lipschitz function @xmath27 on @xmath1 has a finite exponential moment : if @xmath108 and @xmath109 , then @xmath110 moreover , for any locally lipschitz @xmath27 on @xmath1 with @xmath111-mean zero , @xmath112 more precisely , if @xmath113 is in @xmath114 , then so is @xmath27 and ( [ for2.5 ] ) holds true with the standard notation @xmath115 and @xmath116 for @xmath114-norms . the property of being locally lipschitz means that the function @xmath27 has a finite lipschitz seminorm on every compact subset of @xmath1 . in the concentration context , a variant of the first part of the lemma was first established by gromov and milman in @xcite and independently in dimension @xmath117 by borovkov and utev @xcite . here , we follow @xcite , proposition 4.1 , to state ( [ for2.4 ] ) . the second inequality , ( [ for2.5 ] ) , may be derived by similar arguments ; see also @xcite , theorem 4.1 , for an extension to the case of poincar - type inequalities with weight . now , for functions @xmath118 as in ( [ for2.1 ] ) , in view of ( [ for2.2 ] ) , we may write @xmath119 so that @xmath120 , we obtain the following proposition . [ pro2.3 ] under @xmath30 , for any smooth function @xmath37 on @xmath100 such that @xmath101 belongs to @xmath121 , @xmath122 , @xmath123 in addition , if @xmath124 , for all @xmath125 , @xmath126 the empirical poincar - type inequality ( [ for2.3 ] ) can be rewritten equivalently if we integrate by parts the first integral as @xmath127 at this step , it is safe to assume that @xmath37 is continuously differentiable and is constant near @xmath128 and @xmath129 . replacing @xmath101 with @xmath37 , we arrive at @xmath130 for any continuous , compactly supported function @xmath37 on the line . in other words , the integral operator @xmath131 with a ( positive definite ) kernel @xmath132 is continuous and defined on a dense subset of @xmath133 , taking values in @xmath134 . it has the operator norm @xmath135 , so it may be continuously extended to the space @xmath102 without a change of the norm . in the following , we will use a particular case of ( [ for2.6 ] ) . [ cor2.4 ] under @xmath30 , whenever @xmath136 , we have @xmath137 we shall now study the concentration properties of empirical measures @xmath6 around their mean @xmath5 based on poincar - type inequalities . in particular , we shall prove theorem [ teo1.1 ] , which provides a bound on the mean of the kantorovich rubinstein distance @xmath9 note that it is homogeneous of order 1 with respect to the random vector @xmath0 . we first need a general observation . [ lem3.1 ] given distribution functions @xmath5 and @xmath75 , for all real @xmath136 and a natural number @xmath138 , @xmath139 where @xmath140 . let @xmath141 denote the collection of those indices @xmath142 such that in the @xmath142th subinterval @xmath143 , the function @xmath144 does not change sign . let @xmath145 denote the collection of the remaining indices . then , for @xmath146 , @xmath147 in the other case @xmath148 , since @xmath149 changes sign on @xmath150 , we may write @xmath151 where , in the last step , @xmath5 and @xmath75 are treated as probability measures . hence , in this case , @xmath152 . combining the two bounds and using @xmath153 , we get that @xmath154 as the proof shows , the lemma may be extended to an arbitrary partition @xmath155 as follows : @xmath156 let us now apply the lemma to the space @xmath157 satisfying a poincar - type inequality . consider the partition of the interval @xmath158 $ ] with @xmath143 , as in lemma [ lem3.1 ] . by corollary [ cor2.4 ] , @xmath159 by cauchy s inequality , @xmath160 , hence , @xmath161 now , let us rewrite the right - hand side as @xmath162 with parameter @xmath163 and optimize it over @xmath138 . on the half - axis @xmath164 , introduce the function @xmath165 ( @xmath166 ) . it has derivative @xmath167 , therefore @xmath168 is decreasing on @xmath169 $ ] and is increasing on @xmath170 , where @xmath171 . hence , if @xmath172 , we have @xmath173 if @xmath174 , then the argmin lies in @xmath175 . choose @xmath176 + 1 = [ ( 2c)^{2/3 } ] + 1 $ ] so that @xmath177 and @xmath178 . hence , we get @xmath179 thus , in both cases , @xmath180 and we arrive at the following corollary . [ cor3.2 ] under @xmath30 , for all @xmath136 , @xmath181.\ ] ] the next step is to extend the above inequality to the whole real line . here , we shall use the exponential integrability of the measure @xmath5 . proof of theorem [ teo1.1 ] recall that the measure @xmath2 is controlled by using two independent parameters : the constant @xmath25 and @xmath64 , defined by @xmath182 one may assume , without loss of generality , that @xmath183 for all @xmath184 . lemma 2.2 with @xmath185 , @xmath186 and chebyshev s inequality give , for all @xmath125 , @xmath187 therefore , whenever @xmath188 , @xmath189 averaging over all @xmath89 s , we obtain similar bounds for the measure @xmath5 , that is , @xmath190 and @xmath191 . after integration , we get @xmath192 using @xmath193 so that @xmath194 , we get that @xmath195 and similarly for the half - axis @xmath196 . combining this bound with corollary [ cor3.2 ] , with @xmath158 = [ -h , h]$ ] , we obtain that , for all @xmath188 , @xmath197 + 12 \sigma \mathrm{e}^{-(h - a\sigma)/\sigma}.\ ] ] substituting @xmath198 with arbitrary @xmath199 , we get that @xmath200.\ ] ] finally , the choice @xmath201 leads to the desired estimate @xmath202 in addition to the upper bound on the mean of the kantorovich rubinstein distance @xmath96 , one may wonder how to bound large deviations of @xmath96 above the mean . to this end , the following general observation may be helpful . [ lem4.1 ] for all points @xmath203 , @xmath204 in @xmath1 , we have @xmath205 where @xmath206 , @xmath207 . in other words , the canonical map @xmath208 from @xmath1 to the space of all probability measures on the line , which assigns to each point an associated empirical measure , has a lipschitz seminorm @xmath209 with respect to the kantorovich rubinstein distance . as usual , the euclidean space @xmath1 is equipped with the euclidean metric @xmath210 denote by @xmath211 the collection of all ( borel ) probability measures on the real line with finite first moment . the kantorovich rubinstein distance in @xmath211 may equivalently be defined ( cf . @xcite ) by @xmath212 where the infimum is taken over all ( borel ) probability measures @xmath213 on @xmath214 with marginal distributions @xmath75 and @xmath215 . in case of empirical measures @xmath216 , @xmath217 , associated to the points @xmath218 , let @xmath219 be the discrete measure on the pairs @xmath220 , @xmath221 , with point masses @xmath61 . therefore , by cauchy s inequality , @xmath222 this proves lemma [ lem4.1 ] . thus , the map @xmath223 has the lipschitz seminorm @xmath224 . as a consequence , given a probability measure @xmath2 on @xmath1 , this map transports many potential properties of @xmath2 , such as concentration , to the space @xmath211 , equipped with the borel probability measure @xmath225 . note that it is supported on the set of all probability measures with at most @xmath23 atoms . in particular , if @xmath2 satisfies a concentration inequality of the form @xmath226 in the class of all borel sets @xmath64 in @xmath1 with measure @xmath227 ( where @xmath228 denotes an open euclidean @xmath229-neighborhood of @xmath64 ) , then @xmath230 satisfies a similar ( and in fact stronger ) property @xmath231 in the class of all borel sets @xmath232 in @xmath211 with measure @xmath233 ( with respect to the kantorovich rubinstein distance ) . in other words , an optimal so - called concentration function @xmath234 in ( [ for4.1 ] ) for the measure @xmath2 is related to the concentration function of @xmath230 by @xmath235 now , in general , the concentration function has a simple functional description as @xmath236 where the sup is taken over all lipschitz functions @xmath27 on @xmath1 with @xmath109 and where @xmath237 stands for a median of @xmath27 under @xmath2 . ( actually , this holds for abstract metric spaces . ) the concentration function may therefore be controlled by poincar - type inequalities in terms of @xmath25 ( the gromov milman theorem ) . indeed , since the quantity @xmath238 is translation invariant , one may assume that @xmath27 has mean zero . by lemma [ lem2.2 ] with @xmath186 , we get @xmath239 , provided that @xmath240 , which means that any median of @xmath27 satisfies @xmath241 . therefore , again by lemma [ lem2.2 ] , for any @xmath240 , @xmath242 so that @xmath243 the latter also automatically holds in the interval @xmath244 . in fact , by a more careful application of the poincar - type inequality , the concentration bound ( [ for4.3 ] ) may be further improved to @xmath245 ( see @xcite ) , but this is not crucial for our purposes . thus , combining ( [ for4.2 ] ) with ( [ for4.3 ] ) , we may conclude that under pi(@xmath25 ) , @xmath246 now , in the setting of theorem [ teo1.1 ] , consider on @xmath211 the distance function @xmath247 . it is lipschitz ( with lipschitz seminorm @xmath117 ) and has the mean @xmath248 , where @xmath249 . hence , @xmath250 under the measure @xmath230 and for any @xmath125 , @xmath251 we can summarize as follows . [ pro4.2 ] if a random vector @xmath252 in @xmath1 , @xmath72 , has distribution satisfying a poincar - type inequality with constant @xmath25 , then , for all @xmath253 , @xmath254 where @xmath255 and where @xmath17 is an absolute constant . bounds such as ( [ for4.4 ] ) may be used to prove that the convergence holds almost surely at a certain rate . here is a simple example , corresponding to non - varying values of the poincar constants . ( one should properly modify the conclusion when applying this to the matrix scheme ; see section [ sec7 ] . ) let @xmath256 be a random sequence such that for each @xmath23 , @xmath0 has distribution on @xmath1 satisfying @xmath30 with some common @xmath57 . [ cor4.3 ] if @xmath257 , then @xmath258 with probability @xmath117 . note , however , that in the scheme of sequences such as in corollary [ cor4.3 ] , the mean distribution function @xmath22 might also depend on @xmath23 . by a similar contraction argument , the upper bound ( [ for4.4 ] ) may be sharpened , when the distribution of @xmath0 satisfies a logarithmic sobolev inequality . we turn to this type of ( stronger ) hypothesis in the next section . let @xmath259 be a metric space and let @xmath260 be a non - negative continuous function on the product space @xmath261 . given borel probability measures @xmath75 and @xmath215 on @xmath259 , the generalized kantorovich rubinstein or wasserstein ` distance ' with cost function @xmath262 is defined by @xmath263 where the infimum is taken over all probability measures @xmath213 on @xmath261 with marginal distributions @xmath75 and @xmath215 . in the case of the real line @xmath264 with cost function of the form @xmath265 , where @xmath266 is convex , this quantity has a simple description , @xmath267 in terms of the inverse distribution functions @xmath268 ; see , for example , @xcite and @xcite , theorem 2 . if @xmath269 , then we also have the @xmath8-representation for @xmath270 , which we use from the very beginning as our definition . moreover , for arbitrary discrete measures @xmath271 and @xmath272 , as in lemma [ lem4.1 ] , the expression ( [ for4.5 ] ) is reduced to @xmath273 where we assume that @xmath274 and @xmath275 . now , for an arbitrary random vector @xmath276 in @xmath1 , consider the ordered statistics @xmath41 . equation ( [ for4.6 ] ) then yields @xmath277 where @xmath278 are ordered statistics generated by an independent copy of @xmath279 and where @xmath280 are independent copies of the ( random ) empirical measures @xmath6 associated with @xmath279 . by the triangle inequality for the metric @xmath281 , we have @xmath282 it is applied with the mean distribution function @xmath22 . on the other hand , any function of the form @xmath283 is convex on the convex set @xmath211 , so , by jensen s inequality , @xmath284 . the two bounds give @xmath285 by a similar argument , @xmath286 combining ( [ for4.8 ] ) and ( [ for4.9 ] ) and recalling ( [ for4.7 ] ) , we arrive at the two - sided estimate @xmath287 which is exactly the inequality ( [ for1.4 ] ) mentioned in the . similar two - sided estimates also hold for other cost functions in the wasserstein distance . as before , let @xmath0 be a random vector in @xmath288 with joint distribution @xmath2 . similarly to proposition [ pro2.1 ] , now using a log - sobolev inequality for @xmath2 , we arrive at the following ` empirical ' log - sobolev inequality . [ pro5.1 ] under @xmath31 , for any bounded , smooth function @xmath37 on @xmath100 , @xmath289 \leq \frac{2\sigma^2}{n } \int f'^2\,\mathrm{d}f.\ ] ] in analogy with poincar - type inequalities , one may also develop refined applications to the rate of growth of moments and to large deviations of various functionals of empirical measures . in particular , we have the following proposition . [ pro5.2 ] under @xmath31 , for any smooth function @xmath37 on @xmath100 such that @xmath101 belongs to @xmath121 , @xmath122 , @xmath290 in addition , if @xmath124 , then , for all @xmath125 , @xmath291 the proof of the second bound , ( [ for5.2 ] ) , which was already noticed in @xcite in the context of random matrices , follows the standard herbst s argument ; see @xcite and @xcite . the first family of moment inequalities , ( [ for5.1 ] ) , can be sharpened by one inequality on the laplace transform , such as @xmath292 the proof is immediate , by @xcite , theorem 1.2 . however , a major weak point in both poincar and log - sobolev inequalities , including their direct consequences , as in proposition [ pro5.2 ] , is that they may not be applied to indicator and other non - smooth functions . in particular , we can not estimate directly at fixed points the variance @xmath293 or other similar quantities like the higher moments of @xmath294 . therefore , we need another family of analytic inequalities . fortunately , the so - called infimum - convolution operator and associated relations concerning arbitrary measurable functions perfectly fit our purposes . moreover , some of the important relations hold true and may be controlled in terms of the constant involved in the logarithmic sobolev inequalities . let us now turn to the important concept of infimum- and supremum - convolution inequalities . they were proposed in 1991 by maurey @xcite as a functional approach to some of talagrand s concentration results concerning product measures . given a parameter @xmath295 and a real - valued function @xmath27 on @xmath1 ( possibly taking the values @xmath296 ) , put @xmath297 , \\ p_t g(x ) & = & \sup_{y \in\mathbf{r}^n } \biggl [ g(y ) - \frac{1}{2 t } \|x - y\| ^2\biggr].\end{aligned}\ ] ] @xmath298 and @xmath299 then represent , respectively , the infimum- and supremum - convolution of @xmath27 with cost function being the normalized square of the euclidean norm in @xmath1 . by definition , one puts @xmath300 . for basic definitions and basic properties of the infimum- and supremum - convolution operators , we refer the reader to @xcite and @xcite , mentioning just some of them here . these operators are dually related by the property that for any functions @xmath37 and @xmath27 on @xmath1 , @xmath301 . clearly , @xmath302 . thus , in many statements , it is sufficient to consider only one of these operators . the basic semigroup property of both operators is that for any @xmath27 on @xmath1 and @xmath303 , @xmath304 for any function @xmath27 and @xmath305 , the function @xmath299 is always lower semicontinuous , while @xmath298 is upper semicontinuous . if @xmath27 is bounded , then @xmath299 and @xmath298 are bounded and have finite lipschitz seminorms . in particular , both are differentiable almost everywhere . given a bounded function @xmath27 and @xmath305 , for almost all @xmath306 , the functions @xmath307 and @xmath308 are differentiable at @xmath309 and @xmath310 in other words , the operator @xmath311 appears as the generator for the semigroup @xmath312 , while @xmath313 appears as the generator for @xmath314 . as a result , @xmath315 represents the solution to the hamilton jacobi equation @xmath316 with initial condition @xmath317 . below , we separately formulate a principal result of @xcite which relates logarithmic sobolev inequalities to supremum- and infimum - convolution operators . [ lem5.3 ] let @xmath2 be a probability measure on @xmath1 satisfying @xmath31 . for any @xmath2-integrable borel - measurable function @xmath27 on @xmath1 , we have @xmath318 and , equivalently , @xmath319 alternatively , for further applications to empirical measures , one could start from the infimum - convolution inequalities ( [ for5.3 ] ) and ( [ for5.4 ] ) , taking them as the main hypothesis on the measure @xmath2 . they take an intermediate position between poincar and logarithmic sobolev inequalities . however , logarithmic sobolev inequalities have been much better studied , with a variety of sufficient conditions having been derived . now , as in section [ sec2 ] , we apply the relations ( [ for5.3 ] ) and ( [ for5.4 ] ) to functions @xmath320 , where @xmath6 is the empirical measure defined for ` observations ' @xmath321 . by the very definition , for any @xmath305 , @xmath322 \\ & = & \frac{1}{n } \sup_{y_1,\ldots , y_n \in\mathbf{r } } \sum_{i=1}^n \biggl [ f(y_i ) - \frac{1}{2t / n } |x_i - y_i|^2\biggr ] = \int p_{t / n } f\,\mathrm{d}f_n.\end{aligned}\ ] ] similarly , @xmath323 . therefore , after integration with respect to @xmath2 and using the identity @xmath324 , we arrive at corresponding empirical supremum- and infimum - convolution inequalities , as follows . [ pro5.4 ] under @xmath31 , for any @xmath5-integrable borel - measurable function @xmath37 on @xmath100 and for any @xmath305 , @xmath325\,\mathrm{d}f , \\ \label{for5.6}\log\mathbf{e}\mathrm{e}^{t ( \int f\,\mathrm{d}f - \int f\,\mathrm{d}f_n ) } & \leq & t \int[f - q_{t\sigma^2/n } f ] \,\mathrm{d}f.\end{aligned}\ ] ] note that the second inequality may be derived from the first by changing @xmath37 to @xmath326 . in this section , we develop a few direct applications of proposition [ pro5.4 ] to the behavior of empirical distribution functions @xmath327 at a fixed point . such functionals are linear , that is , of the form @xmath328 , corresponding to the indicator function of the half - axis @xmath329 $ ] . when @xmath37 is smooth , proposition [ pro5.2 ] tells us that the deviations of @xmath330 are of order @xmath331 . in the general non - smooth case , the infimum- and supremum - convolution operators @xmath332 and @xmath333 behave differently for small values of @xmath309 and this results in a different rate of fluctuation for @xmath334 . to see this , let @xmath329 $ ] . in this case , the functions @xmath332 and @xmath333 may easily be computed explicitly , but we do not lose much by using the obvious bounds @xmath335 } \leq q_t f \leq p_t f \leq1_{(-\infty , x + \sqrt{2 t } ] } .\ ] ] therefore , ( [ for5.5 ] ) and ( [ for5.6 ] ) yield the following proposition . [ pro6.1 ] under @xmath31 , for any @xmath336 and @xmath305 , with @xmath337 , @xmath338 these estimates may be used to sharpen corollary [ cor3.2 ] and therefore to recover theorem [ teo1.1 ] ( under the stronger hypothesis on the joint distribution @xmath2 , however ) . indeed , for any @xmath305 , @xmath339 taking the logarithm and applying jensen s inequality , we arrive at @xmath340 now , just integrate this inequality over an arbitrary interval @xmath341 , @xmath136 , and use the general relation @xmath342 to obtain that @xmath343 optimization over @xmath309 leads to an improved version of corollary [ cor3.2 ] . [ cor6.2 ] under @xmath31 , for all @xmath136 , @xmath344 note that in both cases of proposition [ pro6.1 ] , for any @xmath345 , @xmath346 hence , the local behavior of the distribution function @xmath5 near a given point @xmath11 turns out to be responsible for the large deviation behavior at this point of the empirical distribution function @xmath6 around its mean . for a quantitative statement , assume that @xmath5 has a finite lipschitz constant @xmath347 , so it is absolutely continuous with respect to lebesgue measure on the real line and has a density , bounded by @xmath48 . it follows from ( [ for6.1 ] ) , with @xmath348 and @xmath349 , that @xmath350 where @xmath351 . by chebyshev s inequality , for any @xmath352 , @xmath353 similarly , @xmath354 . therefore , @xmath355 . changing the variable , we are finished . recall that we use the quantity @xmath356 [ pro6.3 ] assume that @xmath5 has a density , bounded by a number @xmath48 . under @xmath31 , for any @xmath336 and @xmath49 , @xmath357 in particular , with some absolute constant @xmath17 , we have @xmath358 note that ( [ for6.4 ] ) is consistent with the estimate of theorem [ teo1.1 ] . to derive similar bounds on the uniform ( kolmogorov ) distance @xmath359 ( which we discuss in the next section ) , it is better to split the bound ( [ for6.3 ] ) into the two parts , @xmath360 which were obtained in the last step of the proof of ( [ for6.3 ] ) . however , since one might not know whether @xmath5 is lipschitz or how it behaves locally , and since one might want to approximate this measure itself by some canonical distribution @xmath75 , it is reasonable to provide a more general statement . by proposition [ pro6.1 ] , for any @xmath199 , @xmath361 and , similarly , @xmath362 repeating the preceding argument with the random variable @xmath363 and then interchanging @xmath6 and @xmath75 , we get a more general version of proposition [ pro6.3 ] . [ pro6.4 ] under @xmath31 , for any distribution function @xmath75 with finite lipschitz seminorm @xmath364 , for any @xmath336 and @xmath49 , @xmath365 where @xmath366 . in particular , up to some absolute constant @xmath17 , @xmath367 let us stress that in all of these applications of proposition [ pro5.4 ] , only the indicator functions @xmath368}$ ] were used . one may therefore try to get more information about deviations of the empirical distributions @xmath6 from the mean @xmath5 by applying the basic bounds ( [ for5.5 ] ) and ( [ for5.6 ] ) with different ( non - smooth ) functions @xmath37 . for example , of considerable interest is the so - called local regime , where one tries to estimate the number @xmath369 of observations inside a small interval @xmath370 $ ] and to take into account the size of the increment @xmath371 . in case of i.i.d . observations , this may done using various tools ; already , the formula @xmath372 suggests that when @xmath5 is lipschitz , @xmath373 has small oscillations for small @xmath374 ( where @xmath6 and @xmath5 are treated as measures ) . however , the infimum- and supremum - convolution operators @xmath332 and @xmath333 do not provide such information . indeed , for the indicator function @xmath375 , by ( [ for5.5 ] ) , we only have , similarly to proposition [ pro6.1 ] , that @xmath376,\ ] ] where @xmath305 and @xmath337 . here , when @xmath229 is fixed and @xmath377 , the right - hand side is not vanishing , in contrast with the i.i.d . case . this also shows that standard chaining arguments , such as dudley s entropy bound or more delicate majorizing measure techniques ( described , e.g. , in @xcite ) , do not properly work through the infimum - convolution approach . nevertheless , the above estimate is still effective for @xmath374 of order @xmath229 , so we can control deviations of @xmath378 relative to @xmath379 when the intervals are not too small . this can be done with the arguments used in the proof of proposition [ pro6.3 ] or , alternatively ( although with worse absolute constants ) , one can use the inequality ( [ for6.3 ] ) , by applying it to the points @xmath11 and @xmath380 . this immediately gives that @xmath381 changing variables , one may rewrite the above in terms of @xmath382 as @xmath383 with @xmath384 . note that the right - hand side is small only when @xmath385 , which is of order @xmath53 with respect to the number of observations . this can further be generalized if we apply proposition [ pro6.4 ] . [ cor6.5 ] let @xmath75 be a distribution function with density @xmath386 bounded by a number @xmath48 . under @xmath31 , for any @xmath387 and any interval @xmath141 of length @xmath388 , @xmath389 where @xmath366 and @xmath166 is an absolute constant . hence , if @xmath390 and @xmath391 is large , then , with high probability , we have that @xmath392 . as before , let @xmath6 denote the empirical measure associated with observations @xmath321 and @xmath22 their mean with respect to a given probability measure @xmath2 on @xmath1 . in this section , we derive uniform bounds on @xmath393 , based on proposition [ pro6.3 ] , and thus prove theorem [ teo1.2 ] . for applications to the matrix scheme , we shall also replace @xmath5 , which may be difficult to determine , by the well - behaving limit law @xmath75 ( with the argument relying on proposition [ pro6.4 ] ) . let the random variables @xmath18 have joint distribution @xmath2 , satisfying @xmath31 , and assume that @xmath5 has a finite lipschitz seminorm @xmath347 . define @xmath356 proof of theorem [ teo1.2 ] we use the inequalities ( [ for6.5 ] ) and ( [ for6.6 ] ) to derive an upper bound on @xmath394 . ( for the sake of extension of theorem [ teo1.2 ] to theorem [ teo7.1 ] below , we relax the argument and do not assume that @xmath5 is continuous . ) so , fix @xmath352 and an integer @xmath177 . one can always pick up points @xmath395 with the property that @xmath396 note that @xmath397 and @xmath398 . it then follows from ( [ for6.5 ] ) that @xmath399 \geq\beta r\bigr\ } \leq(n-1 ) \mathrm{e}^{-2r^3/27}\ ] ] and , similarly , by ( [ for6.6 ] ) , @xmath400 \geq \beta r\bigr\ } \leq(n-1 ) \mathrm{e}^{-2r^3/27}.\ ] ] hence , for the random variable @xmath401 , \max_{1 \leq i \leq n } [ f(x_{i-1 } ) - f_n(x_{i-1})]\bigr\},\ ] ] we have that @xmath402 now , take any point @xmath336 different from all of the @xmath403 s and select @xmath89 from @xmath404 such that @xmath405 . then , by ( [ for7.1 ] ) , @xmath406 + [ f(x_i- ) - f(x_{i-1 } ) ] \leq \xi_n + \frac{1}{n}.\end{aligned}\ ] ] similarly , @xmath407 + [ f(x_{i-1 } ) - f_n(x_{i-1 } ) ] \leq \xi_n + \frac{1}{n}.\end{aligned}\ ] ] therefore , @xmath408 , which also extends by continuity from the right to all points @xmath403 . thus , @xmath409 and , by ( [ for7.2 ] ) , @xmath410 note that this also holds automatically in the case @xmath411 . choose @xmath412 + 1 $ ] . we then have @xmath413 and get @xmath414 finally , changing @xmath415 into @xmath416 , we arrive at the bound ( [ for1.5 ] ) of theorem [ teo1.2 ] , @xmath417 and so the constant @xmath418 . it remains to derive the bound on the mean @xmath62 . given @xmath419 , we can write , using ( [ for7.3 ] ) , @xmath420 \\[-8pt ] \nonumber & \leq & r_0 + \frac{4}{r_0 } \exp\biggl\{-\frac{2}{27 } \biggl(\frac{r_0}{\beta } \biggr)^3\biggr\}.\end{aligned}\ ] ] this bound also holds for @xmath421 . first , assume that @xmath422 and choose @xmath423 . then , for the last term in ( [ for7.4 ] ) , we have @xmath424 with some constant @xmath232 satisfying @xmath425 . for example , we can take @xmath426 and then , by ( [ for7.4 ] ) , we have @xmath427 as for the values @xmath428 , simple calculations show that the right - hand side of ( [ for7.5 ] ) is greater than 1 , so the inequality ( [ for1.6 ] ) is fulfilled with @xmath429 . theorem [ teo1.2 ] is therefore proved . if @xmath430 in theorem 1.2 were of order 1 , then @xmath62 would be of order at most @xmath431 . note , however , that under @xmath30 , and if all @xmath432 , the quantity @xmath430 is separated from zero and , more precisely , @xmath433 . indeed , by hensley s theorem in dimension @xmath117 @xcite , in the class of all probability densities @xmath434 on the line , the expression @xmath435 is minimized for the uniform distribution on symmetric intervals and is therefore bounded from below by @xmath436 . since @xmath5 is lipschitz , it has a density @xmath437 with @xmath438 . on the other hand , it follows from the poincar - type inequality that @xmath439 . averaging over all @xmath89 s , we get @xmath440 , so @xmath441 . with similar arguments based on proposition [ pro6.4 ] , we also obtain the following generalization of theorem [ teo1.2 ] . [ teo7.1 ] assume that @xmath18 have a distribution on @xmath1 satisfying @xmath31 . let @xmath75 be a distribution function with finite lipschitz seminorm @xmath48 . then , for all @xmath352 , @xmath442 where @xmath366 . in particular , @xmath443 it remains unclear whether or not one can involve the i.i.d . case in the scheme of poincar or logarithmic sobolev inequalities to recover the rate @xmath444 for @xmath62 , even if some further natural assumptions are imposed ( which are necessary , as we know from examples [ ex1 ] and [ ex2 ] ) . in particular , one may assume that the quantities @xmath48 and @xmath57 are of order @xmath117 , and that @xmath68 , @xmath445 . the question is the following : under , say , @xmath31 , is it true that @xmath446 with some absolute @xmath17 ? or at least @xmath447 for individual points ? we shall now apply the bounds obtained in theorems [ teo1.1 ] and [ teo7.1 ] , to the case of the spectral empirical distributions . let @xmath448 be a family of independent random variables on some probability space with mean @xmath449 and variance @xmath450 . put @xmath451 for @xmath452 and introduce a symmetric @xmath453 random matrix , @xmath454 arrange its ( real random ) eigenvalues in increasing order : @xmath21 . as before , we associate with particular values @xmath455 an empirical ( spectral ) measure @xmath6 with mean ( expected ) measure @xmath22 . an important point in this scheme is that the joint distribution @xmath2 of the spectral values , as a probability measure on @xmath1 , represents the image of the joint distribution of @xmath24 s under a lipschitz map @xmath208 with lipschitz seminorm @xmath456 . more precisely , by the hoffman wielandt theorem with respect to the hilbert schmidt norm , we have @xmath457 for any collections @xmath458 and @xmath459 with eigenvalues @xmath0 , @xmath460 , respectively . this is a well - known fact ( @xcite , page 165 ) which may be used in concentration problems ; see , for example , @xcite . in particular ( see proposition [ pra1 ] in the ) , if the distributions of @xmath24 s satisfy a one - dimensional poincar - type inequality with common constant @xmath25 , then @xmath2 satisfies a poincar - type inequality with an asymptotically much better constant @xmath461 . according to theorem [ teo1.1 ] , @xmath462 where @xmath17 is an absolute constant and @xmath463 . since @xmath464 is of order at most @xmath57 , @xmath465 is at most @xmath466 and we arrive at the bound ( [ for1.7 ] ) in theorem [ teo1.3 ] : @xmath467 now , let us explain the second statement of theorem [ teo1.3 ] for the case where the @xmath24 s satisfy a logarithmic sobolev inequality with a common constant @xmath25 , in addition to the normalizing conditions @xmath73 , @xmath468 ( which implies that @xmath469 ) . let @xmath75 denote the standard semicircle law with variance 1 , that is , with density @xmath470 , @xmath471 . in this case , the lipschitz seminorm is @xmath472 . also , @xmath473 for some absolute @xmath474 . therefore , applying theorem [ teo7.1 ] and using @xmath469 , we arrive at the bound ( [ for1.8 ] ) : @xmath475 thus , theorem [ teo1.3 ] is proved . for individual points that are close to the end - points @xmath476 of the supporting interval of the semicircle law , we may get improved bounds in comparison with ( [ for8.1 ] ) . namely , by proposition [ pro6.1 ] ( and repeating the argument from the proof of the inequality of corollary [ cor6.2 ] ) , for all @xmath305 , @xmath477 where @xmath478 . taking the logarithm and applying jensen s inequality , we arrive at @xmath479 using the lipschitz property of @xmath75 only ( that is , @xmath480 ) would yield the previous bound , such as the one in the estimate ( [ for6.7 ] ) of proposition [ pro6.4 ] , @xmath481 however , the real size of increments @xmath482 with respect to the parameter @xmath229 essentially depends on the point @xmath11 . to be more careful in the analysis of the right - hand side of ( [ for8.2 ] ) , we may use the following elementary calculus bound , whose proof we omit . [ lem8.1 ] @xmath483 for all @xmath336 and @xmath125 . since @xmath75 is concentrated on the interval @xmath484 $ ] , for @xmath485 , we have a simple bound @xmath486 . as a result , one may derive from ( [ for8.2 ] ) an improved variant of ( [ for8.3 ] ) . in particular , if @xmath485 , then @xmath487 the more general statement for all @xmath336 is given by the following result . [ teo8.2 ] let @xmath24 @xmath488 be independent and satisfy a logarithmic sobolev inequality with constant @xmath25 , with @xmath73 and @xmath450 . for all @xmath336 , @xmath489,\ ] ] where @xmath17 is an absolute constant . a similar uniform bound may also be shown to hold for @xmath490 @xmath491 ) and @xmath492 @xmath493 ) . note that in comparison with ( [ for8.3 ] ) , there is an improvement for the points @xmath11 at distance not more than @xmath494 from @xmath495 . proof of theorem [ teo8.2 ] according to the bound ( [ for8.2 ] ) and lemma [ lem8.1 ] , for any @xmath125 , we may write @xmath496 , where @xmath497 , @xmath498 . we shall now estimate the minimum of this function . write @xmath499 with parameter @xmath500 to be specified later on . if @xmath501 , then @xmath502 note that the requirement on @xmath386 is equivalent to @xmath503 . thus , we set @xmath504 and take @xmath505 . since @xmath506 , we get @xmath507 . hence , we may apply ( [ for8.5 ] ) . using @xmath508 and @xmath509 , we finally get that @xmath510 this is the desired expression in square brackets in ( [ for8.4 ] ) and theorem [ teo8.2 ] follows . finally , let us comment on the meaning of the general corollary [ cor6.5 ] in the matrix model above . to every interval @xmath141 on the real line , we associate the number @xmath369 of eigenvalues @xmath13 inside it . again , corollary [ cor6.5 ] may be applied to the standard semicircle law @xmath75 with density @xmath386 , in which case @xmath511 . this gives that , under @xmath31 , imposed on the entries @xmath24 , for any @xmath387 and any interval @xmath141 of length @xmath388 , we have that @xmath512 with probability at least @xmath513 . as we have already mentioned , under @xmath30 , one can show that @xmath514 ( @xcite , theorem [ teo1.1 ] ) . therefore , ( [ for8.6 ] ) holds true with high probability , provided that @xmath515 ) . such properties have been intensively studied in recent years in connection with the universality problem . in particular , it is shown in @xcite and @xcite that the restriction ( [ for8.7 ] ) may be weakened to @xmath516 under the assumption that the intervals @xmath141 are contained in @xmath517 $ ] , @xmath518 , that is , ` in the bulk ' . here we recall some facts about poincar - type and log - sobolev inequalities . while lemmas [ lem2.2 ] and [ lem5.3 ] list some of their consequences , one might wonder which measures actually satisfy these analytic inequalities . many interesting examples can be constructed with the help of the following elementary proposition . [ pra1 ] let @xmath519 be probability measures on @xmath100 satisfying @xmath30 ( resp . , @xmath31 ) . the image @xmath2 of the product measure @xmath520 under any map @xmath521 with finite lipschitz seminorm satisfies @xmath522 ( resp . , @xmath523 ) . on the real line , disregarding the problem of optimal constants , poincar - type inequalities may be reduced to hardy - type inequalities with weights . necessary and sufficient conditions for a measure on the positive half - axis to satisfy a hardy - type inequality with general weights were found in the late 1950s in the work of kac and krein @xcite . we refer the interested reader to @xcite and @xcite for a full characterization and an account of the history ; here , we just recall the principal result ( see also @xcite ) . let @xmath2 be a probability measure on the line with median @xmath524 , that is , @xmath525 and @xmath526 . define the quantities @xmath527 , \\ a_1(\mu ) & = & \sup_{x > m } \biggl [ \mu(x,+\infty ) \int_x^{+\infty } \frac{\mathrm{d}t}{p_\mu(t ) } \biggr],\end{aligned}\ ] ] where @xmath528 denotes the density of the absolutely continuous component of @xmath2 ( with respect to lebesgue measure ) and where we set @xmath529 ( resp . , @xmath530 ) if @xmath531 ( resp . , @xmath532 ) . we then have the following proposition . [ proa2 ] the measure @xmath2 on @xmath100 satisfies @xmath533 with some finite constant if and only if both @xmath534 and @xmath535 are finite . moreover , the optimal value of @xmath25 satisfies @xmath536 where @xmath537 and @xmath538 are positive universal constants . necessarily , @xmath2 must have a non - trivial absolutely continuous part with density which is positive almost everywhere on the supporting interval . for example , the two - sided exponential measure @xmath539 , with density @xmath540 , satisfies pi(@xmath25 ) with @xmath541 . therefore , any lipschitz transform @xmath542 of @xmath539 satisfies pi(@xmath25 ) with @xmath543 . the latter property may be expressed analytically in terms of the reciprocal to the so - called isoperimetric constant , @xmath544 where @xmath545 $ ] denotes the distribution function of @xmath2 and @xmath528 the density of its absolutely continuous component . namely , as a variant of the mazya cheeger theorem , we have that pi(@xmath546 is valid with @xmath547 ; see @xcite , theorem 1.3 . to roughly describe the class of measures in the case , where @xmath2 is absolutely continuous and has a positive , continuous well - behaving density , one may note that @xmath548 and the poincar constant are finite , provided that the measure has a finite exponential moment . in particular , any probability measure with a logarithmically concave density satisfies pi(@xmath546 with a finite @xmath57 ; see @xcite . as for logarithmic sobolev inequalities , we have a similar picture , where the standard gaussian measure represents a basic example and plays a similar role as the two - sided exponential distribution for poincar - type inequalities . a full description on the real line , resembling proposition [ proa2 ] , was given in @xcite . namely , for one - dimensional probability measure @xmath2 , with previous notation , we define the quantities @xmath549 , \\ b_1(\mu ) & = & \sup_{x > m } \biggl [ \mu(x,+\infty ) \log\frac{1}{\mu(x,+\infty ) } \int_x^{+\infty } \frac{\mathrm{d}t}{p_\mu(t ) } \biggr].\end{aligned}\ ] ] we then have the following proposition . [ proa3 ] the measure @xmath2 on @xmath100 satisfies @xmath550 with some finite constant if and only if @xmath551 and @xmath552 are finite . moreover , the optimal value of @xmath25 satisfies @xmath553 where @xmath537 and @xmath538 are positive universal constants . in particular , if @xmath2 has a log - concave density , then @xmath550 is satisfied with some finite constant if and only if @xmath2 has sub - gaussian tails . this research was supported by nsf grants dms-07 - 06866 and sfb 701 . bobkov , s.g . and gtze , f. ( 2008 ) . hardy - type inequalities via riccati and sturm liouville equations . in _ sobolev spaces in mathematics , i _ ( v. mazya , ed . ) . * 8 * 6986 . new york : springer .

we investigate the physical properties of the inner gaseous disks of the three , hot , herbig b2e stars , hd76534 , hd114981 and hd216629 , by modelling cfht - espadons spectra using non - lte radiative transfer codes . we assume that the emission lines are produced in a circumstellar disk heated solely by the photospheric radiation from the central star in order to test if the optical and near - ir emission lines can be reproduced without invoking magnetospheric accretion . the inner gaseous disk density was assumed to follow a simple power - law in the equatorial plane , and we searched for models that could reproduce observed lines of hi ( h and h ) , hei , caii and feii . for the three stars , good matches were found for all emission line profiles individually ; however , no density model based on a single power - law was able to reproduce all of the observed emission lines . among the single power - law models , the one with the gas density varying as in the equatorial plane of a 25r ( au ) disk did the best overall job of representing the optical emission lines of the three stars . this model implies a mass for the h-emitting portion of the inner gaseous disk of . we conclude that the optical emission line spectra of these hbe stars can be qualitatively reproduced by aau , geometrically thin , circumstellar disk of negligible mass compared to the central star in keplerian rotation and radiative equilibrium .

1701.06721

herbig aebe ( haebe ) stars are a or b - type , pre - main sequence stars that show optical emission lines , particularly the hydrogen balmer lines , in their spectra and an infrared ( ir ) excess in their spectral energy distributions ( seds ) ; both can be attributed to the presence of circumstellar dust and gas @xcite . haebe systems inherit a large amount of angular momentum from the star formation phase , leading the dust and gas to settle into an equatorial disk . haebe disks have been the subject of many studies at variety of wavelengths due to their role in star and protoplanetary disk formation ( e.g. @xcite ) . as the disks around haebe stars are likely sites of planet formation ( e.g. @xcite ) , they are of great interest in understanding fundamental disk and planet formation physics . haebe stars are the higher mass counterparts to the t tauri stars , which show balmer emission lines in their spectra and an excess in their infrared seds , interpreted as radiation from gas and dust in the form of a circumstellar disk . ] and represent an important link between the formation of low - mass and high - mass stars . the disks of haebe stars can be divided into an inner gaseous disk and an outer dusty disk . due to the high effective temperatures of a and b stars , dust is expected to evaporate near the star . the dust - free , inner gaseous region can be further divided into an inner region of atomic gas and a more distant disk of molecular gas . emission lines in the optical and near - ir are generally used to study the atomic gas , while near - ir molecular emission lines are used to study the outer molecular gas . as the disk temperature decreases with distance from the star , beyond a certain distance dust is stable and settles into a disk , along with the cool gas . this cool , outer region can also be studied using millimeter and sub - millimeter interferometry ( see the review by @xcite and references therein ) . several studies using co overtone emission and millimeter interferometric observation have shown that the gas in these dusty , cool regions follows keplerian rotation @xcite . there are various differences amongst the haebe stars , as well as between haebe stars and t tauri stars . while t tauri stars are believed to accrete magnetospherically ( see @xcite and @xcite for reviews ) , there is little direct observational evidence of accretion amongst the early - type hbe stars . furthermore , if accretion is occurring , it is not clear if the accretion is controlled by large scale magnetic field @xcite . several studies , such as those of @xcite , @xcite , and @xcite , have noted differences in the polarization of herbig stars : cooler hae stars show intrinsic h@xmath0 line polarization , while the hotter hbe stars show line depolarization attributable to dilution by disk h@xmath0 emission . the authors suggest that this difference reflects different underlying types of accretion , in that linear polarization indicates magnetospheric accretion , while depolarization may be explained by disk accretion . recent studies have noted additional differences between hae and hbe stars in the lines of h@xmath0 , h@xmath1 , br@xmath7 , hei @xmath8 , hei @xmath9 , feii @xmath10 , and [ oi]@xmath11 , again suggesting different accretion modes @xcite . @xcite studied the uv balmer excess of the haebe stars and showed that magnetospheric models seem unable to reproduce the excess for early - type hbe stars . all of these lines of evidence suggest that an alternative to magnetospheric accretion may be required for the hbe stars and that studying the region close the stellar surface , i.e. the inner gaseous disk , offers a way to shed some light on the mechanism of how and when material from the disk accretes onto hbe stars . another class of early - type emission - line stars are the classical be stars . classical be stars are b - type , main - sequence stars that show , or have once shown , one or more balmer emission lines in their spectrum ; they can also display modest amounts of emission in metal lines and and posses an ir excess . however , the ir excess is not attributable to dust but rather to free - free emission from an ionized gas in a circumstellar decretion disk @xcite . suggestive similarities exist between hbe stars and the classical be stars in observed features such as polarization level , photometric variability , ir excesses ( which can be small for some hbe stars ) , and h@xmath0 emission line equivalent width ( ew hereafter ) distribution @xcite . in particular , @xcite noted that group iii hbe stars , those with very small ir excesses , closely resemble ( and are sometimes confused with ! ) the classical be stars . these similarities are likely due to the underlying similarity in the gaseous , circumstellar disks themselves , in spite of the obvious difference of origin ( i.e. accretion versus decretion ) . this similarity suggests that using computational methods that have been successfully applied to the disks of the classical be stars may help to better understand and model the physical conditions in the inner gaseous disks of the hbe stars . in this work , we focus on early - type hbe stars as important differences exist between hae and hbe stars ( as stated above ) , including difficulties in applying magnetospheric accretion models to the early - type hbe stars . the computational modelling tools specifically made for classical be stars have been very successful in reproducing the observed data ( spectra , seds , interferometric visibilities ) , as well as providing a general understanding of the density and temperature structure of classical be star disks @xcite . to study the inner gaseous disk of hbe stars , optical and near - ir emission lines are key diagnostics . such lines trace not only the structure of the disk ( its temperature , density , and velocity field ) but also the processes taking place in this region . the strongest emission lines found in the spectra of hbe stars are the hydrogen balmer lines h@xmath0 and h@xmath1 . several other emission lines , such as the caii ir triplet and lines of feii and oi , are of moderate strength and appear in emission in the spectrum . h@xmath0 , caii and feii lines have been previously used as indicators of wind and chromospheric activity @xcite . in contrast , other studies suggest that emission lines such as h@xmath0 and the caii ir triplet are non - photospheric in origin and form in a region at some distance from the stellar surface @xcite . previous studies have utilized hydrogen br@xmath7 spectro - interferometry for several haebe stars with many different types of models , such as magnetospheric models , outflows , and/or keplerian disks ( see @xcite and reference within ) . for h@xmath0 , three hae stars , namely ab aur , hd 172981 , and hd 141529 @xcite , and one hbe star , mwc 361 ( or hd 200775 ) @xcite , have also been studied using spectro - interferometry . in this paper , we attempt to fit the spectral line profiles of three hbe stars using a circumstellar disk model in which no significant accretion is currently occurring and the disk radiates because it is heated and photoionized by the radiation from the star . this is the second paper in our series investigating inner gaseous disks around early - type hbe stars . @xcite , paper i hereafter , modelled optical emission lines of the hb2e star bd+651637 by comparing them to synthetic line profiles computed with the bedisk and beray non - lte circumstellar disk codes @xcite . acceptable matches to the strengths and shapes of all emission lines considered were found based on a geometrically thin disk with a simple , single power - law parametrization of its density structure . no additional heating of the disk was required beyond the photoionization energy input of the central star . gas densities of @xmath12 were required in the disk near the star , with the disk extending to @xmath13 stellar radii . however , distinct regions in the disk density parameter space were required to fit the hydrogen and metal lines , leading to the conclusion that the metal lines ( caii and feii ) require a slower density drop - off in the equatorial plane of the disk compared to the hydrogen balmer lines . in this current paper , three additional hb2e stars and their inner gaseous disks are investigated using the same numerical codes . section [ sec : stars ] provides background information on the three stars , discusses their fundamental parameters , and outlines the observations and data reduction procedures . modeling of the line profiles is discussed in section [ sec : modeling ] . the results are discussed in sections [ sec : results ] through [ sec : bd65 ] , and the paper concludes with key findings in section [ sec : conclusions ] . three hbe stars of spectral type b2 are considered in this study : hd76534 , hd114981 and hd216629 . the fundamental stellar parameters are adopted from @xcite and are given in table [ table1 ] . @xcite derived the effective temperatures by matching synthetic spectra generated by the ` tlusty ` and ` synspec ` codes @xcite to the observed spectra . the synthetic spectra were generated assuming @xmath14 for all stars . the synthetic spectra were rotationally broadened and used to determine the @xmath15 of the stars . using the stellar luminosities and effective temperatures , the stars were placed in the hr diagram and compared to evolutionary tracks calculated using the cesam stellar evolutionary code , version 2k @xcite , to determine the masses and radii . hd76534 a ( v * ou vel ) is a hbe star of spectral type b2e with a visual magnitude of 8.35 located at a distance of 870 pc . it is a member of the vela r2 association and illuminates the surrounding reflection nebula vdb 24 @xcite . this star is the brighter member of a visual binary . hd76534 was first classified as a hbe star by @xcite , and its small infrared excess led it to be classified as group iii by @xcite ; group iii hbe stars have small infrared excesses , similar to the classical be stars . large variations in the equivalent width ( ew ) of h@xmath0 , ranging from -14.3 to + 2.4 , have been reported @xcite . large scale variations are also seen in the observed polarization of hd76534 over time scales of @xmath161 month @xcite . hd114981 ( v * v958 cen ) is a 7.16 visual magnitude star at a distance of 550 pc @xcite . it is a b - type star that shows a small ir excess and has been suggested to be a classical be star . @xcite classified hd114981 as a b - type star with ir excess typical of vega - like stars . @xcite and @xcite classified hd114981 as haebe candidate with spectral type b5 due to h@xmath0 emission ; however @xcite re - classified it as type b2 . @xcite estimated that hd114981 is young with an age of @xmath17 myr . using the recent analysis , and the evidence presented in the analysis ( especially it s young age ) , hd114981 will be treated as a hb2e star . hd216629 ( v * il cep ) is a b2e star of visual magnitude 9.31 located at a distance of 720 pc @xcite . several studies , such as @xcite and @xcite , have associated hd216629 with the cep ob3 association ; however , @xcite argue against this based on the position of hd216629 in the h - r diagram . @xcite studied hd216629 with spectro - astrometry and found that it is a wide , double - lined , spectroscopic binary . @xcite also noted variation in hei lines from one observation to another and indicated that the binary companion might be causing the variations . double - peaked h@xmath0 emission has been observed with the ew ranging from -34.5 to + 4.5 @xcite . the ew of h@xmath1 has been reported to be + 3.4 @xcite . @xcite studied the balmer emission lines of hd216629 and failed to detect any depolarization . because of its binary nature , @xcite do not assign a mass or radius ; therefore , we have adopted average b2 values given by @xcite . the current observational data were obtained with the high - resolution espadons spectropolarimeter on the canadian - france - hawaii telescope ( cfht ) . this instrument covers the wavelength range 3700 to 10500 with a spectral resolution of 65,000 . the hd76534 spectrum , obtained in 2005 ( hjd2453423 ) , can be seen in figure [ fig : hd76534_spec ] . the peak snr per pixel is 221 at 7080 . the spectrum contains not only a strong h@xmath0 emission line but also weaker emission from h@xmath1 and , possibly , from the caii ir triplet . weak emission from several feii multiplets , common in the optical spectra of hbe stars , is also detected . the ews of all the measured line profiles can be found in table [ table2 ] . for hd114981 , two different spectra are available , one from 2005 ( hjd2453422 ) with a peak snr per pixel of 329 at 5150 , and one from 2006 ( hjd2453748 ) with a peak snr per pixel of 633 at 5150 . it is the 2006 spectrum that is shown in in figure [ fig : hd114981_spec ] . again , emission lines of h@xmath0 , h@xmath1 , possibly caii , and feii are detected . a comparison of the two spectra , taken approximately a year apart , can be seen in figure [ fig : hd114981_compare ] , and only small changes are seen . the h@xmath0 emission is a bit stronger in the 2006 observations , and the h@xmath1 line has shallower wings in the 2005 observations . for the feii lines , the 2005 profiles are slightly stronger . all emission lines show symmetric , doubly - peaked profiles . hei @xmath186678 is seen in absorption in both spectra and has a very broad profile . as the 2006 spectrum has a higher snr , we have chosen it for analysis . the ews of the 2006 line profiles for hd114981 can be found in table [ table2 ] . the spectrum for hd216629 was obtained in 2006 ( hjd2454077 ) at peak snr per pixel of 227 at 7080 and can be seen in figure [ fig : hd216629_spec ] . it shows prominent h@xmath0 , h@xmath1 , and caii ir triplet emission lines . two feii lines , @xmath19 and @xmath20 , are also present weakly in emission . h@xmath0 is singly - peaked , while h@xmath1 , the caii ir triplet , and the feii lines are doubly - peaked . h@xmath1 and the feii lines show a stronger red ( r ) component in their doubly - peaked profile compared to the blue ( v ) component . the ews of all the line profiles for hd216629 can also be found in table [ table2 ] . for hd76534 and hd216629 , previous observations have suggested variability @xcite ; the ew of h@xmath0 has ranged from -14.3 to + 2.4 for hd76534 and from -34.5 to + 4.5 for hd216629 . in the current espadons spectra , the h@xmath0 ew for hd76534 is -10.9and for hd216629 , -19.7 ; therefore , the current data set for hd7634 was obtained when the star was in relatively bright phase , while hd216629 was in a more typical phase of its overall variability . the spectra for all three stars were continuum normalized using iraf in order to be able to compare the observed line profiles to the synthetic spectra . the individual orders containing the emission lines of interest were separately normalized using various low order ( 2@xmath21 or 3@xmath22 ) polynomials , with either the ` legendre ' or ` cubic spline ' functions available in iraf . the three caii ir triplet lines ( @xmath188498 , @xmath188542 , @xmath188662 ) are blended with various high - n paschen lines that needed to be subtracted from the composite profiles in order to extract the unblended caii line profiles for analysis . in order to remove the contaminating paschen emission , the two adjacent , unblended paschen lines were averaged , and this average was subtracted from the composite line profile . for example , to deblend caii @xmath188542 from pa15 @xmath188543 , the average profile of pa14 @xmath188596 and pa17 @xmath188465 was subtracted from the composite line . this procedure was performed for all the three caii ir triplets lines for the three stars . an example of the subtraction process , and the result for all three stars , can be seen in figure [ fig : caiisubtraction_c4 ] for the caii @xmath188542 emission line . the subtraction process did not yield a significant caii line profile compared to the continuum noise for two of the three stars , hd76534 and hd114981 . therefore , caii was not considered in the analysis of these stars . the caii line profile for hd216629 , however , shows a strong , doubly - peaked line profile after subtraction , and this profile was retained in the subsequent analysis . to compute the synthetic line profiles used to model the observed emission lines , the non - lte radiative transfer codes bedisk @xcite and beray @xcite , were used . these codes assume an equatorial , non - accreting , gaseous disk in keplerian rotation surrounding the central star . the photoionizing stellar radiation field is the sole source of input energy , and the temperature distribution in the disk is found by enforcing radiative equilibrium . the disk is assumed to be axisymmetric about the star s rotation axis and the disk mid - plane . given a user - supplied disk density structure , bedisk computes the thermal structure of the disk by enforcing of radiative equilibrium in the gas assumed of solar composition at a large number of grid points in the disk , the net rates of heating and cooling due to nine abundant elements ( h , he , c , n , o , mg , si , ca and fe ) over many ionization stages are balanced to determine the radiative equilibrium temperatures . in addition to the temperature structure of the disk , bedisk calculates all of the atomic level populations required to compute the gas emissivity and opacity required for radiative transfer . the atomic level populations are found by solving the equations of statistical equilibrium ( for more details , see @xcite ) . line profiles , spectral energy distributions ( seds ) , and monochromatic images on the sky can be calculated using the beray code @xcite which solves the equation of radiative transfer along a series of rays ( @xmath23 ) that pass through the star+disk system directed at the observer . the disk thermal structure and level populations computed by bedisk are used for the radiative transfer calculations . no incident radiation is assumed for the rays that pass through the disk . for rays that terminate on the stellar surface , an appropriate doppler - shifted , photospheric , lte line profile was used for the boundary condition . thus , the synthetic line profiles produced by beray include contributions from both the stellar photosphere and the disk in a consistent way , and the resulting synthetic line profiles , normalized to the predicted continuum , can directly be compared with the observations . the assumed density structure for the disk was set by two free parameters , a base density @xmath24 and a power law index @xmath25 , in the equation @xmath26 here @xmath27 and @xmath28 are the cylindrical co - ordinates for the disk , @xmath29 is the stellar radius , and @xmath30 is the disk scale height . the disk is assumed to start at the stellar photosphere ( @xmath31 ) and extend out to a radius of r@xmath32 . the disk was assumed to be in vertical , hydrostatic equilibrium , and in this case , the scale height in equation ( [ eq : rho ] ) has the form @xmath33 + where @xmath34 + here , @xmath35 is the stellar mass , @xmath36 is an assumed , hydrostatic equilibrium temperature used _ solely _ to fix the vertical structure of the disk , , but this self - consistent treatment does not lead to large changes in the predicted emission line profiles @xcite . ] and @xmath37 is the mean molecular weight of the gas , taken to be 0.68 . the typical value of @xmath36 in classical be stars is @xmath38 @xcite . a thin disk ( @xmath39 ) is natural consequence of the assumption of vertical hydrostatic equilibrium and the typical radiative equilibrium temperatures found in the disk . however by varying the value of @xmath36 , the scale height of the disk can be changed and thicker disks can be produced . since it is unclear how thick or thin the disks of hbe stars are , both _ thin disk models _ , achieved by setting the @xmath36 to @xmath40 , and _ thick disk models _ , achieved by setting @xmath36 to @xmath41 , were considered . in addition to these two basic disk density models ( thick or thin ) , there is an additional parameter for the adopted disk microturbulence , @xmath42 . the value of microturbulence is set either to zero or the local sound speed in the disk , giving sonic turbulence in the latter case . the models with the microturbulence set to the local sound speed will be referred to as _ turbulent disk models_. when calculating the synthetic line profiles , an additional model parameter is added to the list : the viewing inclination ( @xmath43 ) of the disk , defined such that @xmath44 represents a pole - on star / face - on disk . thus , @xmath24 , @xmath25 , r@xmath32 , @xmath42 , and @xmath43 are all of the input parameters to be varied for the synthetic line profile calculations . the parameters defining the central b2v star , m@xmath3 , r@xmath3 , t@xmath45 , are assumed fixed . the bedisk and beray codes were specifically developed for the gaseous decretion disks of classical be stars . many studies ( see @xcite ) have successfully used the bedisk and beray codes to model be star gaseous disks by matching observables , such as the hydrogen balmer lines , feii lines , ir line fluxes , seds , and optical and near - ir interferometric visibilities . given the apparent similarities between hbe and classical be stars , these computational codes provide a good starting point towards understanding the inner , gaseous disks around hbe stars using the spectral line profile modelling . in the previous work of paper i , the spectrum of the hbe star bd+651637 was successfully modelled using the bedisk and beray codes . to model the observed line profiles of the three hbe stars considered in this work , a large grid of synthetic line profiles for disks surrounding a central b2 star were calculated using the parameters given in the table [ tab : star_param ] . the adopted b2 parameters fall within the uncertainty of the stellar parameters given in table [ table1 ] for all stars . a wide variety of disk parameters , listed in table [ tab : model_param ] , were explored . typical values of @xmath46 and @xmath25 for classical be stars fall in the range of @xmath47 to @xmath48 and 2 to 4 , respectively @xcite . however , for hbe stars , an expanded range of model parameters was considered in order to include more massive disks by calculating higher @xmath46 and/or lower @xmath25 . thus , the models were calculated with the base density parameter , @xmath24 , ranging from @xmath49 to @xmath50 , and the power - law index , @xmath25 , ranging from @xmath51 to @xmath52 . the size of the disk was chosen to be one of three values , r@xmath32=@xmath53 , @xmath54 or @xmath55r@xmath3 , corresponding to @xmath4 , @xmath56 , and @xmath57 au , respectively . all the synthetic line profiles were calculated at four viewing inclinations , which represent the centers of the first four bins of five equal probability bins in a random @xmath58 distribution : 18@xmath59 , 45@xmath59 , 60@xmath59 and 75@xmath59 . the fifth bin , @xmath60 , was not considered as there is no evidence of shell absorption , i.e. deep central reversals in the line profiles caused by disk absorption , in any of the spectra . to compare each synthetic profile in the computed library to an observed line profile , a figure - of - merit , @xmath61 , was computed . this figure - of - merit was defined as @xmath62 where @xmath63 is the observed relative ( continuum normalized ) flux , @xmath64 is the model relative flux , and the sum is over the @xmath65 wavelength points spanning the line . a range of small shifts to the observed wavelength scale were also considered , within the errors of the star s radial velocity , in order to best match the observed profile . the smallest value of @xmath61 was deemed to define the best - fit model for that feature , although the ten profiles with smallest @xmath61 values were visually inspected in order to verify a best match . in cases of asymmetric line profiles , i.e. v / r ( violet to red peak ) ratios other than 1 ( such as h@xmath1 for hd216629 ) , the match was made to only one of the peaks , depending on the shape , strength and width of the line profile . we will first discuss the best - fit models for each individual line of table [ table2 ] for each of the three stars . as will be shown , it is often the case that the best set of model parameters @xmath66 will differ between the various lines of the same star . to address this , we consider global fits , that is how well a disk density model based on a single set of these parameters can reproduce the entire emission line spectrum of each star , in section [ sec : globalfits_c4 ] , and section [ sec : uniquenessoffits ] discusses the uniqueness of the results . disk density parameters for all of the individual and global fits for all three stars are summarized in table [ tab : bestfitmodels_c4 ] . hei ( @xmath67 ) is a photospheric line and is not used to constrain the disk models except for hd216629 ( see section [ sec : linefits_216629 ] for more details . ) the best - fit , synthetic line profiles to the individual observed emission lines for hd76534 are shown in figure [ fig : hd76534_linefits ] . all the observed emission lines are reasonably well matched in strength , shape , and width , with the exception of the hei line , as discussed below . the model that best fits the h@xmath0 line profile is from a @xmath68 , thin disk model with a base density ( @xmath24 ) of @xmath69 and power - law index ( @xmath25 ) of 2.5 viewed at an inclination of @xmath70 . the synthetic line is somewhat narrower at the base when compared to the observed profile , indicating that either slightly more material is required closer to the star and/or the system is viewed at an inclination angle of somewhat more that @xmath71 . however , the fit was not refined along these lines . for h@xmath1 , the best - fit model profile is from a @xmath68 , thick disk model with base density of @xmath72 and power - law index of 3.0 , seen at the same inclination as h@xmath0 , @xmath70 . for the two feii lines , @xmath185169 and @xmath185317 , the model that fits both observed line profiles is from a 25r@xmath3 , thin disk with a base density of @xmath73 and power - law index of 3.0 , again seen at @xmath70 . except between the two feii lines , all of the optimal models for the individual lines differ . as noted before , these best fit models are those with the minimum value of @xmath61 . however , there is a range of other disk density models that fit each line profile nearly as well . the range of fitting models for each feature and the implications will be discussed in section [ sec : uniquenessoffits ] . as for hd76534 s hei absorption line , it presents a special problem . the observed line profile has a wide base in the wings and a narrow absorption core . photospheric synthetic profiles were found to well fit only either the wide wings or the narrow absorption core , depending upon the viewing inclination , with @xmath74 models matching the narrow core and @xmath70 models matching the broad wings . none of the models reproduced the observed line profile as whole . absorption in the line core by the disk seems an unlikely explanation as absorption is not seen in the cores of the hydrogen lines , and the viewing inclination of the system seems close to @xmath75 as based on the other lines . as hei @xmath76 is a transition in helium s singlet spin system , we also do not expect a diffuse , forbidden component to the line profile ( as is the case for some of the triplet system ) . it is important to note that other hei lines , such as @xmath77 , @xmath8 , and @xmath78 , show the same broad base and narrow core as seen in hei @xmath67 . we have explored variations in the adopted stellar parameters ; models were calculated for @xmath79 increased by 2000k ( in order to strengthen the helium lines ) , as well with higher and lower values of @xmath80 . none of these variations were able to improve the fit to the observed hei profile . finally , we investigated the possibility of binarity , postulating that the observed hei line is actually the sum of two stellar components of very similar spectral types . as mentioned earlier , hd76534 is already known to be a visual binary , although the separation is too large for this to be the origin of the addition spectrum . we did manage to produce an acceptable match to the hei line by combining the two photospheric profiles , i.e. a fit to the narrow core with a profile with a @xmath81 of 49 km / s and fit the wide base component of the line profile with a model with a @xmath81 of 222 km / s . the two line profiles were combined assuming an equal flux contribution from each component , and the composite line profile is illustrated in figure [ fig : hei_twomodels ] . thus , there is evidence that the hd76534 spectrum is actually a composite spectrum of two b stars of nearly the same spectral type . we note that we are unable to detect a systematic velocity offset between the two postulated hei components ; therefore , the binary orbit must be either very wide or the orbital plane is in the plane of the sky . despite this circumstantial evidence for a composite spectrum , we will retain the assumption that hd76534 s emission line spectrum is due to a single hbe star in the following analysis . of 222 km / s ( red ) is fit to the wide base and wings , and a photospheric line with @xmath15 of 49 km / s ( green ) is fit to the narrow absorption core . the sum of the two model lines is also shown ( black).,scaledwidth=50.0% ] from table [ tab : bestfitmodels_c4 ] , we see that no single model was able to fit all the emission line profiles for hd76534 . it is likely that a disk density distribution more complex than that of an axisymmetric , single power - law disk , as we have assumed , is necessary . to investigate what range of disk radii contribute to the formation of each line , the cumulative emission for each line was calculated and plotted against radius in the disk for each matching model . to calculate the cumulative disk emission , @xmath82 , a face - on ( @xmath83 ) synthetic image was produced by beray in the light of each emission line using the best - fit disk density model for that line . the intensity was integrated in wavelength over the total width of the line for each image and then this integrated over disk radius as @xmath84 here @xmath85 is the wavelength - integrated line intensity at distance @xmath27 , and @xmath29 is the stellar radius , assumed to be the inner edge of the disk . then @xmath86 can be plotted versus radius of the disk ( @xmath27 ) to determine the contribution to the total intensity of each line throughout the disk . figure [ fig : fvsr_hd76534linefits ] shows the cumulative emission @xmath82 for each individual line . it is important to remember that for this figure , the disk density model used to calculate the @xmath82 for each line is the individual , best fit model for that line given in table [ tab : bestfitmodels_c4 ] . as a result , for example , @xmath86 reaches one at 100r@xmath3 for h@xmath0 and h@xmath1 , but for feii , one is reached by 25r@xmath3 . while h@xmath1 requires a 100r@xmath3 disk to best reproduce the observed profile , most of the emission comes from the region close to the star . h@xmath0 also requires extended disk , but 90% of the emission is produced in the 25r@xmath3 disk near the star . both the feii lines are mostly produced within 5r@xmath3 from the stellar photosphere . for hd114981 , the best - fit model parameters , both individual and global , can be found in table [ tab : bestfitmodels_c4 ] , and the comparison between the model and observed profiles is shown in figure [ fig : hd114981_linefits ] . for this star , all of the observed lines were individually matched reasonably well in terms of strength and shape by synthetic library profiles . the disk density model that best fits the h@xmath0 line profile is a 50r@xmath3 , thick disk with base density ( @xmath24 ) of @xmath87 and power - law index ( @xmath25 ) of 2.0 seen at an inclination of @xmath88 . for h@xmath1 , the model that best fits the observed line was a 100r@xmath3 , thin disk with the base density of @xmath89 and power - law index of 2.5 , again with an @xmath88 inclination . for both balmer lines , the best - fit synthetic profiles show deeper central absorption compared to the observed lines . a change in the inclination angle , intermediate between 45@xmath59 and 60@xmath59 , might improve the fit , but this was not attempted . for the two feii lines , the model that fits both observed profiles well is a 25r@xmath3 , base density of @xmath90 and power - law index of 3.0 when seen at a @xmath70 inclination . the only difference between the fits is that feii @xmath19 requires a thin disk , while feii @xmath20 requires a thick disk . for hei , the photospheric model that best reproduces the observed line profile has a @xmath15 of 222 km / s . the hei line is matched very well in strength and width . the cumulative disk emission as a function of disk radius for is shown in figure [ fig : fvsr_hd114981linefits ] . again , it should be remembered that the individual disk density model for each line ( table [ tab : bestfitmodels_c4 ] ) has been used to compute @xmath86 . h@xmath0 requires region of 50r@xmath3 to reproduce the observed line profile , with most of the emission produced within 30r@xmath3 . h@xmath1 is produced throughout the 100r@xmath3 disk ; however , 90% of the emission comes from the inner 20r@xmath3 . finally , feii @xmath19 is formed over a region closer to the star , with most of the emission coming from within 5r@xmath3 ; feii @xmath20 is formed over a slightly larger range , and 90% of the emission comes from within 10r@xmath3 . the individual , best - fit models for the emission lines of hd216629 are illustrated in figure [ fig : hd216629_linefits ] , and the corresponding disk density parameters for each line can be found in table [ tab : bestfitmodels_c4 ] . unlike hd76534 and hd114981 , the caii infrared - triplet is detected in emission in the spectrum of hd216629 after the paschen line subtraction procedure , and the @xmath91 is available for analysis . as can be seen from figure [ fig : hd216629_linefits ] , the hydrogen and iron emission lines are fit reasonably well by the models , but the best caii ir triplet synthetic line , while fitting the peak emission , is significantly narrower than the observed line profile . the synthetic profile that provides the best fit to h@xmath0 was a 25r@xmath3 thin and turbulent disk with a base density ( @xmath92 ) of @xmath93 and power - law index ( @xmath25 ) of 2.5 seen at @xmath70 however , as can be seen in figure [ fig : hd216629_linefits ] , this model gives a small , double - peaked profile in the core , whereas h@xmath0 is observed to be singly - peaked , and the synthetic line is too narrow at the base near the continuum . the observed h@xmath1 profile shows asymmetric red and blue emission peaks , @xmath94 , superimposed on a wide absorption trough . for modelling , the weaker blue emission peak was chosen , and a good fit to both this peak and the absorption trough was found for a 25r@xmath3 , thin and turbulent disk with the base density of @xmath72 and power - law index of 2.5 , seen at an inclination of @xmath70 . fitting the red peak instead ( not shown in the figure ) leads to a larger base density , @xmath73 , and a larger power - law index , @xmath52 , but seen at an inclination of @xmath88 . the observed caii ir triplet ( @xmath91 ) emission line is very wide , with a full width of @xmath95 at the base of the profile from the apparent edge of one wing to the other . the best - fit synthetic model for this caii line is a 25r@xmath3 , thick and turbulent disk with a base density of @xmath96 and power - law index of 3.0 seen at an angle of @xmath88 . the model reproduces the height of the emission peaks , but fails to match the overall width and strength of the line . feii @xmath19 requires a 25r@xmath3 , thin disk with a base density of @xmath73 and power - law index of 3.0 seen at an inclination of @xmath88 . feii @xmath20 line requires the same disk density parameters , but for a thin and turbulent disk seen at a higher inclination of @xmath97 . the synthetic profile for feii @xmath19 is narrower than the observed profile but matches well in strength . the feii @xmath20 synthetic line profile matches well both in shape and strength . unlike hd76534 and hd114981 , the hei @xmath67 line of hd216629 required a disk model to fit the observed absorption line profile . the model that best fits the observed line is a 50r@xmath3 , thick and turbulent disk with a base density of @xmath98 and a power law index of 1.5 seen at an inclination of 75@xmath59 . the disk is able to produce absorption in the core even at @xmath97 due to the thick disk and low power - law index for this model . the synthetic line profile shows some central emission which can be attributed to a central , quasi - emission ( cqe ) feature , consistent with absorption from a circumstellar disk in keplerian rotation @xcite . as noted in section [ sec : stars ] , hd216629 shows variation in the hei line profile from one observation to another . in addition , hd216629 is a wide , double - lined spectroscopic binary which is possibly the reason behind the variations . a plot of @xmath86 versus disk radius for hd216629 can be found in figure [ fig : fvsr_hd216629linefits ] . similar to hd76534 , most of the emission for the feii lines is produced in the inner most @xmath99 of the disk . for caii , 90% of the flux also comes from within @xmath99 . for h@xmath1 , 90% of the emission is produced within 7r@xmath3 of the disk , and h@xmath0 requires a more extended region and is produced within the first 20r@xmath3 of the disk . to identify a unique density structure for each star s inner gaseous disk , the next step was to search for a density model with a single set of parameters @xmath100 and type ( thin , thick , turbulent ) that could acceptably fit all of the observed line profiles simultaneously . to measure the goodness - of - fit for single disk density model , a weighted sum of all the individual line profile figures - of - merit @xmath61 was computed as @xmath101 where @xmath43 ranges over the lines available , @xmath102 if the caii line is detected in the spectrum with @xmath103 otherwise . initially , the weights @xmath104 were set to one for all lines and such fits will be referred to as the _ all - line fits_. however , additional weightings were considered : _ balmer - line fits _ , in which only the hydrogen balmer lines h@xmath0 and h@xmath1 were retained in the sum , and _ metal - line fits _ , in which only the feii and caii ( when available ) lines were included in the sum . by examining the results of these various weightings , an adopted density model for the inner , gaseous disk for each star is suggested . single , best - fit _ model was adopted from the above three types of line - fits by visually considering the overall fit of the model line profiles to the observations . overall , for the three stars considered , a 25r@xmath3 , thin disk with a base disk density ( @xmath24 ) of @xmath73 and power - law index ( @xmath25 ) of @xmath52 seen at @xmath70 is the model that provides the best qualitative fit to all the lines in the spectrum ( table [ tab : bestfitmodels_c4 ] ) , given the synthetic model library , and is adopted as the _ best - fit , single disk _ model . it is important to note that this model may not necessarily be the best _ all - line fits _ model , where an equal weight was given to all the lines . the mass associated with this disk model is @xmath105 or @xmath106 for the three stars . figure [ fig : hd76534_globalfits_metal ] shows the synthetic line profiles for the above mentioned model when compared to the observed line profiles of hd76534 . as can be seen in the figure , the synthetic line profiles for h@xmath0 and h@xmath1 do not fit the observed line profile very well . they do not possess deep absorption cores because of the @xmath107 viewing angle of the model . the feii lines are much better fit , while the wide base of the hei line is reproduced . this model was the result of _ metal - line fits _ for hd76534 . figure [ fig : hd114981_globalfits ] shows the _ best - fit , single disk model _ for hd114981 , reproduced by both the all - line fit and metal - fit models for this star . as illustrated in the figure , the synthetic profile of h@xmath0 has the right strength and shape , but is narrower in width than the observed emission line profile . the wings of h@xmath1 are reasonably well reproduced , and the line profile has two peaks and the right shape ; however , the synthetic line profile is significantly weaker in the emission peaks than is observed . the synthetic line profile for feii @xmath19 matches to the observed profile well , and for feii @xmath20 , the shape of the observed line profile is well reproduced , but it is weaker in strength in the emission peaks . we note that @xmath20 shows an asymmetry in the emission peaks that is not seen in @xmath19 . lastly , the hei absorption line profile is weaker than the observed profile . figure [ fig : hd216629_globalfits ] illustrates the _ best - fit , single disk _ model for hd216629 , which was reproduced by both _ all - line fit _ as well as _ metal - line fit _ models . this model reproduces the strength of h@xmath0 well ; however , it is narrower at the base . the h@xmath1 synthetic line profile reproduces the observed blue peak of the line very well ; however , the red peak is a poorer match , with the model profile being too narrow and too weak . unlike hd76534 and hd114981 , the caii line is detected in the spectrum of hd216629 . the synthetic caii line of the all - line model is much too weak compared to the observed line . the synthetic line profile for feii @xmath19 is similar in shape and strength , but narrower compared to the observed line profile . on the other hand , the synthetic line for feii @xmath20 matches the observed line quite well . the synthetic hei line matches the observed line well in width but produces emission in the core of the line that is not observed . for hd76534 , two additional models for the _ all - line fit _ and _ balmer - line fit _ were found . figure [ fig : hd76534_globalfits ] shows the synthetic line profiles for _ all - line fits _ model as compared to the observed lines . this model has a 25r@xmath3 , a thin and turbulent disk with a @xmath24 of @xmath108 and @xmath25 of @xmath52 seen at @xmath109 angle . the h@xmath0 synthetic line profile has the right strength and shape , but it is narrower at the base compared to the observed profile . as the selected model is seen close to edge - on at @xmath109 , the core of the synthetic h@xmath0 profile shows a deep absorption feature that is not observed . the synthetic line profile for h@xmath1 produces the correct shape for the two peaks ; however , it also shows a very deep central absorption core that , like h@xmath0 , is not observed . both feii lines produced by this model are stronger than the observed profiles and also show excessive central absorption . the hei synthetic line profile is a poor fit ; however , the possible two - component nature of this line , illustrated in figure [ fig : hei_twomodels ] , is not accounted for in this model . figure [ fig : hd76534_globalfits_balmer ] shows the _ balmer - line fit _ to the observations for hd76534 . for this model , a 25r@xmath3 , a thin disk with a base disk density ( @xmath24 ) of @xmath72 and power - law index ( @xmath25 ) of @xmath110 seen at @xmath70 . the model reproduces the observed line profiles for h@xmath0 and h@xmath1 fairly . the synthetic line profile for h@xmath0 produces a strong line profile and is also narrower at the base . it is , however , able to reproduce the double - peaked emission profile . for h@xmath1 , the synthetic line profile reproduces the core absorption and the shape of the line profile very well , however the strength of the synthetic line is not as strong as the observed . as the metal lines are set to weight - age of zero , the models is not reproducing any of the observed feii line profile . figure [ fig : hd114981_globalfits_balmer ] illustrates the _ balmer - line fit _ for hd114981 . the model parameters can be found in table [ tab : bestfitmodels_c4 ] and this model has a @xmath46 and @xmath25 of @xmath72 and @xmath110 , respectively , for 100r@xmath3 , a thin and turbulent disk seen at @xmath70 . the strength of the h@xmath0 , as well as its overall width , are better reproduced by this model as compared to the all - line fit previously discussed . for h@xmath1 , the width of the line and the broad absorption trough is well reproduced , but again the strength of the emission peaks of the synthetic line profile does not match the observed peaks . as for the two feii lines , the balmer - line model is an extremely poor match , with this model predicting essentially no feii emission . for hei the synthetic line profile is not as the observed , however the shape is reproduced by the model . figure [ fig : hd216629_globalfits_balmer ] shows the best _ balmer - line fit _ model for hd216629 . this model has a @xmath46 and @xmath25 , @xmath72 and @xmath110 respectively , for 25r@xmath3 , a thin and turbulent disk seen at @xmath70 . the h@xmath0 synthetic line profile is reasonably reproduced in strength ; however , the width of the line at the base is narrower when compared to the observed profile . for the h@xmath1 , the strength of the blue peak is reproduced , but the overall the width of the synthetic line and the size of the red peak is not comparable to the observed profile . essentially no feii or caii emission is predicted by this model . the shape of the observed hei line profile is reproduced by the model , however it lacks in strength . remarkably , the adopted , _ best - fit , single disk _ model based on a single set of density parameters for the inner disks of all three stars turns out to be the same : a thin disk with a base density of @xmath111 , a power index of @xmath112 , and an outer disk radius @xmath113 , seen at an inclination of @xmath70 . recall that the only four inclination bins were considered , @xmath114 , @xmath115 , @xmath116 , and @xmath117 , and thus some variation in the viewing inclinations of the three stars is still consistent with the models . the finding of a similar disk model for all three b2 hbe stars is a result of the basic similarity of the strengths and shapes of their emission line spectra , especially the metal lines ; this point is further discussed in section [ sec : bd65 ] . to understand where the emission is produced in the disk by this single model , as well as to help in future work to decide how to structure the disk in order to better reproduce the line profiles , the cumulative line emissions , @xmath86 , are plotted as a function of radius for this adopted global model in the figure [ fig : fvsr_globalfits ] . while this plot applies to all three stars considered , hd76534 , hd114981 and hd216629 , one difference is that caii emission was detected only in hd216629 , despite the common model assigned to all three stars . h@xmath0 is produced in an extended region of the disk , with most of the emission produced within 15r@xmath3 . this is followed by h@xmath1 , where 90% of the emission is produced within 7r@xmath3 of the disk . both the feii lines follow similar curve and are produced mostly in the inner region of the disk , confined to 5r@xmath3 . given that hardly any emission was seen in the caii synthetic line profile , the little emission it produced was from the inner 4r@xmath3 of the disk . as discussed in section [ sec : results ] , the minimum value of the figure - of - merit , @xmath118 , defines the best fitting profile . nevertheless in the complete model library , it is often the case that several models with different combinations of disk parameters @xmath119 ) produce nearly as good fits as @xmath118 to a given line . to quantify the range of models that provide nearly similar fits to each line , all the models with figure - of - merits @xmath61 satisfying @xmath120 have been examined . table [ tab:25fom_models_c4 ] gives the number of such models found for each emission line for each star . the total number of such matching models for each line is seen to vary from one to 17 , depending on the line and the star . lccc emission line & hd76534 & hd114981 & hd216629 + h@xmath0 & 1 & 3 & 2 + h@xmath1 & 4 & 2 & 2 + caii ( @xmath91 ) & & & 3 + feii ( @xmath19 ) & 17 & 1 & 4 + feii ( @xmath20 ) & 10 & 1 & 1 + he i ( @xmath67 ) & & & 11 + in order to illustrate where the ensemble of best - fitting profiles for each line fall in the explored density parameter space , and how well the best fits , @xmath118 , for different lines agree , we have plotted the models satisfying @xmath121 in the plane defined by the disk density parameters @xmath46 and @xmath25 . note that the other model parameters , @xmath43 and @xmath122 and type of disk ( thin , thick , turbulent , etc ... ) , are not distinguished in such a plot . to graphically illustrate ranges of satisfactory models , we adopt the following conventions : a point is used to illustrate a single model ; if there are two selected models , a line connects them , and if there are three models , a triangle is used . if there are more than three models satisfying the condition on @xmath61 , an ellipse is used to enclose the points . for hei , which generally has a photospheric origin and thus does not constrain the disk parameters , an arrow and line of the same color are used to represent the region of density parameters that produce no disk emission . finally , in these figures , the adopted global disk model for all stars is at the point @xmath123 . figure [ fig : hd76534_ellipseplot ] shows the @xmath124 plane for hd76534 . only some of the models for the different emission lines overlap in the disk density parameter space , illustrating the conclusion of table [ tab : bestfitmodels_c4 ] that there is no single set of disk parameters in a single power - law for the disk s equatorial density for which all of the lines can be well - fit . for hd76354 , two separate regions in parameter space can be seen , one in which the feii lines are best fit and another where the fits to the two hydrogen balmer lines are found . the feii lines select a large region in parameter space corresponding to @xmath125 , whereas the balmer lines select a much smaller region that is of comparable base density but larger power - law index . the plot illustrating the @xmath124 plane for hd114981 can be seen in figure [ fig : hd114981_ellipseplot ] . the regions selected are much smaller than for hd76534 . again the balmer and feii lines select quite different regions , although in this case , the balmer lines favor both lower base density and power - law index as compared to the feii lines . finally , figure [ fig : hd216629_ellipseplot ] illustrates the models for hd216629 . for this star , h@xmath0 requires somewhat smaller base densities than the region selected by the other lines . hd216629 is the single star which includes the caii ir triplet which favors the region selected by h@xmath1 and the feii lines . hd216629 also the only star in the study that requires disk models to fit the observed line profile for the hei line profile , rather than a photospheric profile . the hei covers regions of low density at variety of power law indices , which overlaps with only the h@xmath0 profile . finally note that the adopted global model for all three stars would fall at the point @xmath126 and @xmath127 in these plots . in general , this is the region selected by the metal lines feii and caii , when available . comparing to the hbe star analyzed in paper i of this series , bd+651637 , also a b2e star , there are some similarities and differences among the keys findings . similar to bd+651637 , it was possible to find in almost all cases synthetic line profiles that match the observed emission lines of the three hbe stars considered here . the exceptions are the hei line of hd76534 , for which we suggest a composite spectrum indicative of an unresolved binary , the caii line of hd216629 , for which the observed line profile is wider than any of the model lines , and the hei line of hd216629 , which required a disk model to fit the observed line profile , as oppose a line of photospheric origin . in paper i , the individual line fit for the caii profile of bd+651637 was much better , so the failure in the case of hd216629 does not seem to represent an inability of the present codes to model this line . as with bd+651637 , it was not possible for any of the three hbe stars studied to find a single , global disk density model based on a single power - law that predicted emission line profiles in good accord with all observed lines . therefore , as in paper i , it must be concluded that disk models with more complex density variations are required in order to be able to reproduce all the observed profiles with a unified model . it was also noted in paper i that bd+651637 required higher density and lower power - law index models for the metal lines ( caii & feii ) as compared to the balmer lines . the trend is not so clear in the present work : hd76534 requires the feii lines to have higher @xmath46 and lower @xmath25 compared to the balmer lines , whereas hd114981 requires the feii lines to have higher @xmath46 and higher @xmath25 . the situation is even less clear for hd216629 where only h@xmath0 disagrees with the location in the @xmath128 ) plane selected by the metal lines , including caii in this case , and h@xmath1 . interestingly , the best global fit model for bd+651637 , excluding the caii lines , agrees with the single , global model found in the present work : @xmath129 . however , it must be restated that the global model line fits are , in general , much poorer than the individual line fits for all four stars . finally , the disk size for bd+651637 was found to be somewhat larger at @xmath130 compared to the @xmath131 required for the three stars of the present work . a plot of all the emission lines for all four stars is given in figure [ fig : allstars_compared ] which shows how similar the emission line strengths and shapes are for the three hbe stars studied in the present work . bd+651637 is seen to have significantly stronger emission lines , and this seems to be due to the larger size of its inner gaseous disk @xmath132 ) as compared to the three stars studies here ( @xmath133 ) . among the hei lines of the four stars , only the profile of hd76534 is clearly different , which has been very tentatively attributed to an unresolved contribution from a binary companion . as noted in section [ sec : intro ] , the h@xmath0 emission line of several haebe stars has been studied using spectro - interferometric observations taken with the vega instrument on the chara array . @xcite studied the hae star ab aur and modelled h@xmath0 using a combination of magneto - centrifugal acceleration and a wind . their model suggests that the h@xmath0 emission arises from a disk of 2.3au ( and as close as 0.2au from the star ) . two other hae stars , hd 179218 and hd 141569 , were studied by @xcite and estimated to have h@xmath0 emitting regions of size smaller than 0.21 and 0.12 au respectively . only one hbe star , mwc 361 ( hd 200775 ) , a known magnetic hbe star , has been studied using spectro - interferometric observations . @xcite found that the h@xmath0 line is emitted from disk beyond the regions generally inferred from the accretion process . they estimated the size of the h@xmath0 emitting region to be 0.22au . the values estimated for the h@xmath0 emitting regions for the hae stars may not be relevant to the three hbe stars considered in this study . however , the result of mwc 361 can be directly compared to that of the stars examind in this study . the individual fits to the h@xmath0 line profile for these three stars range anywhere from 25r@xmath3 ( 0.78 au ) to 100r@xmath3 ( 3.1 au ) . given that observations for only one early - type herbig be star are available to compare to our results , it would be very valuable to have further interferometry of the early - type herbig be stars to extend the comparisons . the key findings of this study of the inner gaseous disk of three hbe stars , hd76534 , hd216629 and hd114981 , are : * the emission line spectra of all three stars can be adequately reproduced , in terms of both line strength and shape , from a small ( @xmath134 ) , geometrically thin , gaseous circumstellar disk heated solely by the available photoionizing radiation field from the photosphere of central b star . * the equatorial density in the disks varies roughly as @xmath135 . * the size and mass of the emitting disk are @xmath136 and @xmath137 . * the models , however , are not successful in reproducing the line profiles of hi , hei , caii and feii simultaneously based on an equatorial disk density varying as a single power - law with radius . a possible resolution is to consider equatorial density distributions that vary in a more complex way . * the equatorial density distribution of the three stars is the same as found for the hbe b2 star bd+651637 analyzed in paper i , although their disks are approximately a factor of two smaller in outer radius . we note that the current study focused exclusively on line emission from the inner gaseous disk ; however , continuum emission is also expected , at least in principle , either in the near - ir from mostly hydrogen free - free emission or in the uv from hydrogen bound - free emission . it would be important to quantify the expected level of this emission from the disks above to see if this continuum emission can make a significant contribution to either ( 1 ) the near - ir excess of hbe stars , perhaps addressing why the hbe stars seem undersized compared to the predictions of the disk size luminosity correlation found by @xcite , or ( 2 ) the uv excess seen in some hbe stars that does not seem to be able to be reproduced by magnetospheric accretion models @xcite . finally , an important limitation of the current study is that the three hbe stars analyzed here plus bd+651637 of paper i are all group iii herbig ae / be objects according to the @xcite classification , that is , objects with small infrared excesses resembling classical be stars . we intend to extend our current analysis to group i and ii objects that show unambiguous signs of dust in their infrared spectral energy distributions . we are also extending this analysis to both hotter and cooler hbe stars . the authors would like to thank the anonymous referee for many helpful comments . this work is supported by the canadian natural sciences and engineering research council ( nesrc ) through discovery grants to t. a. a. sigut and j. d. landstreet . alecian , e. , wade , g. a. ; catala , c. , folsom , c. , grunhut , j. , donati , j .- f . , petit , p. , bagnulo , s. , marsden , s. c. , ramirez velez , j. c. , landstreet , j. d. , boehm , t. , bouret , j .- c . & silvester , j. , 2008 , coska , 38 , 235a . sigut , t. a. a. , 2011 , in iau symposium 272 : active ob stars : structure , evolution , mass loss , and critical limits , ed . c. neiner , g. wade , g. meynet , & g. peters ( cambridge : cambridge univ . press ) , 426 .

millisecond pulsars are rather weakly - magnetized neutron stars which are thought to have been spun up by disc accretion , with magnetic linkage between the star and the disc playing a key role . their spin history depends sensitively on details of the magnetic field structure , but idealized models from the 1980s and 1990s are still commonly used for calculating the magnetic field components . this paper is the third in a series presenting results from a step - by - step analysis which we are making of the problem , starting with very simple models and then progressively including additional features one at a time , with the aim of gaining new insights into the mechanisms involved . in our first two papers , the magnetic field structure in the disc was calculated for a standard shakura / sunyaev model , by solving the magnetic induction equation numerically in the stationary limit within the kinematic approximation ; here we consider a more general velocity field in the disc , including backflow . we find that the profiles of the poloidal and toroidal components of the magnetic field are fairly similar in the two cases but that they can be very different from those in the models mentioned above , giving important consequences for the torque exerted on the central object . in particular we find that , contrary to what is usually thought , some regions of the disc outward of the co - rotation point ( rotating more slowly than the neutron star ) may nevertheless contribute to _ spinning up _ the neutron star on account of the detailed structure of the magnetic field in those parts of disc . [ firstpage ] accretion , accretion discs magnetic fields mhd turbulence methods : numerical x - rays : binaries .

1310.7012

accretion discs around magnetized neutron stars can greatly influence the stellar magnetic field , and magnetic field deformations can in turn have an effect on the angular momentum exchange between the disc and the central object . here we investigate these deformations and their effect on the neutron star spin . we focus mostly on the progenitors of millisecond pulsars , which are commonly believed to be descendants of normal neutron stars that have been spun up and recycled back as radio pulsars by acquiring angular momentum from their companion via an accretion disc during the low - mass x - ray binary phase @xcite . the importance of magnetic fields in accretion discs around neutron stars was recognized already in the early paper by @xcite . it was already clear then that the most relevant character in the play is the profile of the toroidal component of the magnetic field , @xmath0 . however , it was only in @xcite that a detailed model was developed . they pointed out that turbulent motion , reconnection and the kelvin helmholtz instability allow the stellar field to penetrate the disc and hence to be affected by its velocity field . by assuming a dipole profile for the poloidal component of the magnetic field , they calculated a corresponding profile for @xmath0 . one of the main consequences of having a non - zero lorentz force inside the disc is the creation of a magnetic torque , which can drastically influence the spin history of the central object and also change the shape of the disc . @xcite used the solution obtained in @xcite to compute the torque exerted on the neutron star by a keplerian disc . however in @xcite it was shown that the solution obtained by @xcite violated the induction equation and a consistent analytic profile for the toroidal field was instead calculated , as a stationary solution of the induction equation in axisymmetry . in that model the poloidal component of the magnetic field was assumed to be a pure dipole ; the disc was assumed to be keplerian , to be in corotation with the neutron star at the upper and lower surfaces and not to have any motion along the poloidal direction ; also , all horizontal derivatives were neglected with respect to vertical derivatives . at almost the same time , @xcite developed a similar one - dimensional model where , in addition , a non - keplerian profile for the disc was considered and where the horizontal terms in the induction equations were also included . both authors agreed in writing the toroidal component of the magnetic field in the following form , in cylindrical coordinates @xmath1 : @xmath2 where @xmath3 is the amplification factor , @xmath4 and @xmath5 are the disc and stellar angular velocities , respectively , and @xmath6 is the dissipation time scale . for a dipole @xmath7 , equation ( [ eq : bp_an ] ) implies @xmath8 , where @xmath9 . in these models , the amplification factor @xmath3 was taken to be a constant not much greater than unity ( it depends on the steepness of the transition in @xmath10 , from disc motion in the equatorial plane to corotation at the disc surface ) . the precise profile of @xmath6 depends on what is the dominant mechanism for dissipating the magnetic field . @xcite considered three different cases , with @xmath6 being dominated by the alfvn velocity , turbulent diffusion and magnetic reconnection , respectively . both @xcite and @xcite employed equation ( [ eq : bp_an ] ) with a dipole vertical field to calculate the magnetic torque . they found that the magnetic contribution to the torque has the same sign as @xmath11 , i.e. the regions of the disc inwards of the corotation point rotate faster than the star , push the magnetic field lines forward and try to speed the star up , while the regions outwards of the corotation point rotate more slowly , drag the field lines backwards and spin the star down . @xcite relaxed the approximation of zero poloidal velocity and studied the magnetic field structure only in the inner region of the disc . his analysis confirmed the results of the previous models . @xcite addressed the problem from a complementary viewpoint , considering the influence of a given stellar magnetic field on the structure of the accretion disc in terms of its height and surface density . they also considered the back reaction on the magnetic field , solving both the induction equation ( in 2d ) and the disc - diffusion equation ( in 1 + 1d ) , using different time steps . they showed that , when @xmath7 is a dipole , equation ( [ eq : bp_an ] ) is valid only to within a factor of @xmath12 @xmath13 on average ; this factor changes with radius and can be as small as @xmath14 for large radii . @xcite carried out numerical calculations to look for steady - state axisymmetric configurations of a force - free magnetosphere corotating with the central star , treating the interaction with the disc by means of a simple model in which the disc velocity profile had only an azimuthal component ( i.e. no accretion flow ) and horizontal derivatives in the disc were neglected with respect to vertical ones . doing this , they found field - line inflation occurring immediately outside the corotation radius , caused by interaction between the stellar dipole field and the disc , with the field lines spreading out and the local field strength becoming weaker . calculating the resulting torque on the star in a way similar to @xcite , they found that this can then be up to two orders of magnitude smaller than what is obtained if the vertical field is taken to be a dipole ( depending on the boundary conditions and the size of the numerical domain ) . this phenomenon of magnetic field - line inflation had previously been discussed by a number of other authors in the 1990s ( see @xcite , ) . it arises due to initially dipolar field lines being dragged round by interaction with the differentially rotating disc matter ; the field - line twisting produced by this leads to inflation of the magnetosphere structure in response to increased magnetic pressure coming from the resulting toroidal component of the field . there can be associated line opening , leading to loss of part of the magnetic linkage between the disc and the star and to a wind being driven out along the open field lines . this has been studied in detail by @xcite in the context of classical ttauri stars . in @xcite ( @xcite , hereafter referred to as paper i and paper ii , respectively ) , we commenced a systematic step - by - step study of the how the original magnetic field of the neutron star is distorted by its interaction with the disc . the strategy was to start with very simple models and then progressively include additional features one at a time , with the aim of gaining new insights into the mechanisms involved . we started off by focusing entirely on the structure of the magnetic field inside the disc , imposing simple dipolar boundary conditions at the top of it and taking pure vacuum outside . consideration of interaction with a non - vacuum magnetosphere ( including the phenomena of magnetic - line inflation mentioned earlier ) is reserved for a subsequent stage of the step - by - step programme . in papers i and ii , the stationary induction equation was solved numerically in two dimensions throughout the interior of the disc , calculating all of the magnetic field components ( i.e. the poloidal field was not forced to be dipolar there ) . as in earlier work , axisymmetry was assumed and the kinematic approximation was used ( i.e. the velocity field was specified and then kept fixed during the calculation without including back reaction from the magnetic field ) . the velocity field was not taken to be purely azimuthal but also had a radial component , as given by the standard @xmath15-disc model of @xcite ( @xcite , hereafter s&s ) , and a theta component , introduced so as to match the upper and lower boundary conditions ( imposed through the use of a coronal layer above and below the disc ) was the same as that of the @xmath15-disc ; @xmath16 was zero in the disc , while in the corona it smoothly changed to match the boundary conditions ; and @xmath10 was keplerian in the main part of the disc , smoothly changing to corotation at the inner edge and in the corona . ] . the results of this analysis demonstrated inward dragging of the poloidal field by the accreting matter and showed that the profile of the toroidal component of the magnetic field can deviate significantly from that given by equation ( [ eq : bp_an ] ) , depending on the values of the magnetic reynolds numbers . in this paper , we follow the same methodology as in papers i and ii but apply it now to the disc velocity profile calculated by @xcite ( @xcite , hereafter referred to as k&k ) . in their paper , k&k presented a self - consistent analytic solution for the general structure of the three - dimensional velocity field inside a stationary axisymmetric @xmath15-disc , and we use this here as the basis for our discussion . we again calculate all of the components of the magnetic field inside the disc : as in papers i and ii , we find significant deviations of the poloidal component away from a pure dipole and of the toroidal component away from equation ( [ eq : bp_an ] ) whereas , when comparing between the results for the s&s and k&k velocity prescriptions ( which are very different ) , we find that the main overall behaviour is qualitatively rather similar . we also compute the contributions to the net torque acting on the neutron star which would come from magnetic linkage with the disc . instead of using a simplified expression for the torque , as in the early models , we calculate it numerically , within our model assumptions , by computing the moment of the lorentz force using the magnetic field configurations obtained as above . this calculation will need to be revisited when we subsequently include improved boundary conditions representing a join on to a non - vacuum magnetosphere , but already we are seeing here striking results concerning which regions of the disc spin the star up or down . in the next section , we describe our model in more detail . in section [ sec : equ ] , we discuss the equations used for our calculations : we write down the induction equation , describe the velocity field and derive the equation for the torque calculation . in section [ sec : mfs ] , we present our results for the magnetic field structure , while those regarding the torque are presented in section [ sec : tor ] . in section [ sec : cmp ] , we compare our results with those from papers i and ii and finally , we summarize our analysis in section [ sec : con ] . our model consists of a neutron star with a dipole magnetic field which is aligned with the common rotation axis of the star and the ( corotating ) disc , everything being axisymmetric . as mentioned above , at this stage of the project , we are focusing entirely on the structure of the magnetic field inside the disc , imposing simple dipolar boundary conditions at the top of it and taking pure vacuum outside . the magnetic field is taken to be sufficiently weak in the main part of the disc , so that it does not produce any significant back reaction on the fluid flow . the approximation of neglecting this back reaction ( the kinematic approximation ) enormously simplifies the set of equations that one needs to solve in order to study the magnetic field structure . in fact , one then has to solve only the magnetic induction equation , which we do by looking for a stationary configuration of the magnetic field . it is reasonable to suppose that close to the stellar surface the magnetic field is strong and dipolar : we are considering a surface field of @xmath17 g , which is typical for millisecond pulsars ( see e.g. @xcite ) . because of the @xmath18 dipolar fall off , as one moves away from the star , the magnetic field intensity decreases outwards quite rapidly and the field lines become progressively more distorted by the matter in the disc . as shown in papers i and ii , this deformation is twofold : ( 1 ) the angular rotation stretches any dipolar field line in the azimuthal direction , thus creating a toroidal magnetic - field component , and ( 2 ) the radial motion likewise has an effect by tending to pull any field line inwards with the accretion flow ( or outwards in the case of a backflow ) . close to the stellar surface , the magnetic field is strong and completely dominates the dynamics of the matter , i.e. this is the opposite regime to the one in which the kinematic approximation holds . however , for exactly that reason , in this region we can safely assume that the magnetic field is close to dipolar . our strategy is therefore not to make any detailed study of what happens in the close vicinity of the neutron star but instead to take the magnetic field there as just being a dipole . we place the inner edge of the disc at the alfvn radius in the radius of the inner edge . ] and solve the induction equation from there outwards . the innermost part of the disc is a transition region where the magnetic field does not completely dominate the dynamics of the matter , but where the magnetic torque is sufficient to alter the angular velocity of the matter flow away from keplerian . we treat this region ( which we refer to as the ` inner disc ' ) by imposing a suitable sub - keplerian profile for the angular velocity there . the solution of the induction equation in this region is used only to provide suitable boundary conditions for the solution further out , where the kinematic approximation becomes reasonable . there is another region where the kinematic approximation may not be suitable , and this is near to the surface of the disc . although the magnetic field intensity decreases quite rapidly with @xmath19 and can be very small for large @xmath19 , the matter pressure here is even weaker since it decreases sharply with @xmath20 , because of the change in density from the disc to the exterior ( here taken to be vacuum ) . we model this aspect by introducing a boundary layer , which we call the corona , at the top and bottom surfaces of the disc . in this corona , the weak magnetic field is less prone to follow the motion of the low density matter and we reproduce this behaviour in the model by using a larger value of the turbulent magnetic diffusivity there , which helps in reducing distortions away from a dipole ( i.e. away from the boundary conditions ) . we use this ` by - hand - enhancement ' of the diffusivity also in the inner part of the disc mentioned earlier . throughout this paper , we consider a canonical @xmath21 neutron star with radius @xmath22 km , rotation period @xmath23 ms and a surface magnetic field of @xmath24 g , corresponding to a magnetic dipole moment of @xmath25 g @xmath26 . the mass accretion rate is taken to be @xmath27 g s@xmath28 . a sketch ( not to scale ) of the model is given in fig . [ fig : model ] , where the two regions with larger diffusivity are clearly marked : they are the ` inner disc ' and the ` corona ' . the inner disc extends from the inner edge @xmath29 to the transition radius @xmath30 ( where @xmath31 cm is the schwarzschild radius ) ; the corona extends from @xmath32 to @xmath33 for all radii . the corotation radius is at @xmath34 and the light cylinder at @xmath35 . to avoid strong dependence on the outer boundary conditions , our numerical domain of integration extends much further out than the light cylinder ( dashed lines in the figure ; more details are given in sections [ ssec : bcs ] and [ ssec : sol ] ) . and @xmath30 . for a @xmath22 ms pulsar ( with mass @xmath36 ) , @xmath37 and @xmath38 . the opening angles are @xmath39 for the disc alone and @xmath40 for the disc plus corona . the dashed lines indicate that the outer disc extends much further out than the main region shown here : the grid continues until @xmath41 when we are solving for the toroidal component of the magnetic field and to @xmath42 when we are solving for the poloidal one.,scaledwidth=40.0% ] in this section , we derive the equations for calculating the magnetic field structure and comment on the profile of the velocity field and on the turbulent diffusivity as well as saying something further about the boundary conditions . we also derive equations for calculating the magnetic torque exerted by the disc on the neutron star and briefly comment on the solution method . as mentioned previously , we are making our calculations using the kinematic approximation , within which the velocity field of the disc matter is specified separately ( here following the k&k solution ) rather than being solved for together with the magnetic field calculation . once the velocity field has been specified , the magnetic field structure can then be obtained by solving the induction equation : @xmath43 where @xmath44 and @xmath45 are the mean components of the magnetic field and velocity field respectively , and @xmath46 is the turbulent magnetic diffusivity . resolving the full structure of turbulence and studying accretion driven by the magneto - rotational instability goes well beyond the scope of this paper . interested readers are referred to @xcite where the first global axisymmetric simulations of that kind are described . here , we use average quantities and follow a phenomenological approach for the angular momentum transfer . as in papers i and ii , we are neglecting the so - called @xmath15-effect , which involves the appearance of a large scale electromotive force because of the interaction between the turbulent fluctuations in the magnetic fields and the velocity fields . this term creates a growing solution for the magnetic field and is fundamental when one wants to study dynamo effects . in line with our approach , we want to use a simple model where one can unambiguously understand the roles played by the different contributing factors . only afterwards will additional ingredients ( such as dynamo action ) be added and their effects studied by comparison with the previous analyses . we use spherical coordinates ( @xmath19 , @xmath20 , @xmath47 ) , with the origin of the coordinate system placed at the centre of the neutron star and the unit vector @xmath48 pointing in the direction of the magnetic dipole axis ( and the stellar rotation axis ) . in our model we are assuming axisymmetry ( @xmath49 = 0 $ ] ) and we want to find a stationary solution ( @xmath50 = 0 $ ] ) . the three scalar components of equation ( [ eq : ind ] ) are then : @xmath51\right ] \right\ } \label{eq : ind_r } \ , \,{\rm , } \\ 0 & = & \partial_r \left\ { r \hspace{0.5cm}\left [ v_r b_\theta - v_\theta b_r - \frac{\eta}{r}[\partial_r(rb_\theta ) - \partial_\theta b_r ] \right ] \right\}\label{eq : ind_t } \ , \,{\rm , } \\ 0 & = & \partial_r \left\ { r \left [ v_\phi b_r - v_rb_\phi + \frac{\eta}{r } \partial_r(rb_\phi ) \right ] \right\ } -\nonumber\\ & & \partial_\theta \left\ { v_\theta b_\phi - v_\phi b_\theta - \frac{\eta}{r\sin\theta}\partial_\theta ( b_\phi \sin\theta ) \right\ } \,\ , \rm{. } \label{eq : ind_p}\end{aligned}\ ] ] the first two equations are independent of @xmath0 , and so we can solve these first for the poloidal components of the magnetic field and then solve the third equation afterwards for the toroidal component . in order to solve for @xmath52 and @xmath53 , one needs to specify profiles for @xmath54 , @xmath16 and @xmath46 and to solve for @xmath0 one also needs a prescription for @xmath55 . then , given a set of consistent boundary conditions , the set of equations ( [ eq : ind_r])([eq : ind_p ] ) has a single unique solution . k&k made a calculation of the three - dimensional structure for an @xmath15-disc , taking the disc to be suitably thin and taylor expanding the equations of viscous hydrodynamics in terms of a small parameter @xmath56 , where @xmath57 is a characteristic vertical scaleheight in the disc and @xmath58 is a characteristic radius ( taken , in practice , to be the radius of the neutron star ) . the equation of state of the accreting medium was taken to be polytropic with constant specific entropy , implying that for any comoving fluid element , increases in entropy due to viscous dissipation are exactly balanced by losses due to radiation and conduction . they were able to find analytical solutions for the lowest order terms in their taylor expansions and these demonstrated a flow structure which was entirely an inward accretion flow at the top and bottom of the disc but could have an outward backflow near to the mid plane at radii beyond a certain value . a similar flow pattern , including equatorial zone backflow , was found by @xcite in a calculation including radiative transfer in an @xmath15-disc modelled with a perfect gas equation of state . clearly , this flow pattern is very different from that in standard s&s discs and so it was of interest to repeat our earlier calculations of magnetic field distortions by s&s discs also for these k&k ones - viscosity prescription with @xmath59 , for which the issue of large turbulent cells need not arise . indeed , having them would completely change the spectra and character of the s&s solutions . more sophisticated mri treatments for the effective viscosity have smaller - scale turbulence and _ require _ stratification for successful operation . ] . the general form of the taylor expansions used by k&k is given in section 2.1 of their paper . we use here their prescriptions for the poloidal components of the velocity field , @xmath54 and @xmath60 , leaving the rotational velocity @xmath10 as being keplerian in the main part of the disc . due to parity considerations and mass conservation , some of the terms appearing in the general form of the expansions are zero , and one is left with the leading - order terms in the expressions for @xmath54 and @xmath60 as being : @xmath61 where @xmath62 is a characteristic sound speed in the disc . the analytic expressions for @xmath63 and @xmath64 are given by equations 2.53 and 2.56 of k&k , respectively . for implementing this velocity field within our calculation , we first choose the geometrical profile of the disc by specifying the disc height @xmath65 as a function of radius @xmath19 ( note that here we are using @xmath65 and @xmath19 to represent physical lengths and not the scaled quantities used in k&k ) . we choose this profile as being @xmath66 , where @xmath67 is the co - latitude of the disc surface ( for which we take @xmath68 ) . we sharply truncate the disc at an inner radius @xmath69 ( for which we take @xmath70 ) . we then specify a quantity which we call @xmath71 , which is the value of the radial velocity @xmath54 at the point @xmath72 . doing this turns out to be all that is required in order to uniquely determine the velocity field in physical units ( i.e. as unscaled quantities ) . changing the value of @xmath71 is equivalent to changing the value of @xmath15 . from equation 2.55 of k&k , one has ( neglecting @xmath73 corrections ) : @xmath74 where @xmath75 is the keplerian linear velocity and all quantities are calculated at @xmath72 . we follow the k&k poloidal velocity profile through all of the disc and then modify it in the corona . there we first modify @xmath16 via an error function in order to match the boundary conditions smoothly and analytically . then we modify @xmath54 in such a way that the total local kinetic energy remains constant . the resulting velocity - field vectors are shown in figs [ fig : pol_ref_vfield ] and [ fig : pol_ref_vfield_zoom ] . note the appearance of a ` stagnation surface ' where both @xmath54 and @xmath16 go to zero , marking a change in direction of the flow . having the surfaces where @xmath76 and @xmath77 coinciding is a particular property of the height profile that we are using here ( with @xmath78 ) . and @xmath16 plotted using cartesian coordinates @xmath79 and @xmath80 measured in units of the schwarzschild radius . ( note that we use these cartesian coordinates for plots throughout this paper . ) the dashed line separates the inflow region from the backflow one . the heavy solid line marks the boundary between the corona and the external vacuum ; the other solid line shows the boundary between the disc and the corona . ] around the region where the flow changes direction . the region from @xmath81 to the equatorial plane is shown.,scaledwidth=45.0% ] as regards the angular rotation velocity , we follow the same profile as in paper ii ( see section 3.2 and fig . 2 of that paper ) . here , we repeat the details again for completeness : @xmath82 , r\in[r_{\rm tr } , r_{lc}]\\ & \\ \mbox{smooth join in $ \theta$ } & \hspace{0.3 cm } \mbox{in the corona:}\\ & \hspace{0.3 cm } \theta\in[\theta_{\rm cor } , \theta_{\rm disc}]\\ & \\ \mbox{smooth join in $ r$ } & \hspace{0.3 cm } \mbox{in the inner disc:}\\ & \hspace{0.3 cm } r\in[r_{\rm in } , r_{\rm tr}]\\ & \\ \omega_{\rm s } & \hspace{0.3 cm } \mbox{at the ghost zones : } \theta = \theta_{\rm cor } - \delta \theta\\ & \hspace{0.3 cm } \mbox{at the inner edge : } r = r_{\rm in } \end{array } \right.\ ] ] where @xmath83 is the upper surface of the corona , @xmath84 is the angular grid resolution , @xmath5 is the stellar spin rate , @xmath85 is the keplerian angular velocity and the two smooth connections are made using the error functions given in equations ( [ eq : erf_t ] ) and ( [ eq : erf_r ] ) below . in the @xmath20 , direction we write @xmath86\\ & = & [ \omega_k(r ) - \omega_{\rm s } ] \ , f_1(\theta ) + \omega_{\rm s } \ , \,{\rm , } \end{aligned}\ ] ] where @xmath87 \ , \,{\rm , } \ ] ] with @xmath88 rad ( i.e. @xmath89 ) . similarly , in the radial direction : @xmath90 \\ & = & [ \tilde{\omega}(r,\theta ) - \omega_{\rm s } ] \ , f_2(r ) + \omega_{\rm s } \ , \ , { \rm , } \end{aligned}\ ] ] where @xmath91 \ , \ , { \rm , } \ ] ] with @xmath92 and @xmath93 . this angular velocity profile is keplerian in all of the outer disc ( see fig . [ fig : model ] ) , while in the corona and in the inner disc it deviates from keplerian in order to match corotation at the boundaries . on the equatorial plane , @xmath10 starts to become sub - keplerian at @xmath94 ( where deviations are of the order of @xmath95 ) and reaches its maximum at @xmath96 , after which it decreases towards corotation . for the diffusivity , we follow the same approach as in paper ii , writing @xmath97 \ , \left [ \frac{\eta_{\rm c}}{\eta_0 } - 1 \right ] \right\ } \,\ , { \rm , } \ ] ] where @xmath98 is the value in the main disc region and @xmath99 is the value in the corona and inner disc ( see fig . [ fig : model ] ) . we here take @xmath100 @xmath101 s@xmath28 and @xmath102 @xmath101 s@xmath28 as standard values . the magnitude of the diffusivity is often discussed in terms of the turbulent magnetic prandtl number , which links it with that of the turbulent viscosity . there is a lot of uncertainty about which values to take and the one used here as our standard for @xmath98 may seem rather low . we have already discussed this issue in paper ii and note also the discussions by @xcite and others . we chose our value because it gives an interesting amount of field distortion in the disc , but it is clearly important to investigate the effects of varying it , and we also discuss this in the following sections . for joining @xmath103 and @xmath104 between the low@xmath46 and high@xmath46 regions , we use joining functions of the form @xmath105 \ , \ , { \rm , } \ ] ] where @xmath106 and @xmath19 for @xmath107 and @xmath108 respectively , with @xmath109 and @xmath110 in the two cases ; @xmath111 is the width of the transition in the error function , for which we use @xmath112 in the radial direction and @xmath113 in the theta direction . we have already commented on the boundary conditions being used for the present stages of our calculations : vacuum everywhere outside of the disc ( plus corona ) and the neutron star , with a dipolar magnetic field throughout those vacuum regions . this is highly idealized , of course , but it is our working approximation for the moment . in order to have a smooth transition between the disc and the external vacuum , we include a boundary layer , a corona , and then impose boundary conditions at the top of it . we also impose dipole conditions at the inner and outer radial edges of the disc . at the inner edge , this is done because the magnetic field is strong there and so it is very hard to get any distortions there at all . what is done at the outer edge is irrelevant anyway because we are placing it so far away that it does not influence what is happening inside the light cylinder , and so we make the simple choice of imposing dipole conditions there as well . in paper i , we tested various kinds of boundary conditions ( spherical lines , vertical lines and lines with @xmath114 inclination ) and found that for poloidal magnetic field lines entering the extended disc ( i.e. disc plus corona ) at the same locations , their shape within the disc varied hardly at all among the different cases . however , the values for the _ field strength _ as a function of distance out along the surface of the extended disc will also vary depending on the structure of the external magnetic field and this will clearly influence the magnitude of the magnetic torque exerted by the disc . we assume equatorial symmetry as well as axisymmetry and solve for only one quadrant . at our lower boundary in @xmath20 ( i.e. at the equatorial plane ) , we require @xmath53 to be symmetric about the equatorial plane ( i.e. @xmath115 ) , while @xmath52 and @xmath0 have antisymmetric conditions there ( i.e. @xmath52 and @xmath0 are zero at this boundary , and at the first ghost grid - point beyond it they have values opposite to those at the last active grid - point before it ) of paper i for more details . ] . as regards the poloidal component of the velocity field , we follow k&k exactly for the inner and lower boundaries . this means that on the equatorial plane @xmath77 and @xmath54 is symmetric . this condition is equivalent to choosing a profile for @xmath53 at @xmath116 satisfying : @xmath117 at the top boundary , we have to modify the k&k profile . in paper i ( section 3.2 ) , we have already shown that if one requires the magnetic field to be a dipole , then the induction equation gives the following two conditions for the velocity field , @xmath118 and @xmath119 everywhere . ] : @xmath120 where @xmath121 can take any value . we use the first condition directly to calculate @xmath16 in the coronal layer and for @xmath10 we use corotation , which is consistent with the second condition with @xmath122 . the magnitude of the torque exerted by the accreting matter on the neutron star , is clearly affected by the behaviour of the magnetic field in the magnetosphere . in the present simple model , we are taking vacuum above and below the extended disc so that the magnetic field there remains strictly dipolar , but in the general realistic case there can be field - line inflation , as mentioned earlier , which may also lead to breaking of some of the magnetic linkage between the disc and the star . it has been suggested that associated opening of field lines might give rise to jets and winds : see , for example , @xcite concerning magnetospheric ejections . in this subsection , we consider the various contributions to the torque , first giving a rather general discussion and then specializing it to our simplified model . the disc can exert a torque on the central object via the link provided by magnetic field lines anchored at one end to the neutron star while the other end threads the disc , where the interaction with the moving matter creates a lorentz force which can have a non - zero moment with respect to the central object . we call this contribution to the total torque @xmath123 . as matter moves inwards , it encounters a progressively stronger magnetic field which eventually comes to completely dominate the motion of the plasma . this happens at the inner edge of the disc , where we suppose that matter leaves the disc following the magnetic field lines . if the magnetic field is then rigidly connected with the neutron star , it will force the matter to move with the same angular velocity as the central object . however , during this motion along the magnetic field lines , the matter is approaching the neutron star , and so although its angular velocity remains constant , there is a progressive decrease in its angular momentum ( the cylindrical radial distance is decreasing ) . this occurs due to a torque exerted on it by the magnetic field lines , associated with which there is an equal and opposite torque tending to spin up the neutron star . we call this contribution @xmath124 . finally , when the accreting matter hits the neutron star surface , it transfers its residual angular momentum to the star . the torque concerned in this is typically very small , however , and we will neglect it here . the total torque exerted on the star can then be written as : @xmath125 with @xmath126 where @xmath127 is the moment of the lorentz force ( @xmath128 ) exerted on the neutron star by the local fluid element ; @xmath129 is the volume containing all of the fluid elements associated with the disc which are magnetically linked to the neutron star and for which @xmath130 is non - zero ; @xmath131 is the accretion rate ; @xmath132 is the neutron star spin period ; @xmath133 is the distance from the rotation axis to the inner edge of the disc , while @xmath134 is the corresponding distance for the point where the matter hits the surface of the neutron star and @xmath135 is the unit vector along the rotation axis . to the best of our knowledge , this is the first work where the magnetic torque is calculated by computing the integral of the lorentz force through the whole disc , i.e. by means of equation ( [ eq : t_bd_def ] ) . in the usual approach , only the value at the disc surface is considered . throughout this work we express torques with respect to the star , i.e. a positive value means that the torque is spinning the star up , while a negative one produces spin down . the sign of @xmath136 is always positive : it always acts so as to spin the star up ( positive torque ) because the angular momentum of the matter is decreasing when going from the inner edge of the disc to the neutron star surface . the sign of @xmath137 , however , is not obvious until the calculations are performed ( see equation [ eq : torque ] below ) and in general it can be either positive or negative . we now consider in more detail the contribution to the torque due to the integrated moment of the lorentz force ( i.e. that coming from equation [ eq : t_bd_def ] ) . we are calculating the torque with respect to the origin of the coordinate system , and so the radial component of the force has zero moment . the force acting along the @xmath20 direction is always perpendicular to the position vector and so , if it is non - zero , it will always give a non - zero torque along the @xmath47 direction . however , since this component is antisymmetric with respect to the equatorial plane ( while the position vector is symmetric ) , the integral of its moment will be zero . the only non - zero contribution to the net torque comes from the moment of the @xmath47 component . if we use @xmath138 to indicate the position vector , and write @xmath139 with @xmath140 , @xmath141 and @xmath142 being unit vectors in the @xmath19 , @xmath20 and @xmath47 directions , we have @xmath143 where in the last equality we have changed the volume integral into three integrals over the three coordinate directions , integrated over @xmath47 ( giving the @xmath144 factor ) and have explicitly written the direction of the torque as being along the vertical axis ( this implies a change of sign because @xmath141 and @xmath135 have opposite directions ) . from maxwell s equations in cgs gaussian units , we can write the general expression for the lorentz force per unit volume as @xmath145 whose @xmath47 component in spherical coordinates is @xmath146 when considering axisymmetric models ( as we are doing here ) , the third term can be neglected . the second term is also usually neglected for thin disc models because it involves derivatives with respect to the radius , while the first one involves vertical derivatives . the strategy of this work is to fully consider both radial and vertical structure and so we keep both terms in the torque calculation . substituting equation ( [ eq : fl_phi ] ) into ( [ eq : t_bdisc ] ) one obtains @xmath147 r^2 \right\ } dr d\theta \,\ , \mathbf{\hat{z } } \,\ , { \rm .}\end{aligned}\ ] ] we write the torque as the sum of two contributions in order to facilitate the comparison with the thin disc models that neglect the radial term : @xmath148 with @xmath149 dr d\theta \,\ , \mathbf{\hat{z } } \,\ , { \rm , } \\ % \label{eq : t_bdr } \mathbf{t}_{b-{\rm disc}}^{r } & = & \iint\limits_{r,\theta } -\frac{1}{2 } \left [ b_r \partial_r(b_\phi\,r ) \,r^2\sin\theta \,\right ] dr d\theta \,\ , \mathbf{\hat{z } } \ , \,{\rm .}\end{aligned}\ ] ] as we have mentioned earlier ( see section [ sec : mod ] ) , in the inner part of the disc our kinematic approximation breaks down because the magnetic pressure becomes non - negligible there with respect to the gas pressure ( although it still remains smaller than the gas pressure until @xmath150 is reached ) . in this part of the disc , we expect the angular velocity to become sub - keplerian and finally to reach corotation with the neutron star at @xmath150 . since we can not reproduce this profile self - consistently ( because we are using the kinematic approximation and are not solving an equation of motion ) , we change the @xmath10 profile by hand in that region , as described in section [ ssec : vel ] . this should actually mimic the real situation rather well . for this region , instead of using equation ( [ eq : t_bd_def ] ) to calculate the torque contribution , we instead calculate the back reaction on the field lines corresponding to the angular momentum loss by the matter ; this would then be expected to be mainly transmitted to the neutron star ( although it could also go into a jet ) . we calculate this torque contribution as @xmath151 where @xmath152 is the location on the equatorial plane where the angular velocity differs from the keplerian value by @xmath95 ( @xmath153 with our model settings ) , and @xmath154 is the vertically averaged value of the angular velocity at that radial location . more rigorously one should consider local values for @xmath131 and @xmath10 and integrate equation ( [ eq : t_inner ] ) over the vertical direction . + we then calculate the magnetic torque from the disc as the sum of two components : @xmath155 where @xmath156 is given by equation ( [ eq : t_inner ] ) and @xmath157 is given by equation ( [ eq : t_bd_def ] ) integrated from @xmath152 out to the point where the contribution to the integral becomes negligible . we calculate @xmath124 from equation ( [ eq : t_acc_def ] ) , taking @xmath133 as being the distance from the rotation axis to the uppermost part of the inner edge of the disc ( i.e. @xmath158 ) and getting @xmath134 ( the corresponding distance for the point where the infalling matter hits the surface of the neutron star ) by assuming that the matter concerned follows dipole magnetic field lines after leaving the inner edge of the disc ; this then gives @xmath159 . equation ( [ eq : ind ] ) has been solved numerically using a gauss seidel iterative scheme with an evenly spaced grid in @xmath19 and @xmath20 . as in papers i and ii , all variables and coordinates were written in a dimensionless form for the purposes of the calculation . because of the partial decoupling between the poloidal and toroidal components of the magnetic field , we have first solved for @xmath160 ( measured in units of @xmath161 g ) , following the procedure described in paper i , and then for @xmath0 ( measured in units of @xmath162 , where @xmath163 is the radius of the neutron star in units of @xmath164 ) following the procedure described in paper ii . details of the numerical scheme , including tests , have been given in papers i and ii . from a technical point of view , the calculations performed here for obtaining the magnetic field structure are very similar to those of paper ii , the only difference being in the functions used for the velocity components . we recall that we use a grid resolution of @xmath165 and @xmath166 for both @xmath160 and @xmath0 . in both cases , we use a numerical domain which is much larger than the physical domain of interest in order to avoid dependence of the results on the outer boundary conditions . in addition to solving for the magnetic field structure , in this paper we are also solving for the torque , which involves calculating the integral given in equation ( [ eq : torque ] ) . we do this by using a trapezoidal scheme , as described in section [ sec : tor_disc ] . we have studied a range of different models , so as to investigate the effect of varying parameters , but first we present results for a representative fiducial model to provide a standard against which to compare the others . our fiducial run is characterized by intermediate values of the magnetic reynolds numbers , that are defined as @xmath167 , where @xmath168 , @xmath169 and @xmath170 are characteristic values for the length , velocity and diffusivity , respectively . for @xmath168 , we take the radius of the inner edge of the disc ( i.e. @xmath171 cm ) ; for @xmath170 , we take @xmath98 , as defined above ; for @xmath169 , we take either a representative radial velocity ( for which we use @xmath71 , as defined earlier ) or a representative azimuthal velocity ( for which we use the value of @xmath55 at the corotation radius , which we denote as @xmath172 ) . depending on which of these velocities is used , we get either the poloidal or toroidal magnetic reynolds number ( @xmath173 or @xmath174 , respectively ) . for our neutron star model , @xmath175 cm s@xmath28 . the values that we use for our fiducial run are @xmath176 and @xmath177 , which we obtain by taking @xmath178 cm s@xmath28 and @xmath179 @xmath101 s@xmath28 . the resulting structure of the poloidal component of the magnetic field @xmath160 is shown in fig . [ fig : ref_bpol ] . in fig . [ fig : ref_bphi ] , we show the contour plot of the toroidal component @xmath0 . we find significant deviations of both @xmath160 and @xmath0 from the profiles which are usually considered , i.e. a dipole and the one given in equation ( [ eq : bp_an ] ) , respectively . contour plot for the fiducial run , in units of @xmath180 . solid contours indicate positive values , while dashed lines are used for the negative ones . the solid grey line is the boundary between the disc and the corona ; the dashed grey line here marks the stagnation surface and a white dot dashed line is used for the zero - level contour ( which is just inwards of the corotation point).,scaledwidth=45.0% ] the reason for the deviations of @xmath160 away from the dipole are clear : when field lines enter the disc , they are pushed inwards by the inward motion of the plasma and a negative @xmath52 appears . as one moves towards the equatorial plane , the radial component of the velocity field decreases , goes through the stagnation surface and then becomes positive in the region of backflow ( as shown in fig . [ fig : pol_ref_vfield_zoom ] ) . as a consequence , we see that the magnetic field lines are bent backwards , with @xmath52 going through a zero and then becoming positive again . in the inner disc region , the magnetic field follows the plasma motion to a lesser degree , as it should if one assumes that the magnetic stresses eventually truncate the disc ( in fact @xmath52 always stays positive in this region ) . we recall that we obtain this effect because in this region we have a larger characteristic value for the turbulent diffusivity ( giving a smaller local magnetic reynolds number ) . the toroidal component of the magnetic field has two extrema : the first being positive and inwards of the corotation radius , and the second being negative and outwards of it . this is exactly what one would expect if @xmath0 were to follow equation ( [ eq : bp_an ] ) , with the vertical field being the same as for a dipole . however , if one analyses the profile in more detail one sees that while the first peak follows closely that of @xmath181 , the second one appears at a much larger radius . a similar behaviour was also obtained in paper ii and it was explained there by considering the magnetic distortion function @xmath182 ( see section 5.2 of that paper ) . the distortion function was introduced in paper i : it is defined in the same way as the magnetic reynolds number but , instead of taking characteristic values for the velocity and the diffusivity , it takes local values . depending on which velocity component one considers , one can have a poloidal or toroidal magnetic distortion function : @xmath183 as in paper ii we find that the position of the second peak in @xmath0 follows the peak in the distortion functions ( see figs . [ fig : pol_ref_dmcont ] and [ fig : tor_ref_dmcont ] ) . . its peak is at about the same radial location as that of @xmath184 ( see fig . [ fig : tor_ref_dmcont ] ) , and that of the second peak in the @xmath0 contour plots ( see figs . [ fig : ref_bphi ] , [ fig : ref_b3cont ] and [ fig : v0_0_eta0_1e9_b3cont]).,scaledwidth=45.0% ] . its peak is at about the same radial location as that of @xmath185 ( see fig . [ fig : pol_ref_dmcont ] ) , and that of the second peak in the @xmath0 contour plots ( see figs . [ fig : ref_bphi ] , [ fig : ref_b3cont ] and [ fig : v0_0_eta0_1e9_b3cont]).,scaledwidth=45.0% ] in fig . [ fig : pol_eta0_4e10_bpolavg ] , we report the shell averages of the poloidal component of the magnetic field calculated in half of the disc , with the dipole ones for comparison . distortions are present in both poloidal components , with the ones for @xmath52 being the more evident . this component even changes sign , at about @xmath186 ( the absolute value of the shell average is being plotted ) and stays negative from that radius outwards , tending to zero from negative values . for the @xmath20 component ( whose intensity is larger by about one order of magnitude ) , there is a weak but extended amplification in the inner part , between @xmath187 and @xmath188 . this is the region where magnetic field lines accumulate ( see fig . [ fig : ref_bpol ] ) . we have seen the same behaviour also in papers i and ii and it is what one might expect in the transition region , where the magnetic field goes from being dominated by the plasma ( large distortions ) to dominating over the plasma ( small distortion , dipole field ) . ( dotted ) and @xmath53 ( solid ) . @xmath52 changes sign at @xmath186 . a dipole field is shown for comparison : its @xmath19 component is shown with the dot dashed line , while its @xmath20 component is shown with the dashed line.,scaledwidth=41.0% ] fig . [ fig : tor_eta0_4e10_b3cmp_avg ] compares the @xmath0 shell average with the results of @xcite and @xcite ) , with a dipole field at @xmath116 being taken for @xmath7 and then normalizing to the maximum of the absolute value . ] and includes also the analytic prescription suggested in paper ii ( see section 5.3 of that paper ) that we report here for clarity : @xmath189 \,\,{\rm .}\ ] ] if we apply this expression to the model considered in the present calculation , the agreement is not as good as it was in paper ii . this is due to the breaking down of one of the assumptions made in deriving equation ( [ eq : paperiibp ] ) . in paper ii , we assumed that the @xmath0 generation rate @xmath190 can be written as @xmath191 / r$ ] and that the loss term @xmath192 can be written as @xmath193 and we then equated the two terms ( because we were looking for a stationary solution ) . in the present model , there are some regions of the disc where the approximation for the loss term represents only a lower estimate . nevertheless , in fig . [ fig : tor_eta0_4e10_b3cmp_avg ] , one can see that the analytic expression still has a similar behaviour to that of the numerical solution and the agreement is quite good in the interpeak region , although it overestimates @xmath0 in the inner part and underestimates it in the outer one . all of the curves have the same general behaviour , with two peaks ( the first positive and inwards of corotation , the second negative and outwards of corotation ) , and they all go through zero at approximately the same location ( close to the corotation point ) . however , the relative amplitudes of the two extremal points are very different . the models of the 1980s , i.e. equation ( [ eq : bp_an ] ) with a dipole @xmath7 , predict the first ( positive ) peak to be the strongest one , whereas our present calculations are showing just the opposite , with the second ( negative ) peak having an amplitude more than four times larger than the first one . our analytic prescription can reproduce this behaviour , with the ratio between the amplitudes of the second and first peaks being larger than @xmath194 , although that is still smaller ( i.e. @xmath195 ) than what is required . ) with a dipole @xmath7 and the dotted line is our analytic prescription as obtained in paper ii and reported in equation ( [ eq : paperiibp ] ) . the scale for the fiducial shell average is given on the left y - axis , while the other two curves are normalized to the maximum of their absolute value ( scale on the right ) . the 1980s result has its zero for @xmath0 at corotation ( @xmath196 ) , whereas the fiducial configuration has it at @xmath197 and the analytic profile has it at @xmath198.,scaledwidth=45.0% ] in addition to the fiducial case , we analysed a range of other models to investigate the changes occurring in the magnetic field distortions when the parameters @xmath71 and @xmath98 ( or @xmath173 and @xmath174 ) were varied over a wide range . overall , we considered @xmath199 configurations , which we label ` mf ' to denote magnetic field structure . these are all listed in table [ tab : conf_mf ] . for each configuration , we report the value of the turbulent magnetic diffusivity , @xmath98 , the scale factor of the velocity field in the poloidal direction , @xmath71 , the poloidal magnetic reynolds number , @xmath173 , the toroidal magnetic reynolds number , @xmath174 , and the value of @xmath15 corresponding to @xmath71 ( see equation ( [ eq : v0_alpha ] ) ) . .parameter values for the mf configurations . [ cols="^,^,^,^,^,^ " , ] [ tab : torque_cmp ] this paper is part of a step - by - step analysis of the interaction between the magnetic field of a central rotating neutron star and an encircling accretion disc , focusing particularly on millisecond pulsars . our strategy has been to start with a very simple model and then to progressively include additional features one at a time , so as to understand the effects of each of them . for the model presented here , we have considered a neutron star with a dipole magnetic field aligned with the rotation axis , surrounded by a disc which is truncated at the alfvn radius and which has a coronal layer above and below it , treated as a boundary layer between the disc and its exterior . the external region is here taken to be vacuum and the magnetic field there is taken to be a perfect dipole . ( effects coming from interfacing on to a non - vacuum magnetosphere will be considered in subsequent work . ) we have used the kinematic approximation , in which the velocity field of the matter is kept fixed , and have solved the induction equation for mean fields looking for a stationary configuration . we solve self - consistently for all of the magnetic field components , employing a velocity field having all three components different from zero , and do not neglect horizontal derivatives . the code and the procedure used for this analysis have been described in detail in @xcite , where we used a specific three - dimensional extension of the @xcite velocity profile ( i.e. a velocity field with three non - zero components , each depending at most on the two coordinates @xmath19 and @xmath20 ) . here , instead , we consider a more general prescription for the velocity field , as described in @xcite . the key features of this velocity field are ( 1 ) a fully three - dimensional axisymmetric profile and ( 2 ) the presence of a flow reversal ( backflow ) in a region close to the equatorial plane . the magnetic configuration that we find is consistent with the conceptual picture described in papers i and ii . as soon as magnetic field lines enter the corona , they start to be pushed inwards by the accreting matter . however , as one proceeds downwards in the disc , they start to be bent back again , as a consequence of the change in direction of the velocity field from inflow to backflow ( see fig . [ fig : ref_bpol ] ) . when interaction with a non - vacuum magnetosphere is included , one expects to see the field lines above and below the disc being spread out with respect to the dipole configuration . it is interesting to compare our results with ones from those calculations which show this line inflation and also resolve the region inside the disc ( e.g. @xcite ) ; it can be seen that our solutions for the magnetic field - line structure inside the disc are rather similar to those . the toroidal component of the magnetic field in our present calculations shows a simpler profile than that in paper ii ( see fig . [ fig : ref_bphi ] ) . it is positive ( dragged forwards with respect to the neutron star ) inwards of the corotation point and negative ( dragged backwards with respect to the neutron star ) outwards of that , just as envisaged by @xcite and @xcite ( i.e. as given by equation [ eq : bp_an ] when a dipole field is used for @xmath7 ) . however , we find some crucial differences with respect to those models : ( 1 ) the relative strength of @xmath0 at the two extrema is reversed and , most importantly , ( 2 ) @xmath0 decreases with @xmath19 much more slowly , and this has consequences for the torque ( see fig . [ fig : tor_eta0_4e10_b3cmp_avg ] ) . we find it convenient to introduce two magnetic reynolds numbers to describe the magnetic field structure . one , @xmath173 , is defined using a characteristic poloidal velocity and the other one , @xmath174 , uses the keplerian linear velocity at the corotation point . the former fully determines the structure of @xmath160 and the shape of @xmath0 , while the latter fixes the magnitude of @xmath0 . it is only when two configurations have the same pair of reynolds numbers that one obtains the same results for all of the three components of the magnetic field . when the profile of the velocity or of the magnetic diffusivity is changed , the magnetic reynolds numbers lose their predictive power and one should instead consider the magnetic distortion functions @xmath185 and @xmath184 ( i.e. a two - dimensional generalization of the magnetic reynolds numbers ) , meaning that configurations with the same @xmath200 but different @xmath182 give different results . as in paper ii , we find that the magnetic distortion functions determine the position of the second peak in @xmath0 ; while the first peak is connected with @xmath201 . we have also made preliminary calculations of the total torque exerted on the neutron star . we split this into two parts : ( 1 ) the torque produced by the interaction between the magnetic field and the disc @xmath202 and ( 2 ) that produced by the accreting matter , leaving the disc at the inner edge and reaching the ns surface along magnetic field lines , @xmath203 . the standard reference for calculating @xmath202 is @xcite . here we have modified that approach in two ways : ( 1 ) we drop some of the simplifying assumptions made in calculating the analytic expression for the torque ( i.e. we retain both horizontal and vertical derivatives ) and ( 2 ) for the magnetic field we use our numerical results , rather than the approximated profile of earlier models , although we recall our caveats about our present neglect of line - inflation and possible breaking of magnetic linkage beyond a certain radius in the disc . regarding the dipolar boundary conditions : in paper i we showed that the shape of the field - lines inside the disc does not depend on the chosen boundary conditions . however , the magnitude of the field is affected , and the dilution introduced by line inflation would be expected to decrease the overall torque from the outer parts of the disc . ( we will comment further on this below . ) regarding the breaking of magnetic linkage : we note that in the results of @xcite , the point where magnetic linkage is lost comes at beyond twice their co - rotation radius . the situation which they are considering is rather different from ours ( they are studying t tauri stars ) , but if one transfers this result to our case , it would correspond to loss of linkage only at the very edge of the domain used in our plots , where the magnetic torque contribution is negligible anyway . in the standard scenario developed in the 1980s and 1990s , and still widely used , the magnetic torque is taken to be proportional to @xmath0 , which in turn is taken to be proportional to the difference between the local @xmath10 of the disc matter and that of the central object , so that the disc inwards of the corotation radius gives a spin - up contribution , while the remaining part gives a spin - down . we find that this description , although rather reasonable , in fact fails to reproduce the behaviour of a simple two - dimensional disc model , even when the vertical averages of @xmath0 and the magnetic torque are considered ( see figs . [ fig : tor_eta0_4e10_b3cmp_avg ] and [ fig : torque_avg ] ) . for any given configuration , there is a vertical structure , such that at any radius there are both spin - up and spin - down regions ( at different heights in the disc , as shown in fig . [ fig : integrand_cont ] ) whose locations vary with the values of the magnetic reynolds numbers . the sign of the torque mainly follows that of the vertical derivative of the toroidal field , @xmath204 . we find similar results also when calculating the torque for the models of papers i and ii , with an s&s velocity profile . within our present model , the torque contributions from the inner part of the disc and from the accreting matter inwards of its inner edge are both positive ( acting in the sense of spinning up the neutron star ) while that from the rest of the disc can be either positive or negative . we have shown that the behaviour of the latter is complicated , and needs to be calculated with care rather than relying on the standard simplified prescriptions . however , for our present restricted model , we find that the net contribution from this region would still be negative ( acting in the direction of spin - down ) , except in extreme cases , and that it would generally dominate over the positive contributions . this can not be right for the spinning - up of millisecond pulsars , and so we infer that the action of field - line inflation and reconnections in weakening the negative torque from the outer parts of the disc probably plays a crucial role in allowing the overall spin - up . in this paper , we have extended our previous work , where we used the s&s velocity profile , by using instead the more sophisticated profile given by k&k , and we have also made calculations for the torque exerted by the disc on the neutron star . the next step of our programme of work will be to replace the present vacuum boundary conditions by a consistent treatment of the interface with a non - vacuum force - free magnetosphere . ln has been supported by a chinese academy of sciences fellowship for young international scientists ( grant number 2010y2jb12 ) . the chinese academy of sciences and the national astronomical observatory of china ( naoc ) of cas have supported this work within the silk road project ( cas grant number 2009s1 - 5 ) and it has also been partially supported by the polish ncn grant nn203381436 . 99 agapitou v. , papaloizou j. c. b. , 2000 , mnras , 317 , 273 alpar m. a. , cheng a. f. , ruderman m. a. , shaham j. , 1982 , nature , 300 , 728 bardou a. , heyvaerts j. , 1996 , a&a , 307 , 1009 campbell c. g. , 1987 , mnras , 229 , 405 campbell c. g. , 1992 , geophys . astrophys fluid dyn . , 63 , 179 elstner d. , rdiger g. , 2000 , a&a , 358 , 612 ghosh p. , lamb f. k. , 1979 , apj , 232 , 259 ghosh p. , lamb f. k. , 1979 , apj , 234 , 296 kluniak w. , kita d. , 2000 , preprint ( astro - ph/0006266 ) ( k&k ) kluniak w. , rappaport s. , 2007 , apj , 671 , 1990 lii p. , romanova m. & lovelace r. , 2012 , mnras , 420 , 2020 lovelace r.v.e . , romanova m.m . & bisnovatyi - kogan g.s . , 1995 , mnras , 275 , 244 lynden - bell d. & boily c. , 1994 , mnras , 267 , 146 miller k. a. , stone j. m. , 1997 , a&a , 489 , 890 naso l. , miller j. c. , 2010 , a&a , 521 , a31 ( paper i ) naso l. , miller j. c. , 2011 , a&a , 531 , a163 ( paper ii ) pringle j. e. , rees m. j. , 1972 , a&a , 21 , 1 radhakrishnan v. , srinivasan g. , 1982 , curr . , 51 , 1096 regev o. , gitelman l. , 2002 , a&a , 396 , 623 reyes - ruiz m. , stepinski t.f . , 1996 , apj , 459 , 653 romanova m. m. , ustyugova g. v. , koldoba a. v. & lovelace r. v. e. , 2011 , mnras , 416 , 416 shakura n. i. , sunyaev r. a. , 1973 , a&a , 24 , 337 ( s&s ) wang y .- m . , 1987 , a&a , 183 , 257 wang y .- m . , 1995 , apj , 449 , l153 zanni c. & ferreira j. , 2009 , a&a , 508 , 1117 zanni c. & ferreira j. , 2013 , a&a , 550 , a99 zhang c .- m . , kojim y. , 2006 , mnras , 366 , 137

this is the first in a series of papers studying the variable stars in large magellanic cloud globular clusters . the primary goal of this series is to better understand how the rr lyrae stars in oosterhoff - intermediate systems compare to those in oosterhoff i / ii systems . in this paper we present the results of our new time - series photometric study of ngc 1466 . a total of variables were identified in the cluster , of which are new discoveries . the variables include rrab stars , rrc s , rrd s , candidate rr lyrae , long - period variables , potential anomalous cepheid , and variables of undetermined classification . we present photometric parameters for these variables . for the rr lyrae stars physical properties derived from fourier analysis of their light curves are presented . the rr lyrae stars were used to determine a reddening - corrected distance modulus of . we discuss several different indicators of oosterhoff type and find ngc 1466 to be an oosterhoff - intermediate object .

1107.5515

studies of rr lyrae stars in the milky way globular clusters reveal what is traditionally known as the oosterhoff dichotomy ; the tendency for these clusters to be either oosterhoff i ( oo - i ) or oosterhoff ii ( oo - ii ) objects with a relatively clear zone of avoidance between these two groups . the left panel of figure 5 in @xcite strikingly illustrates the oosterhoff dichotomy in a plot of the average period of the rr lyrae stars of bailey type ab ( rrab ) versus cluster metallicity for milky way globular clusters . when one looks at the nearby dwarf galaxies and their globular clusters , one does not see an oosterhoff dichotomy as these objects fall not only into the oo - i / ii groups but also into the gap between those two groups . in fact the distribution of these extragalactic objects seems to peak in the zone of avoidance @xcite . the existence of oosterhoff - intermediate objects ( oo - int ) , objects that fall into the zone of avoidance , in the nearby dwarf galaxies poses a significant challenge to the hierarchical merger model for the formation of the milky way halo . the lack of oo - int clusters in the halo means that the objects that were accreted to form the halo could not have had properties entirely like the present - day nearby dwarf galaxies . this is the first of a series of papers focusing on variable stars in globular clusters in the large magellanic cloud ( lmc ) . the goal of this series is to better understand how the rr lyrae stars in oo - int systems compare to those in oo - i / ii systems . the lmc is an ideal place for this study because of the large number of oo - int clusters that it contains @xcite ; of the twelve lmc globular clusters that contain at least @xmath10 rrab variables , five of these clusters are oo - int ( ngc 1466 , ngc 1853 , ngc 2019 , ngc 2210 , and ngc 2257 ) . in this paper we identify , classify , and discuss the variable stars found in the lmc globular cluster ngc 1466 . subsequent papers will focus on the variable stars in other lmc globular clusters . ngc 1466 is an old globular cluster that is located relatively far from the center of the lmc . it has a metal abundance of [ fe / h ] @xmath11 , see discussion in section [ sec : abprop ] , and is not very reddened , e(b - v ) = @xmath12 @xcite . @xcite used color - magnitude diagrams obtained with the hubble space telescope to determine that the age of ngc 1466 is within @xmath6 gyr of the ages of the milky way globular clusters m3 ( ngc 5272 ) and m92 ( ngc 6341 ) . ngc 1466 features a well - populated horizontal branch that extends through the instability strip ( see figure 5 in walker 1992 ) ; thus it is expected to contain a significant number of rr lyrae stars . rr lyrae stars were first found in ngc 1466 by @xcite , and additional rr lyrae were discovered by @xcite . the most recent study of variables in ngc 1466 was conducted by @xcite , who found @xmath13 rr lyrae stars ; @xmath14 rrab and @xmath15 rrc stars . due to the density of unresolved stars in the cluster core , walker did not perform photometry of stars within a radius of @xmath16 arcsec from the cluster center , suggesting that there are probably additional rr lyrae stars that could not be detected in his study . the core radius of ngc 1466 is @xmath17 arcsec , as measured by @xcite using images from the hubble space telescope , which is entirely within the region of the cluster where walker was unable to perform photometry . advances in instrument resolution and image - subtraction techniques now make it easier for us to search for variable stars in more crowded regions . in this paper we present the results of a new search for variable stars in ngc 1466 and use the rr lyrae in ngc 1466 to confirm the distance modulus and oosterhoff classification of the cluster . we also present fourier fits to the light curves and derive physical properties for the stars from these fits . a total of @xmath18 @xmath19 and @xmath20 @xmath21 images of ngc 1466 were obtained using the soi imager on the soar @xmath22-m telescope in february and december of 2008 . an additional @xmath23 @xmath19 and @xmath24 @xmath21 images were obtained using andicam on the smarts @xmath25-m telescope operated by the smarts consortium from september 2006 to january 2007 . this represents a larger data set than the 35 bv pairs obtained by @xcite . exposure times for the soar observations were between @xmath3s and @xmath26s for @xmath19 and between @xmath27s and @xmath26s for @xmath21 , but were usually @xmath28s and @xmath29s for the @xmath19 and @xmath21 observations , respectively . smarts exposures were @xmath30s for each filter . the images from both telescopes were bias subtracted and flat - field corrected using iraf . as noted in @xcite , ngc 1466 is located near two very bright stars which create problems with scattered light across some of the images , especially in many of the images obtained with the smarts telescope . we used the procedure described in detail by @xcite to remove this scattered light ; this was the same method employed by @xcite . peter stetson s daophot ii / allstar packages @xcite were used to locate , measure , and subtract stars from all images ; this produced images that only contained the sky background and scattered light . the subtracted images were then smoothed using a @xmath18x@xmath18 pixel median filter . the average sky level in the smoothed images was determined and then the smoothed images were subtracted from the original images , removing the background light . the average sky values were then added back into the new images , creating a final set of images with a constant sky background . stetson s daophot ii / allstar packages were run on the new set of images in to order obtain instrumental magnitudes for each star . observations of the landolt standard fields pg0231 , ru149 , sa95 , and sa98 @xcite were used to determine an initial transformation from instrumental magnitudes to the standard system . @xcite showed walker s photometry to be in good agreement with hst / wfpc2 observations of ngc 1466 , after these observations were transformed from the on - board hst system to the standard johnson - cousins system . we compared the results of our initial transformation to the standard system using walker s photometry of isolated non - variable stars across a range of brightnesses within the cluster . we made small corrections to the zero points and color terms from our initial transformation equations in order to achieve better agreement with walker s photometry . this produced an average difference between our data and walker s of @xmath31 and @xmath32 . this is a similar level of scatter to that obtained by @xcite in their comparison with walker . the profile fitting photometry used in daophot ii / allstar does not work well in very crowded regions , such as the center of ngc 1466 . in order to locate variable stars in this region , the soar data on the cluster center was also searched using the isisv2.2 image subtraction program @xcite . the original images were used for the isis processing as the scattered light is less of a problem when using image subtraction techniques . the light curves produced by isis are in relative fluxes . there is a method to transform the relative fluxes obtained by isis into magnitudes in the standard system by using daophot to obtain magnitudes of the stars in the reference frame created by isis @xcite . four of our variable stars were found only by isis ( v@xmath33 , v@xmath34 , v@xmath35 , v@xmath27 ) . these stars are located deep in the cluster center and are too crowded to obtain accurate profile fitting photometry , even in the reference image . these four light curves are shown in relative fluxes ; for the other variables we present their light curves obtained from daophot ii / allstar . figure [ cmd ] shows our cmd for ngc 1466 . the cluster features a well - defined horizontal branch ( hb ) that extends across the instability strip . the majority of the hb stars are blue , with few hb stars located redward of the instability strip . as was noted by @xcite , the rr lyrae stars appear to be less luminous than the blue hb stars near the blue edge of the instability strip . as can be seen in the figure , there are some non - variable stars in the instability strip region . these are due to either poor photometry due to blending in the cluster center or are line - of - sight field stars . we used the @xmath19-band time series , which had more phase points than the @xmath21-band one , to identify candidate variable stars using peter stetson s allframe / trial package @xcite . period searches on the candidate variables were carried out using supersmoother @xcite , period04 @xcite , and a discrete fourier transform @xcite as implemented in the peranso software suite . this produced light curves which were then examined by eye to confirm the reality of each variable star candidate . light curves of potential rr lyrae stars were then fit to template light curves @xcite and checked by eye in order to confirm variable classification and period . resulting primary periods are typically good to @xmath36 or @xmath37 days while errors for the secondary periods of rrd stars are slightly worse but on the same order of magnitude . a similar search of the light curves produced with isis was done to find additional variables . lccccccccccc v1 & 03:44:44.7 & @xmath3871:40:44.9 & rrab - bl & 0.55035 & 0.64 & 0.77 & 19.380 & 19.793 & 0.421 & wa415 & 2454094.6963 + v4 & 03:44:20.3 & @xmath3871:39:57.1 & rrab - bl & 0.57239 & 0.87 & 1.16 & 19.406 & 19.804 & 0.425 & wa67 & 2454094.7100 + v5 & 03:44:26.6 & @xmath3871:39:48.9 & rrab & 0.57863 & 0.67 & 0.82 & 19.391 & 19.781 & 0.403 & wa108 & 2454510.5801 + v6 & 03:44:30.8 & @xmath3871:39:54.8 & rrab - bl & 0.52805 & 1.19 & 1.49 & 19.290 & 19.672 & 0.410 & wa182 & 2454502.2325 + v8 & 03:44:26.3 & @xmath3871:40:06.5 & rrab - bl & 0.51970 & 0.89 & 1.2 & 19.273 & 19.599 & 0.358 & wa106 & 2454094.7502 + v9 & 03:44:26.7 & @xmath3871:40:17.2 & rrab & 0.52260 & 1.24 & 1.52 & 19.414 & 19.743 & 0.362 & wa109 & 2454086.9323 + v12 & 03:44:28.9 & @xmath3871:40:15.0 & rrab & & & & & & & wa146 & + v13 & 03:44:33.8 & @xmath3871:40:48.1 & rrc & 0.33936 & 0.55 & 0.69 & 19.298 & 19.564 & 0.276 & wa262 & 2454094.5656 + v14 & 03:44:36.2 & @xmath3871:40:48.7 & rrab - bl & 0.56414 & 0.93 & 1.29 & 19.402 & 19.803 & 0.421 & wa320 & 2454501.3551 + v15 & 03:44:35.4 & @xmath3871:40:44.2 & rrab & 0.56708 & 0.81 & 1.05 & 19.209 & 19.675 & 0.487 & wa303 & 2454094.9557 + v16 & 03:44:34.9 & @xmath3871:40:45.1 & rrab & 0.69427 & 0.53 & 0.69 & 19.360 & 19.817 & 0.466 & wa290 & 2454092.6208 + v18 & 03:44:26.7 & @xmath3871:40:14.5 & ? & 0.48837 & 0.15 & 0.29 & 18.903 & 20.168 & 1.268 & wa110 & 2454510.7987 + v19 & 03:44:47.1 & @xmath3871:40:39.8 & rrab & 0.61620 & 0.61 & 0.79 & 19.360 & 19.785 & 0.435 & wa426 & 2454097.6522 + v20 & 03:44:30.7 & @xmath3871:41:22.5 & rrab & 0.59084 & 1.05 & 1.33 & 19.280 & 19.648 & 0.396 & wa183 & 2454094.6908 + v21 & 03:44:51.9 & @xmath3871:39:53.4 & rrab & 0.52150 & 1.22 & 1.54 & 19.315 & 19.632 & 0.359 & wa448 & 2454048.0181 + v22 & 03:44:35.5 & @xmath3871:41:05.7 & rrc & 0.33081 & 0.51 & 0.65 & 19.200 & 19.543 & 0.353 & wa301 & 2454094.7036 + v23 & 03:44:38.6 & @xmath3871:41:24.7 & rrab & 0.49342 & 1.25 & 1.51 & 19.291 & 19.681 & 0.421 & wa361 & 2454094.4714 + v26 & 03:44:34.2 & @xmath3871:40:51.4 & rrab & 0.51364 & 1.03 & 1.38 & 19.339 & 19.702 & 0.400 & wa274 & 2454094.7460 + v28 & 03:44:21.3 & @xmath3871:40:38.1 & rrc & 0.32039 & 0.57 & 0.7 & 19.338 & 19.652 & 0.324 & wa71 & 2454094.7532 + v30 & 03:44:31.3 & @xmath3871:39:52.0 & rrab & 0.69445 & 0.66 & 0.91 & 19.273 & 19.677 & 0.421 & wa194 & 2454095.2544 + v31 & 03:44:32.1 & @xmath3871:39:50.5 & ? & 0.34272 & & & & & & & 2454510.6763 + v32 & 03:44:29.4 & @xmath3871:40:13.7 & ac ? & 0.49661 & 0.42 & 0.70 & 18.118 & 18.614 & 0.505 & & 2454094.7437 + v33 & 03:44:29.0 & @xmath3871:40:43.9 & rrc & 0.27841 & 0.51 & 0.57 & 19.355 & 19.600 & 0.247 & wa148 & 2454094.8536 + v34 & 03:44:28.0 & @xmath3871:39:60.0 & ? & 0.61594 & 0.31 & 0.45 & 18.194 & 18.896 & 0.709 & wa129 & 2454040.9638 + v35 & 03:44:29.7 & @xmath3871:40:42.5 & rrab - bl & 0.58172 & 0.6 & 0.78 & 19.248 & 19.668 & 0.432 & wa164 & 2454094.6090 + v38 & 03:44:49.0 & @xmath3871:40:48.7 & rrc & 0.34858 & 0.44 & 0.54 & 19.354 & 19.682 & 0.335 & wa437 & 2454092.5444 + v39 & 03:44:30.6 & @xmath3871:40:07.2 & rrab & 0.56248 & 1.12 & 1.37 & 19.156 & 19.538 & 0.407 & wa179 & 2454510.8795 + v40 & 03:44:31.0 & @xmath3871:40:01.1 & rrab & 0.53606 & & & & & & wa185 & 2454092.8046 + v41 & 03:44:39.4 & @xmath3871:40:21.3 & rr ? & & & 0.59 & & 19.598 & & & + v42 & 03:44:30.4 & @xmath3871:41:00.6 & rrab & 0.63482 & 0.51 & 0.66 & 19.384 & 19.865 & 0.489 & wa177 & 2454094.6637 + v44 & 03:44:22.0 & @xmath3871:38:13.1 & rrab & & & & & & & wa75,s27 & + v45 & 03:44:23.1 & @xmath3871:41:16.0 & rrc & 0.31043 & 0.38 & 0.49 & 19.233 & 19.503 & 0.275 & wa82,s27 & 2454094.4913 + v46 & 03:44:27.4 & @xmath3871:39:14.0 & rrab & 0.68909 & 0.38 & 0.49 & 19.352 & 19.813 & 0.466 & wa121 & 2454094.3662 + v48 & 03:44:29.4 & @xmath3871:40:00.0 & rrc & 0.37697 & 0.44 & 0.54 & 19.288 & 19.580 & 0.316 & wa153,s3 & 2454092.6531 + v49 & 03:44:30.2 & @xmath3871:40:09.5 & rrab & 0.68635 & 0.53 & 0.7 & 19.255 & 19.773 & 0.528 & & 2454510.2159 + v50 & 03:44:31.1 & @xmath3871:40:28.7 & rrab - bl & 0.58194 & 1.05 & 1.24 & 19.444 & 19.778 & 0.356 & & 2454092.5885 + v51 & 03:44:31.6 & @xmath3871:40:12.7 & ? & 0.35892 & & & & & & & 2454410.5428 + v52 & 03:44:31.6 & @xmath3871:40:34.6 & rrc & & & & & & & & + v53 & 03:44:31.9 & @xmath3871:40:15.8 & lp & & & & & & & & + v54 & 03:44:32.1 & @xmath3871:40:12.9 & ? & 0.55608 & & & & & & & 2454510.3605 + v55 & 03:44:32.5 & @xmath3871:40:13.3 & rrab - bl & 0.60951 & & & & & & & 2454510.6458 + v56 & 03:44:33.4 & @xmath3871:40:01.8 & rrab & 0.62238 & 0.73 & 0.96 & 19.189 & 19.701 & 0.527 & & 2454510.5354 + v57 & 03:44:33.5 & @xmath3871:39:51.7 & rrc & 0.35692 & 0.42 & 0.5 & 19.295 & 19.685 & 0.395 & wa249,s29 & 2454094.6595 + v58 & 03:44:33.5 & @xmath3871:39:59.0 & rrab & 0.56752 & 1.05 & 1.28 & 19.368 & 19.905 & 0.570 & & 2454510.4638 + v59 & 03:44:33.7 & @xmath3871:40:20.2 & ? & & 0.46 & 0.54 & 18.194 & 18.651 & 0.460 & & 2454511.0736 + v60 & 03:44:33.9 & @xmath3871:40:11.4 & ac?,rrab ? & 0.52210 & & & & & & & 2454510.1325 + v61 & 03:44:33.9 & @xmath3871:40:01.7 & rrab - bl & 0.66927 & 0.86 & 1.13 & 19.510 & 19.985 & 0.497 & & 2454510.6914 + v62 & 03:44:35.5 & @xmath3871:40:35.8 & ? & 0.50414 & 0.71 & 1.08 & 18.758 & 19.266 & 0.531 & & 2454094.2817 + v63 & 03:44:35.6 & @xmath3871:39:54.0 & rrab & 0.58760 & 0.82 & 0.99 & 19.415 & 19.890 & 0.490 & wa306,s31 & 2454510.6730 + v64 & 03:44:38.4 & @xmath3871:40:39.5 & rrab & 0.68767 & 0.49 & 0.59 & 19.307 & 19.809 & 0.508 & wa356,s30 & 2454092.8083 + v65 & 03:44:39.0 & @xmath3871:40:03.7 & lp & & & & & & & & + v66 & 03:44:40.0 & @xmath3871:41:10.3 & rrc & 0.33633 & 0.52 & 0.67 & 19.277 & 19.608 & 0.342 & wa382,s17 & 2454094.7809 + v67 & 03:44:45.1 & @xmath3871:42:13.1 & ? & & & & & & & & + v68 & 03:44:56.5 & @xmath3871:38:54.3 & rrc & 0.34914 & 0.47 & 0.65 & 19.410 & 19.749 & 0.348 & wa458 & 2454040.8262 + [ vartable ] a total of @xmath1 variables were found , including @xmath39 rr lyraes , @xmath6 additional candidate rr lyrae , @xmath7 long - period variables , a candidate anomalous cepheid , and @xmath8 variables of unknown classification . we were able for the first time to identify double - mode ( rrd ) variables among the rr lyrae population of ngc 1466 . of the confirmed rr lyrae stars , @xmath3 were rrab stars , @xmath4 were rrc stars , and @xmath5 were rrd s . the rrd variable stars are discussed separately in section [ sec : rrd ] . figures [ abcurves]-[othercurves ] show the light curves for the rrab , rrc , and the other variable stars . table [ vartable ] lists the identified variable stars , except for the rrd stars , as well as their classification , period , @xmath19 and @xmath21 amplitudes , intensity - weighted @xmath19 and @xmath21 mean magnitudes , and magnitude - weighted mean @xmath40 color . the intensity - weighted mean magnitudes and the magnitude - weighted mean color were obtained through the fitting of the light curves with template light curves @xcite . ( for the relation between these average quantities and the color of the equivalent static star , the reader is referred to bono et al . notes on some of the individual stars are in the following subsections . table [ phottable ] contains the photometric data for the variable stars . lccccc v01 & @xmath19 & @xmath41 & @xmath42 & @xmath43 & @xmath44 + v01 & @xmath19 & @xmath45 & @xmath46 & @xmath47 & @xmath48 + v01 & @xmath19 & @xmath49 & @xmath50 & @xmath51 & @xmath52 + v01 & @xmath19 & @xmath53 & @xmath54 & @xmath55 & @xmath56 + v01 & @xmath19 & @xmath57 & @xmath58 & @xmath59 & @xmath60 + [ phottable ] we use a naming system that is an extension of the one used in @xcite . walker identified the variable stars found by wesselink by their number in that paper , preceded by a `` we '' . however , walker did not use an extension of wesselink s numbering scheme for the new variables that he discovered . instead , he denoted the new variables by their number in his photometry table 3 . it is , however , convenient to have an identification system that refers only to the known variable stars in ngc 1466 . therefore we have extended wesselink s numbering system , adding the additional variables that are not included in @xcite to the end of wesselink s orignal list of variables . all variables are named in the form of vxx , with stars up through v@xmath13 being ones that were originally found by wesselink . a cross identification of the original walker numbers and our variable star numbers is included in table [ vartable ] . figures [ findout ] and [ findin ] show the locations of all of the variable stars listed in the table . we found @xmath61 of walker s ( 1992 ) @xmath13 rr lyrae stars along with @xmath62 new rrab stars , @xmath6 new rrc , and @xmath7 new rrd stars . six of the previously identified rr lyrae stars in ngc 1466 are reclassified as rrd stars . we did not find walker s variable @xmath63 , which appears to be a blend of two stars in our images , neither of which display variability . the additional rr lyrae stars are all located in the central region of the cluster , which was not searched by walker due to crowding . the rrab stars have intensity weighted mean magnitudes of @xmath64 and @xmath65 , while the mean magnitudes for the rrc stars are @xmath66 and @xmath67 . walker found @xmath68 and @xmath69 for the rrab stars while the rrcs had @xmath70 and @xmath71 , which are in good agreement with our results . in general our results agreed with those of walker for the stars that appeared in both of our lists , with the exceptions noted below . walker @xmath72 , v@xmath73 , is listed as an unclassified variable for reasons that are stated in the next subsection . we were unable to determine periods for walker s ( 1992 ) variables @xmath74 ( v@xmath75 ) and @xmath76 ( v@xmath77 ) due to not having enough data points for accurate period determination . these two rrab stars were not detected in all images , but the shape of the partial light curves are entirely consistent with their classification as rrab stars . v@xmath61 was classified only as a candidate rr lyrae star due to aliasing issues which prevented us from determining a period or classification as rrab or rrc . @xcite was unable to measure this star due to its proximity to a brighter star . no significant differences between our periods and those calculated by @xcite were found . v@xmath78 was the only star that did not turn out to be a misclassified rrd to display a difference in period greater than @xmath79 days . walker found a period of @xmath80 days while we found a period of @xmath81 days . we ran a period search on walker s light curve for this star and found that there is a secondary signal for a period around @xmath82 days , suggesting that this disagreement in periods is due to aliasing in walker s light curve . the limited accuracy of the periods determined by both walker and ourselves prevents the identification of any evolutionary rr lyrae period changes . we did not obtain enough observations to definitively determine if any of the rr lyrae stars have the blazhko effect . however , based on the scatter in their light curves , v@xmath6 , v@xmath22 , v@xmath62 , v@xmath5 , v@xmath83 , v@xmath84 , and v@xmath85 are potential blazhko stars . v@xmath86 and v@xmath87 also show scatter in their light curves that could be due to the blazhko effect , but these two stars are located in the crowded region near the cluster center so the scatter could also be due to contamination . v@xmath18 features a light curve that has a fair amount of scatter as well as a shape that is unusually symmetric for an rrab star . despite its unusual shape , the light curve was best fit with a period of @xmath88 days and thus is tentatively classified as an rrab . the increased number of observations ( over wesselink 1971 and walker 1992 ) allowed for the identification of eight rrd stars in ngc 1466 . table [ rrdtable ] lists the identified rrd stars , their fundamental mode and first overtone periods and amplitudes , and their period ratios . fits to the two periods were calculated using the period04 program @xcite . the secondary periods are typically good to @xmath89 or @xmath90 days while the primarly periods are typically slightly more accurate . lccccccccccccccc v2 & 03:44:27.4 & @xmath3871:40:59.7 & rrd & 0.4751 & 0.3534 & 0.7451 & 0.12 & 0.42 & 0.15&0.53 & 19.375 & 19.758 & 0.389 & wa124 & 2454094.4975 + v3 & 03:44:20.2 & @xmath3871:40:00.2 & rrd & 0.4717 & 0.3505 & 0.7431 & 0.13 & 0.41 & 0.15 & 0.53 & 19.368 & 19.779 & 0.416 & wa66 & 2454021.7034 + v7 & 03:44:28.4 & @xmath3871:39:13.3 & rrd&0.4859 & 0.3615 & 0.7440 & 0.35 & 0.43 & 0.42 & 0.52 & 19.326 & 19.642 & 0.310 & wa135 & 2454094.4294 + v10 & 03:44:27.4 & @xmath3871:40:19.0 & rrd&0.4730 & 0.3518 & 0.7437 & 0.26 & 0.48 & 0.30 & 0.52 & 19.425 & 19.797 & 0.370 & & 2454094.5513 + v24 & 03:44:19.9 & @xmath3871:41:25.5 & rrd & 0.4938 & 0.3675 & 0.7442 & 0.12 & 0.43 & 0.15 & 0.60 & 19.292 & 19.642 & 0.357 & wa65 & 2454094.4549 + v25 & 03:44:36.4 & @xmath3871:39:53.9 & rrd&0.4715 & 0.3508 & 0.7440 & 0.20 & 0.44 & 0.40 & 0.65 & 19.360 & 19.705 & 0.350 & wa323 & 2454029.6612 + v29 & 03:44:15.5 & @xmath3871:40:11.9 & rrd&0.4824 & 0.3589 & 0.7440 & 0.45 & 0.58 & 0.36 & 0.38 & 19.428 & 19.750 & 0.320 & wa53 & 2454084.3778 + v47 & 03:44:28.9 & @xmath3871:40:27.8 & rrd&0.4733 & 0.3519 & 0.7435 & 0.42 & 0.58 & 0.48 & 0.60 & 19.410 & 19.760 & 0.350 & & 2454094.6365 + [ rrdtable ] six of the rr lyrae identified by @xcite turned out to be rrd stars . five were originally identified as rrc stars and one as an rrab . v@xmath7 , v@xmath91 , v@xmath92 , and v@xmath14 all have first overtone periods that agree with the rrc period that was found by walker . v@xmath93 was previously classified as an rrc with a period of @xmath94 days ; however , we obtained a first overtone period of @xmath95 days for it . walker classified v@xmath96 as an rrab star with a period of @xmath97 days , which does not correspond with either the fundamental or first overtone period that we found . of the two new rrd stars , v@xmath98 was thought to be a potential variable by @xcite . @xcite did not classify this star as a variable due to its proximity to a blue hb star , which made photometry difficult . figure [ peterson ] shows the ratio of the first overtone period ( @xmath99 ) to the fundamental mode period ( @xmath99 ) vs fundamental period ( petersen diagram ) for the rrd stars in ngc 1466 ; the rrd stars in the lmc field @xcite are also plotted . the ngc 1466 rrd stars fall in the same area of the petersen diagram as the majority of the lmc field stars . this is the same region where rrd stars in milky way oosterhoff - i clusters tend to fall ( see , e.g. , fig . 1 in @xcite , and fig . 13 in @xcite ) . two long - period variables , v@xmath100 and v@xmath101 , were found in our study of ngc 1466 . we were unable to determine periods for either of these stars ; in fact , we do not have enough light curve information to say for sure that their variability is periodic . both stars are located near the tip of the giant branch on the cmd , as can be seen in figure [ cmd ] , and have several possible periods in the range of @xmath27 to @xmath102 days . v@xmath103 appears to be a potential anomalous cepheid ( ac ) , having an average @xmath19 luminosity that is approximately @xmath104 magnitudes brighter in @xmath19 than the rr lyrae stars . the star has a potential period of either @xmath105 days or @xmath106 days . the @xmath106 day period would be consistent with the period - luminosity diagram for acs from @xcite . based on equations @xmath7 through @xmath10 in @xcite and the distance modulus that we derive for ngc 1466 , v@xmath103 would most likely be a first overtone pulsator . if the @xmath105 day period is correct , the star would be too bright to be an ac , based on the period - luminosity diagram , and would most likely be a foreground rr lyrae . the shape of v@xmath103 s light curve is similar to an rrab star , which would be inconsistent with a @xmath105 day period . based on this we feel that the @xmath106 day period is correct and that this star is either an ac or , less likely , a foreground rrab . nine variables of unknown classification were found . @xmath22 of these stars were only found by isis as they were either located in the crowded cluster center ( v@xmath33 , v@xmath34 , and v@xmath35 ) or near the bright star in the southeast corner of the images ( v@xmath107 ) . thus their light curves are in differential fluxes , which does not allow us to obtain positive classifications for them . while periods and light curve shape can be used to determine potential classifications , without having a magnitude we can not distinguish between a cluster rr lyrae , a foreground rr lyrae , or an ac , for example . notes on the other stars with unknown classification follow . - v@xmath73 ( walker @xmath72 ) is slightly brighter than the horizontal branch but is very red , @xmath108 . it has a period of @xmath109 days and an amplitude of @xmath110 mag in @xmath19 . @xcite noted that in his images this star had an elongated shape that suggested a likely unresolved companion . our images also show this star to be elongated . the hst images of @xcite were checked in order to investigate the possibility of a close companion , however v@xmath73 did not fall within the hst field . - v@xmath111 ( walker @xmath112 ) is about one magnitude brighter than the rr lyrae level and has a @xmath113 . @xcite had labeled this star as a variable but @xcite considered it a constant rhb star . the period , @xmath114 days , and light curve shape of this star suggest that it could be an rrab star whose color and brightness are affected by contamination . - star v@xmath115 is located in the cluster center and has an isis light curve that suggests several possible periods . its daophot light curves suggest that it has a similar color to the rr lyrae stars but is about one magnitude brighter ; this is possibly due to contamination . our images and the hst images of @xcite confirm the existence of neighbor stars that could be effecting the photometry of v@xmath115 through blending . however , v@xmath115 displays a @xmath21 to @xmath19 amplitude ratio that is similar to the other rr lyraes in the cluster ( fig . [ bvamp ] ) . v@xmath115 could also potentially be an ac or a type ii cepheid . - v@xmath27 is located in the crowded cluster center and has a clean isis light curve that suggests a period of @xmath116 days . its light curve from daophot suggests a luminosity approximately two magnitudes brighter than the rr lyrae stars but the light curve is messy , possibly due to blending . - star v@xmath1 is located in the cluster center and has a light curve that suggests that it is a variable with a period near @xmath117 days . the period and the light curve shape suggest that v@xmath1 is possibly an rr lyrae but the light curve is messy and it is brighter than the other rr lyrae stars , likely due to contamination by nearby stars . the contamination explanation is supported by the fact that v@xmath1 has a @xmath21 amplitude that is slightly too large large for its @xmath19 amplitude , compared to the rr lyrae stars in ngc 1466 . the hst images of @xcite confirm a close neighbor for v@xmath1 which does not appear as a separate star on our images . the rrab light curves were fit with a fourier series of the form @xmath118 where @xmath119 is the magnitude as a function of time and @xmath120 , where @xmath121 is the pulsation period . the rrc light curves were fit in a similar fashion but a cosine series was used instead of the sine series . fourier decomposition was not carried out on the rrd stars as after we deconvolved their light curves there were not enough data points for accurate fourier analysis . the fourier fitting was done using a fourier fitting code originally written by geza kovcs with an analysis wrapper written by nathan de lee . error analysis for the fits was done based on @xcite . a number of fits were performed for each light curve , with a varying number of terms , and the resulting fits were plotted and then checked by eye to determine if it was a successful fit . not all light curves were fit successfully , and the optimum number of terms to fit each individual light curve varied . @xmath16 rrab and @xmath93 rrc light curves were successfully fit . tables [ abcoeff ] and [ ccoeff ] give the amplitude ratios , @xmath122 , and phase differences , @xmath123 , for the low - order terms ( @xmath124 ) , and the number terms used in the fourier fit , for the rrab and rrc stars , respectively . for the rrab stars , we also list the jurcsik - kovcs @xmath125 value @xcite . this value helps to separate rrab stars with `` regular '' light curves from those with `` anomalous '' light curves , with lower @xmath125 values representing more `` regular '' light curves . lccccccccc v4 & 0.346 & 0.498 & 0.301 & 0.155 & 2.16 & 4.77@xmath1260.18 & 1.12 & 30.46 & 8 + v5 & 0.243 & 0.380 & 0.361 & 0.206 & 2.31 & 5.12@xmath1260.21 & 1.69 & 49.89 & 8 + v6 & 0.422 & 0.322 & 0.303 & 0.174 & 2.68 & 5.30@xmath1260.27 & 1.83 & 9.62 & 6 + v8 & 0.349 & 0.402 & 0.324 & 0.194 & 2.47 & 5.02@xmath1260.37 & 1.39 & 40.96 & 7 + v9 & 0.463 & 0.416 & 0.244 & 0.099 & 2.30 & 4.52@xmath1260.17 & 1.10 & 7.63 & 7 + v14 & 0.355 & 0.440 & 0.351 & 0.205 & 2.25 & 4.84@xmath1260.14 & 1.34 & 14.33 & 9 + v19 & 0.252 & 0.432 & 0.177 & 0.117 & 1.94 & 4.93@xmath1260.37 & 1.17 & 6.53 & 7 + v20 & 0.363 & 0.502 & 0.356 & 0.271 & 2.26 & 4.99@xmath1260.09 & 1.26 & 45.97 & 6 + v23 & 0.424 & 0.456 & 0.361 & 0.237 & 2.25 & 4.66@xmath1260.08 & 1.04 & 47.39 & 10 + v26 & 0.367 & 0.369 & 0.249 & 0.139 & 2.26 & 4.80@xmath1260.19 & 0.78 & 34.99 & 7 + v30 & 0.232 & 0.457 & 0.305 & 0.153 & 2.36 & 5.22@xmath1260.34 & 1.93 & 23.20 & 10 + v42 & 0.204 & 0.411 & 0.251 & 0.098 & 2.46 & 5.51@xmath1260.16 & 2.20 & 10.83 & 10 + v56 & 0.276 & 0.400 & 0.282 & 0.135 & 2.50 & 5.38@xmath1260.12 & 1.67 & 5.31 & 8 + [ abcoeff ] lcccccccc v22 & 0.272 & 0.179 & 0.085 & 0.067 & 4.67 & 2.55@xmath1260.30 & 0.89 & 6 + v28 & 0.276 & 0.252 & 0.103 & 0.126 & 4.73 & 3.15@xmath1260.37 & 1.55 & 6 + v33 & 0.246 & 0.255 & 0.039 & 0.011 & 4.38 & 2.72@xmath1260.75 & 4.95 & 8 + v38 & 0.228 & 0.114 & 0.064 & 0.028 & 4.58 & 4.16@xmath1260.57 & 1.74 & 6 + v45 & 0.199 & 0.090 & 0.042 & 0.055 & 4.41 & 2.71@xmath1260.79 & 1.53 & 6 + v57 & 0.222 & 0.139 & 0.080 & 0.045 & 4.71 & 3.97@xmath1260.60 & 2.01 & 7 + v66 & 0.259 & 0.181 & 0.090 & 0.033 & 4.30 & 3.93@xmath1260.54 & 3.24 & 9 + [ ccoeff ] @xcite , @xcite , and kovcs & walker ( 1999 , 2001 ) provide formulae , determined empirically , for calculating the metallicity , absolute magnitude , and effective temperature of rrab stars using the fourier coefficients from their @xmath19 light curves . [ fe / h ] , @xmath127 , @xmath128 , and @xmath129 are determined using the following equations , which come from equations ( 1),(2),(5 ) , and ( 11 ) in @xcite , respectively : @xmath130 } = -5.038 - 5.394\,p+1.345\,\phi_{31 } , \ ] ] @xmath131 @xmath132 and @xmath133}.\ ] ] the values for [ fe / h ] derived through this method are in the scale of @xcite , which can be transformed to the more common @xcite scale by the relation @xmath134_{j95 } } = 1.431{\rm [ fe / h]_{zw84}}+0.880 $ ] @xcite . the dereddened @xmath40 and @xmath135 color indices were calculated using equations ( 6 ) and ( 9 ) from @xcite : @xmath136 and @xmath137 these color indices are also used to calculate effective temperatures via equations ( 11 ) and ( 12 ) in @xcite , @xmath138\end{gathered}\ ] ] and @xmath139 , \end{gathered}\ ] ] where @xmath140 is obtained from equation ( 12 ) in @xcite , @xmath141 assuming a mass of @xmath142 . the resulting physical properties for the rrab stars are given in table [ abphysical ] . the mean metallicity of the rrab stars is @xmath134_{j95}}=-1.40\pm0.09 $ ] which in the zinn & west scale is @xmath134_{zw84}}=-1.59\pm0.06.$ ] previously published metallicities for ngc 1466 showed a wide range of values . @xcite reported a metallicity of @xmath134}=-2.17 $ ] based on spectroscopic observations of the calcium triplet in two giant stars in the cluster whose individual metallicities were @xmath134}=-1.85 $ ] and @xmath143 . @xcite argued that the more metal - poor of these two stars was not a cluster member based on photometric results . walker instead reported that the cluster metallicity was @xmath134}=-1.85 $ ] , based on the remaining star from olszewski et al . and comparisons of the rr lyrae stars and the cluster cmd to that of other clusters . @xcite used stellar popular synthesis models to find metallicities of @xmath134}=-1.30 $ ] and @xmath144 , depending on whether they used the full spectrum or just the cn spectrum . a more metal poor result of @xmath134}=-2.25 $ ] was adopted by @xcite for their work with cluster surface brightness profiles . a far more metal - rich value of @xmath134}=-1.64 $ ] was found by @xcite using the equivalent widths of metal lines in integrated spectra . our value of @xmath134}=-1.59 $ ] is most consistent with the santos & piatti result . @xcite demonstrated that the fourier parameters of the @xmath19-band light curves of rrc stars could be used to calculate various physical properties for these stars . unlike the empirically determined relationships between physical properties and fourier parameters for rrab stars , the relationships for rrc stars were determined using light curves that were created with hydrodynamic pulsation models . the mass , luminosity , temperature , and helium parameter of rrc stars can be calculated using the equations : @xmath145 @xmath146 @xmath147 and @xmath148 which are equations ( 2),(3),(6 ) , and ( 7 ) in @xcite , respectively . it should be noted that the helium abundance parameter , @xmath149 , is not equal to the helium abundance , @xmath150 . equation ( 10 ) in @xcite , @xmath151 allows us to calculate the absolute magnitude for the rrc stars . the metallicity in the @xcite scale can be calculated using equation ( 3 ) from @xcite : @xmath152_{zw84}}=52.466\,p^{2}-30.075\,p+0.131\,\phi_{31}^{2}\\+0.982\,\phi_{31}-4.198\,\phi_{31}p+2.424.\end{gathered}\ ] ] it should be noted that while these equations are commonly used in the literature , the equations for mass , luminosity , and temperature violate the pulsation equation @xcite . while the values calculated for these properties can not all be physically correct , they can be used for the purpose of making comparisons with the physical properties calculated the same way for rrc stars in other clusters . the calculated physical properties for the rrc stars are given in table [ cphysical ] . lcccccccc v4 & -1.717 & 0.748 & 1.164 & 3.805 & 0.336 & 3.810 & 0.490 & 3.811 + v5 & -1.278 & 0.824 & 1.153 & 3.805 & 0.365 & 3.804 & 0.526 & 3.802 + v6 & -0.764 & 0.828 & 1.008 & 3.819 & 0.313 & 3.825 & 0.461 & 3.818 + v8 & -1.085 & 0.846 & 1.054 & 3.815 & 0.330 & 3.817 & 0.482 & 3.813 + v9 & -1.774 & 0.736 & 1.125 & 3.810 & 0.295 & 3.824 & 0.438 & 3.825 + v14 & -1.577 & 0.762 & 1.149 & 3.806 & 0.338 & 3.811 & 0.493 & 3.811 + v19 & -1.725 & 0.749 & 1.209 & 3.800 & 0.354 & 3.804 & 0.513 & 3.805 + v20 & -1.511 & 0.737 & 1.139 & 3.807 & 0.341 & 3.810 & 0.496 & 3.810 + v23 & -1.434 & 0.810 & 1.064 & 3.815 & 0.314 & 3.821 & 0.462 & 3.819 + v26 & -1.359 & 0.823 & 1.057 & 3.816 & 0.317 & 3.820 & 0.466 & 3.818 + v30 & -1.759 & 0.679 & 1.292 & 3.791 & 0.378 & 3.795 & 0.544 & 3.797 + v42 & -1.056 & 0.805 & 1.175 & 3.802 & 0.374 & 3.802 & 0.538 & 3.799 + v56 & -1.155 & 0.775 & 1.136 & 3.806 & 0.357 & 3.807 & 0.517 & 3.804 + mean & -1.400@xmath1260.088 & 0.779@xmath1260.013 & 1.133@xmath1260.021 & 3.808@xmath1260.002 & 0.339@xmath1260.007 & 3.812@xmath1260.003 & 0.494@xmath1260.009 & 3.810@xmath1260.002 + [ abphysical ] lcccccc v22 & -1.970 & 0.657 & 0.724 & 1.762 & 3.859 & 0.254 + v28 & -1.669 & 0.591 & 0.611 & 1.713 & 3.864 & 0.272 + v33 & -1.420 & 0.788 & 0.633 & 1.675 & 3.871 & 0.281 + v38 & -1.419 & 0.696 & 0.495 & 1.693 & 3.864 & 0.284 + v45 & -1.764 & 0.720 & 0.672 & 1.724 & 3.864 & 0.266 + v57 & -1.614 & 0.666 & 0.526 & 1.715 & 3.862 & 0.276 + v66 & -1.425 & 0.711 & 0.515 & 1.690 & 3.865 & 0.283 + mean & -1.612@xmath1260.079 & 0.690@xmath1260.023 & 0.597@xmath1260.033 & 1.710@xmath1260.011 & 3.86@xmath1260.001 & 0.274@xmath1260.004 + [ cphysical ] it has been noted in the literature that rrc masses calculated using the @xcite equations can at time be too small,@xmath153 , approaching the mass of a degenerate helium core at the helium flash @xcite . table [ cphysical ] shows that this occurs for three of our stars . this is likely due to a problem with the equations used but still produces values that are suitable for comparison purposes . the mean metallicity of the rrc stars is @xmath134_{zw84}}=-1.61\pm0.08 $ ] , which is in good agreement with the value obtained for the rrab stars . from the fourier fits we obtain a mean @xmath19 absolute magnitude of @xmath154 for the rrc stars and @xmath155 for the rrab stars . the zero points used for these calibrations are different and there is some debate as to which one , if either , is correct . instead we use the calibration of the absolute magnitude - metallicity relation for rr lyrae stars by @xcite . the average of our fourier derived metallicities for the rrab and rrc stars is @xmath134_{zw84}}=-1.60\pm0.05 $ ] , which gives us an absolute magnitude of @xmath156 . the average apparent magnitude of the rrab and rrc stars are @xmath64 and @xmath66 , respectively . since the average magnitude for the rrab and rrc stars are essentially the same , we combine them for the purpose of calculating the distance , resulting in an average magnitude of @xmath157 . using the reddening value of @xmath158 from @xcite and a standard extinction law with @xmath159 , we obtain a reddening - corrected distance modulus of @xmath160 . this distance modulus is approximately equal to the distance modulus of @xmath161 that @xcite derived for the lmc . the average periods for the rr lyrae stars in ngc 1466 are @xmath162 days and @xmath163 days . we found @xmath3 rrab , @xmath4 rrc , and @xmath5 rrd stars , giving the cluster a @xmath164 . all three of these values are consistent with an oosterhoff - intermediate ( oo - int ) classification for ngc 1466 . the minimum period for rrab stars has also been shown to be a good indicator of oosterhoff class . ngc 1466 has a minimum period for rrab stars of @xmath165 days , which is consistent with it being an oo - int object ( catelan et al . 2011 , in preparation ) . -band amplitude ( top panel ) , for the rr lyrae stars in ngc 1466 . solid red lines and dashed green lines indicate the typical position for rr lyrae stars in oosterhoff i and oosterhoff ii clusters , respectively @xcite . rrab stars for which we obtained a value of @xmath166 are indicated with filled circles , those with a @xmath167 are indicated with open circles , and the rrab stars for which we were unable to obtain a good fourier fit are indicated by crosses . the bottom panel is the same as the top panel but using the @xmath21-band amplitudes.,title="fig:",scaledwidth=45.0% ] -band amplitude ( top panel ) , for the rr lyrae stars in ngc 1466 . solid red lines and dashed green lines indicate the typical position for rr lyrae stars in oosterhoff i and oosterhoff ii clusters , respectively @xcite . rrab stars for which we obtained a value of @xmath166 are indicated with filled circles , those with a @xmath167 are indicated with open circles , and the rrab stars for which we were unable to obtain a good fourier fit are indicated by crosses . the bottom panel is the same as the top panel but using the @xmath21-band amplitudes.,title="fig:",scaledwidth=45.0% ] another indicator of oosterhoff status is the position of the rr lyrae stars in the bailey period - amplitude diagram . figure [ vperamp ] shows the diagrams for ngc 1466 using both the @xmath19 and @xmath21-band amplitudes . the typical location of the rrab and rrc stars in oosterhoff i ( oo - i ) and ii ( oo - ii ) clusters are indicated by solid red lines and dashed green lines , respectively @xcite . in both diagrams the rrab stars display a wide scatter . many are located between the oo - i and o - ii loci , which is consistent with an oo - int object . however , there is a significant number of rrab stars that are clustered around the oo - i line , which is more typical for an oo - i object . the rrc stars fall almost exclusively between the reference lines , as is expected for an oo - int object . we found an hb type of @xmath168 for ngc 1466 . this was determined using the stars in an annulus with an inner radius of @xmath98 arcsec and an outer radius of @xmath74 arcsec ; this represented a region for which we felt our sample of the hb stars was complete . this hb type , combined with the metallicity of @xmath134}\approx-1.60 $ ] that we obtained from the rr lyrae stars , places the cluster outside of the `` forbidden region '' where oo - int objects tend to fall in the hb type - metallicity diagram ( fig . 7 in catelan 2009 ) . if one used a lower metallicity similar to what was favored by @xcite or @xcite , ngc 1466 would be located much closer to the forbidden region . a lower metallicity would also be consistent with the theoretical predictions by @xcite for oo - int behavior , whereas the higher metallicity would indicate an hb morphology that is too red with respect to the predictions of the models ( see fig . 8 in catelan 2009 ) . figure [ hist ] shows the distribution of both the raw and the fundamentalized periods for the rr lyrae stars in ngc 1466 . a sharply peaked fundamentalized period distribution is not present , in contrast to what was found in @xcite in a number of ooi and ooii clusters . the distribution of the raw periods for the rrab stars displays two peaks , a broad one around @xmath169 days and a sharper peak at @xmath170 days . this is similar to the distribution seen in ngc 1835 @xcite although the shorter period peak for ngc 1466 is broader and its longer period peak occurs at a slightly longer period . soszyski suggested that this two - peaked distribution may be an indicator of a sum of oo - i and oo - ii characteristics , but our bailey diagram shows no evidence of an oo - ii population . tables 7 and 8 in @xcite show the mean physical parameters for rrc and rrab stars , respectively , in a selection of previously studied globular clusters . the parameters that we calculated for ngc 1466 fall between those for oo - i and oo - ii clusters , consistent with ngc 1466 having an oo - int classification . a more thorough comparison of the physical parameters for the rr lyrae stars in ngc 1466 and the other globular clusters in our study will be the subject of a future paper in this series . a photometric study of the lmc globular cluster ngc 1466 was conducted in order to locate variable stars , with a specific interest in the cluster s rr lyrae population . @xmath39 rr lyrae stars and @xmath6 candidate rr lyrae were found . we discovered @xmath8 rr lyrae stars which were not found by @xcite , previously the most extensive search for rr lyrae in ngc 1466 . of our @xmath39 rr lyrae stars , @xmath5 are rrd stars , making this the first time double - mode rr lyrae stars were found in this cluster . a candidate anomalous cepheid variable was found in ngc 1466 . acs are rare in globular clusters , having previously been found only in ngc 1786 , ngc 5466 , and @xmath171 cen @xcite . interestingly , these three systems are ooii , though this of course may be just a coincidence , particularly since it is believed that acs in globular clusters should be the progeny of blue straggler stars ( e.g. , sills et al . 2009 and references therein ) , which in principle should not be related to the oosterhoff dichotomy . we performed fourier analysis on the rr lyrae light curves and used the resulting fourier parameters to calculate physical properties for the stars which will be compared to the properties of rr lyrae stars in other clusters in a future paper in this series . we obtained a reddening - corrected distance modulus of @xmath172 from the rrab and rrc stars . the average periods for the rrab and rrc stars and the ratio of number of rrc to total number of rr lyrae all suggest an oosterhoff - intermediate classification for ngc 1466 . the position of the rrab stars in the bailey diagram provides for a less clear oosterhoff determination . we wish to thank the referee , j.m . nemec , for his comments and suggestions which helped improve this paper . support for m.c . is provided by proyecto fondecyt regular # 1110326 ; by proyecto basal pfb-06/2007 ; and by fondap centro de astrofsica 15010003 . m.c . and j.b . are supported by the ministry for the economy , development , and tourism s programa inicativa cientfica milenio through grant p07 - 021-f , awarded to the milky way millennium nucleus . support for h.a.s . and is provided by nsf grants ast 0607249 and ast 0707756 . alard , c. 2000 , a&as , 144 , 363 bono , g. , caputo , f. , & stellingwerf , r. f. 1994 , , 423 , 294 bono , g. , caputo , f. , & stellingwerf , r. f. 1995 , , 99 , 263 cacciari , c. , corwin , t. m. , & carney , b. w. 2005 , , 129 , 267 catelan , m. 2004a , , 600 , 409 catelan , m. 2004b , in variable stars in the local group , asp conf . ser . , 310 , ed . kurtz , & k.r . pollard ( san francisco : asp ) , 113 catelan , m. 2009 , , 320,261 catelan , m. , & corts , c. 2008 , , 676 , l135 clement , c.m . et al . 2001 , , 122 , 2587 clementini , g. , corwin , t. m. , carney , b. w. , & sumerel , a. n. 2004 , , 127 , 938 contreras , r. , et al . 2010 , , 140 , 1766 corwin , t.m . , catelan , m. , smith , h.a . , borissova , j. , ferraro , f.r . , & raburn , w.s . 2003 , , 125 , 2543 deb , s. , & singh , h.p . 2010 , , 402 , 691 ferraz - mello , s. 1981 , , 86 , 619 johnson , j.a . , bolte , m. , stetson , p.b . , hesser , j.e . , & somerville , r.s . 1999 , , 527 , 1999 jurcsik , j. 1995 , aca , 45 , 653 jurcsik , j. 1998 , , 333 , 571 jurcsik , j. , & kovcs , g. 1996 , , 312 , 111 jurcisk , j. , & kovcs , g. 1999 , newa rev . , 43 , 463 kovcs , g. 1998 , , 69 , 49 kovcs , g. , & walker , a.r . 1999 , , 512 , 271 kovcs , g. , & walker , a.r . 2001 , , 371 , 579 landolt , a.u . 1992 , , 104 , 340 layden , a.c . 1998 , , 115 , 193 lenz , p. , & breger , m. , 2005 , commun . asteroseismol . , 146 , 53 mackey , a.d . , & gilmore , g.f . 2003 , , 338 , 85 mochejska , b. j.,kaluzny , j. , stanek , k. z. , sasselov , d. d. , & szentgyorgyi , a. h. 2001 , , 121 , 2032 ; erratum : , 121 , 3284 morgan , s.m . , wahl , j.n . , & wieckhorst , r.m . 2007 , , 374 , 1421 nemec , j.m . , nemec , a.f.l , & lutz , t.e . 1994 , , 108 , 222 nemec , j.m . , walker , a.r . , jeon , y.b . 2009 , , 138 , 1310 olsen , k.a.g . , et al . 1998 , , 300 , 665 olszewski , e.w . , schommer , r.a . , suntzeff , n.b . , & harris , h.c . 1991 , , 101 , 515 peterson , j.o . 1986 , , 170 , 59 popielski , b. l. , dziembowski , w. a. , & cassisi , s. 2000 , aca , 50 , 491 pritzl , b.j . , armandroff , t.e . , jacoby , g.h . , & da costa , g.s . 2002 , , 124 , 1464 pritzl , b.j . , armandroff , t.e . , jacoby , g.h . , & da costa , g.s . 2005 , , 129 , 2232 reimann , j.d . 1994 , phd thesis ( university of california ) santos jr . , j.f.c . , & piatti , a.e . 2004 , , 428 , 79 sills , a. , karakas , a. , & lattanzio , j. 2009 , , 692 , 1411 simon , n.r . , & clement , c.m . 1993 , , 410 , 526 soszyski , i. , et al . 2003 , aca , 53 , 93 stetson , p.b . 1987 , , 99 , 191 stetson , p.b . 1992 , in asp conf . 25 , astronomical data analysis software and systems i , ed . worrall , c. biemesderfer , & j. barnes ( san francisco : asp ) , 297 stetson , p.b . 1994 , , 106 , 250 stetson , p.b . , & harris , w.e . 1988 , , 96 , 909 thackeray , a.d . , & wesselink , a.j . 1953 , nature , 171 , 693 walker , a.r . 1992 , , 104 , 1395 wesselink , a.j . 1971 , , 152 , 159 wolf , m.j . , drory , n. , gebhardt , k. , & hill , g.j . 2007 , , 655 , 179 zinn , r. , & west , m.j . 1984 , , 55 , 45 zorotovic , m. , et al . 2010 , , 140 , 912

this review is based on the talk presented at the susy 2012 ( beijing ) . the new particle around 125 gev observed at the large hadron collider ( lhc ) is almost consistent with the standard model higgs boson , except that the diphoton decay mode may be excessive . we summarize a number of possibilities . while at the lhc the dominant production mechanism for the higgs boson of the standard model and some other extensions is via the gluon fusion process , the alternative vector - boson fusion is more sensitive to electroweak symmetry breaking . using the well known dijet - tagging technique to single out the vector - boson fusion mechanism , we investigate potential of vector - boson fusion to discriminate a number of models suggested to give an enhanced inclusive diphoton production rate .

1211.6823

in this talk , we are going to summarize a few models that have been suggested to explain the newly observed particle of about 125 gev at the large hadron collider ( lhc ) @xcite . it is of very high expectation that the observed particle is the long - sought higgs boson , which was proposed in 1960s @xcite . before the lhc era there have been many speculations of the breaking of electroweak symmetry ( ewsb ) . there are two known scales in particle physics the electroweak scale and the planck scale . the fundamental higgs boson of order 100 gev is unstable against the radiative corrections . the so - called gauge hierarchy problem requires new physics has to come in around tev scale in order that the unnatural cancellation between the bare mass term and the higher - order terms of the higgs boson mass is under control . historically , there are two categories of models : one with a strongly - coupled ewsb sector and one with a weakly - coupled ewsb sector . the most studied model for strong ewsb is the technicolor - type model @xcite while that for weakly - coupled ewsb model is the supersymmetry @xcite . in technicolor models , the standard model ( sm ) is simply an effective model below tev scale , at which the theory is replaced by another strong dynamics . therefore , the cutoff scale now becomes just tev . on the other hand , supersymmetric models predict another set of particles , which differ from their sm counter parts by a half - integral spin . quadratically divergent contributions to the higgs boson radiative corrections are cancelled among the sm particles and the corresponding supersymmetric particles . if the superconducting super collider ( ssc ) were built the higgs boson could have been discovered in early 2000s with perhaps a short running of the machine . since the higgs boson has been hiding for such a long time , many interesting alternatives were proposed in the last 1015 years . around the turn of the century extra dimension models became very popular , not to mention there have been a large number of varieties large extra dimension @xcite , universal extra dimension @xcite , randall - sundrum models @xcite , etc . there are also the little higgs type models @xcite . in contrast to supersymmetry the new particles have the spin as their sm counter parts . contributions to the higgs boson radiative corrections are cancelled among the sm particles and the corresponding new particles . perhaps , more and more models would have been proposed if the higgs boson kept hiding . would nt it be more fascinating for theorists ? the past few years the high energy community has been very excited with a number of experimental anomalies from the tevatron , lhc , and dark matter ( dm ) experiments . the long - time inconsistency among the dama results @xcite ( also the cogent @xcite ) and the other direct detection experiments has motivated a number of unconventional dark matter models @xcite , such as inelastic dm @xcite , isospin - violating dm @xcite , multi - component dm @xcite , etc . the top - quark forward - backward asymmetry observed by the cdf and d collaborations is rather puzzling too @xcite . many models such as flavor - changing @xmath0 , unusual @xmath1 , axigluon , etc @xcite were proposed , but the new lhc results seem ruled out almost all of these models @xcite . the cdf @xmath2 anomaly in 2011 @xcite also stimulated a large number of theoretical or phenomenological models to account for the observation @xcite . however , with non - observation of the resonance from d and cms @xcite the excitement gradually died out . at the end of 2011 , both the atlas and cms @xcite experiments at the lhc have seen some excess of events of a possible higgs candidate in the decays of @xmath3 , @xmath4 and @xmath5 channels . finally , the discovery was jointly announced in july 2012 by atlas @xcite and cms @xcite . all the observed channels , @xmath6 , @xmath7 and @xmath8 are consistent with the predictions of the sm higgs boson , except that the @xmath8 rate is somewhat higher than expectation . the @xmath9 and @xmath10 channels are not confirmed yet , because of large uncertainties . the diphoton production rate is about a factor of @xmath11 higher than that of the standard model higgs boson , while the @xmath12 and @xmath13 rates are consistent with the sm higgs boson within uncertainties . nevertheless , the observed rates are consistent with either the sm higgs boson or some other higgs models . a large number of models have been put forward to account for the observed particle at 125 gev , including the sm , mssm , nmssm , umssm and other mssm - extended models , fermiophobic higgs , 2hdm , rs radion , inert - higgs doublet , triplet higgs models , etc . the @xmath14 events collected by cms and atlas can be divided into two categories : _ inclusive _ @xmath15 and _ exclusive _ @xmath16 ( though both experiments have more refined sub - divisions among various classes of events ) . presumably , the inclusive @xmath15 events include all production channels such as gluon fusion , vector - boson fusion , associated production , etc , among which gluon fusion dominates for production of the sm higgs boson and most of the models considered in this talk , except for the fermiophobic higgs boson . on the other hand , exclusive @xmath16 events mainly come from vector - boson fusion and associated production , which can be further disentangled by jet - tagging techniques . the vector - boson fusion produces energetic forward jets while associated production with a @xmath17 or @xmath18 produces jets with @xmath19 to @xmath20 . the current evidence of the higgs boson in the diphoton channel comes mainly from inclusive @xmath15 events , simply because the inclusive event rate is much higher than the exclusive @xmath21 event rate . the main goal of this talk is to summarize all the models and find the parameter space of each model that have been proposed to explain the excess in the inclusive higgs diphoton events , and attempt in distinguishing the models using the exclusive @xmath21 channel in vector - boson fusion ( vbf ) @xcite . the exclusive vbf events with @xmath21 in the final state are selected using the forward jet - tagging techniques which will be explained shortly . we will first choose the parameter - space region of each model that can account for the excess in the inclusive diphoton rate , and then in that region of parameter space we calculate the exclusive @xmath22 vbf production rates . we found that the exclusive @xmath21 production rate in vbf channel can give more information to help in distinguishing a number of popular models . we summarize a number of models that have been used to account for the excess in the inclusive @xmath15 data as follows . 1 . the sm higgs boson @xcite is still believed to be the most desirable candidate . it is still consistent with the data within uncertainty . 2 . the lighter higgs boson of the minimal supersymmetric standard model ( mssm ) can acquire a large radiative correction from the top - stop sector to achieve a mass of 125 gev , though it has been shown rather difficult to achieve an enhanced diphoton rate @xcite . however , it is possible when one of the staus is light enough , just above the lep limit , and so the diphoton branching ratio is enhanced @xcite . 3 . one of cp - even higgs bosons in the next - to - minimal supersymmetric standard model ( nmssm ) can account for the observed 125 gev boson with an enhanced diphoton rate @xcite . it could be the lightest or the second lightest one . the @xmath23-extended mssm ( umssm ) @xcite and other extensions @xcite are also possible to account for the observed boson . the analyses for these extended mssm models are much more involved and deserve dedicated studies . the lighter cp - even higgs boson of various types of the two - higgs - doublet models ( 2hdm ) @xcite , which has enough free parameters in the model that allows one to achieve a large branching ratio into @xmath8 . 5 . in the fermiophobic ( fp ) higgs boson model , the higgs boson is only responsible to generate the masses to @xmath17 and @xmath18 bosons while the fermion masses are generated by some other means . since the fp higgs boson does not couple to the quarks , it can not be produced via gluon fusion at hadronic colliders , but only through the vbf and the associated production with a vector - boson . nevertheless , the fp higgs boson lighter than 130 gev has a much larger branching ratio into diphoton , such that it can still account for the observed inclusive diphoton rate at the lhc @xcite . 6 . in ref . @xcite , it was pointed out that the randall - sundrum ( rs ) radion , with enhanced couplings to @xmath24 and @xmath8 due to trace anomaly , can explain the excess in the inclusive diphoton production rate and suppressed @xmath6 and @xmath7 rates , which provides the most economical alternative solution to explain the observed data . the inert - higgs - doublet model ( ihdm ) @xcite , which is a special case of 2hdm , in which one of the doublets entirely decouples from the leptons , quarks , and gauge bosons while the other one takes on the role of the sm higgs doublet . the production rate of the higgs boson is the same as the sm one . however , the decay width of @xmath25 can be enhanced by the presence of the charged higgs boson in the loop . it was shown @xcite that the diphoton production rate can be enhanced by a factor of about @xmath26 . there may also be some possibilities that the sm - like higgs boson first decays into two light scalar or pseudoscalar bosons , followed by subsequent decays into collimated pairs of photons , which appear as two photons in the final state @xcite . on the other hand , instead of top - down approaches , it would also be useful to reversely determine the couplings and the nature of the observed 125 gev particle by studying all the available data @xcite . the disadvantage of gluon fusion is that it is not clear what particles and their masses running in the triangle loop . in some models , the contribution from a particular charged particle can increase or decrease the diphoton decay width , depending on the relative signs . for example , in the supersymmetric models , there are additional sfermions , charginos , charged higgs bosons running in the loop , and therefore resulting in complicated dependence on the model parameters . on the other hand , the advantage of using @xmath6 fusion or associated production with a @xmath17 or a @xmath18 boson is that the production diagram is clean and directly testing the couplings of @xmath27 and @xmath28 . furthermore , the @xmath6 fusion has a cross section at least a factor of 2 larger than the associated production . we therefore focus on @xmath6 fusion . the @xmath6 fusion can be extracted by the presence of two energetic forward jets . we can impose selection cuts to select jets in forward rapidity and high energy region @xcite . by combining the production rates in the inclusive @xmath15 and exclusive @xmath21 channels , one can obtain useful information about the nature of the 125 gev new particle recently observed at the lhc . we calculate the event rates in the @xmath6 fusion channel for a number of models that have been used to interpret the current lhc data of the 125 gev `` higgs boson '' . the theoretical cleanliness of @xmath6 fusion has been explained in the last paragraph . we believe that the @xmath6 fusion channel can provide useful information to discriminate various models . the organization of this paper is as follows . we briefly highlight a number of models in the next section , and the @xmath6 fusion and selection cuts in sec . we give the decay branching ratios in sec . iv and production rates in sec . v. we conclude in sec . vi . the sm higgs boson @xcite is still the most favorable candidate to interpret the observed boson , though the experimental data showed slightly excess in inclusive @xmath29 events over the prediction of the sm higgs boson @xcite . production of the sm higgs boson is dominated by gluon fusion , which is an order of magnitude larger than the next important mechanism there are two higgs doublets instead of just one in the 2hdm . in order to avoid dangerous tree level flavor - changing neutral currents , the popular 2hdms are imposed a discrete symmetry . in the type i , all of the fermions couple to a single higgs doublet , and do not couple to the second doublet ; while in the type ii , one doublet couples only to down - type quarks and another doublet couples to the up - type quarks . in this talk , we focus on the type ii , which has the same higgs sector as the mssm . the higgs sector consists of two higgs doublets @xmath30 where the subscripts @xmath31 denote the right - handed quark singlet field that the higgs doublet couples to . the electroweak symmetry is broken when the higgs doublet fields develop the following vevs : @xmath32 physically , there are two cp - even , one cp - odd , and a pair of charged higgs bosons after electroweak symmetry breaking ( ewsb ) , and the @xmath17 and @xmath18 bosons as well as the sm fermions , except for neutrinos , acquire masses . the yukawa couplings and masses for fermions can be obtained from the following yukawa interactions after ewsb @xmath33 where @xmath34 . the parameters of the model in the cp - conserving case include @xmath35 where @xmath36 is the mixing angle between the two cp - even higgs bosons . there are enough free parameters in the higgs potential such that all the above parameters are free inputs to the model , in contrast to the mssm where the higgs potential is highly restricted by supersymmetry in addition to gauge symmetry . the couplings of the two lighter and heavier cp - even higgs bosons @xmath37 and @xmath38 respectively and the cp - odd higgs boson @xmath39 to the top , bottom quarks , and taus are given by , with a common factor of @xmath40 being suppressed , @xmath41 while the charged higgs @xmath42 couples to @xmath43 and @xmath44 via @xmath45 \;.\ ] ] other relevant couplings in @xmath6 fusion are those to gauge bosons are given by , @xmath46 dominant production of the light cp - even higgs boson @xmath37 at the lhc is via gluon fusion , similar to the sm higgs boson with the top quark running in the loop ; however , in the large @xmath47 region the bottom - quark contribution can also be substantial . since the bottom - yukawa coupling can be substantially enhanced , the gluon fusion cross section can be larger than the sm . on the other hand , since the couplings of the @xmath37 to the @xmath6 and @xmath7 are simply the sm values multiplied by @xmath48 , @xmath6 fusion cross sections are in general smaller than the sm . the decay into two photons is somewhat more complicated than the sm . besides the couplings @xmath27 , @xmath49 , and @xmath50 are different , there are also the charged higgs bosons running in the loop . the charged higgs boson couples to the light cp - even higgs with the coupling @xcite @xmath51 however , the @xmath52 and b meson mixing constraints require the charged higgs boson mass @xmath53 gev for intermediate to large values of @xmath47 @xcite . we will choose @xmath54 gev in our analysis below . the overall diphoton production rate @xmath55 in gluon fusion can easily vary between @xmath56 of the sm prediction depending on parameters @xcite . it was shown in ref . @xcite that the enhancement in branching ratio can be obtained roughly along @xmath57 near zero for all @xmath47 in the type ii model . we choose parameter space points there to illustrate . in order to achieve a mass of 125 gev for the lighter cp - even higgs boson , a very large radiative correction is needed , which essentially comes from top - stop loop . the approximate formula for the lighter cp - even higgs boson is given by @xcite @xmath58\ ] ] where @xmath59 and @xmath60 tev is the susy scale . a large @xmath61 is needed to generate a large correction . here we follow the findings in ref . @xcite for the parameter space : we choose @xmath62 gev , @xmath63 tev , @xmath64 tev , and @xmath65 . a detailed analysis of the mssm parameter space based on bayesian statistical analysis in light of the new observation of the 125 gev higgs candidate is also presented recently in ref . the reason behind such a large @xmath47 is the stau contribution to the diphoton branching ratio explained below @xcite . in the production part via gluon fusion , the difference between the sm and supersymmetric models is that squarks also run in the triangle loop . as the experimental data have pushed the squark masses of the first two generations to be quite heavy but not the third generation ( stop and sbottom ) the change in production rates could be substantial , especially in large @xmath47 . on the other hand , the decay into diphoton is more involved in susy models . all charged particles , including squarks , sleptons , charginos , charged higgs boson can flow in the triangle loop . with the present constraints from experiments , the production rate into diphoton ( equal to production cross section times the branching ratio into diphoton ) in the mssm is shown to be very similar to the sm one and that the diphoton production rate can hardly be enhanced by more than a factor of @xmath66 @xcite . the formulas for the decay of the higgs boson into two photons as well as production via gluon fusion can be found in ref . the couplings of the lighter cp - even higgs boson to the @xmath6 or @xmath7 are given by the sm ones multiplied by @xmath67 . therefore , the production rate in the @xmath6 fusion is in general similar to or smaller than the sm prediction . we look at the parameter space in which the diphoton production rate would be larger than the sm value in the mssm . it was shown in ref . @xcite that diphoton production rate can be larger than the sm one if one pushes the stau to be very light , just above the lep limit . in addition to the above mentioned soft parameters , the other parameters are @xmath68 , @xmath69 , and the @xmath70 . without loss of generality we choose @xcite @xmath71 in which we can scan for the diphoton production rate @xmath72 to be larger than the sm rate . the region essentially gives a light stau , which can enhance the @xmath73 . we will scan the region according to eq . ( [ stau ] ) and require the mass of the lighter cp - even higgs boson around 125 gev and the diphoton production rate larger than the sm value . with the name `` fermiophobic '' ( fp ) the higgs boson only couples to the vector bosons at tree level , though higher - loop corrections can induce small couplings to fermions . in this case , the yukawa couplings and masses of fermions are generated by some other mechanisms , which are not of concern in this talk . the coupling strength of the fp higgs boson to vector bosons is the same as that of the sm higgs boson . we write the interactions as @xmath74 since the fp higgs boson does not couple to quarks , it can not be produced dominantly by gluon fusion at hadronic colliders , but only through the vector - boson fusion and the associated production with a @xmath75 boson . the corresponding production cross sections are the same as the vbf of the sm higgs boson . nevertheless , the fp higgs boson lighter than 130 gev has a much larger branching ratio into diphoton , such that it can still account for the observed diphoton rate at the lhc @xcite . there are two reasons : ( i ) the fp higgs boson decay into fermions is highly suppressed with only the loop - induced couplings , and ( ii ) the decay into photons is via a loop of @xmath17 boson without the negative interference from the top quark . thus , the branching ratio into diphoton can be enhanced by more than an order of magnitude . overall , the diphoton production rate at the lhc is comparable to the sm higgs boson , as was used to account for the observed boson @xcite . an earlier study of fp higgs boson at the lhc can be found in ref . there is basically no free parameters in this model . the rs model @xcite used a warped 5d space - time , a slice of the symmetric space @xmath76 , to explain the gauge hierarchy problem . the extra dimension @xmath77 is a single @xmath78 orbifold with one hidden and one visible 3-brane localized at @xmath79 and @xmath80 , respectively . it was pointed out by goldberger and wise @xcite that the original rs model has a four - dimensional massless scalar ( the modulus or radion ) which does not have a potential and therefore the extra dimension can not be stabilized . a stabilization mechanism was proposed in @xcite by adding a bulk scalar field propagating in the background solution which can generate a potential to stabilize the modulus field . the minimum of the potential can be arranged to give the desired value of @xmath81 to solve the gauge hierarchy problem without extreme fine tuning of parameters . as a consequence , the lightest excitation mode of the modulus field is the radion , which has a mass of the order of 100 gev to a tev , and the strength of its coupling to the sm fields is of the order of @xmath82 @xcite . phenomenology of the stabilized radion and its effects on the background geometry were studied in @xcite . the interactions of the stabilized modulus ( radion ) @xmath83 with the sm particles on the visible brane are completely determined by 4-dimensional general covariance . thus the radion lagrangian is given by @xmath84 where @xmath85 is of the order of tev and @xmath86 is the trace of the sm energy - momentum stress tensor , which has the following lowest order terms @xmath87 the coupling of the radion to a pair of gluons ( photons ) is induced at one loop level , with the dominated contributions coming from the heavy top quark ( top quark and @xmath17 ) as well as from the trace anomaly in qcd ( qed ) . the expressions of the induced couplings can be found in ref.@xcite . similar to the sm higgs boson , the dominant production channel for the radion is via @xmath24 fusion , followed by vbf @xcite . in addition , @xmath88 gets substantial enhancement from the trace anomaly . for the decay of the radion , it is dominated by the @xmath24 mode instead of @xmath9 at the low mass region , while its diphoton branching ratio is merely a fraction of the sm value of @xmath89 . overall , the diphoton production rate @xmath90 can be larger than the sm rate if the scale @xmath91 is small enough , and as long as it is consistent with the search for rs graviton . if we do not concern about naturalness , the scale @xmath91 can be as small as 0.8 tev . here we fix the scale @xmath91 to be @xmath92 tev , which can enhance the diphoton production rate in gluon fusion by a factor of @xmath93 @xcite relative to the sm rate . note that the branching ratios of the radion is independent of @xmath91 . ihdm is a special case of 2hdm , in which one of the doublets takes on the role of the sm higgs doublet , while the other one is inert which means that it entirely decouples from the sm leptons , quarks , and gauge bosons . the model also has an additional @xmath94 symmetry , for which all sm particles are even , except for the particle content of the second inert higgs doublet . the lightest @xmath94-odd particle of the second doublet can work as a candidate of dark matter . the higgs sector consists of @xmath95 the electroweak symmetry is broken solely by one vev : @xmath96 the higgs potential is given by @xmath97\ ] ] physically , there are 2 cp - even scalars @xmath98 , 1 cp - odd scalar @xmath99 , and a pair of charged higgs @xmath100 . the @xmath37 plays the role of the sm higgs boson while the others are inert . a list of parameters of the model includes @xmath101 and @xmath102 . production via gluon fusion and via @xmath6 fusion are the same as the sm higgs boson . however , the decay into @xmath8 receives additional contributions from the @xmath103 running in the loop . if kinematically allowed the higgs boson @xmath37 can also decay into @xmath104 and @xmath105 . for simplicity and to achieve a large enough branching ratio into @xmath8 we set the masses of @xmath106 to be above the threshold @xmath107 . in this model , the coupling between the charged higgs boson and the sm higgs boson @xmath37 is given by @xmath108 it is clear that from this equation the sign of the charged higgs contribution to the triangular loop can be positive or negative , depending on the sizes of @xmath109 and @xmath110 . thus , if we set @xmath111 , the charged higgs contribution vanishes and so the diphoton branching ratio becomes the same as the sm one . it was shown in ref . @xcite that in gluon fusion the diphoton production rate is determined by the product of gluon - fusion cross section and the @xmath8 branching ratio @xmath112 which can be varied from about @xmath113 to @xmath114 . the charged higgs contribution to the diphoton branching ratio depends on the sizes of @xmath115 and @xmath110 . it was shown in ref . @xcite that when @xmath116 the diphoton branching ratio is enhanced . together with other theoretical constraints one can find the region to be @xmath117 gev and @xmath118 . a feynman diagram showing the vector - boson fusion into sm higgs , fp higgs or radion.,width=384 ] the most distinguished feature of vbf at hadronic colliders is the appearance of two energetic forward jets separated by a large @xmath119 , where @xmath120 is the pseudo - rapidity and @xmath83 is the azimuth angle . the feynman diagram is shown in fig . [ feyn ] . each of the initial quarks radiates a @xmath75 boson , which further annihilates into the higgs boson or some other particles under consideration . the unique feature of this process is that the @xmath75 bosons participating in the fusion process are close to on - shell @xcite , and irrespecive whether the @xmath75 is longitudinal or transverse , the scattered quark carries almost all the energy of the incoming quark and goes in the forward direction . " @xcite this fact justifies the use of effective @xmath17 approximation in all calculations for vbf in the early days . based on this feature we impose the following experimental cuts in selecting the dijet events coming dominantly from the vbf : @xmath121 and @xmath122 where the subscript `` 1 '' denotes the most energetic jet . in the cut of eq . ( [ jcuts ] ) , we choose either the energy of the most energetic jet @xmath123 gev or the invariant mass of the jet pair @xmath124 gev . this set of cuts is similar to that used by cms @xcite and atlas @xcite in their searches for fp higgs boson . the vector - boson fusion is well - known that it allows to probe the direct coupling between the vector bosons and the higgs boson or other particles under consideration . this is in contrast to the @xmath24 fusion , because any colored particles can flow in the triangular loop and affect the production rate . for example , in mssm all squarks can circulate inside the loop . .[talk - tab1 ] decay branching ratio @xmath73 for the sm higgs boson @xmath125 , the fermiophobic higgs boson @xmath126 , the radion @xmath83 , the inert higgs doublet model , the two higgs doublet model , and the mssm . [ cols="<,^,^,>",options="header " , ] we calculate the vbf cross sections for @xmath127 for various models under consideration . we impose the selection cuts for energetic forward jets as in eqs . ( [ jcut1 ] ) and ( [ jcuts ] ) . we let the higgs boson decay into @xmath8 , and impose the following cuts on the photons : @xmath128 and implementing the selection cuts for the forward jets and the diphoton . the production rates of diphoton for various models are then obtained by multiplying the corresponding diphoton branching ratios . we use pythia @xcite for parton showering and hadronization . during the parton showering we turn on the initial and final state qed and qcd radiations ( isr and fsr ) , and fragmentation / hadronization according to the lund model . we use pgs @xcite for detector - simulation with the most general settings for the lhc . electromagnetic and hadronic calorimeter resolutions are set at @xmath129 and @xmath130 , respectively . we use the cone algorithm for jet finding with a cone size of @xmath131 , sagitta resolution of @xmath132 m ( @xmath133 ) , track - finding efficiency of @xmath134 , and minimum track @xmath135 of 1 gev . more details can be found in refs . @xcite . the numerical results for lhc-7 , lhc-8 , and lhc-14 are listed in table [ talk - prod1 ] for the sm , fp , radion , ihdm , 2hdm , and the mssm . it would be more transparent to show the production rate relative to the sm one @xmath136 where @xmath137 stands for the sm higgs or any other higgs - like candidate in various models . we found that this ratio is quite robust against various cuts ( mjjcut or ejcut as in eq . ( [ jcuts ] ) ) and against the energy of the collision . in the upper panel of fig . [ ratio ] , we first show the ratio for the _ inclusive _ diphoton production rate @xmath138 of each model , which is dominated by gluon fusion , except for the fp higgs boson . the parameter space of each model is chosen such that this inclusive diphoton rate is equal to or larger than the sm rate except for the fp higgs boson which has no free parameter . for the same parameter space , we show in the lower panel the ratio for the _ exclusive _ @xmath22 production rate @xmath139 for each model . the figure is valid for lhc-7 , lhc-8 and lhc-14 , and for using either mjjcut cut on both forward jets or ejcut cut on the most energetic jet . it is clear that the models can be employed to explain the excess in the inclusive diphoton rates in some parameter space region , but for the same region of parameter space the ratio of exclusive vbf production would be different among the models . the fp higgs would be a number of times larger than the sm in the vbf channel , but the rs radion would give negligible exclusive vbf production . on the other hand , the ihdm , 2hdm , and the mssm would give similar ratios in both inclusive and exclusive production . the 2hdm can give a somewhat smaller ratio in the exclusive vbf . _ upper : _ the ratio of _ inclusive _ diphoton production rates @xmath140 for various models . the parameter space is chosen such that the ratio is equal to or larger than 1 , except for the fp higgs boson which has no free parameter . _ lower : _ the ratio of the _ exclusive _ @xmath22 production rate @xmath139 for each model in the corresponding parameter space . , title="fig:",width=643 ] _ upper : _ the ratio of _ inclusive _ diphoton production rates @xmath140 for various models . the parameter space is chosen such that the ratio is equal to or larger than 1 , except for the fp higgs boson which has no free parameter . _ lower : _ the ratio of the _ exclusive _ @xmath22 production rate @xmath139 for each model in the corresponding parameter space . , title="fig:",width=643 ] the experimental details for vbf in the context of searching for the fp higgs boson are given in ref . the dominant backgrounds to @xmath25 consist of i ) the irreducible background from the prompt diphoton production , and ii ) the reducible backgrounds from @xmath141 and @xmath142 , when one or more of the jets hadronize into ( typically neutral ) pions and deposit energy in the electromagnetic calorimeter . these reconstructed objects are generally referred to as fake photons . isolation is a very useful tool to reject the non - prompt background coming from electromagnetic showers originating in jets , because these fake photons are often accompanied by single and multiple @xmath143s . the dijet tag , as defined in eqs . ( [ jcut1 ] ) and ( [ jcuts ] ) , together with the higgs boson decaying into diphoton selects a special class of events consisting of two photons and two forward energetic jets . a signal - to - background ratio of order 1 can be achieved @xcite . the signal from @xmath25 will be identified as a sharp peak in the @xmath144 distribution , where the background is in continuum . in ref . @xcite , the background model is derived from data . in the new run at 8 tev , we expect the same treatment is applied to the background . the estimation of significance of each signal is beyond the scope of the present paper . given that we have obtained the event rates of each model in the tables , which have been under experimental cuts and simulations ( pgs ) , it is straight - forward to estimate the required luminosity to probe various scenarios . the continuum background at the 7 tev and 8 tev lhc in vbf channel has been obtained in refs . @xcite and @xcite . since the sm higgs boson has been seen above the background with some significance level in the vbf channel in the current lhc runs ( about 5 fb@xmath145 at 7 tev and about 5 fb@xmath145 at 8 tev ) , the scenarios with higher event rates than the sm higgs boson should be detectable . the projected integrated luminosity at the end of 2012 is about @xmath146 fb@xmath145 for each experiment , it would not be a problem to investigate the scenarios in the vbf channel . lhc is expected to confirm if there is a new particle at 125 gev by the end of this year . the likelihood for a new discovery is rather high . nevertheless , whether this new particle is the sm higgs boson is not an easy question to answer . here the scenario of this 125 gev particle produced by vector - boson fusion , instead of gluon fusion as the dominant production mechanism for the standard model higgs boson , is studied in details . by using the forward dijet tagging technique , one can single out the vector - boson fusion mechanism . we studied a number of popular new physics models that have been employed to interpret the observed particle at 125 gev , including fermiophobic higgs boson , the randall - sundrum radion , inert - higgs - doublet model , two - higgs - doublet model , and the mssm . since the inclusive diphoton channel showed an excess over the sm predictions , we first selected the parameter space of each model that can give an inclusive diphoton rate larger than or equal to the sm rate . then , we calculate the exclusive @xmath22 diphoton production rate in vbf for that parameter space . if the diphoton mode excess seen at lhc-7 can be firmly established by the new lhc-8 data , it will be the utmost task to identify the nature of this particle . perhaps , it is simply the sm higgs boson with some level of statistical fluctuation , but it could also be the rs radion @xcite , fermiophobic higgs boson @xcite , the light cp - even higgs boson of the 2hdm @xcite , the higgs boson of the ihdm @xcite , or one of the cp - even higgs bosons in other extensions of the mssm @xcite , all of which can allow an enhancement to the @xmath8 production rate . on the other hand , the vector - boson fusion , as singled out by the dijet tag , provides useful information in helping to differentiate among various models . it is not hard by browsing through fig . [ ratio ] to conclude the following * if a similar rate is seen in inclusive production but no large excess is seen in the exclusive vbf production it would unlikely be a fermiophobic higgs boson . * if a similar rate or excess is seen in inclusive production but some events are seen in exclusive vbf production rate , it would unlikely be the rs radion . * if moderate excess is seen in both inclusive production and exclusive vbf production , it could be the higgs boson of the ihdm , 2hdm , or the mssm . however , if the excess is over 60% it will pose severe challenge to the mssm . it seems easy to rule out either fermiophobic higgs boson or rs radion , providing that we see no large excess or some events in vbf channel , respectively . however , it is still difficult to distinguish the other models when the inclusive and exclusive rates are similar to or slightly larger than the sm values . if the production rate of the diphoton mode at 125 gev lines up with the sm prediction eventually , it is still premature to conclude this is coming from the sm higgs boson . other alternatives in mssm , nmssm , or other susy models can also be mimicking the sm higgs boson , depending on the parameter space of the new physics model . in any case , once the signals at 125 gev are confirmed further studies including all possible decay modes are to be taken into account in order to discriminate these many alternatives beyond the standard model . vector - boson fusion is the next most important production mechanism that must be taken into account to fully identify the newly discovered particle . this work was supported in part by the national science council of taiwan under grants no . 99 - 2112-m-007 - 005-my3 and no . 101 - 2112-m-001 - 005-my3 as well as the wcu program through the kosef funded by the mest ( r31 - 2008 - 000 - 10057 - 0 ) . g. aad _ et al . _ [ the atlas collaboration ] , arxiv:1207.7214 [ hep - ex ] . s. chatrchyan _ et al . _ [ the cms collaboration ] , arxiv:1207.7235 [ hep - ex ] . p. w. higgs , phys . * 13 * , 508 ( 1964 ) ; f. englert and r. brout , phys . * 13 * , 321 ( 1964 ) ; g. s. guralnik , c. r. hagen and t. w. b. kibble , phys . * 13 * , 585 ( 1964 ) . e. farhi and l. susskind , phys . rept . * 74 * , 277 ( 1981 ) . h. e. haber and g. l. kane , phys . rept . * 117 * , 75 ( 1985 ) . n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . d * 59 * , 086004 ( 1999 ) [ hep - ph/9807344 ] . t. appelquist , h. -c . cheng and b. a. dobrescu , phys . rev . d * 64 * , 035002 ( 2001 ) [ hep - ph/0012100 ] . l. randall and r. sundrum , phys . lett . * 83 * , 3370 ( 1999 ) [ hep - ph/9905221 ] . n. arkani - hamed , a. g. cohen , e. katz and a. e. nelson , jhep * 0207 * , 034 ( 2002 ) [ hep - ph/0206021 ] . r. bernabei _ et al . _ [ dama and libra collaborations ] , eur . phys . j. c * 67 * , 39 ( 2010 ) [ arxiv:1002.1028 [ astro-ph.ga ] ] . c. e. aalseth _ [ cogent collaboration ] , phys . lett . * 106 * , 131301 ( 2011 ) [ arxiv:1002.4703 [ astro-ph.co ] ] . z. ahmed _ et al . _ [ cdms - ii collaboration ] , science * 327 * , 1619 ( 2010 ) [ arxiv:0912.3592 [ astro-ph.co ] ] . e. aprile _ et al . _ [ xenon100 collaboration ] , arxiv:1207.5988 [ astro-ph.co ] . d. tucker - smith and n. weiner , phys . d * 64 * , 043502 ( 2001 ) [ hep - ph/0101138 ] . j. l. feng , j. kumar , d. marfatia and d. sanford , phys . b * 703 * , 124 ( 2011 ) [ arxiv:1102.4331 [ hep - ph ] ] . k. m. zurek , phys . rev . d * 79 * , 115002 ( 2009 ) [ arxiv:0811.4429 [ hep - ph ] ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . * 101 * , 202001 ( 2008 ) [ arxiv:0806.2472 [ hep - ex ] ] ; t. aaltonen _ et al . _ [ cdf collaboration ] , phys . d * 83 * , 112003 ( 2011 ) [ arxiv:1101.0034 [ hep - ex ] ] ; v. m. abazov _ et al . _ [ d0 collaboration ] , phys . rev . d * 84 * , 112005 ( 2011 ) [ arxiv:1107.4995 [ hep - ex ] ] . s. jung , h. murayama , a. pierce and j. d. wells , phys . rev . d * 81 * , 015004 ( 2010 ) [ arxiv:0907.4112 [ hep - ph ] ] ; k. cheung , w. -y . keung and t. -c . yuan , phys . b * 682 * , 287 ( 2009 ) [ arxiv:0908.2589 [ hep - ph ] ] ; j. shu , t. m. p. tait and k. wang , phys . d * 81 * , 034012 ( 2010 ) [ arxiv:0911.3237 [ hep - ph ] ] ; p. h. frampton , j. shu and k. wang , phys . lett . b * 683 * , 294 ( 2010 ) [ arxiv:0911.2955 [ hep - ph ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 709 * , 28 ( 2012 ) [ arxiv:1112.5100 [ hep - ex ] ] ; j. a. aguilar - saavedra and m. perez - victoria , jhep * 1109 * , 097 ( 2011 ) [ arxiv:1107.0841 [ hep - ph ] ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . lett . * 106 * , 171801 ( 2011 ) [ arxiv:1104.0699 [ hep - ex ] ] . k. cheung and j. song , phys . lett . * 106 * , 211803 ( 2011 ) [ arxiv:1104.1375 [ hep - ph ] ] ; a. e. nelson , t. okui and t. s. roy , phys . d * 84 * , 094007 ( 2011 ) [ arxiv:1104.2030 [ hep - ph ] ] . x. -p . wang , y. -k . wang , b. xiao , j. xu and s. -h . zhu , phys . d * 83 * , 115010 ( 2011 ) [ arxiv:1104.1917 [ hep - ph ] ] . f. yu , phys . d * 83 * , 094028 ( 2011 ) [ arxiv:1104.0243 [ hep - ph ] ] ; m. r. buckley , d. hooper , j. kopp and e. neil , phys . d * 83 * , 115013 ( 2011 ) [ arxiv:1103.6035 [ hep - ph ] ] . p. j. fox , j. liu , d. tucker - smith and n. weiner , phys . d * 84 * , 115006 ( 2011 ) [ arxiv:1104.4127 [ hep - ph ] ] ; e. j. eichten , k. lane and a. martin , phys . lett . * 106 * , 251803 ( 2011 ) [ arxiv:1104.0976 [ hep - ph ] ] . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . * 107 * , 011804 ( 2011 ) [ arxiv:1106.1921 [ hep - ex ] ] ; s. chatrchyan _ et al . _ [ cms collaboration ] , arxiv:1208.3477 [ hep - ex ] . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 710 * , 49 ( 2012 ) [ arxiv:1202.1408 [ hep - ex ] ] ; s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 710 * , 26 ( 2012 ) [ arxiv:1202.1488 [ hep - ex ] ] . j. chang , k. cheung , p. -y . tseng and t. -c . yuan , arxiv:1206.5853 [ hep - ph ] . m. carena , s. gori , n. r. shah and c. e. m. wagner , jhep * 1203 * , 014 ( 2012 ) [ arxiv:1112.3336 [ hep - ph ] ] ; m. carena , s. gori , n. r. shah , c. e. m. wagner and l. -t . wang , jhep * 1207 * , 175 ( 2012 ) [ arxiv:1205.5842 [ hep - ph ] ] . h. baer , v. barger and a. mustafayev , phys . d * 85 * , 075010 ( 2012 ) [ arxiv:1112.3017 [ hep - ph ] ] ; s. heinemeyer , o. stal and g. weiglein , phys . b * 710 * , 201 ( 2012 ) [ arxiv:1112.3026 [ hep - ph ] ] ; a. arbey , m. battaglia , a. djouadi , f. mahmoudi and j. quevillon , phys . lett . b * 708 * , 162 ( 2012 ) [ arxiv:1112.3028 [ hep - ph ] ] ; p. draper , p. meade , m. reece and d. shih , phys . d * 85 * , 095007 ( 2012 ) [ arxiv:1112.3068 [ hep - ph ] ] ; s. akula , b. altunkaynak , d. feldman , p. nath and g. peim , phys . d * 85 * , 075001 ( 2012 ) [ arxiv:1112.3645 [ hep - ph ] ] ; m. kadastik , k. kannike , a. racioppi and m. raidal , jhep * 1205 * , 061 ( 2012 ) [ arxiv:1112.3647 [ hep - ph ] ] ; j. cao , z. heng , d. li and j. m. yang , phys . b * 710 * , 665 ( 2012 ) [ arxiv:1112.4391 [ hep - ph ] ] ; n. d. christensen , t. han and s. su , arxiv:1203.3207 [ hep - ph ] ; j. l. feng and d. sanford , phys . d * 86 * , 055015 ( 2012 ) [ arxiv:1205.2372 [ hep - ph ] ] ; j. -j . cao , z. -x . heng , j. m. yang , y. -m . zhang and j. -y . zhu , jhep * 1203 * , 086 ( 2012 ) [ arxiv:1202.5821 [ hep - ph ] ] ; t. li , j. a. maxin , d. v. nanopoulos and j. w. walker , phys . b * 710 * , 207 ( 2012 ) [ arxiv:1112.3024 [ hep - ph ] ] ; v. barger , m. ishida and w. -y . keung , arxiv:1207.0779 [ hep - ph ] . u. ellwanger , jhep * 1203 * , 044 ( 2012 ) [ arxiv:1112.3548 [ hep - ph ] ] ; j. f. gunion , y. jiang and s. kraml , phys . b * 710 * , 454 ( 2012 ) [ arxiv:1201.0982 [ hep - ph ] ] ; s. f. king , m. muhlleitner and r. nevzorov , nucl . b * 860 * , 207 ( 2012 ) [ arxiv:1201.2671 [ hep - ph ] ] ; j. cao , z. heng , j. m. yang , y. zhang and j. zhu , jhep * 1203 * , 086 ( 2012 ) [ arxiv:1202.5821 [ hep - ph ] ] ; u. ellwanger and c. hugonie , arxiv:1203.5048 [ hep - ph ] ; d. a. vasquez , g. belanger , c. boehm , j. da silva , p. richardson and c. wymant , phys . d * 86 * , 035023 ( 2012 ) [ arxiv:1203.3446 [ hep - ph ] ] . c. -f . chang , k. cheung , y. -c . lin and t. -c . yuan , jhep * 1206 * 128 ( 2012 ) , [ arxiv:1202.0054 [ hep - ph ] ] . m. hirsch , w. porod , l. reichert and f. staub , arxiv:1206.3516 [ hep - ph ] ; h. an , t. liu and l. -t . wang , arxiv:1207.2473 [ hep - ph ] ; p. athron , s. f. king , d. j. miller , s. moretti and r. nevzorov , arxiv:1206.5028 [ hep - ph ] ; k. s. jeong , y. shoji and m. yamaguchi , jhep * 1209 * , 007 ( 2012 ) [ arxiv:1205.2486 [ hep - ph ] ] ; f. boudjema and g. d. la rochelle , phys . d * 86 * , 015018 ( 2012 ) [ arxiv:1203.3141 [ hep - ph ] ] ; p. fileviez perez , phys . b * 711 * , 353 ( 2012 ) [ arxiv:1201.1501 [ hep - ph ] ] . p. m. ferreira , r. santos , m. sher and j. p. silva , phys . d * 85 * , 077703 ( 2012 ) [ arxiv:1112.3277 [ hep - ph ] ] ; g. burdman , c. e. f. haluch and r. d. matheus , phys . d * 85 * , 095016 ( 2012 ) [ arxiv:1112.3961 [ hep - ph ] ] ; e. cervero and j. -m . gerard , phys . b * 712 * , 255 ( 2012 ) [ arxiv:1202.1973 [ hep - ph ] ] ; p. m. ferreira , r. santos , m. sher and j. p. silva , phys . rev . d * 85 * , 035020 ( 2012 ) [ arxiv:1201.0019 [ hep - ph ] ] ; n. chen and h. -j . he , jhep * 1204 * , 062 ( 2012 ) [ arxiv:1202.3072 [ hep - ph ] ] ; a. arhrib , r. benbrik and c. -h . chen , arxiv:1205.5536 [ hep - ph ] ; h. s. cheon and s. k. kang , arxiv:1207.1083 [ hep - ph ] ; w. altmannshofer , s. gori and g. d. kribs , arxiv:1210.2465 [ hep - ph ] ; n. craig and s. thomas , arxiv:1207.4835 [ hep - ph ] ; d. s. m. alves , p. j. fox and n. j. weiner , arxiv:1207.5499 [ hep - ph ] ; s. chang , s. k. kang , j. -p . lee , k. y. lee , s. c. park and j. song , arxiv:1210.3439 [ hep - ph ] ; y. bai , v. barger , l. l. everett and g. shaughnessy , arxiv:1210.4922 [ hep - ph ] ; a. drozd , b. grzadkowski , j. f. gunion and y. jiang , arxiv:1211.3580 [ hep - ph ] ; t. abe , n. chen and h. -j . he , arxiv:1207.4103 [ hep - ph ] . e. gabrielli , b. mele and m. raidal , arxiv:1202.1796 [ hep - ph ] ; e. l. berger , z. sullivan and h. zhang , arxiv:1203.6645 [ hep - ph ] ; e. gabrielli , k. kannike , b. mele , a. racioppi and m. raidal , phys . rev . d * 86 * , 055014 ( 2012 ) [ arxiv:1204.0080 [ hep - ph ] ] . k. cheung and t. -c . yuan , phys . lett . * 108 * , 141602 ( 2012 ) [ arxiv:1112.4146 [ hep - ph ] ] ; v. barger , m. ishida and w. -y . keung , phys . lett . * 108 * , 101802 ( 2012 ) [ arxiv:1111.4473 [ hep - ph ] ] ; _ ibid . d * 85 * , 015024 ( 2012 ) [ arxiv:1111.2580 [ hep - ph ] ] ; b. grzadkowski , j. f. gunion and m. toharia , phys . b * 712 * , 70 ( 2012 ) [ arxiv:1202.5017 [ hep - ph ] ] ; y. tang , arxiv:1204.6145 [ hep - ph ] ; s. matsuzaki and k. yamawaki , phys . rev . d * 85 * , 095020 ( 2012 ) [ arxiv:1201.4722 [ hep - ph ] ] ; h. de sandes and r. rosenfeld , phys . d * 85 * , 053003 ( 2012 ) [ arxiv:1111.2006 [ hep - ph ] ] . a. arhrib , r. benbrik and n. gaur , phys . d * 85 * , 095021 ( 2012 ) [ arxiv:1201.2644 [ hep - ph ] ] ; l. wang and x. -f . han , jhep * 1205 * , 088 ( 2012 ) [ arxiv:1203.4477 [ hep - ph ] ] . p. draper and d. mckeen , phys . d * 85 * , 115023 ( 2012 ) [ arxiv:1204.1061 [ hep - ph ] ] . d. carmi , a. falkowski , e. kuflik and t. volansky , arxiv:1202.3144 [ hep - ph ] ; a. azatov , r. contino and j. galloway , jhep * 1204 * , 127 ( 2012 ) [ arxiv:1202.3415 [ hep - ph ] ] ; j. r. espinosa , c. grojean , m. muhlleitner and m. trott , jhep * 1205 * , 097 ( 2012 ) [ arxiv:1202.3697 [ hep - ph ] ] ; v. barger , m. ishida and w. -y . keung , phys . lett . * 108 * , 261801 ( 2012 ) [ arxiv:1203.3456 [ hep - ph ] ] ; p. p. giardino , k. kannike , m. raidal and a. strumia , jhep * 1206 * , 117 ( 2012 ) [ arxiv:1203.4254 [ hep - ph ] ] ; j. ellis and t. you , jhep * 1206 * , 140 ( 2012 ) [ arxiv:1204.0464 [ hep - ph ] ] ; a. azatov , r. contino , d. del re , j. galloway , m. grassi and s. rahatlou , jhep * 1206 * , 134 ( 2012 ) [ arxiv:1204.4817 [ hep - ph ] ] ; m. klute , r. lafaye , t. plehn , m. rauch and d. zerwas , arxiv:1205.2699 [ hep - ph ] . j. bagger , v. d. barger , k. -m . cheung , j. f. gunion , t. han , g. a. ladinsky , r. rosenfeld and c. -p . yuan , phys . d * 52 * , 3878 ( 1995 ) [ hep - ph/9504426 ] . d. l. rainwater and d. zeppenfeld , jhep * 9712 * , 005 ( 1997 ) [ hep - ph/9712271 ] . k. cheung and o. c. w. kong , phys . d * 68 * , 053003 ( 2003 ) [ hep - ph/0302111 ] . s. akula , p. nath and g. peim , arxiv:1207.1839 [ hep - ph ] ; p. nath , arxiv:1210.0520 [ hep - ph ] . j. f. gunion , h. e. haber , g. l. kane and s. dawson , front . phys . * 80 * , 1 ( 2000 ) . k. hagiwara , j. s. lee and j. nakamura , arxiv:1207.0802 [ hep - ph ] . a. g. akeroyd , j. phys . g g * 24 * , 1983 ( 1998 ) [ hep - ph/9803324 ] . w. goldberger and m. wise , phys . lett . * 83 * , 4922 ( 1999 ) . c. cski , m. graesser , l. randall and j. terning , phys . d * 62 * , 045015 ( 2000 ) ; c. cski , m. l. graesser and g. d. kribs , phys . rev . d * 63 * , 065002 ( 2001 ) [ hep - th/0008151 ] . k. -m . cheung , phys . d * 63 * , 056007 ( 2001 ) [ hep - ph/0009232 ] ; w. d. goldberger , b. grinstein and w. skiba , phys . lett . * 100 * , 111802 ( 2008 ) [ arxiv:0708.1463 [ hep - ph ] ] . m. s. chanowitz , ann . nucl . part . sci . * 38 * , 323 ( 1988 ) . cms collaboration , `` search for the fermiophobic model higgs boson decaying into two photons '' , cms - pas - hig-12 - 002 , march 2012 . m. frank , t. hahn , s. heinemeyer , w. hollik , h. rzehak and g. weiglein , jhep * 0702 * , 047 ( 2007 ) [ hep - ph/0611326 ] . madgraph : j. alwall , m. herquet , f. maltoni , o. mattelaer and t. stelzer , jhep * 1106 * , 128 ( 2011 ) [ arxiv:1106.0522 [ hep - ph ] ] . t. sjostrand , s. mrenna and p. z. skands , comput . commun . * 178 * , 852 ( 2008 ) [ arxiv:0710.3820 [ hep - ph ] ] . the mg / me pythia - pgs package http://madgraph.hep.uiuc.edu/index.html , j. alwall and the cp3 development team .

scale - free percolation is a percolation model on which can be used to model real - world networks . we prove bounds for the graph distance in the regime where vertices have infinite degrees . we fully characterize transience vs. recurrence for dimension 1 and 2 and give sufficient conditions for transience in dimension 3 and higher . finally , we show the existence of a hierarchical structure for parameters where vertices have degrees with infinite variance and obtain bounds on the cluster density .

1604.08180

random graphs are mathematical models commonly used to study real - world networks such as the world - wide web , social , financial , neural , and biological networks . many real - world networks exhibit the following two properties : * _ the small - world property : _ distances within the network are very small in comparison to the number of nodes . with `` small '' we mean distances are at most of order of an iterated logarithm . some real - world networks are even _ ultra - small , _ meaning that the distances are at most a double logarithm . * _ the scale - free property : _ the number of connections per node behave statistically like a power - law . this implies that the variation is typically very high . an example of a random graph model with these properties is the _ norros - reittu random graph _ @xcite ( see figure [ figurenorros ] ) . this model produces a random graph @xmath1 on a fixed set of vertices @xmath2 , but with a random edge set @xmath3 as follows : every vertex @xmath4 is assigned an i.i.d . random weight @xmath5 . conditioned on the weights of its end - vertices , the edge @xmath6 is present in @xmath7 with probability @xmath8 , independently of the status of other possible edges ( here @xmath9 is a normalizing constant ) . see @xcite for more results on inhomogeneous random graphs . these two properties are important , but the structure of many real - life networks , such as social networks , often have other features that influence the structure and formation of networks : * _ geometric clustering : _ in social networks this manifests itself because people who are geographically close to each other are more likely to know each other , giving rise to formation of locally concentrated clusters within the network . * _ hierarchies : _ again in social networks , the more ` important ' people are , the more likely they know other important people , even if those people might be far away , giving rise to hierarchies within the network . a well - known model that has geometric clustering and the connections over long distances required for the existence of hierarchies is _ long - range percolation _ ( lrp , see figure [ figurelrp ] ) @xcite . lrp is a percolation model that produces random subgraphs of the graph @xmath10 wherein an edge @xmath11 is ( independently ) _ retained _ with probability @xmath12 for some positive constants @xmath13 and @xmath14 , and removed otherwise . thus , the connection probabilities are monotonically decreasing in @xmath14 , and increasing in @xmath13 . for many choices of @xmath15 and @xmath14 , lrp has a _ percolation phase transition _ in @xmath13 , meaning that there exists @xmath16 such that when @xmath17 there exists an infinite cluster almost surely , whereas when @xmath18 , all clusters are almost surely finite . when @xmath19 , this model has the clustering property , as well as something akin to the small - world property @xcite . it is , however , clearly not scale - free , since the decay of the degree distribution is faster than exponential . various models have been introduced in the recent years that combine three or four of the network properties described above . we mention , for instance , the models introduced by aiello _ et al . _ @xcite , flaxman , frieze , and vera @xcite , and jacob and mrters @xcite . in this paper we consider another model that has all four properties : _ scale - free percolation _ ( sfp , also known as _ heterogeneous long - range percolation _ ) . sfp interpolates between long - range percolation and the norros - reittu random graph ( see figure [ figurescalefree ] ) . sfp was introduced by deijfen , van der hofstad , and hooghiemstra in @xcite . we start with a formal definition of the model . [ sfpdef ] consider the graph @xmath10 for some fixed @xmath20 . assign to each vertex @xmath21 an i.i.d . weight @xmath22 , where the weights follow a power - law distribution with parameter @xmath23 : @xmath24 where @xmath25 is a slowly - varying function ( i.e. , @xmath26 for all @xmath27 as @xmath28 , so the law of @xmath22 is @xmath29-regularly varying ) . conditionally on the weights , an edge @xmath30 is retained independently of all other edges with probability @xmath31 where @xmath32 and @xmath33 are positive constants of the model .- norm because it is a practical metric , but defining sfp with respect to any @xmath34-norm with @xmath35 $ ] gives qualitatively similar results . ] the edge is removed otherwise . we call retained edges _ open _ , and removed edges _ closed . _ we denote the joint probability measure of edge occupation and weights by @xmath36 ( where the subscript @xmath37 refers to the _ law _ of the weights , not the actual values ) and write just @xmath38 if the parameters are clear from the context . .48 .48 .96 before we proceed with our results , let us briefly summarize some important features of sfp , as proved by deijfen , van der hofstad , and hooghiemstra @xcite , and by deprez , hazra , and wthrich @xcite . it turns out that the following parameter is frequently useful to describe the behaviour of sfp concisely : @xmath39 like long - range percolation , sfp on @xmath0 with parameter @xmath14 and i.i.d . vertex weights whose law @xmath37 is @xmath29-regularly varying has a percolation phase - transition in @xmath13 at @xmath40 this phase transition is non - trivial , except when @xmath41 and @xmath42 , in which case @xmath43 , and when @xmath44 , @xmath45 , and @xmath46 , in which case @xmath47 @xcite . in the regime where sfp percolates , the infinite cluster @xmath48 is almost surely unique @xcite . _ show that the percolation density of sfp is continuous when @xmath49 : at @xmath50 there is no infinite cluster almost surely @xcite . by the choice of the power - law distribution , this model is scale - free indeed , the degrees @xmath51 follow a power - law of the form @xmath52 for some slowly varying function @xmath53 @xcite . this shows that the model behaves differently from long - range percolation . many real - world networks are believed to have infinite variance degree distributions . sfp has infinite variance degrees when @xmath42 . when @xmath54 , sfp locally behaves like an ultra - small world @xcite . under the assumption that the weights are bounded away from 0 , the probability that an edge is open in scale - free percolation with parameters @xmath55 and @xmath13 stochastically dominates the probability that an edge is open in long - range percolation with parameters @xmath14 and some @xmath56 . et al . _ @xcite use this domination to show that sfp locally has the small - world and clustering properties when @xmath49 , analogous to long - range percolation @xcite . given a graph @xmath57 , the _ graph distance on @xmath58 _ between any @xmath59 is defined as @xmath60 with the conventions that @xmath61 and @xmath62 if @xmath63 and @xmath64 are not in the same connected component of the graph . we define the _ diameter _ of @xmath58 as the maximal distance between two vertices in @xmath1 , i.e. , @xmath65 . the infinite random subgraph @xmath66 of @xmath10 corresponding to the infinite component of supercritical sfp thus naturally produces a random metric on @xmath0 . we write @xmath67 for this metric . we write @xmath68 for the minimum of @xmath63 and @xmath64 . our first result is the proof of a conjecture by deijfen _ et al . _ @xcite . [ thmgraphdiameter ] consider sfp on @xmath0 with @xmath69 , @xmath70 , and with i.i.d . vertex weights whose law @xmath37 satisfies for some @xmath71 and some @xmath72 , @xmath73 then @xmath74 almost surely when @xmath75 , and @xmath76 almost surely when @xmath77 . note that implies @xmath78 , thus the weights are bounded away from @xmath79 . see figure [ phasesdistances ] for an overview of the graph distances in which we combine the results of the present paper and those of @xcite . theorem [ thmgraphdiameter ] thus complements the characterization of distances . our proof for the case @xmath80 is based on the proof of a similar result for long - range percolation with @xmath80 by benjamini , kesten , peres , and schramm @xcite . for the norros - reittu random graph a similar result to theorem [ thmgraphdiameter ] is known : van den esker _ et al . _ @xcite prove that when the weights are distributed as an infinite - mean power - law , then the diameter of the graph is almost surely @xmath81 or @xmath82 ( more precise results are obtained under extra conditions ) . graph distances are one way of characterizing the geometry of a graph . another way of doing this is by studying the behaviour of random walk on the graph . the notions of _ transience _ and _ recurrence _ are particularly relevant : a simple random walk on a locally finite graph @xmath1 is a sequence @xmath83 with @xmath84 where @xmath85 is chosen uniformly at random from the `` neighbours '' of @xmath86 , i.e. , @xmath87 independently of @xmath88 . a graph is called _ recurrent _ if for every @xmath89 a random walk returns almost surely to its starting point @xmath89 . a graph is called _ transient _ if it is not recurrent . we prove the following two theorems , the results of which are summarized in the phase diagram in figure [ phasesrandomwalk ] . [ transientrandomthm ] consider sfp on @xmath0 with @xmath69 , i.i.d . vertex weights whose law @xmath37 satisfies , either @xmath90 or @xmath91 , or both , and @xmath92 . then the infinite cluster of sfp is transient almost surely . recall plya s theorem , which states that the lattice of @xmath0 with nearest neighbour edges is recurrent if and only if @xmath93 , and transient otherwise . therefore , transience in these dimensions shows a dramatic difference to regular lattices . berger @xcite proved for lrp that the random walk is transient in one or two dimensions if and only if @xmath49 . for sfp the result is stronger : for any @xmath94 , there exists @xmath95 such that the infinite cluster is transient . [ recurrentrandomwalk2d ] consider sfp on @xmath0 with @xmath96 , i.i.d . vertex weights whose law @xmath37 satisfies @xmath97 for some @xmath71 and @xmath72 , @xmath98 , and @xmath99 , and such that either @xmath100 or @xmath101 , or both . then the infinite percolation cluster is recurrent @xmath102-almost surely . note that , as mentioned before , in dimension 1 when @xmath45 and @xmath103 there is no infinite cluster almost surely @xcite , so in this case a random walk is trivially recurrent . we therefore give a full characterization of recurrence and transience of sfp in dimension one and two , while for @xmath104 we only characterize it when @xmath105 or @xmath42 . for nearest - neighbour percolation it is known that the infinite cluster is transient @xcite . it would be interesting to verify whether this is true for other percolation models on @xmath0 , in particular for scale - free percolation or long - range percolation . we show that sfp has the geometric clustering property not only for @xmath49 as shown by deprez _ et al . _ * theorem 6 ) , but also when @xmath106 . moreover , these clusters can be organized in a hierarchical structure , a phenomenon that is also present in some real - life networks ( see for example ( * ? ? ? * chapter 13 ) or ( * ? ? ? * chapter 9 ) ) . these hierarchical structures are not only present in finite boxes , they extend throughout @xmath0 . indeed , the infinite component of sfp contains an infinite subgraph exhibiting a prescribed hierarchy . we introduce the notion of a _ hierarchically clustered tree_. [ def : hct ] fix @xmath107 and @xmath21 . let @xmath108^d\cap { \mathbb{z}^d}$ ] . consider the set of trees @xmath109 of all unrooted , connected , cycle - free subgraphs of @xmath110 ( i.e. , trees on @xmath111 ) , where each vertex @xmath112 in such a tree is endowed with a weight @xmath113 . fix @xmath114 $ ] and @xmath115 . we call an element @xmath116 an _ @xmath117-hierarchically clustered tree _ if @xmath118 ( i.e. , @xmath119 is the isolated vertex @xmath63 with a weight ) . for @xmath120 , we call an element @xmath121 an _ @xmath122-hierarchically clustered tree _ if the following four properties hold : 1 . _ [ positive density ] _ @xmath119 contains at least a fraction @xmath123 of all the vertices in the box @xmath111 : @xmath124 2 . _ [ ultra - small world ] _ @xmath119 is an ultra - small world in the sense that @xmath125 3 . _ [ ordered weights ] _ if we root @xmath119 at its maximum - weight vertex , then , for any vertex in the tree , the weights decrease step - by - step along the path from the root to that vertex . [ spatial clustering ] _ if we remove any given edge from @xmath119 , then there exists an @xmath126 ( depending on @xmath119 and the removed edge ) such that the two trees @xmath127 and @xmath128 that remain satisfy 1 . at least one ( say @xmath129 ) is an @xmath130-hierarchically clustered tree for some @xmath131 , and 2 . the other ( say @xmath132 ) has its vertex set @xmath133 disjoint with the box on which @xmath129 is defined : @xmath134 note that condition @xmath135 together with condition @xmath136 implies that there exists @xmath137 , such that for all @xmath107 @xmath138 so hierarchically clustered trees combine a topological and a spatial version of the ultra - small world property . [ thm : trees ] consider sfp on @xmath0 with @xmath69 , i.i.d . vertex weights whose law @xmath37 satisfies , with @xmath139 , and any @xmath140 . let @xmath141 denote the sfp configuration _ inside _ the cube @xmath142^d$ ] . there exist @xmath143 , a density @xmath144 and a constant @xmath145 , such that 1 . for all @xmath146 , @xmath147 2 . the infinite component @xmath48 contains a.s . an infinite , connected , cycle - free subgraph @xmath148 such that if we remove any given edge from @xmath148 , a finite and infinite connected component remain and there exist @xmath21 and @xmath107 such that the finite connected component is an @xmath149-hierarchically clustered tree . this paper combined with @xcite gives bounds on the graph distance for every value in the parameter space , but the picture is not yet complete . we were not able to prove a non - trivial lower bound on the diameter in the regime where @xmath42 and @xmath150 , and it is not clear to us that the upper bound is sharp . and in the regime @xmath49 where @xmath45 , there is a gap in the bounds on graph distances , since there the best known bounds are @xcite : @xmath151 what is the right asymptotics of @xmath152 in this regime ? in section [ sectionclusters ] below we determine that the bound on @xmath153 in theorem [ thm : trees ] is @xmath154 . biskup @xcite shows a result rather similar to theorem [ thm : trees ] on the clustering density for long - range percolation when @xmath49 , where @xmath155 . the corresponding range for @xmath153 for scale - free percolation would be @xmath156 . it might be possible to extend theorem [ thm : trees ] to hold for this regime of @xmath153 . scale - free percolation is defined as a model on the infinite lattice @xmath0 . a challenging question is the study of scale - free percolation , and in particular its critical behaviour , on the finite torus . working on the torus keeps the translation invariance and provides the opportunity to compare the model to its non - spatial counterparts , such as the norros - reittu random graph @xcite . scale free percolation on finite boxes is strongly related to geometric variants of the norros - reittu model or the chung - lu model . for example , bringmann , keusch , and lengler @xcite introduce _ geometric inhomogenous random graphs _ , which generalise a certain class of hyperbolic random graphs , and which could be described as `` continuous sfp on the torus '' . indeed , they do not use a grid , but place the points randomly . in a fairly general setup , where , contrary to our model , the connection probability does not need to approach to 1 as @xmath157 goes to infinity , these authors prove that such graphs are ultra - small @xcite . moreover , they claim that their results also carry over to finite boxes . because of this more general setup , it would be interesting ( but possibly not straightforward ) to see whether in their setting hierarchically clustered trees are also present . the proofs of the main results partly rely on a number of elementary properties of the vertex weights . we begin by proving these properties in section [ sectionpreliminary ] . in section [ sectiondistance ] we prove the boundedness of the graph distance for @xmath158 and @xmath159 . in section [ sectionrw ] we prove the random walk results , and in section [ sectionclusters ] we prove theorem [ thm : trees ] on hierarchical clustering . we start by introducing some basic notation and definitions . given two percolation configurations @xmath160 , we write @xmath161 if @xmath162 when @xmath163 for all @xmath164 , i.e. , all edges that are open in @xmath165 are also open in @xmath166 . we say that an event @xmath167 is _ increasing _ if @xmath168 implies @xmath169 for all @xmath161 . given two random variables @xmath170 and @xmath171 , we say that @xmath171 _ stochastically dominates _ @xmath170 if for every @xmath172 the inequality @xmath173 holds , and we write @xmath174 . [ obs : increasing ] let @xmath37 and @xmath175 be random variables such that @xmath176 . for any increasing event @xmath167 , @xmath177 this lemma can be proved with a straightforward coupling argument that we leave to the reader . we commonly use lemma [ obs : increasing ] to simplify the law of @xmath37 : if the law of @xmath37 satisfies and the law of @xmath175 satisfies @xmath178 with the same constant @xmath179 as in , then holds . the upcoming lemmas allow us to construct a coarse - graining argument in the proofs of theorems [ transientrandomthm ] and [ thm : trees ] . [ powerlawmultiplyscalar ] let @xmath37 be a random variable with law given by . let @xmath180 be a random variable with law given by @xmath181 then , for @xmath182 , the conditional law of @xmath37 given @xmath183 is the same as the law of @xmath184 , i.e. , @xmath185 . for @xmath186 @xmath187 [ maximumweight ] let @xmath188 be an i.i.d . sequence of random variables with law given by . then , for all @xmath189 and all @xmath190 , @xmath191 using that the weights are i.i.d . , that @xmath192 , and that @xmath193 , we can bound the left - hand side by @xmath194 [ transientlemmabigdegrees ] fix an integer @xmath41 and @xmath195 such that @xmath196 . assign to each vertex in @xmath197^d \subset { \mathbb{z}^d}$ ] an i.i.d . random variable with law satisfying . let @xmath198 be the event that the box @xmath197^d$ ] contains at least @xmath199 vertices with weight larger than @xmath200 . then , for all @xmath201 , @xmath202 let @xmath171 denote the number of vertices in @xmath197^d$ ] with weight exceeding @xmath203 . by and independence of the weights we have @xmath204 , where @xmath205 . note that since @xmath54 we have @xmath206 = c \beta^{-(\tau-1 ) } n^{d(1-\gamma/2 ) } \gg \log n$ ] and var@xmath207 ^ 2 $ ] . it follows by the paley - zygmund inequality that ( when @xmath208 is sufficiently large ) , @xmath209 - \log n)^2}{\mathrm{var}(x ) + { \mathbb{e}}[x]^2 } \longrightarrow 1 . \qed\ ] ] we call any set that is a translate of @xmath197^d \subset { \mathbb{z}^d}$ ] an _ @xmath208-box . _ we say that two @xmath208-boxes @xmath210^d$ ] and @xmath211^d$ ] are `` @xmath212 boxes away from each other '' if @xmath213 ( where we recall that @xmath214 denotes the @xmath215-norm ) . [ maxdistance ] let @xmath69 and @xmath216 . consider two @xmath208-boxes @xmath217 and @xmath218 that are @xmath212 boxes away from each other . for arbitrary @xmath219 and @xmath220 , @xmath221 let @xmath222 and @xmath223 be such that @xmath210^d$ ] and @xmath211^d$ ] . applying the triangle inequality twice , one obtains @xmath224 [ transientlemmaconnectivity ] fix @xmath225 and let @xmath226 and @xmath227 be two @xmath208-boxes that are @xmath212 boxes away from each other such that @xmath228^d$ ] and @xmath229^d$ ] with @xmath230 . let @xmath201 be given , the weights @xmath231 be i.i.d . according to a law satisfying , and @xmath232 be i.i.d . with law . for @xmath233 write @xmath234 then @xmath235 let @xmath236 unif@xmath237 $ ] denote a standard uniform random variable with cdf @xmath238 for @xmath239 $ ] . then , by definition [ sfpdef ] , @xmath240 where the probability measure @xmath241 on the right - hand side is with respect to @xmath242 and @xmath243 , which are i.i.d . with the same law as the elements of @xmath231 , and an independent random variable @xmath244 unif@xmath237 $ ] . using lemmas [ powerlawmultiplyscalar ] and [ maxdistance ] we bound the right - hand side from below by @xmath245 where the probability measure @xmath246 is with respect to @xmath247 and @xmath248 which are i.i.d . with the same law as the elements of @xmath232 and an independent random variable @xmath244 unif@xmath237 $ ] . on the other hand , since @xmath249 , by definition [ sfpdef ] we also have @xmath250 the claim thus follows . by translation invariance of the model , it suffices to show @xmath251 . since we assumed that the law of @xmath37 satisfies , there exists a @xmath252 such that @xmath253 for all @xmath254 almost surely . fix @xmath21 . for @xmath255 , let @xmath256 denote the box centred at @xmath257 with sides of length @xmath258 and let @xmath259 with @xmath260 . note that there are @xmath261 vertices in @xmath262 . we prove that the probability that the vertex with maximal weight for every @xmath262 is connected to both 0 and @xmath63 is strictly greater than some positive constant and let the result follow by borel - cantelli . for each @xmath263 , let @xmath264 be the vertex in @xmath262 with maximal weight and let @xmath265 be the event that @xmath264 is connected by an open edge to both 0 and @xmath63 . let @xmath266 and denote @xmath267 , which is an increasing event . using lemma [ obs : increasing ] and lemma [ maximumweight ] with @xmath268 and @xmath269 , we can bound @xmath270 where the measure @xmath271 refers to a model where all weights are distributed as in . the right hand side of is bounded below by some @xmath272 uniformly in @xmath212 . observe that @xmath273 and recall that @xmath95 and @xmath159 . write @xmath274 . we can bound the probabilities on the events @xmath265 by conditioning on @xmath275 as follows : @xmath276 since this bound is independent of @xmath212 and of the weights @xmath231 , it follows that @xmath277 , and therefore , @xmath278 observe that the events @xmath265 are independent of each other , hence we obtain the result for @xmath159 using the lemma of borel - cantelli . by translation invariance it again suffices to show that @xmath279 for all @xmath254 . recall that the assumption on the law of @xmath37 implies that @xmath280 almost surely . note that @xmath281 is an increasing event . hence by lemma [ obs : increasing ] , @xmath282 observe that sfp with constant vertex weights is equivalent to long - range percolation with the same @xmath15 and @xmath14 and some possibly different parameter @xmath283 . benjamini _ _ et al . _ _ ( * ? ? ? * example 6.1 ) show that the diameter of the infinite cluster in long - range percolation with @xmath77 for any @xmath284 is equal to @xmath285 almost surely . our claim about sfp therefore follows . the proof of theorem [ transientrandomthm ] is inspired by berger s proof of transience for long - range percolation ( * ? ? ? * theorem 1.4(ii ) ) . we use in particular a multiscale ansatz which roots back to the work of newman and schulman @xcite for long - range percolation . * the case @xmath106 . * in view of lemma [ obs : increasing ] , we may assume rather than without loss of generality . we show that the infinite cluster of sfp almost surely contains a transient subgraph . the proof has two steps : 1 . we first assume that @xmath13 is large enough . with small probability we remove some vertices from the graph independent of each other . then we use a multiscale ansatz : we group vertices into finite boxes , and call boxes ` good ' or ` bad ' according to the weights and edge structure _ inside _ the box . we iterate this process by considering larger boxes , which we call good or bad according to the number of good boxes in them and the edges between vertices in those boxes . this will imply transience for large values of @xmath13 . to couple the original model ( for _ any _ @xmath286 ) to the model of the first step , we use a coarse - graining argument : we ` zoom out ' by considering large boxes of vertices and only considering the vertices with maximum weight in the boxes . we show that , with high probability , the weights of these vertices are so high , that the graph , only defined on these vertices , dominates a graph as described in the first step . we use ( * ? ? ? * lemma 2.7 ) , which describes a sufficient structure for a graph to be transient . to this end , we introduce the notion of a `` renormalized graph '' : we start with some notation . given a graph @xmath1 and a sequence @xmath287 let @xmath288 with @xmath289 and @xmath290 be a subset of the vertex set @xmath2 . now let for @xmath291 @xmath292 we call the sets @xmath293 _ bags _ , and the numbers @xmath294 _ bag sizes . _ we say that the graph @xmath1 is _ renormalized for the sequence _ @xmath287 if we can construct an infinite sequence of graphs such that the vertices of the _ @xmath295-th stage graph _ are labelled by @xmath296 for all @xmath297 , and such that for every @xmath298 , every @xmath299 , and all pairs of distinct @xmath300 and @xmath301 there is an edge in @xmath58 between a vertex in @xmath302 and a vertex in @xmath303 the underlying intuition is that every @xmath304-th stage bag contains @xmath294 @xmath305-stage bags , which contains again @xmath306 @xmath307-stage bags . every pair of @xmath307-stage bags in an @xmath304 stage bag is connected by an edge between one of the vertices in the bags ( see figure [ renormalizedgraph ] ) . [ renormalizedtransient ] a graph renormalized for the sequence @xmath294 is transient if @xmath308 . the lemma follows from the proof of ( * ? ? ? * lemma 2.7 ) . [ lemmatransientlargelambda ] consider scale - free percolation with @xmath42 and weight distribution satisfying . independently of this , perform an i.i.d . bernoulli site percolation on the vertices of @xmath0 , colouring a vertex `` green '' with probability @xmath309 $ ] . then the subgraph of the infinite scale - free percolation cluster that is induced by the green vertices has a ( unique ) infinite component @xmath310 . there exists @xmath311 and @xmath312 , such that @xmath310 is transient for @xmath313 and @xmath314 almost surely . the proof exploits a multiscale technique . indeed , we proceed by showing that @xmath310 contains a renormalized subgraph that is transient . therefore , @xmath310 is also transient . for all @xmath315 , let @xmath316 and @xmath317 we partition the lattice @xmath0 into disjoint boxes of side length @xmath318 , so that each such box contains @xmath319 vertices , and call these the _ 1-stage _ boxes . ( by convention we call vertices of @xmath0 the _ 0-stage _ boxes . ) we view these boxes as the vertices of a renormalized lattice . now cover the lattice again , grouping together @xmath320 1-stage boxes to form _ 2-stage _ boxes with sides of length @xmath321 . continue in this fashion , so that the _ @xmath304-stage _ boxes form a covering of @xmath0 by translates of @xmath322^d$ ] . we call a 0-stage box `` good '' if the vertex associated with it is green . for every stage @xmath323 , we define rules for a box to be `` good '' or `` bad '' , depending only on the weights @xmath22 and the edges of @xmath66 inside the box . this implies that disjoint boxes are good or bad independently of each other . a 1-stage box is good if it contains at least @xmath324 good 0-stage boxes and one of the vertices in these boxes has weight at least @xmath325 . for each good @xmath326-stage box , call the maximum - weight vertex , having weight at least @xmath325 , and call it _ 1-dominant . _ for @xmath327 say that an @xmath304-stage box @xmath328 is good if the following three conditions are satisfied : 1 . at least @xmath294 of the @xmath305-stage boxes in @xmath328 are good . 2 . for any good @xmath305-stage box @xmath329 , the @xmath307-dominant vertices in @xmath330 form a _ clique _ ( i.e. , every two @xmath307-dominant vertices in @xmath330 are connected by an edge in @xmath66 ) . 3 . there is an @xmath305-dominant vertex in one of its good @xmath305-stage boxes , with weight at least @xmath331 . for each good @xmath304-stage box , choose the maximum weight vertex and call it the _ @xmath304-dominant _ vertex if its weight is at least @xmath331 . ( a vertex may be dominant for different values of @xmath304 . ) see figure [ renormalizationtransient ] for a sketch of this definition . note that by construction , the subgraph of @xmath66 induced by the vertices that are in a good @xmath304-stage box for every @xmath332 is a graph renormalized by a sequence of bag sizes @xmath333 that satisfies the transience condition of lemma [ renormalizedtransient ] . our aim is therefore to show that almost surely such a subgraph exists . define @xmath334 and @xmath335 to be the events that conditions ( e ) , ( f ) and ( g ) hold for the @xmath304-stage box containing the vertex @xmath112 . to simplify notation , define @xmath336 and @xmath337 . we write @xmath338 and @xmath339 for the events that the @xmath304-th stage boxes containing @xmath112 and @xmath79 , respectively , are good . by translation invariance it is sufficient to show that @xmath340 the events @xmath339 are positively correlated , hence it is sufficient to show that @xmath341 we bound @xmath342 first , we give an upper bound for @xmath343 . recall that we use the @xmath215-norm for distance in the definition of the edge - probabilities of sfp . the @xmath215-distance between two vertices in the same @xmath304-stage box is at most @xmath344 the probability that two good @xmath307-stage boxes are _ not _ connected by an open edge between its @xmath307-dominant vertices ( which have weight at least @xmath345 ) is therefore at most @xmath346^\alpha\right)^22^{-n\alpha}((n+1)!)^{-2\alpha}\right)\\ & = \exp(-\lambda n^{d(2-\gamma)}).\end{aligned}\ ] ] there are @xmath347 pairs of @xmath307-stage boxes inside an @xmath304-stage box , so there can be at most @xmath348 edges between @xmath307-dominant vertices inside a good @xmath304-stage box . it follows by taking the union bound that @xmath349 we proceed by establishing an upper bound on @xmath350 . there exists a constant @xmath351 such that @xmath352 note that any good @xmath304-stage box contains at least @xmath353 @xmath305-dominant vertices that all have weight larger than @xmath354 . using , lemma [ maximumweight ] , and @xmath355 , gives for some @xmath356 that @xmath357 the last term we bound is @xmath358 . all @xmath305-stage boxes are good independent of each other with probability @xmath359 . let @xmath360 be binomially distributed , so that @xmath361 . we use chernoff s bound that if @xmath362 , @xmath363 , then @xmath364 for our model , this obtains @xmath365 combining , , , and , gives that @xmath366 if @xmath13 large enough ( say larger than @xmath367 ) , there exists @xmath368 such that for @xmath369 @xmath370 define the sequence @xmath371 and observe that @xmath372 for any fixed @xmath373 , we can find @xmath374 and @xmath311 such that @xmath375 , because @xmath376 depends only on the weights and edges inside a _ finite _ box . we further bound for all @xmath377 by @xmath378 and choose @xmath379 so large , that the last bound in holds . thus , using , and @xmath380 for all @xmath304 , yields that @xmath381 with probability 1 the graph contains a cluster of good vertices that can be renormalized for the sequence @xmath294 . by lemma [ renormalizedtransient ] this cluster is transient itself , since showing transience for a subgraph is enough for transience on the whole graph ( * ? ? ? * section 9 ) . * the case @xmath382 . * we need two lemmas from the literature , which are complementary to the case @xmath49 of proposition [ lemmatransientlargelambda ] and lemma [ transientlemmabigdegrees ] . [ clustersizetransient]assume @xmath383 and let @xmath49 . choose @xmath286 and let @xmath384 . for every @xmath385 and @xmath201 there exists @xmath386 such that for all @xmath387 @xmath388 where @xmath389 is the largest connected component in @xmath142^d$ ] . note that ( * ? ? ? * lemma 9 ) is proven for the exact power law distribution of the weights in . this is not a problem , since lemma [ obs : increasing ] implies that the result extends to a weight distribution satisfying when @xmath390 . for @xmath391 , and percolation parameter @xmath56 , we observe that the model is equivalent to the case where @xmath392 and @xmath393 , since if the law of @xmath37 satisfies for some @xmath252 and for @xmath175 it holds for @xmath394 that @xmath395 , then @xmath396 hence , we can scale the parameters such that @xmath392 and apply lemma [ obs : increasing ] . [ sitebondlongrangelemma ] let @xmath20 , @xmath49 and @xmath284 . consider the long - range percolation model on @xmath0 in which every two vertices , @xmath63 and @xmath64 , are connected by an open edge with probability @xmath397 independently of other edges , and every vertex is good with probability @xmath398 , independently of all other vertices . there exist @xmath399 and @xmath400 , such that if @xmath401 and @xmath402 , the infinite cluster on the good vertices is transient . [ largeconnectivity ] consider scale - free percolation with weight distribution satisfying . let @xmath226 and @xmath227 be two @xmath208-boxes that are @xmath212 boxes away from each other . let @xmath201 be given . moreover , assume that @xmath226 and @xmath227 contain connected components @xmath403 and @xmath404 , respectively , of size at least @xmath200 . then @xmath405 since we assumed , all weights are at least @xmath406 . by lemma [ maxdistance ] , we get for arbitrarily chosen vertices @xmath407 that @xmath408 since both clusters contain at least @xmath200 vertices , there are at least @xmath409 possible edges . we obtain @xmath410 note that this is the @xmath49-counterpart of lemma [ transientlemmaconnectivity ] . the previous lemmas readily imply the result for sufficiently large @xmath13 , and we are left to extend this to all @xmath286 , which we achieve via coarse - graining . when @xmath42 , let @xmath367 and @xmath411 be the values that we obtain from proposition [ lemmatransientlargelambda ] . to apply the proposition for all @xmath412 , we partition @xmath0 into ( hyper)cubes of side length @xmath208 ( for some large @xmath208 to be determined below ) , which we call @xmath208-boxes . in every @xmath208-box we identify the vertex in it with maximum weight and call it the dominant vertex . we now choose @xmath413 large enough so that @xmath414 . second , we call those @xmath208-boxes good that contain a vertex with weight at least @xmath200 . we choose @xmath208 large enough so that the probability that an @xmath208-box is good is larger than @xmath411 using lemma [ transientlemmabigdegrees ] . lemma [ transientlemmaconnectivity ] implies that the probability that the dominant vertices in two good @xmath208-boxes , being _ @xmath212 boxes away from each other _ , are connected , is bounded from below by @xmath415 , where @xmath230 such that @xmath249 , and where @xmath416 are i.i.d . distributed according to . thus , the status of the edges between dominant vertices in good @xmath208-boxes stochastically dominates an sfp model on @xmath0 with parameters @xmath14 , @xmath367 and weight - law @xmath180 , combined with a site percolation of intensity @xmath411 , exactly as described in proposition [ lemmatransientlargelambda ] . we now apply proposition [ lemmatransientlargelambda ] to obtain the result for the case @xmath54 . the case @xmath417 is analogous : when @xmath49 , let @xmath400 and @xmath418 be the values that we obtain from lemma [ sitebondlongrangelemma ] . to apply the lemma , we partition @xmath0 into @xmath208-boxes again . choose @xmath413 large enough , using lemma [ largeconnectivity ] , such that two @xmath208-boxes being @xmath212-boxes away from each other , having clusters @xmath403 and @xmath404 with size at least @xmath200 are connected by an open edge between @xmath403 and @xmath404 with probability at least @xmath419 . call the @xmath208-boxes that contain a cluster of size at least @xmath200 the good boxes . choose @xmath208 large enough so that the probability that an @xmath208-box is good is larger than @xmath418 , using lemma [ clustersizetransient ] . we thus find that the status of the edges between the dominant vertices of good @xmath208-boxes stochastically dominates an lrp model on @xmath0 with independent edge probabilities @xmath420 , combined with a site - percolation of intensity @xmath418 . an application of lemma [ sitebondlongrangelemma ] thus obtains the result for the case @xmath417 . we conclude that in both cases we found a subgraph of the infinite cluster on which the random walk is transient , and hence the random walk is transient on the infinite cluster itself , cf . * section 9 ) . we verify that we can apply the following lemma . [ lemma2drecurrent ] let @xmath421 and let @xmath422 be a family of probabilities , such that @xmath423 consider a shift invariant percolation model on @xmath424 on which the bond between @xmath425 and @xmath426 is open with marginal probability @xmath427 . if there exists an infinite cluster , then this cluster is recurrent . to bound the marginal probabilities we need a bound on the expectation of the product of the weights . [ upperboundproductweights ] assume that the weight - distribution satisfies . let @xmath428 be two independent copies of the random variable @xmath37 . if @xmath429 , there exists a constant @xmath430 , such that for @xmath431 : @xmath432\leq c u^{-1}.\ ] ] if @xmath433 , then there exists a constant @xmath430 , such that @xmath432\leq c \log(u)u^{-(\tau-1)}.\ ] ] the proof for @xmath429 is straightforward . observe that @xmath434<\infty$ ] , hence @xmath432\leq \mathbb{e}[w]^2/u.\ ] ] for @xmath433 , we prove the claim for weights that satisfy for some @xmath390 . lemma [ obs : increasing ] then implies that the claim holds for weights that satisfy . let @xmath435 denote the distribution function of @xmath436 . from ( * lemma 4.3 ) we have for some @xmath437 that @xmath438 by @xmath439 we obtain the result @xmath440&=1-h(u)+\frac{1}{u}\int_1^u vdh(v)\\ & \leq 1-h(u)+\frac{1}{u}\int\limits_1^u ( 1-h(v))dv\\ & \leq c'\log(u)u^{-(\tau-1)}+\frac{c'}{u}\int\limits_1^u\log(v)v^{-(\tau-1)}dv\\ & \leq c'\log(u)u^{-(\tau-1)}+\frac{c''}{u}\log(u)u^{-(\tau-2)}\\ & \leq c \log(u)u^{-(\tau-1)}. \qed \end{split}\ ] ] observe that the scale - free percolation measure @xmath102 is indeed shift invariant . according to lemma [ lemma2drecurrent ] , we need to prove that @xmath441 whenever @xmath442 . for convenience , we treat only @xmath443 , the other cases follow analogously . lemma [ upperboundproductweights ] and the bound @xmath444 give @xmath445\\ & \leq \limsup_{k\rightarrow\infty } k^4\mathbb{e}\left[1-e^{-\lambda\frac{w_{(k,0)}w_{(0,0)}}{k^\alpha}}\right]\\ & \leq \limsup_{k\rightarrow\infty } k^4\mathbb{e}\left[\left(\lambda\frac{w_{(k,0)}w_{(0,0)}}{k^\alpha}\right)\wedge1\right].\end{aligned}\ ] ] for @xmath429 , recall that @xmath446 , and therefore @xmath447 \leq \limsup_{k\rightarrow\infty}c \lambda k^{4-\alpha}\leq c \lambda<\infty.\ ] ] for @xmath433 and @xmath45 , @xmath448&\leq \limsup_{k\rightarrow\infty}k^4c \lambda^{\tau-1 } \log(k^\alpha/\lambda)k^{-\alpha(\tau-1)}\\ & \leq \limsup_{k\rightarrow\infty } c'\log(k)k^{4-\alpha(\tau-1)}\\ & = \limsup_{k\rightarrow\infty } c ' \log(k)k^{4 - 2\gamma}\\ & = 0 . \qed\end{aligned}\ ] ] for the proof of theorem [ thm : trees ] we largely use the same two steps as for the proof of theorem [ transientrandomthm ] . first we prove the result for large values of @xmath13 on an sfp model combined with an i.i.d . bernoulli site percolation , but only along a sequence @xmath449 diverging to infinity . then , we prove the result in this sfp model for all sufficiently large @xmath450 . lastly , by a coarse graining argument , we extend the result to hold for all @xmath451 . we assume the weights satisfy . lemma [ obs : increasing ] then implies that the claim holds for weights that satisfy . we start with stating and proving a proposition in which we consider a site - percolated version of sfp . [ lemlargestcluster ] consider sfp on @xmath0 with @xmath41 , @xmath42 , and weight distribution that satisfies . independently of this , perform an i.i.d . bernoulli site percolation on the vertices of @xmath0 , colouring a vertex `` green '' with probability @xmath309 $ ] . denote by @xmath310 the ( unique ) infinite subgraph of the infinite scale - free percolation cluster induced by the green vertices . we call this the site - percolated sfp . 1 . the probability that the site - percolated sfp configuration with parameters @xmath456 and @xmath457 contains a @xmath458-hierarchically clustered tree inside the box @xmath459^d$ ] , is bounded from below by @xmath460 2 . site - percolated sfp with @xmath457 and @xmath456 has an infinite component @xmath461 almost surely . @xmath461 contains a.s . an infinite , connected , cycle - free subgraph @xmath148 such that removing an arbitrary edge yields a finite and infinite connected component and there exist @xmath21 and @xmath462 such that the finite connected component is an @xmath122-hierarchically clustered tree . let @xmath318 be a large integer to be determined later , and let @xmath463 , with @xmath464}$ ] , be a sequence also to be determined later , such that @xmath465 let @xmath466 the bound @xmath467 implies @xmath468 for all @xmath469 , let @xmath470 , so that we have the telescoping product @xmath471 and let @xmath472 we give a sequence of graphs for @xmath0 by the same procedure as in proposition [ lemmatransientlargelambda ] . we call the vertices the @xmath79-stage boxes . we partition the lattice @xmath0 into boxes of side - length @xmath318 , so that each box contains @xmath319 vertices , and call these the @xmath326-stage boxes . iteratively , we group @xmath473 @xmath305-stage boxes into @xmath304-stage boxes , so that the @xmath304-stage boxes form a covering of @xmath0 by translates of @xmath474^d$ ] . we call a @xmath79-stage box `` good '' if the associated vertex is green , and we call this vertex `` @xmath79-dominant '' . for every stage @xmath475 , we define rules for a box to be `` good '' or `` bad '' and for a vertex to be `` @xmath304-dominant '' depending only on the weights @xmath22 and colours of the vertices and on the edges of @xmath476 inside the box . this implies that disjoint boxes are good or bad independently of each other . 1 . at least @xmath294 of the @xmath305-stage boxes it contains are good . the maximum weight @xmath305-dominant vertex in one of its good @xmath305-stage boxes has weight at least @xmath331 . call this vertex @xmath304-dominant . ( a vertex can be dominant for multiple stages . ) all @xmath305-dominant vertices in the good @xmath305-stage boxes are connected to the @xmath304-dominant vertex by an edge in @xmath310 . define @xmath334 and @xmath335 to be the events that respectively ( e ) , ( f ) and ( g ) hold for the @xmath304-stage box containing the vertex @xmath112 . denote by @xmath338 the event that the @xmath304-stage box containing @xmath112 is good , i.e. , @xmath478 . to simplify notation we write @xmath479 and @xmath480 . 1 . start with the set of all @xmath79-dominant vertices inside the @xmath304-stage box containing @xmath112 . 2 . for every @xmath481 connect each @xmath482-dominant vertex @xmath112 to the @xmath483-dominant vertex inside the @xmath483-stage box that contains @xmath112 ( unless this creates a self - loop ) . we sequence @xmath484 and obtain a bound : @xmath485 we claim that the constructed connected component of the @xmath304-dominant vertex in the @xmath304-stage box of @xmath112 is a tree that satisfies the conditions given in definition [ def : hct ] . since holds , the event @xmath486 implies that the intersection of the site - percolated sfp configuration and the cube @xmath487^d$ ] contains a tree with at least @xmath488 vertices , which verifies condition ( 1 ) of definition [ def : hct ] . we obtain condition ( 2 ) for some @xmath115 because the box - size @xmath489 as well as the number of vertices in @xmath459^d$ ] both grow double exponentially fast in @xmath304 . conditions ( 3 ) and ( 4 ) follow straightforwardly from the construction . we therefore conclude that proposition [ lemlargestcluster](1 ) follows if we show that @xmath490 and that proposition [ lemlargestcluster](2 ) follows if we show that @xmath491 the events @xmath339 are positively correlated , hence it is sufficient for assertion ( 2 ) of the theorem that @xmath492 , which follows from . it thus remains to prove . we start with proving two bounds on @xmath343 . the conditioning on @xmath494 gives that the sum of cluster sizes in the good @xmath305-stage boxes is at least @xmath495 . using lemma [ maximumweight ] with @xmath496 , the definition of @xmath497 yields that @xmath498 similarly , the conditioning on @xmath494 gives that there are at least @xmath294 vertices with weight at least @xmath354 . using lemma [ maximumweight ] with @xmath499 , the definition of @xmath497 yields that @xmath500 note that it is not a - priori clear which of these two bounds is better , since this may depend on the choice of @xmath501 and the sequence @xmath502 . therefore we use both bounds . we move on to @xmath503 . for @xmath504 define @xmath505 since @xmath506 by definition , it follows that @xmath507 substituting the values of @xmath508 , and gives @xmath509 it follows that for some @xmath252 , @xmath510 the distance between two @xmath305-dominant vertices in the same @xmath304-stage box is at most @xmath511 . there are at most @xmath512 good @xmath305-dominant vertices ( having weight at least @xmath354 ) . recalling , by the union bound and the conditioning on @xmath494 and @xmath513 , we thus obtain that @xmath514 since @xmath515 grows double exponentially fast and @xmath516 , we obtain for @xmath304 sufficiently large ( larger than @xmath368 , say ) , that @xmath517 we move on to @xmath358 . all @xmath305-stage boxes are good independently of each other with probability @xmath359 . let @xmath518 . then , @xmath519 as in , we apply chernoff s bound and obtain @xmath520 whenever @xmath521 . we now choose @xmath522 we will show below that @xmath523 for all @xmath524 and that @xmath525 , as required by . assuming these inequalities we have @xmath526 so that using @xmath527 , @xmath528 , , and yields @xmath529 which gives the desired bound . for later reference we note that if instead we apply , , and , then we obtain for @xmath304 sufficiently large @xmath530 all that remains is to show that @xmath523 for all @xmath524 and that @xmath525 . for positivity of @xmath531 , it is by sufficient to show that for some @xmath532 and @xmath318 sufficiently large @xmath533 since @xmath534 , @xmath535 and @xmath536 by , we obtain @xmath537 we show that @xmath538 by definition @xmath539 . so we derive , after rearranging terms and dividing by @xmath540 , that @xmath541 squaring and substituting @xmath355 yield @xmath542 so that after rearranging @xmath543 hence , we obtain the result after dividing by @xmath544 and , using , substituting @xmath545 : @xmath546 which can be inverted to give @xmath538 hence , there exists a constant @xmath532 such that , when we choose @xmath318 sufficiently large , @xmath547 by we have @xmath548 , so follows , and we may conclude that @xmath549 . observe that @xmath550 if and only if @xmath551 , since @xmath531 approaches @xmath552 double exponentially fast . moreover , combining and gives @xmath553 so that , using , we can bound @xmath554 define the sequence @xmath555 and observe that @xmath556 for any fixed @xmath373 , we can find @xmath374 and @xmath311 such that @xmath375 , because @xmath376 depends only on the weights and edges inside a _ finite _ box . since @xmath557 grows double exponentially fast , we further bound for all @xmath377 by @xmath558 we thus choose @xmath379 so large that the last bound in holds . using , and @xmath380 for all @xmath304 , yields that @xmath381 to prove theorem [ thm : trees ] we need extend the above claims from the specific sequence @xmath449 to _ all _ ( sufficiently large ) @xmath559 . this extension is the content of the next lemma . after this lemma we extend the claim to hold for all @xmath284 and weights following a power - law given by . [ lemma : density ] consider sfp on @xmath0 with @xmath560 , and weight distribution that satisfies . independently of this , perform i.i.d . bernoulli site percolation on the vertices of @xmath0 , colouring a vertex `` green '' with probability @xmath561 $ ] . denote by @xmath562 the sfp - realization induced by the green vertices in the box @xmath563^d$ ] . we call this the site - percolated sfp . then there exist a density @xmath564 and constants @xmath565 , and @xmath566 , such that for @xmath146 , and parameters @xmath567 , the probability that @xmath562 with parameters contains a @xmath568-hierarchically clustered tree inside the box @xmath142^d$ ] is bounded from below by @xmath569 whenever @xmath570 . let the constants @xmath571 , the sequences @xmath572 and @xmath573 be as in proposition [ lemlargestcluster ] , and @xmath545 as in . assume @xmath574 , and let @xmath450 be large enough ( how large precisely will be determined in several steps ) . we define @xmath575 both depending on @xmath450 , and note that @xmath576 as @xmath577 . partition the box @xmath578^d$ ] into @xmath579 boxes of side - length @xmath489 . we call these the @xmath304-boxes . let @xmath580 be the vertex in @xmath578^d$ ] with maximum weight . we use the same definition of good boxes and dominant vertices as in the proof of proposition [ lemlargestcluster ] . so in particular , an @xmath304-box is good if its @xmath304-dominant vertex has weight at least @xmath581 . we define @xmath7 to be the event that at least @xmath582 @xmath304-boxes are good , @xmath583 the event that @xmath584 , and @xmath58 the event that every good @xmath304-box s @xmath304-dominant vertex is connected by an open edge to @xmath580 . let @xmath585 . observe that the event @xmath25 implies indeed that there exists a @xmath586-hierarchically clustered tree . properties ( 1 ) , ( 3 ) and ( 4 ) of definition [ def : hct ] readily follow from proposition [ lemlargestcluster ] . for property ( 2 ) , we observe that , for any @xmath587 , the diameter of the constructed tree , after connecting the separate trees via @xmath580 , only increases by 2 in comparison to proposition [ lemlargestcluster ] . hence there exists @xmath137 , such that property ( 2 ) is satisfied . we bound @xmath588 and analyze the three summands term by term . the @xmath591-distance between two vertices in @xmath578^d$ ] is bounded above by @xmath592 , and the number of @xmath304-dominant vertices is at most @xmath579 . therefore , @xmath593 we claim that @xmath594 indeed , by our choice of @xmath304 and @xmath212 in , @xmath595 , and @xmath596 by their definitions in and , we have @xmath597 iterating this gives @xmath598 recall that by @xmath599 , and by @xmath600 , so @xmath601 hence , we can further estimate for some @xmath602 @xmath603 so that @xmath604 and finally @xmath605 from which follows . consequently , by , @xmath606 where we choose @xmath13 sufficiently large for the second bound ( this determines the value of @xmath607 ) . it remains to bound @xmath608 . by proposition [ lemlargestcluster ] , there exists @xmath564 , such that for @xmath304 large , @xmath304-boxes are good independently of each other with probability at least @xmath609 . let @xmath610 . writing out the binomial distribution , using @xmath611 and @xmath612 , further bounding the sum by its maximum and the number of terms , we obtain @xmath613 for any @xmath614 with @xmath615 , we can take @xmath450 large enough so that ( using @xmath616 ) @xmath617 combining , , , and gives that for @xmath450 sufficiently large , @xmath618 from the construction it follows that @xmath619 so that for @xmath450 large @xmath620 which finishes the proof . recall that we consider sfp models with @xmath90 and any @xmath140 , and that in this setting , @xmath621 @xcite let @xmath418 and @xmath400 be the values that we obtain from lemma [ lemma : density ] . to apply lemma [ lemma : density ] , we partition @xmath0 into @xmath208-boxes . in every @xmath208-box we only consider the vertex with maximum weight and call it the dominant vertex . choose @xmath413 large enough , using lemma [ transientlemmaconnectivity ] , such that two @xmath208-boxes that are @xmath212-boxes apart , with dominant vertices @xmath325 and @xmath622 having weight at least @xmath200 , are connected by an open edge between @xmath325 and @xmath622 with probability at least @xmath623 for @xmath230 such that @xmath624 and weights @xmath180 with law given by . define the @xmath208-boxes that contain a vertex with weight at least @xmath200 to be the good boxes . choose @xmath208 large enough so that the probability that an @xmath208-box is good is larger than @xmath418 using lemma [ transientlemmabigdegrees ] . thus , the status of the edges between dominant vertices in good @xmath208-boxes in the sfp model with parameters @xmath625 and weight - law @xmath37 stochastically dominates an sfp model on @xmath0 with parameters @xmath14 , @xmath400 and weight - law @xmath180 combined with a site percolation of intensity @xmath418 , exactly as described in lemma [ lemma : density ] . let @xmath626 and @xmath627 be the constants we obtain from lemma [ lemma : density ] . observe that @xmath628 then the assertions of theorem [ thm : trees ] follows if we set @xmath629 and @xmath630 . jj thanks the erasmus+ programme for funding and the lmu munich for hospitality during a three - month stay in autumn 2015 . the work of th is supported by the netherlands organisation for scientific research ( nwo ) through the gravitation networks grant 024.002.003 . we thank remco van der hofstad for suggesting the model to us and thomas beekenkamp for comments and corrections on an earlier version of the manuscript .

we study slow - roll inflation with the gauss - bonnet and chern - simons corrections . we obtain general formulas for the observables : spectral indices , tensor - to - scalar ratio and circular polarization of gravitational waves . the gauss - bonnet term violates the consistency relation . particularly , blue spectrum and scale invariant spectrum of tensor modes are possible . these cases require the gauss - bonnet coupling function of . we use examples to show new - inflation - type potential with symmetry breaking scale and potential with flat region in lead to observationally consistent blue and scale invariant spectra , respectively . hence , these interesting cases can actually be realized . the chern - simons term produce circularly polarized tensor modes . we show an observation of these signals supports existence of the chern - simons coupling function of . thus , with future observations , we can fix or constrain the value of these coupling functions , at the cmb scale .

1008.2724

currently , the concept of slow - roll inflation is widely accepted by cosmologists , due to many precise observations , such as the wilkinson microwave anisotropy probe ( wmap)@xcite . we can explain all observations simply , using single field slow - roll inflation . however , future observations might detect non - standard signals , which will be evidence for a non - standard inflationary theory . the energy scale of inflation is considered to be very high compared to current experimental scale . thus , we can say that a new inflationary theory might contain information with regard to high energy physics , such as a superstring theory . therefore it is important to study theories beyond standard single field slow - roll inflation . one of candidates for a modified inflationary model is inflation with the gauss - bonnet and chern - simons corrections . these are the only two meaningful combinations of second order curvature terms , as a low energy effective theory@xcite , and can also be derived from some superstring models@xcite . cosmology with the gauss - bonnet term can affect background evolution and result in a non - singular cosmological model@xcite or other interesting solutions@xcite . in inflation with the chern - simons term , primordial gravitational waves are circularly polarized@xcite , which is considered as an interesting target for some future observations@xcite . many authors studied gauss - bonnet and chern - simons modified inflation . however , _ slow - roll _ inflation with the gauss - bonnet terms is not much studied , except in the papers@xcite . taking current observations into consideration , slow - roll inflation is the most plausible paradigm for inflation . therefore investigating slow - roll inflation with the gauss - bonnet and chern - simons corrections is an important subject . we have studied this topic in the previous paper@xcite . in this paper , we re - formalize slow - roll inflation in gravity with the gauss - bonnet and chern - simons corrections . we derive formulas for the observables , namely the scalar spectral index @xmath7 , the tensor spectral index @xmath8 , the tensor - to - scalar ratio @xmath9 and the circular polarization ratio @xmath10 , in our inflationary model . the gauss - bonnet term violates the consistency relation @xmath0 . we show if this violation is observationally confirmed , the derivative of the gauss - bonnet coupling function at the observation scale , namely the cosmic microwave background ( cmb ) scale , is fixed . we study the two typical cases , blue and scale invariant spectra of tensor modes . because blue and scale invariant mean @xmath1 and @xmath2 , respectively , these cases strongly violate this consistency relation . if these are confirmed in future observations , the gauss - bonnet coupling function must be @xmath3 , at the cmb scale . of course , even if no such effects are observed in future , we can get the constraint @xmath11 , at least . however , it is unclear whether these interesting cases are consistent with current observations . therefore , we perform calculations in some concrete examples . we show new - inflation - type potential with symmetry breaking scale of @xmath4 leads to an observationally consistent blue spectrum ; potential with flat region in @xmath12 leads to a consistent scale invariant spectrum . in both cases , the almost constant @xmath13 is required . the chern - simons term leads to circular polarization of gravitational waves . we show if this circular polarization is detected , we can fix the derivative of the chern - simons coupling function , at the cmb scale . it might be @xmath6 . therefore , the value of the gauss - bonnet and chern - simons coupling functions at the cmb scale will be fixed , or constrained , with future observations . we organize this paper as follows : in section ii , the action we considered is shown and slow - roll inflation in this action is studied , calculations of perturbations in slow - roll inflation is in section iii , interesting features of our model are mentioned in section iv , and section v is for conclusion . we consider the following action : @xmath14 % \nonoumber \\ % & \qquad - \frac{1}{16 } \int { \rm d}^4 x \sqrt{-g } \xi(\phi ) r_{\rm gb}^2 + \frac{1}{16 } \int { \rm d}^4 x \sqrt{-g } \omega(\phi ) r \tilde{r } \ , \label{s}\end{aligned}\ ] ] where @xmath15 is the reduced planck mass , @xmath16 and @xmath17 are the gauss - bonnet and chern - simons combinations . these combinations are defined by @xmath18 where @xmath19 is the levi - civita anti - symmetric tensor . hence , the chern - simons correction represents parity violation in gravity . this action is composed of the einstein - hilbert term and a canonical scalar field which couples to the gauss - bonnet and chern - simons terms through the coupling functions @xmath20 and @xmath21 . note that only these two types of higher curvature corrections , namely the gauss - bonnet and chern - simons terms , are essential as a low energy effective theory@xcite . here , we take these terms as small corrections . in principle , coupling functions must be fixed by quantum gravity , such as a superstring theory . however , it is impossible to determine these coupling functions , because we havent succeeded to construct quantum gravity . therefore , we treat these as free functions and attempt to fix or constrain these coupling functions from cosmological observations . through these attempts , we may get new information with regard to quantum gravity , because these coupling functions are related to properties of quantum gravity . here , as a background metric , we choose the flat friedmann - robertson - walker(frw ) metric : @xmath22 where @xmath23 denotes a scale factor which describes cosmic expansion . variating eq.([s ] ) in the frw metric , eq.([frw ] ) , we get background equations : @xmath24 where a dot denotes a time derivative and the hubble parameter @xmath25 is defined by @xmath26 . because of parity symmetry of eq.([frw ] ) , the chern - simons correction can not affect these background equations . we take @xmath27 throughout this paper . here we focus on the case in which scalar field is slowly rolling and its friction term is dominating eq.([eom0 ] ) . in other words , we take slow - roll approximations : @xmath28 in this paper , we consider the gauss - bonnet term as a _ small _ correction . hence , it must be much smaller than the einstein - hilbert term , energetically speaking . the following inequality , hence , must be satisfied : @xmath29 under eqs.([slow - roll ] ) and ( [ gb_is_small ] ) , the background equation , eq.([friedmann0 ] ) , become @xmath30 this equation does nt contain the gauss - bonnet correction since it is energetically negligible , of course . the time derivative of the hubble parameter becomes @xmath31 the another background equation , eq.([eom0 ] ) , is @xmath32 the last term in the right - hand side can be written as @xmath33 hence , this term is negligible compared with the first term in the right - hand side of eq.([gb_is_small ] ) , and eq.([eom0 ] ) becomes @xmath34 this equation shows that the gauss - bonnet coupling function @xmath35 works as effective potential . in our inflationary model , slow - roll equations are eq.([friedmann_slow ] ) and ( [ eom_slow ] ) . we can see only the equation of motion , eq.([eom_slow ] ) , is modified . here , we define the new function @xmath36 : @xmath37 eq.([eom_slow ] ) becomes @xmath38 in our model , the expression for the e - folding number @xmath39 is written as @xmath40 where we assume that potential has a minimum at @xmath41 . we must consider consistency conditions for eqs.([friedmann_slow ] ) and ( [ dphizeta ] ) . first , we check slow - roll conditions , eqs.([slow - roll ] ) : @xmath42 second , the condition for smallness of the gauss - bonnet term , eq.([gb_is_small ] ) , is @xmath43 here , we define three slow - roll parameters : @xmath44 it is clear that if @xmath45 are realized , all consistency conditions are satisfied and slow - roll inflation occurs . in the case of conventional slow - roll inflation , @xmath46 is realized . hence , @xmath47 and @xmath48 becomes conventional @xmath49 , and @xmath50 becomes @xmath51 . for later convenience , we derive expressions for @xmath52 , @xmath53 and @xmath54 in slow - roll parameters . from eqs.([doth_slow ] ) , ( [ dphizeta ] ) and ( [ sigma ] ) , we get @xmath55 remember that we are taking @xmath56 . as in a conventional scenario , the time derivative of the hubble parameter is small . in the case that @xmath47 can be treated as a constant , we can solve the equation for @xmath25 easily : @xmath57 where @xmath58 is a constant of integration . because we choose @xmath59 , @xmath60 for @xmath61 and @xmath62 for @xmath63 . the most important difference between our model and a conventional inflation is that , in our model super inflation can be realized , while in a conventional one it can not . in our model , the slow - roll parameter @xmath47 can take a negative value . from eqs.([doth_param ] ) , we can see this leads to super inflation , @xmath64 . we can expect that if this super inflation phase corresponds to the cmb scale , an observationally interesting signal could be detected in future experiments , and this signal might be regarded as evidence for existence of gravitational higher curvature terms . here , we show an interesting effect of the gauss - bonnet correction in slow - roll inflation . let us consider general potential @xmath65 , which does not satisfies conventional slow - roll conditions . if we set @xmath66 @xmath67 is realized . this makes all slow - roll parameters zero , and inflation occurs . this shows that , even if we use extremely steep potential , slow - roll inflation can be achieved due to the gauss - bonnet correction . in this section , we perform perturbative calculations on the background of slow - roll inflation . first , we study scalar and tensor modes of perturbations . note that no vector modes plays an important role since there is no source for vectors . second , we derive expressions for reconstructing the value of potential and coupling functions using observables . finally , we briefly comment on the lyth bound in our model . here , we consider scalar perturbations in the flat frw background . the metric is @xmath68 \ , \end{aligned}\ ] ] where @xmath69 is the conformal time @xmath70 and a bar denotes a spacial derivative . latin indices run from @xmath71 to @xmath72 . in the case of slow - roll inflation , @xmath69 becomes @xmath73 where @xmath74 . we fix gauge as @xmath75 , where @xmath76 is a perturbation of scalar field @xmath77 . this gauge is useful in this case , because the perturbative quantity @xmath78 can be treated as a master variable and @xmath79 coincide with the gauge invariant comoving curvature perturbation @xmath80 . of course , this is a complete choice of gauge . here , we concentrate on @xmath78 , because what we need is the power spectrum of @xmath78 in super - horizon scale . expanding @xmath78 in fourier modes , @xmath81 the action for @xmath78 becomes @xmath82 where a prime denotes a conformal time derivative and @xmath83 \ , \end{aligned}\ ] ] and @xmath84 is defined by @xmath85 this action contains only @xmath35 . this is because scalar perturbations in the flat frw background have parity symmetry . let us define the new variable @xmath86 , by @xmath87 , and the above action becomes @xmath88 this leads to the equation of motion : @xmath89 here , we impose slow - roll approximation , and derive the expressions for @xmath90 and @xmath91 up to first order in slow - roll parameters . remember that the time derivative of the hubble parameter @xmath25 is first order . in computing primordial power spectra , we treat slow - roll parameters as constants of time . we get the following equations : @xmath92 and @xmath93 let us consider the effective potential @xmath94 . for this purpose , we derive some equations : @xmath95 hence @xmath96 where @xmath97 is defined by @xmath98 eq.([eom_s0 ] ) becomes @xmath99 to quantize the quantity @xmath100 , we replace @xmath100 with the operator @xmath101 and expand this as @xmath102 where @xmath103 and @xmath104 are annihilation and creation operators . the mode function @xmath105 obeys the same equation as @xmath86 does , namely eq.([eom_s1 ] ) : @xmath106 for solving this equation , we must impose an initial condition . because effective potential vanishes in asymptotic past , we can choose the bunch - davies vacuum at @xmath107 : @xmath108 under this condition , eq.([eom_s2 ] ) has the analytic solution @xmath109 where @xmath110 is the first kind of hankel function . here , we derive the expression for the power spectrum of scalar perturbation @xmath111 in super - horizon , which is defined by @xmath112 in super - horizon , @xmath113 , @xmath105 has the asymptotic form @xmath114 hence @xmath111 becomes @xmath115 the scalar power spectrum is of order @xmath116 , and this is easily understood because @xmath117 , in which we use eq.([doth_param ] ) . the scalar spectral index @xmath7 becomes @xmath118 if we take @xmath119 , the slow - roll parameters @xmath47 and @xmath50 coincide with the conventional ones @xmath49 and @xmath51 , respectively . hence , in this case , the expression for @xmath7 takes ordinary form . in this subsection , we compute perturbations of tensor modes . the metric is @xmath120 \ , \end{aligned}\ ] ] where @xmath121 denotes a transverse - traceless tensor on a constant - time hypersurface . hence it satisfies @xmath122 . we expand @xmath121 , which is gauge - invariant because tensor modes have no gauge freedom , in fourier modes and circular polarization : @xmath123 where @xmath124 is the polarization tensor for circular polarization and @xmath125 denotes the helicity of each polarization mode . note that @xmath126 and @xmath127 are for later convenience . the action for @xmath128 becomes @xmath129 where @xmath130 the mass scale @xmath131 corresponds to the uv cut - off scale of our model and @xmath132 is defined by @xmath133 note that we omit @xmath125 symbols in the expressions for @xmath134 and @xmath135 for simplicity . the chern - simons coupling function @xmath136 does nt appear in the background and scalar perturbations , but , at the same time , this appears in tensor perturbations , because _ circular _ modes clearly violate parity symmetry . from eqs.([atct ] ) , we can see that if @xmath137 , @xmath134 takes a negative value in one of helicity modes at the cut - off scale @xmath138 . this is very important because it implies the appearance of ghost modes in gravitational waves . since we do nt know proper treatment for ghost modes , we impose the condition @xmath139 . for simplicity of following calculation , we take @xmath132 as a constant , which is justified in many models because @xmath77 is slowly rolling . we can always treat the chern - simons correction @xmath140 as a small correction , because @xmath139 and @xmath141 . of course , @xmath137 model might lead to very drastic results and such model is interesting itself . however , we do nt analyze this situation in this paper . introducing the new variable @xmath142 , eq.([s_t0 ] ) can be rewritten as @xmath143 hence the equation of motion for @xmath144 is @xmath145 where we also omit @xmath125 in a @xmath144 expression . considering slow - roll inflation , we compute spectra up to first order in slow - roll parameters and @xmath132 . the quantities @xmath146 and @xmath147 become @xmath148 and we get @xmath149 effective potential becomes @xmath150 where @xmath151 is defined by @xmath152 hence , eq.([eom_t0 ] ) becomes @xmath153 let us quantize perturbations . we promote @xmath144 to the operator @xmath154 and expand in the annihilation and creation operators @xmath103 and @xmath155 : @xmath156 the equation for @xmath157 is the same as that for @xmath158 , eq.([eom_t1 ] ) : @xmath159 as in the case of scalar modes , effective potential vanishes at asymptotic past . thus we can use the bunch - davies vacuum as the initial condition for eq.([eom_t2 ] ) . we take at @xmath160 : @xmath161 hence , the solution for eq.([eom_t2 ] ) is @xmath162 where @xmath163 is the confluent hypergeometric function . this solution has the asymptotic form in super - horizon scale : @xmath164 we define the power spectra for each polarization mode @xmath165 and the total power spectrum @xmath166 : @xmath167 in super - horizon , these @xmath168 and @xmath166 become @xmath169 the tensor spectral index @xmath8 is @xmath170 this expression becomes the same as an ordinary one in the @xmath171 limit . remember that in our model , @xmath47 can takes a negative value in which super inflation , @xmath64 , is realized . hence , this expression shows that in super inflation , a blue spectrum for tensor modes , @xmath1 , is realized . the detection of this blue signal might be evidence for this inflationary model , because @xmath172 is always satisfied in a conventional model . we have derived both scalar and tensor spectra . thus , we can get the observationally important quantity , namely the tensor - to - scalar ratio @xmath9 : @xmath173 note that @xmath174 in the @xmath171 limit . there is the consistency relation @xmath0 in ordinary single field slow - roll inflation , or @xmath175 in multi - filed one . however , these consistency relation does nt hold in our case , because @xmath47 and @xmath48 are in principle different quantities . the violation of this relation might be an observationally interesting feature of our model . note that a blue spectrum of tensor modes can also be interpreted as a strong violation of this relation . the another important feature of our model is a violation of parity symmetry . eq.([ptpm ] ) shows that the parity violating chern - simons correction results in difference between power spectra of each helicity mode . we quantify this difference using the ratio @xmath10 which is defined by @xmath176 this quantity @xmath10 takes naturally a small value , because @xmath177 and @xmath178 . however , we can still expect @xmath10 is of order a few percent , which is considered to be observable in future experiments@xcite . note that we have got no observational constraint on @xmath136 and @xmath132 , because these only appear in tensor modes , which we havent detected . in a conventional model , the transformation of potential @xmath179 , where @xmath180 is constant , has no effect on spectral indices , tensor - to - scalar ratio and e - folding number . hence , an overall factor of potential is fixed by the scalar power spectrum @xmath111 . in our model , from eqs.([slow - roll_param ] ) and ( [ e - fold_0 ] ) , it is clear that these observables depend on the transformation @xmath181 . in this case , the transformations which keep @xmath182 and @xmath39 constants are @xmath181 and @xmath183 . slow - roll inflation in the gauss - bonnet modified gravity violates the consistency relation @xmath184 . however , in the case of standard multi field inflation , this relation also modified as @xmath175 . therefore , the case of over production of gravitational waves , @xmath185 , is important . from eqs.([ntex ] ) and ( [ rex ] ) , we can see over - production is realized if @xmath186 is satisfied . taking a look at the definition of slow - roll parameters and @xmath36 , eqs.([slow - roll_param ] ) and ( [ def_zeta ] ) , we can see in the situation that @xmath187 is realized , gravitational waves will always be overly produced . thus , we can expect over production in many models . if the violation of this relation is observationally confirmed , we can reconstruct the value of the derivative of the gauss - bonnet coupling function @xmath13 . from eqs.([slow - roll_param ] ) @xmath188 the potential @xmath189 and the slow - roll parameter @xmath48 are related by the scalar power spectrum @xmath111 . from eq.([psex ] ) @xmath190 hence , @xmath13 is written as @xmath191 let us parameterize the violation ratio of consistency relation by using @xmath192 : @xmath193 the expression for @xmath13 becomes @xmath194 this equation clearly shows that we can get the value of @xmath13 , or constrain @xmath13 at least , with observing the tensor - to - scalar ratio @xmath9 and the violation ratio of consistency relation @xmath192 . in other words , we can predict this violation ratio of consistency relation by using this equation for each gauss - bonnet inflationary model . the expression for the value of potential , at the cmb scale , is the same as usual . from eq.([psex2 ] ) @xmath195 however , the derivative of potential @xmath196 suffers modification . from the definition of slow - roll parameters , eq.([slow - roll_param ] ) @xmath197 using eqs.([xiphi]),([v_o ] ) and ( [ dv_o ] ) , we can fix the value of @xmath198 and @xmath13 using observables . the expression for the scalar spectral index @xmath7 contains @xmath199 and @xmath200 . however , we can not separate these quantities in our model . thus , from these observables , we can not fix the value of @xmath199 and @xmath200 . to this point , we have studied a general case , placing special emphasis on a violation of consistency relation . however , there are other models which also can violate this relation . particularly , the model which overly produce tensor modes exists@xcite . therefore in the next section , we will investigate an unique phenomenon in our model , namely a blue spectrum of tensor mode . next , let us study the chern - simons coupling function . considering future observations , deriving the expression for @xmath201 by using observable is meaningful . from the definition of @xmath132 , eq.([def_omega ] ) , @xmath201 is @xmath202 thus , we can see that for confirming existence of the chern - simons term , it is important to detect circular polarization in tensor modes . we can also use this expression to predict this polarization ratio . because @xmath203 is required by the definition , eq.([def_pi ] ) , we can get the constraint on the value of @xmath201 : @xmath204 of course , if we allow an appearance of ghost in tensor modes , a situation might be changed drastically . however , it exceeds the scope of this paper . here , we comment on the lyth bound@xcite . the statement of this bound is following : if primordial gravitational waves are detected , the variation of inflaton field must exceed a scale on the order of planck mass . in our model , this bound also holds . from eq.([e - fold_0 ] ) @xmath205 and @xmath206 this can be easily understood : large @xmath9 corresponds to small @xmath53 , and this leads to small @xmath207 . note that effect of gravitational waves on the cmb multi pole is relevant in approximately @xmath208 , which corresponds to @xmath209 . this means that if gravitational waves with tensor - to - scalar ratio @xmath9 are observed , the field value of inflaton has changed its value approximately @xmath210 during @xmath209 . hence , from this equation , we can set the lower bound on variation of inflaton field @xmath211 during whole inflation : @xmath212 we can see that the lyth bound holds in our case , while an equation of motion is modified due to the gauss - bonnet correction . therefore , extremely small field inflation , namely @xmath213 , can not produce an observable amount of tensor modes . in our model , a spectrum of primordial gravitational waves can be blue or scale invariant , and these cases strongly violate the consistency relation @xmath0 . hence , these ( especially , a blue spectrum ) are important for confirming our model with future observations . in this section , we use concrete examples to check these interesting situations are consistent with the current wmap result . we show characteristics of potential and the gauss - bonnet coupling function which realize a blue or scale invariant spectrum . we also consider the case in which potential is cancelled out by the gauss - bonnet effective potential . we take e - folding number as @xmath214 for calculations in examples . the most interesting and unique phenomenon in our model might be a blue spectrum of tensor modes . note that it is impossible to achieve a blue spectrum in a conventional model . for achieving a blue spectrum , @xmath63 is required . hence @xmath215 is necessary because @xmath27 . in a case of blue spectrum , a scalar field is climbing up potential , not rolling down . of course , this blue situation must be over at some time , or inflation lasts forever , because @xmath64 is satisfied in this blue case . a climbing up situation can be understood using eq.([eom_slow ] ) , which shows that the coupling function @xmath35 works as effective potential . hence , if the effective potential @xmath35 is steep enough , @xmath77 can climbs up potential . we show this in fig.[blue ] . and the coupling function @xmath35 , which achieve a blue spectrum , is shown . we need a climbing - up situation to realize a blue spectrum . therefore the following situation is required : at early stage , the effective potential @xmath35 makes @xmath77 climb up potential and a blue spectrum is realized . at late stage , @xmath77 rolls down potential as usual . ] in a blue spectrum case , @xmath13 can be calculated by employing eq.([xiphi ] ) with @xmath216 . hence , there is a necessary condition for a blue spectrum : @xmath217 and combining with eq.([v_o ] ) @xmath218 remember @xmath219 and @xmath220 . to observe tensor modes , @xmath9 can not take an extremely small value . hence , @xmath192 can not exceed a value of order unity and @xmath221 must hold . in a consistent blue spectrum case , hence , we need the derivative of the gauss - bonnet coupling function @xmath13 on the order of @xmath222 and @xmath223 . it is important that both too small and too large a value of @xmath13 can not result in a desirable blue spectrum . in other words , if a blue spectrum is detected , @xmath13 at the cmb scale must be on the order of @xmath222 . in the case that a blue spectrum of tensor modes is confirmed , we can reconstruct the value of @xmath13 , which must be on the order of @xmath222 . however , it is unclear whether or not an observationally consistent blue spectrum case can actually be realized ; what kind of potential and the gauss - bonnet coupling function is likely to realize a blue spectrum . to show these , we use two types of examples of blue spectra . first one is a simple , but analytically solvable model and second one is a somewhat realistic model , which is calculated numerically . we take the following potential and coupling function : @xmath224 in this case , the e - folding number @xmath39 can be written analytically : @xmath225 where @xmath226 note that @xmath227 must be in the range @xmath228 and @xmath229 represents effect of the gauss - bonnet correction . we use @xmath227 as a variable instead of @xmath77 and @xmath230 corresponds to a climbing - up situation and a blue spectrum . to show a realization of blue spectrum , we express @xmath227 in @xmath229 and @xmath39 : @xmath231 where @xmath232 here , we have two parameters in this model , namely @xmath229 and @xmath233 . let us consider varying @xmath229 to realize a blue spectrum in fixed @xmath233 , which corresponds to fixed @xmath234 . in the situation that @xmath235 , eq.([x0 ] ) becomes @xmath236 hence , with sufficiently large @xmath229 , @xmath230 and blue spectrum can be realized for any @xmath234 value . the condition for a blue spectrum is @xmath237 where @xmath238 this inequality is satisfied , only if @xmath239 . we define @xmath240 so that @xmath241 is realized at @xmath242 . note that @xmath243 and @xmath244 correspond to red and blue spectra , respectively . because @xmath245 is necessary for a blue spectrum , the condition for @xmath240 is @xmath246 here , we define @xmath247 and @xmath248 in the same manner as @xmath240 . from eqs.([defy ] ) , ( [ thetal0 ] ) and ( [ defq ] ) , we can relate these variables : @xmath249 note that for @xmath250 model , @xmath251 ( and @xmath252 ) is the solution for above equations . the scalar spectral index @xmath7 , the tensor spectral index @xmath8 and the tensor - to - scalar ratio @xmath9 become @xmath253 \\ n_{\rm t } & = \frac{y_0}{n } \frac{-x}{1-x^2 } \left [ x + \theta ( 1-x^2 ) \right ] = \frac{y_0}{n } \frac{\theta x}{1-x^2 } ( x - x_1)(x - x_2 ) \\ r & = \frac{8 y_0}{n } \frac{1}{1-x^2 } \left[x + \theta ( 1-x^2)\right]^2 = \frac{8 y_0}{n } \frac{\theta^2}{1-x^2 } ( x - x_1)^2 ( x - x_2)^2 \ .\end{aligned}\ ] ] let us consider the large @xmath229 limit , in which @xmath254 is satisfied . in this limit , the following equation is realized : @xmath255 hence , @xmath227 approaches to @xmath256 . we define new variable @xmath257 : @xmath258 because @xmath259 is realized , @xmath260 holds . note that @xmath261 approaches to zero exponentially as a function of @xmath229 , and this leads to @xmath262 . observables become @xmath263 note that @xmath9 becomes small in large @xmath229 , because @xmath264 . this shows that in any model , this large @xmath229 limit results in small @xmath9 and extremely red @xmath7 . the reason for small @xmath9 is that @xmath265 corresponds to the fixed point , at which potential is cancelled out by effective potential . hence , @xmath266 and @xmath267 are realized . since this is not a desirable case , we consider the small @xmath229 case , in which @xmath268 holds . in this case , @xmath269 is realized , and observables become @xmath270 note that @xmath248 and @xmath233 are not independent variables , because @xmath248 is fixed by choosing @xmath233 . remember that @xmath251 is the solution for @xmath250 model . let us substitute these : @xmath271 here , we get good result : @xmath7 and @xmath9 are consistent with current observations and a blue spectrum of tensor modes is obtained . therefore , desirable solutions might exist in the vicinity of @xmath272 and @xmath273 : @xmath274 let us consider the model , in which @xmath234 is large enough to satisfy @xmath275 . from eqs.([defy ] ) and ( [ cony ] ) , @xmath276 is realized , and @xmath9 at @xmath241 becomes @xmath277 because @xmath278 , this large field model results in @xmath9 , which is too large . this is because , in this large field model , inflaton @xmath77 must roll large field value between the hill - top of potential and the vacuum , within @xmath214 e - fold . this requires large @xmath53 , in other words , large @xmath9 . in addition , from eq.([ob0 ] ) , this large field model also leads to a flat spectrum of scalar modes . in the small field model , which satisfies @xmath279 , the inequality @xmath280 always holds . hence , as in the above large @xmath229 limit , this results in small @xmath9 and extremely red @xmath7 . note that larger field models are corresponding to larger values of @xmath9 , and this is consistent with the consideration in the lyth bound . we show @xmath229-trajectories of solutions on a @xmath281 plane in fig.[trjsin ] . we plot the 68% and 95% c.l . contours from the wmap 7-year result@xcite . we choose eight @xmath234 values surrounding @xmath282 . thick dotted and thick solid lines correspond to @xmath283 and @xmath284 blue spectra . as stated above , models with @xmath285 have observationally consistent blue regions ; large field models correspond to large @xmath9 and small or no consistent blue spectrum region ; small field models have extremely small @xmath9 ; in large @xmath229 limit , @xmath9 becomes zero and @xmath7 becomes extremely red . it is clear that a consistent blue spectrum model can actually be constructed . to get the value of @xmath13 , we can use eq.([xiphi ] ) . of course , as mentioned above , @xmath3 is realized in a blue region . is varying , of analytic blue models are shown on a @xmath281 plane . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw eight lines , which are corresponding to @xmath286 ( red , bottom ) to @xmath287 ( blue , top ) with an even interval . thick dotted and thick solid lines denote the blue spectra of @xmath283 and @xmath284 , respectively . models with @xmath285 are consistent with observations and realizes blue spectra . ] next , we consider a realistic model . to realize a blue spectrum , new - inflation - type potential , as depicted in fig.[blue ] , is required . therefore , we use forth - order double - well potential and an exponential coupling function : @xmath288 ^ 2 \ , \quad \xi(\phi ) = s \frac{b}{\rho } \exp(s a \phi / m_{\rm pl } ) \ , \end{aligned}\ ] ] where @xmath289 , @xmath180 will be fixed by the scalar power spectrum @xmath290@xcite and @xmath291 . we take @xmath292 as our vacuum . we use this exponential coupling , since it appears in many theories , such as a dilatonic case or an one - loop correction from a superstring motivated model@xcite . we show @xmath293-trajectories on a @xmath281 plane in fig.[trj1a ] . we plot the 68% and 95% c.l . contours from the wmap 7-year result . we use nine values of @xmath294 , which varies from @xmath295 to @xmath71 with an even logarithmic interval , and three value of @xmath234 , @xmath296 and @xmath297 . we choose @xmath234 value so that @xmath7 and @xmath9 values are inside this 68% c.l . contour in the @xmath298 limit . thick dotted and thick solid lines in fig.[trj1a ] denote the blue spectra of @xmath283 and @xmath284 , respectively . we can see it is possible to realize a consistent blue spectrum . the case of @xmath299 is appropriate to achieve a consistent blue spectrum . small @xmath234 leads to small @xmath9 and large @xmath234 leads to a small consistent blue region , and these are essentially consistent with the above analytic case . due to this exponential coupling , even large field models , namely @xmath300 models with @xmath301 , have blue spectrum regions and small field models , namely @xmath302 models with @xmath303 , have an observable amount of @xmath9 . this figure shows that an order unity value of @xmath294 leads to a non - desirable result , which has too small @xmath9 or no blue region . therefore @xmath304 is appropriate to obtain a blue spectrum . moreover , if @xmath294 is large in @xmath301 models , it implies the gauss - bonnet effect becomes stronger toward the end of inflation . in this case , standard reheating scenarios seem to be broken . therefore , large @xmath294 in @xmath301 models is also not desirable in this sense . note that with eq.([xiphi ] ) , we can get the value of @xmath13 . is varying , of some blue models are shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and top in right panels ) to @xmath71 ( blue , right in left panels and bottom in right panels ) with an even logarithmic interval . thick dotted and thick solid lines denote @xmath283 and @xmath284 blue spectra , respectively . the model with @xmath299 is appropriate to achieve blue spectrum . an exponential coupling makes large @xmath234 ( @xmath300 ) model result in a blue spectrum , and small @xmath234 ( @xmath302 ) model achieve an observable amount of @xmath9 . , title="fig : " ] @xmath306 is varying , of some blue models are shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and top in right panels ) to @xmath71 ( blue , right in left panels and bottom in right panels ) with an even logarithmic interval . thick dotted and thick solid lines denote @xmath283 and @xmath284 blue spectra , respectively . the model with @xmath299 is appropriate to achieve blue spectrum . an exponential coupling makes large @xmath234 ( @xmath300 ) model result in a blue spectrum , and small @xmath234 ( @xmath302 ) model achieve an observable amount of @xmath9 . , title="fig : " ] @xmath307 is varying , of some blue models are shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and top in right panels ) to @xmath71 ( blue , right in left panels and bottom in right panels ) with an even logarithmic interval . thick dotted and thick solid lines denote @xmath283 and @xmath284 blue spectra , respectively . the model with @xmath299 is appropriate to achieve blue spectrum . an exponential coupling makes large @xmath234 ( @xmath300 ) model result in a blue spectrum , and small @xmath234 ( @xmath302 ) model achieve an observable amount of @xmath9 . , title="fig : " ] @xmath308 is varying , of some blue models are shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and top in right panels ) to @xmath71 ( blue , right in left panels and bottom in right panels ) with an even logarithmic interval . thick dotted and thick solid lines denote @xmath283 and @xmath284 blue spectra , respectively . the model with @xmath299 is appropriate to achieve blue spectrum . an exponential coupling makes large @xmath234 ( @xmath300 ) model result in a blue spectrum , and small @xmath234 ( @xmath302 ) model achieve an observable amount of @xmath9 . , title="fig : " ] @xmath309 is varying , of some blue models are shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and top in right panels ) to @xmath71 ( blue , right in left panels and bottom in right panels ) with an even logarithmic interval . thick dotted and thick solid lines denote @xmath283 and @xmath284 blue spectra , respectively . the model with @xmath299 is appropriate to achieve blue spectrum . an exponential coupling makes large @xmath234 ( @xmath300 ) model result in a blue spectrum , and small @xmath234 ( @xmath302 ) model achieve an observable amount of @xmath9 . , title="fig : " ] @xmath310 is varying , of some blue models are shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and top in right panels ) to @xmath71 ( blue , right in left panels and bottom in right panels ) with an even logarithmic interval . thick dotted and thick solid lines denote @xmath283 and @xmath284 blue spectra , respectively . the model with @xmath299 is appropriate to achieve blue spectrum . an exponential coupling makes large @xmath234 ( @xmath300 ) model result in a blue spectrum , and small @xmath234 ( @xmath302 ) model achieve an observable amount of @xmath9 . , title="fig : " ] @xmath311 from the above analytic model and this realistic one , we can say that inflationary models with observationally consistent blue spectra can be constructed . hence , the gauss - bonnet coupling function will actually be fixed or constrained with future observations . new - inflation - type potential with the symmetry breaking scale of @xmath312 and an almost constant form of @xmath13 might lead to a consistent blue spectrum . of course , these are not general conditions . however , we can expect many models are also consistent with these conditions . note that to obtain blue spectrum , @xmath3 must be realized in any model at the cmb scale . in inflation with a standard gravitational action , a scale invariant spectrum of gravitational waves corresponds to almost purely de sitter inflation and this requires the tensor - to - scalar ratio @xmath9 extremely small . however , inflation with the gauss - bonnet term behaves differently . let us consider the case , in which the gauss - bonnet term dominate the scalar equation of motion , eq.([dphizeta ] ) . the function @xmath36 becomes @xmath313 and this requires @xmath314 in this case , @xmath47 becomes negligible . hence @xmath315 these equations means that if the gauss - bonnet correction dominates this equation of motion , the tensor - to - scalar ratio @xmath9 can take an observable value although @xmath316 is realized . because of the consistency relation @xmath0 , this scale invariant spectrum is impossible in a conventional model . therefore , if scale invariant primordial gravitational waves are detected , it implies existence of the gauss - bonnet term in an inflationary theory and @xmath13 is determined by eq.([xiphi ] ) with @xmath317 : @xmath318 and using eq.([v_o ] ) @xmath319 as in a blue case , @xmath3 and @xmath320 must be satisfied in this flat spectrum case , too . both too large and too small a value of @xmath13 can not lead to a consistent scale invariant spectrum . from eqs.([zetadv ] ) and ( [ obs_flat ] ) , the inequality for @xmath9 holds : @xmath321 this shows that potential must be nearly flat for the tensor - to - scalar ratio @xmath9 to remain small . here , we check whether or not an observationally consistent scale invariant model can be constructed . in this subsection , we also use two types of examples , namely a simple and analytic one and a realistic one . let us start with an analytic model : @xmath322 the e - folding number @xmath39 becomes @xmath323 where @xmath227 and @xmath229 are defined by @xmath324 we use @xmath227 instead of @xmath77 , and @xmath227 is @xmath325 \ .\end{aligned}\ ] ] the observables , namely the scalar spectral index @xmath7 , the tensor spectral index @xmath8 and the tensor - to - scalar ratio @xmath9 become @xmath326 in this model , we also variate @xmath229 and fix @xmath234 . let us consider the limit case , in which @xmath327 is satisfied , and @xmath227 becomes @xmath328 hence , observables are @xmath329 we can see that in this small @xmath229 limit , all models give observationally consistent values . next , we investigate the large @xmath229 limit , @xmath330 . the variable @xmath227 takes an exponentially large value in this limit . hence , observables become @xmath331 these show that in this limit , @xmath7 and @xmath8 approach to flat one and @xmath9 becomes large . remember that @xmath227 behaviors exponentially as a function of @xmath229 . if this large @xmath229 limit is realized in the wmap consistent region , such as @xmath332 , the result which has an observable amount of @xmath9 and flat @xmath8 is achieved . here , we consider @xmath333 case , which is the boundary case between large and small @xmath229 . in this case , @xmath334 is realized . the observables @xmath8 and @xmath9 become @xmath335 if @xmath336 is satisfied , @xmath337 and @xmath338 hold . this @xmath9 might be too large to realize that large @xmath229 limit in @xmath332 . therefore , to achieve desirable results , @xmath339 at this @xmath333 case might be necessary . in other words , @xmath340 the physical meaning of this condition can be easily understood : large field model requires large @xmath53 and inconsistently large @xmath341 , to reach a flat potential region within @xmath214 e - fold . this interpretation is essentially the same as that in the condition for blue spectrum . the difference is that small field model can also achieve large @xmath9 in this case , because _ small _ does nt means the smallness of the variation of the field value of the inflaton @xmath77 . it means the smallness of the field value between the vacuum and a flat region of potential . hence , this is consistent with the lyth bound . if this condition is satisfied , sufficiently large @xmath229 , namely @xmath342 , in other words @xmath343 leads to a consistent flat spectrum . we show @xmath229-trajectories of solutions on a @xmath281 plane in fig.[trjtan ] . we plot the 68% and 95% c.l . contours from the wmap 7-year result . we choose five @xmath234 values , @xmath344 and @xmath4 . thick lines denote the flat region , in which @xmath345 is realized . note that @xmath0 holds in conventional single field inflation . therefore large @xmath346 value implies a flat spectrum . we can see that small @xmath234 models , namely @xmath347 and @xmath348 models , result in wmap - consistent flat spectra . this is consistent with the criterion at the above calculation , namely @xmath349 . as mentioned above , @xmath7 also becomes flat , in a flat tensor region . note that the smaller field model , namely @xmath347 model , has larger consistent scale invariant region . of course , @xmath13 can be reconstructed from eq.([xiphi ] ) , and @xmath350 must hold in a flat region . is varying , of analytic flat models are shown on a @xmath281 plane . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw five lines , which are corresponding to @xmath351 ( red , bottom ) to @xmath4 ( blue , top ) with an even interval . thick lines denote the regions , which realize @xmath352 . small @xmath234 models , namely @xmath347 and @xmath348 models , realize observationally consistent flat spectra . ] here , we consider a realistic example of a flat spectrum model . nearly flat potential results in a gauss - bonnet dominating situation . therefore we take the following potential and coupling function : @xmath353 \ , \quad \xi(\phi ) = s\frac{b}{\rho } \exp(sa \phi / m_{\rm pl } ) \ , \end{aligned}\ ] ] where @xmath289 , @xmath180 will be fixed by @xmath290 , and @xmath291 . in the @xmath354 limit , this potential becomes extremely flat . we use the same exponential coupling in this example as in previous subsection . we show @xmath293-trajectories on a @xmath281 plane in fig.[trjflat ] . we plot the wmap 68% and 95% c.l . contours . we use three values of @xmath234 , @xmath355 and @xmath4 , and the same set of @xmath294 in this case as in the realistic blue spectrum case . thick lines denote the regions @xmath345 . as predicted from the previous analytic models , smaller @xmath234 models has larger regions of flat spectra . note that @xmath301 models have larger consistent flat region than that of @xmath301 models . an order unity value of @xmath294 leads to extremely small @xmath9 in both @xmath289 models , and considering reheating , large @xmath294 in @xmath301 models is not desirable . therefore also in this case , @xmath356 is appropriate . eq.([xiphi ] ) is valid to get the value of @xmath13 . is varying , of some flat models is shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and right in right panels ) to @xmath71 ( blue , right in left panels and left in right panels ) with an even logarithmic interval . in this case , tensor spectrum is always red . thick lines denote the regions of flat spectra , @xmath345 . this figure shows that small @xmath234 leads to a flat spectrum . note that in @xmath301 models , trajectories approach to some points , in large @xmath293 limit . , title="fig : " ] @xmath357 is varying , of some flat models is shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and right in right panels ) to @xmath71 ( blue , right in left panels and left in right panels ) with an even logarithmic interval . in this case , tensor spectrum is always red . thick lines denote the regions of flat spectra , @xmath345 . this figure shows that small @xmath234 leads to a flat spectrum . note that in @xmath301 models , trajectories approach to some points , in large @xmath293 limit . , title="fig : " ] @xmath358 is varying , of some flat models is shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and right in right panels ) to @xmath71 ( blue , right in left panels and left in right panels ) with an even logarithmic interval . in this case , tensor spectrum is always red . thick lines denote the regions of flat spectra , @xmath345 . this figure shows that small @xmath234 leads to a flat spectrum . note that in @xmath301 models , trajectories approach to some points , in large @xmath293 limit . , title="fig : " ] @xmath359 is varying , of some flat models is shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and right in right panels ) to @xmath71 ( blue , right in left panels and left in right panels ) with an even logarithmic interval . in this case , tensor spectrum is always red . thick lines denote the regions of flat spectra , @xmath345 . this figure shows that small @xmath234 leads to a flat spectrum . note that in @xmath301 models , trajectories approach to some points , in large @xmath293 limit . , title="fig : " ] @xmath360 is varying , of some flat models is shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and right in right panels ) to @xmath71 ( blue , right in left panels and left in right panels ) with an even logarithmic interval . in this case , tensor spectrum is always red . thick lines denote the regions of flat spectra , @xmath345 . this figure shows that small @xmath234 leads to a flat spectrum . note that in @xmath301 models , trajectories approach to some points , in large @xmath293 limit . , title="fig : " ] @xmath361 is varying , of some flat models is shown on a @xmath281 plane . in the @xmath298 limit , all trajectories converges to one point . two contours roughly corresponds to the 68% and 95% confidence level of the wmap 7-year result . we draw nine lines , which are corresponding to @xmath305 ( red , left in left panels and right in right panels ) to @xmath71 ( blue , right in left panels and left in right panels ) with an even logarithmic interval . in this case , tensor spectrum is always red . thick lines denote the regions of flat spectra , @xmath345 . this figure shows that small @xmath234 leads to a flat spectrum . note that in @xmath301 models , trajectories approach to some points , in large @xmath293 limit . , title="fig : " ] @xmath362 in @xmath301 models , trajectories approach to some points for each @xmath294 in large @xmath293 limit , in fig.[trjflat ] . this is understood from eq.([e - fold_0 ] ) . in the large @xmath293 limit of @xmath301 models , only a large @xmath77 region contribute to the integration for an e - folding number . because @xmath363 in this region , we can omit @xmath364 term : @xmath365 this shows that , in this large @xmath293 limit , @xmath13 approaches to @xmath366 . this also fixes the value of @xmath200 and all slow - roll parameters . the observables @xmath7 and @xmath9 become @xmath367 hence , trajectories approach to some points in large @xmath293 . this shows that @xmath368 model leads to very small @xmath9 because @xmath9 approaches to @xmath369 at @xmath370 . from above results , we can say that an observationally consistent flat spectrum can be obtained . hence , confirming existence of the gauss - bonnet term is possible , with observing a scale invariant tensor mode . in the cases we considered , @xmath371 and nearly constant @xmath13 might lead to a consistent flat spectrum . note that @xmath234 denotes the difference between the filed value of a flat region and the vacuum of potential . we can expect these conditions are also satisfied in many other cases . to realize a flat spectrum , @xmath3 is necessary in any model . as mentioned in the end of section [ sec_slow - roll ] , if we take @xmath372 , potential is cancelled out by the gauss - bonnet effective potential . considering this fact , let us take @xmath35 as follows : @xmath373 where @xmath374 is a constant and we assume @xmath375 . the function @xmath36 becomes @xmath376 eq.([dphizeta ] ) becomes @xmath377 hence if @xmath378 is realized , a scalar field always rolls up potential toward a hill - top . this is not a desirable situation . therefore we take @xmath379 . slow - roll parameters becomes @xmath380 where we used @xmath381 . in this case , @xmath48 becomes negligible . note if the gauss - bonnet term is dominating , @xmath47 becomes negligible . in this case , the observables @xmath7 , @xmath8 and @xmath9 are @xmath382 because the tensor - to - scalar ratio @xmath9 is zero , the value of the spectral index @xmath8 is meaningless . therefore observationally speaking , this situation is not interesting . however , because @xmath383 , three slow - roll conditions are automatically satisfied . hence , even if we use extremely steep potential , inflation can be achieved . in this paper , we studied slow - roll inflation with the gauss - bonnet and chern - simons corrections . we defined three slow - roll parameters , while we used five parameters in the previous paper@xcite . we derived expressions for the scalar spectral index @xmath7 , the tensor spectral index @xmath8 , the tensor - to - scalar ratio @xmath9 and the circular polarization ratio @xmath10 by using these parameters . we showed that in our model , the consistency relation @xmath0 is not automatically satisfied . if this violation is observationally confirmed , we can determine the derivative of the gauss - bonnet coupling function @xmath13 at the cmb scale . in our model , even both blue and scale invariant spectra of gravitational waves can be realized . because blue and scale invariant mean @xmath1 and @xmath384 , respectively , these cases violate this consistency relation strongly . therefore these are the key for confirming our model in future observations . we showed that if either a blue spectrum or a scale invariant one is observed , it supports existence of the gauss - bonnet coupling function of @xmath350 at the cmb scale . we checked whether or not these blue and scale invariant spectra are consistent with current observations . for this purpose , we used concrete examples . we showed that new - inflation - type potential with @xmath4 symmetry breaking scale and potential with flat region in @xmath12 might result in observationally consistent blue and scale invariant spectra , respectively . an almost linear form of the gauss - bonnet coupling function is appropriate in both cases . these tell us that the detection of a blue or scale invariant spectrum of tensor modes can be actually expected with future observations . we also showed that if circular polarization of gravitational waves is detected , the derivative of the chern - simons coupling @xmath201 must be on the order of @xmath222 . thus , we can say that existence of gravitational higher coupling terms in a inflationary model will be confirmed , or at least constrained , with future experiments . these higher curvature terms are expected to appear in non - gaussian part of perturbations . therefore , calculating non - gaussianity in our model is important future work . s. kawai and j. soda , arxiv : gr - qc/9906046 . s. kawai and j. soda , phys . b * 460 * , 41 ( 1999 ) [ arxiv : gr - qc/9903017 ] . j. c. hwang and h. noh , phys . d * 61 * , 043511 ( 2000 ) [ arxiv : astro - ph/9909480 ] . c. cartier , j. c. hwang and e. j. copeland , phys . d * 64 * , 103504 ( 2001 ) [ arxiv : astro - ph/0106197 ] . s. kawai , m. a. sakagami and j. soda , phys . b * 437 * , 284 ( 1998 ) [ arxiv : gr - qc/9802033 ] . j. soda , m. a. sakagami and s. kawai , arxiv : gr - qc/9807056 . s. kawai , m. a. sakagami and j. soda , arxiv : gr - qc/9901065 . m. gasperini , phys . rev . d * 56 * , 4815 ( 1997 ) [ arxiv : gr - qc/9704045 ] . c. cartier , e. j. copeland and m. gasperini , nucl . b * 607 * , 406 ( 2001 ) [ arxiv : gr - qc/0101019 ] . b. m. leith and i. p. neupane , jcap * 0705 * , 019 ( 2007 ) [ arxiv : hep - th/0702002 ] . z. k. guo , n. ohta and s. tsujikawa , phys . d * 75 * , 023520 ( 2007 ) , [ arxiv : hep - th/0610336 ] . a. lue , l. m. wang and m. kamionkowski , phys . * 83 * , 1506 ( 1999 ) [ arxiv : astro - ph/9812088 ] . k. choi , j. c. hwang and k. w. hwang , phys . d * 61 * , 084026 ( 2000 ) [ arxiv : hep - ph/9907244 ] . s. alexander and j. martin , phys . d * 71 * , 063526 ( 2005 ) [ arxiv : hep - th/0410230 ] . d. h. lyth , c. quimbay and y. rodriguez , jhep * 0503 * , 016 ( 2005 ) [ arxiv : hep - th/0501153 ] . m. satoh , s. kanno and j. soda , phys . d * 77 * , 023526 ( 2008 ) , arxiv:0706.3585[astro - ph ] . s. saito , k. ichiki and a. taruya , jcap * 0709 * , 002 ( 2007 ) , arxiv:0705.3701 [ astro - ph ] . n. seto , phys . lett . * 97 * , 151101 ( 2006 ) [ arxiv : astro - ph/0609504 ] . n. seto and a. taruya , phys . lett . * 99 * , 121101 ( 2007 ) , arxiv:0707.0535 [ astro - ph ] . s. weinberg , phys . d * 77 * , 123541 ( 2008 ) , arxiv:0804.4291[hep - th ] . m. satoh and j. soda , jcap * 0809 * , 019 ( 2008 ) , arxiv:0806.4594[astro - ph ] . z. k. guo and d. j. schwarz , phys . d * 81 * , 123520 ( 2010 ) , arxiv:1001.1897[hep - th ] . v. f. mukhanov and a. vikman , jcap * 0602 * , 004 ( 2006 ) [ arxiv : astro - ph/0512066 ] . d. h. lyth , phys . 78 * , 1861 ( 1997 ) [ arxiv : hep - ph/9606387 ] .

we examined the anisotropic point spread function ( psf ) of suprime - cam data utilizing the dense star field data . we decompose the psf ellipticities into three components , the optical aberration , atmospheric turbulence and chip - misalignment in an empirical manner , and evaluate the amplitude of each component . we then tested a standard method for correcting the psf ellipticities used in weak lensing analysis against mock simulation . we found that , for long - exposure data , the optical aberration has the largest contribution to the psf ellipticities , which could be modeled well by a simple analytic function based on the lowest - order aberration theory . the statistical properties of psf ellipticities resulting from the atmospheric turbulence are investigated by using the numerical simulation . the simulation results are in a reasonable agreement with the observed data . it follows from those findings that the spatial variation of psf ellipticities consists of two components ; one is a smooth and parameterizable component arising from the optical psf , and the other is a non - smooth and stochastic component resulting from the atmospheric psf . the former can be well corrected by the standard correction method with polynomial fitting function . however , for the later , its correction is affected by the common limitation caused by sparse sampling of psfs due to a limited number of stars . we also examine effects of the residual psf anisotropies on suprime - cam cosmic shear data ( 5.6-degree of-band data ) . we found that the shape and amplitude of b - mode shear variance are broadly consistent with those of the residual psf ellipticities measured from the dense star field data . this indicates that most of the sources of residual systematic are understood , which is an important step for cosmic shear statistics to be a practical tool of the precision cosmology .

1304.4992

weak gravitational lensing has now became a unique and practical tool to probe dark matter distribution irrespective of its relation to luminous objects , thanks to great progress in technique for weak lensing analysis as well as in a wide field imaging instrument ( for reviews see mellier 1999 ; bartelmann & schneider 2001 ; refregier 2003 ; hoekstra & jain 2008 ; munshi et al . measuring the statistics of weak lensing shear , also known as the cosmic shear statistics , has been recognized as one of most powerful probes of cosmological parameters and is employed as the primary science application by many on - going / future wide field survey projects ( e.g. , the panoramic survey telescope & rapid response system ( pan - starrs ) ; dark energy survey ( des ) ; hyper suprimecam survey ; the kilo - degree survey ( kids ) ; the large synoptic survey telescope ( lsst ) ; wide field infrared survey telescope ( wfirst ) ; euclid mission ) . it should be , however , noted that the full potential of cosmic shear statistics to place a constraint on the cosmological parameters is archived only if systematic errors are reduced down to a required level ( huterer et al . 2006 ; cropper et al . 2013 ; massey et al . 2013 ) . in the process of weak lensing analysis , the correction for the point spread function ( psf ) is of fundamental importance , because it has two serious effects on any galaxy shape measurement : one is the circularization of the images , which systematically lowers the lensing shear . the other is a coherent deformation of the images caused by anisotropy of the psf , which may mimic a weak lensing shear . much effort has been made to develop techniques and software to correct the psf effects , and the weak lensing community has conducted blind tests using mock data to evaluate their performance ( heymans et al . 2006a ; massey et al . 2007 ; bridle et al . 2010 ; kitching et al . 2012a , b ; and references therein ) . most of the software is designed to correct the psf , regardless of its origins . major sources of psf , recognized so far , except for diffraction , include the atmospheric turbulence , optical aberration , the misalignment of ccd chips on a focal plane and pixelization . an alternative and complementary approach to this issue is to investigate the properties of the psf for a specific instrument , while focusing on a specific cause(s ) , which may help to optimize the psf correction scheme and to understand a level of residual systematics ( hoekstra 2004 ; jarvis & jain 2004 ; wittman 2005 ; jarvis , schechter & jain 2008 ; jee et al . 2007 ; 2011 ; rhodes et al . 2007 ; jee & tyson 2011 ; heymans et al 2012 ; chang et al . 2003 ) . the latter approach is exactly what we consider in the present paper . the purpose of this study is twofold : first , we look at the properties of anisotropic psfs of the subaru prime focus camera ( suprime - cam , miyazaki et al . 2002 ) . we pay a special attention to evaluating the amplitude and spatial correlation of psf anisotropies caused by different sources . to do this , we used a set of dense star field images from which we can sample the psf densely , enabling us to investigate the spatial variation of the psf in detail . second , using mock simulation data of psfs , we examined how well the psf anisotropy can be corrected using the standard psf correction scheme . the motivation of this study concerns the existence of a non - zero b - mode in the cosmic shear correlation functions measured from suprime - cam data . the measurement was made by hamana et al . ( 2003 ) , they analyzed a 2 degree@xmath0 data and found statistically significant b - mode correlations ( see [ sec : scambmode ] for an updated but similar result ) . since the b - mode shear is not produced by gravitational lensing in the standard gravity theory at the lowest order of the perturbative treatment , its existence indicates any non - lensing process(es ) taking place . the origin(s ) of the b - mode was unclear ; it could be simply due to remaining systematics in the data reduction and analysis . also it may arise from an intrinsic alignment of galaxy shapes ( e.g. , croft & metzler 2000 ; catelan , kamionkowski & blandford 2001 ; crittenden et al . 2002 ; jing 2002 ; hirata & seljak 2004 ; heymans et al . more importantly , the b - mode may be produced by a non - standard gravity ( e.g. , yamauchi , namikawa & taruya 2012 ; 2013 ) . thus the b - mode shear , in principle , provides us with valuable information on the gravity theory . it is thus important to first understand the level of residual systematic in the data analysis ( see , for a similar approach , hoekstra 2004 ; van waerbeke , mellier & hoekstra 2005 ) . the outline of this paper is as follows ; in section 2 , data reduction and psf measurements of dense star field data are described . the amplitude and spatial correlation of psf anisotropies caused by different sources are evaluated separately in section 3 . the psf correction scheme adopted in our cosmic shear analysis is tested against the mock simulation data in section 4 . in section 5 , we assess the impact of the residual psf anisotropy on the cosmic shear analysis . finally , a summary and discussion are given in section 6 . in appendices , we summarize the properties of psf ellipticities caused by third - order optical aberrations ( appendix 1 ) , and describe a numerical study of psf ellipticities caused by atmospheric turbulence ( appendix 2 ) . we used @xmath1-band data of a dense stellar field taken with the suprime - cam on 2002/9/30 , 2003/6/30 and 2003/7/1 . the field is located at the galactic coordinate of @xmath2 . we collected 289 shots of 60 seconds exposure and 70 shots of 30 seconds exposure from the data archive smoka . the seeing fwhms of those data range from 0.46to 0.88with a median of 0.63 . each ccd data was reduced using the sdfred software ( yagi et al . 2002 ; ouchi et al . note that we conservatively used data only within 15 arcmin radius from the field center of suprime - cam , because at the outside of that , the point spread function ( psf ) becomes elongated significantly , which may make the correction for the psf inaccurate . then , mosaicking of 10 ccds was performed with scamp ( bertin 2006 ) and swarp ( bertin et al . 2002 ) . in addition to the individual exposure images , we also generated stacked images using swarp . note that the images obtained during a same night were taken with the same pointing ( thus no dithering was made ) . since differences in pointings of images taken on different nights are very small ( less than the gaps between ccds ) , no data from different ccds was added . object detections were performed with sextractor ( bertin & arnouts 1996 ) and _ hfindpeaks _ of imcat software ( kaiser , squires & broadhurst 1995 ) , and two catalogs were merged by matching positions with a tolerance of 1 arcsec . from the image of stars , we measured an anisotropy of the psf . we use the following ellipticity estimator ( exactly speaking the polarization , see kaiser et al . 1995 ) to characterize the anisotropy of psfs : @xmath3 @xmath4 where @xmath5 is the surface brightness of an object and @xmath6 is the gaussian window function . getshapes _ of imcat for the actual computation of these quantities . from catalogs of detected objects ( almost all of them are stars ) , we generated two kinds of sub - catalogs by mimicking weak lensing analysis : one is a `` star - role '' catalog which contains 700 ( the number density of 1 arcmin@xmath7 ) of the highest sn non - saturated stars . the other is a `` galaxy - role '' catalog which contains 30,000 ( 42 arcmin@xmath7 ) of the next highest sn non - saturated stars . the minimum flux sns for star - role and galaxy - role catalogs are sufficiently high ; @xmath8 and @xmath9 , respectively . an anisotropy of psf results from several sources including the atmospheric turbulence , the optical aberration , the misalignment of ccd chips on a focal plane , the pixelization , and tracking error of the telescope . our attempt in this section is to examine the psf on a source - by - source basis using physically motivated models . we evaluate the statistical properties of psf anisotropies using a two - point correlation function related estimator , called the e- and b - mode aperture mass variances defined by ( schneider et al . 1998 ; we follow the notation by schneider , van waerbeke & mellier 2002 ) @xmath10,\ ] ] and @xmath11,\ ] ] where @xmath12 and @xmath13 are defined in schneider et al . ( 2002 ) , and @xmath14 and @xmath15 are the two - point correlation functions of the ellipticities computed by @xmath16 in the above expressions , the summation is taken over all pairs of objects with a distance within a width of a bin considered @xmath17 , @xmath18 is the number of pairs in the bin , and @xmath19 and @xmath20 are the tangential and @xmath21 rotated ellipticity components in the frame defined by the line connecting a pair of objects . some authors adopt @xmath22 instead of @xmath23 as an estimator of the spatial correlation of psf ellipticities and its residual , which remains after the correction ( jee & tyson 2011 ; chang et al 2012 ; 2013 ) . the reason for our choice of @xmath24 is as follows . in the case of an ellipticity field with power spectrum with a spectrum index more negative than @xmath25 ( this is exactly our case as will be shown in subsequent subsections ) , @xmath26 becomes almost flat shape , because the amplitude of not only the large - scale , but also the small - scale correlation function , is dominated by the power on larger scales . also , even if only the larger scale psf ellipticy correlation is corrected , not only the large - scale but also the small - scale amplitude decreases . therefore @xmath22 is not suited for quantifying the spatial correlation of our psf ellipticity data . on the other hand , @xmath24 is a kind of bandpass filtered variance , thus it is more suitable estimator for our current purpose . although psf ellipticities arise from several causes and vary temporally , there is a visibly identifiable component that most likely results from optical aberrations . in the top - left panel of figure [ fig : esp_emap1 ] and [ fig : esp_emap2 ] , we show two examples of _ whisker plots _ of psf ellipticities where characteristic features of optical aberration can be clearly observed : we also show one case in figure [ fig : esp_emap3 ] where an optical aberration is almost invisible . we fit the spatial variation of psfs with the following fitting function of the optical psf ellipticities , which is motivated from the lowest - order optical aberration theory ( see appendix [ appendix : aberration ] for details ) , @xmath27 where @xmath28 and @xmath29 . we determine the model parameters @xmath30 , @xmath31 and @xmath32 by the standard least - square method with observed psf ellipticities . this model consists of the axis - symmetric , asymmetric and constant components , which are displayed in the bottom - left panel , bottom - right panel , and top - right corner of top - panels , respectively . note that in the top - left panel , the psf ellipticities measured from `` star - role '' stars , but after the constant component being subtracted , are plotted . it is clearly seen from figure [ fig : esp_emap1 ] and [ fig : esp_emap2 ] that the simple three - component model can well reproduce the observed psfs . it may therefore be said that the optical aberration is one major origin of the psf anisotropy . note that a cause(s ) of the constant component is uncertain . it arises not only from an optical aberration ( the misalignment coma , see appendix [ appendix : aberration ] ) but also from , e.g. , a tracking error . since the constant ellipticity component does not contribute to the aperture mass variances , we do not consider it in the following discussion . , [ fig : esp_emap2 ] and [ fig : esp_emap3 ] are marked with an open circle , open square and open triangle , respectively . dotted boxes define groups used for statistical analyses . [ fig : gamt_prof],width=302 ] having found that the optical aberration is one major source of the psf ellipticities and can be well modeled with one constant and two spatially variable components , we introduce estimators that quantify the amplitude of the symmetric and asymmetric components : @xmath33 and @xmath34 , where the rms scatter @xmath35 is computed from the model ellipticities of 700 star - role psfs for each exposure , and `` sign '' in @xmath36 is @xmath37 or @xmath38 for the tangentially ( the mean tangential ellipticity has the plus sign ) or radially elongated cases , respectively . the results for all exposures are plotted in figure [ fig : gamt_prof ] . it can be observed from the figure that there is a `` v - shaped '' trend between strengths of the symmetric component and asymmetric component . in the lowest - order aberration theory , the symmetric aberration is caused by a shift in the focal plane position from an ideal position , whereas the asymmetric one is caused by the decenter and/or tilt between the axes of the focal plane and the corrector ( lenses ) . the result indicates that those two deviations happen in a mutually related manner . in figure [ fig : gamt_prof ] one may find that there are some exposures that are located somewhat outside of the `` v - shaped '' trend . it is speculated from a visual inspection that the aberration model fitting of those cases is affected due to the atmospheric psf . in our model fitting procedure , such contamination is unavoidable . this issue should be noticed when one uses the model fitting for evaluating the optical psf component from observed psf ellipticities . , and the plotted aperture mass variances are average values taken over all exposures in each group . [ fig : mapvar_groups],width=302 ] next we examine the statistical properties of the psf ellipticities using the aperture mass variance , eqs . ( [ eq : mape ] ) and ( [ eq : mapb ] ) while especially focusing on contribution from the optical aberration . to do so , we classified data using @xmath36 and @xmath39 into 9 groups as shown in figure [ fig : gamt_prof ] . we computed the average aperture mass variances for each group , and show the results in figure [ fig : mapvar_groups ] . the left panels of the figure show the results based on the observed data , from which it is found that the e - mode variance has a larger amplitude on larger scales , whereas in most cases b - mode variance is smaller than the e - mode especially on larger scales . if we assume a power - law spectrum for ellipticity power spectrum where @xmath40 is the aperture function . thus @xmath41 for a power - law power spectrum of @xmath42 . ] @xmath42 , the measured e - mode variance indicates that @xmath43 . the right panels of figure [ fig : mapvar_groups ] show the aperture mass variances computed from psf ellipticities of the optical aberration model . comparing those with observed variance , plotted in the left panels , we can estimate the amount of the contribution from the optical aberration to the psf ellipticity . we first look at the e - mode . it follows from the comparison between the aperture mass variance measured from the observed ellipticities and the model ( the left and right panels figure [ fig : mapvar_groups ] ) that for large aberration cases such like g1 - 3 , g8 and g9 , the model variance accounts for most of the observed one , indicating that the optical aberration is the dominant component of the psf ellipticities , as expected from the visual impression ( figs . [ fig : esp_emap1 ] and [ fig : esp_emap2 ] ) . for small aberration cases ( g4-g7 ) , one may see the aperture mass variances of the model are slightly smaller than the observed ones . in the following subsections , we argue that the residual variances mostly arise from atmospheric turbulence and chip misalignment . let s turn to the b - mode . it follows from a comparison between the left and right panels that the variances measured from the model are smaller than the observed ones except for g1 . this also suggests the existence of other components than the optical component . next , we examine the psf ellipticities remaining after subtracting the optical aberration component . let us start with a visual impression . in figures [ fig : emap1 ] and [ fig : emap2 ] , the psf ellipticities of the observed data ( left ) , the optical aberration model ( center ) , and the residual of those two ( @xmath44 ) are plotted . note that the mosaicked data consists of 2-row@xmath455-column ccd chips with a masked outer region ( beyond 15 arcmin radius from the field center ) . in those residual plots , there are two apparent features that should be noticed : one is discontinuities between chips , and the other is ripple - like patterns . the latter is a characteristic feature of psf ellipticites caused by atmospheric turbulence which was also observed in cfht data ( heymans et al . 2012 ) , and was also found in numerical simulations ( jee & tyson 20011 ; chang et al . 2012 ; 2013 ; also see appendix [ appendix : atmos ] ) , which we examine in detail in the next subsection [ sec : atmospheric ] . in this subsection , we focus on psf ellipticities caused by misalignment of ccd chips . to do so , we first need to create data that are minimally affected by the optical aberration and atmospheric turbulence . the former could be minimized by using data with small values of both @xmath46 and @xmath39 . the latter could be suppressed by stacking many images as the amplitude of the atmospheric psf decreases , with the exposure time being @xmath47 ( de vries et al 2007 ; chang et al . 2012 ; heymans et al . 2012 ; see also appendix [ appendix : atmos ] ) . thus we generate a coadded image of 52 shots of 60-second exposures taken from the g5 group where the effect of the optical aberration is at a minimum level and the largest number of shots ( 52 shots ) is contained . the stacking was done using swarp , and the object detection and ellipticity measurement were performed in the same manner as that described in section [ sec : data ] . . the solid line is for the observed psfs , dashed line is for the aberration model , the dotted line is for the residual of those two ( @xmath44 ) and the long - dashed for chip - misalignment psf model . [ fig : mapvar_chip],width=302 ] the psf ellipticities of the coadded data are shown in figure [ fig : emap ] . one can see psf discontinuities between the chips even in the observed psf ( top - left panel ) as well as in the residual psfs ( top - right panel ) . we attempt to extract the chip - misalignment component from the residual psf ellipticities by fitting the residual psfs with the 2nd order bi - polynomial function on the chip - by - chip basis . the derived psf models are shown in the bottom - right panel of figure [ fig : emap ] in which it can be seen that characteristic features of chip - misalignment psf ellipticities ( such as discontinuities between the chips and the chip - scale smooth variation ) are well reproduced by the model . we measured the aperture mass variances of the psf ellipticities ( show in figure [ fig : mapvar_chip ] ) . note that the b - mode variances are too small to properly analyze , and thus we do not present them . although our model may not extract only the chip - misalignment psfs , but may be affected by other sources of psf ellipticity , it may still be reasonable to consider that the model allows us to assess an approximate amplitude of the chip - misalignment psf ellipticies . we read from the figure [ fig : mapvar_chip ] that the amplitude of the chip - misalignment psf ellipticies is roughly @xmath48 . it should , however , be noted that the chip - misalignment psf is a phenomenon related to the optical aberration , and thus its amplitude may vary with deviation in the focal plane position . therefore the above value should be considered as a characteristic size . it is also found from figure [ fig : mapvar_chip ] that the aperture mass variance of the residual psfs ( @xmath49 ) is slightly larger than that of the chip - misalignment psf model . the amplitude of the excess is roughly @xmath50 . the origin of this excess is not clear , though it could be due to some imperfection in the models , and/or due to atmospheric psfs , and/or due to other causes such as pixelization effects caused in a process of resampling ( hamana & miyazaki 2008 ) . if the latter is the case , it can be settled by recently developed shape measurements schemes that do not involve a resampling process ( e.g. , miller et al . 2007 ; 2013 ; and miyatake et al 2013 ) . thus we leave it for future work . , width=332 ] arcmin as a function of the exposure time . circles and triangles are for e- and b - mode , respectively . the expected exposure time dependence of @xmath51 ( plotted by the solid line ) is clearly seen . [ fig : mapvar_texp],width=302 ] let us turn to the psf ellipticity arising from atmospheric turbulence . the properties of the atmospheric psf ellipticity has been examined experimentally ( wittman 2005 ; heymans et al . 2012 ) or using numerical simulations ( de vries et al 2007 ; jee & tyson 20011 ; chang et al . 2012 ; 2013 ) . a major characteristic of the atmospheric psf ellipticity found from previous studies is that the amplitude of psf ellipticities decreases with the exposure time as @xmath47 ( de vries et al 2007 ; chang et al . 2012 ; heymans et al . 2012 ) . also , the shape of the power spectrum and aperture mass variance of the atmospheric psf ellipticities was investigated numerically ( see appendix [ appendix : atmos ] ) . we investigate the atmospheric psfs using data from the g5 group . in top - panels of figure [ fig : mapvar_diff ] , we show the aperture mass variances for different exposure times ; from top to bottom , single 30-second , 60-second exposures , coadded 4@xmath4560 , 13@xmath4560 and 52@xmath4560-second exposures . as is shown in the plots , the amplitude of the aperture mass variance decreases as the exposure time increases . however the minimum amplitude of the e - mode is set by the optical aberration psfs and chip - misalignment psfs , as shown in the last two subsections , whereas the b - mode variance decreases down to @xmath52 . this indicates that the optical and chip - misalignment psf largely contribute to the e - mode psf ellipticity . in order to evaluate the amplitude of the atmospheric psf ellipticities separately from other components , we make a working hypothesis that the amplitude of the deepest data ( 52@xmath53-sec coadded data ) represents the amplitude of ( quasi-)static psf components ( such like the optical and chip - misalignment psfs ) ; the difference from that gives an estimate of atmospheric psfs ellipticities . in the bottom - panels of figure [ fig : mapvar_diff ] and figure [ fig : mapvar_texp ] , we plot the differences , from which one finds that the aperture mass variances have both the exposure time dependence and a shape similar to those obtained from the numerical simulation ( appendix [ appendix : atmos ] ) . also , the e- and b - mode variances have similar amplitudes , as expected from the numerical simulation . although the above results were derived relaying on the working hypothesis , those nice agreements would support a reasonableness of the hypothesis . we may therefore say that we successfully captured characteristic features of the atmospheric psf ellipticities in the suprime - cam data . here we consider how well the psf ellipticities can be corrected by the standard correction scheme , which we describe below . the aim of this analysis is twofold : one is to understand what sets a lower limit of the anisotropic psf correction ; the other is to evaluate the amplitude of residual psf ellipticities . to do this , we use both the dense star field data and mock data that we describe below . before going into details , we briefly describe a basic procedure concerning the anisotropic psf correction implemented in actual weak lensing studies so far ( see for details , heymans et al 2006 ; massey et al 2007 and references therein ) : firstly , psf ellipticities ( or other estimators of psf ) are measured from high - sn star images . in usual weak lensing analyses , a typical number density of stars used for psf sampling is @xmath54 arcmin@xmath7 . secondly , the spatial variation of psf ellipticities is modeled by fitting the psf ellipticities with an analytic function , such as a polynomial . then finally , an artificial shape deformation in galaxy images caused by the psf is corrected using the psf model delivered by the analytic function . although an actual implementation of the above procedure depends on technique of weak lensing analysis ( heymans et al 2006 ; massey et al 2007 ) , there is one common limitation which comes from a sparse sampling of stars , namely , the spatial variation of psfs on scales smaller than the mean star separation is poorly sampled , and as a result small - scale components in the psf anisotropy are hardly corrected . our psf correction scheme is based on the ksb algorithm ( kaiser et al 1995 ; hamana et al . 2003 ; heymans et al 2006 ) . we refer the reader to the above references for details . here , we only describe some details concerning implementation , which are specific to this study . we use bi - variable polynomial function for modeling the spatial variation of psf ellipticites . we divide data into @xmath55 regions or into each ccd chip , and a psf model is generated for each sub - region . the order of the polynomial is set to 4th for a @xmath55-division and 2nd or 3rd for the chip - basis case . the @xmath55-division is taken because in our actual cosmic shear analysis the psf correction is made on a sub - region basis with a similar area , whereas the chip - basis analysis is used to test its ability of improving the anisotropic psf correction . -division , whereas the dotted line is for the chip - basis . from left to right , the exposure time is 1 , 5 and 25 min , respectively . [ fig : mapvar_e3_texp_chip],width=332 ] let us start with a simpler situation using mock psf ellipticity catalogs . the mock data have the same geometry and number of objects as those of the dense star field data , that is 700 `` star - role '' objects and 30,000 `` galaxy - role '' objects ( see section [ sec : data ] ) . instead of generating mock image , we simply generate object catalogs . each object has various items ; the positions , and 3-component psf ellipticities , including optical , atmospheric , and ccd - misalignment components denoted by @xmath56 , @xmath57 and @xmath58 , respectively . the optical component is made by using the optical aberration - motivated model , equation ( [ eq : psfaberrat ] ) , introduced in appendix [ appendix : aberration ] ; we adopted model parameters for g5 group in subsection [ sec : aberration ] . for the ccd - misalignment component , we adopt the 2nd - order polynomial models obtained in subsection [ sec : chip ] . the atmospheric component is modeled by the random field with the power spectrum having a power - law index of @xmath59 based on our finding ( see section [ appendix : atmos ] and appendix 2 ) . its amplitude is set by the empirical relation of equations ( [ eq : ps - scaling ] ) for exposure times of 1 , 5 , and 25 minutes . we compute the total ellipticity by simply adding the three components , instead of treating in the convolution manner . this is a reasonable approximation , because each ellipticity component is small . then , the anisotropic psf correction is made for mock catalogs in the procedure described above . the results from a mock simulation are presented in figure [ fig : mapvar_e3_texp_chip ] , from which the following three points can be seen . first , as mentioned above , the mean separation of stars ( @xmath54 arcmin ) , which are used to model the spatial variation of psfs , sets a fundamental limitation on the psf correction . in the case of the aperture mass , its window function is most sensitive to fluctuations on scales of @xmath60 of the aperture size ( schneider et al . 1998 ) , this defines a characteristic aperture scale of @xmath61 arcmin ; on scales larger than that , the correction works very well but on smaller scale it does very poorly ( this was argued by hoekstra et al . 2004 ) . this is especially the case for the atmospheric psf . second , in spite of the above limitation , the optical components are well corrected , even on smaller scales . the reason of this is the smoothness of the optical component ; as shown in subsection [ sec : aberration ] , the spatial variation of the optical psfs can be well fitted by a polynomial - based model . thus even with a sparse sampling , it can be well - modeled with the polynomial function . third , for @xmath55-division case , the ccd - misalignment component combined with the optical component sets a limit on the anisotropic psf correction . this is a natural consequence that even if each individual component is smooth and is modeled by a polynomial function , the mixed psfs may no longer be simple and thus may not be corrected by the polynomial model . this can be partly avoided by adopting the chip - basis correction , as can be seen in the right panel of figure [ fig : mapvar_e3_texp_chip ] , though the residual variance does not reach to the level of the atmospheric psf . a further improvement of the psf correction may be achieved by recently proposed correction schemes ( e.g. , chang et al . 2012 ; berg et al . 2012 ; gentile , courbin & meylan 2013 ) ; we leave it for future study . the above three findings provide us with a clue to interpret a result obtained from an analysis of real data . sub - regions except for the magenta curve which is for the chip - basis correction . [ fig : mapvar_abbgroup_eb.g5cor],width=332 ] let us turn to the analysis of dense star field data . in figure [ fig : mapvar_abbgroup_eb.g5cor ] , the aperture mass variances measured from the g5 group data ( defined in figure [ fig : gamt_prof ] ) are presented for different exposure times , where an anisotropic psf correction was made on the basis of @xmath55 sub - regions , except for the magenta curve , to which the chip - basis correction was applied . from the figure , the following four findings are obtained : firstly , it is found that after the psf correction , the e- and the b - mode are almost equally partitioned , regardless of properties of raw psf ellipticities . a plausible reason for this is that the e- and b - mode are mixed up in the process of the psf correction . secondly , the result of 60-sec exposure is in a good agreement with that of the mock simulation . but at a closer look at larger scales , it is found that the residual aperture mass variance falls less quickly than the mock result , indicating that the spatial variation of real psf ellipticities is more complex than that assumed in the mock simulation . thirdly , as expected from the result of the mock simulation , the residual aperture mass variance decreases with increasing the exposure time . however , the decrement is less than what can be expected from the mock simulation . finally , it is found from the comparison of two different correction schemes for 52-coadded data that the chip - basis correction makes an only a small improvement over the other case . it is speculated from the last two points combined with the findings from the mock simulation ( that is , an existence of a non - smooth component sets a lower limit of the correction on scales smaller than mean separation of stars ) that there exists unidentified psf ellipticity component(s ) whose amplitude is greater than the chip component mentioned in subsection [ sec : chip ] ( see also figure [ fig : mapvar_chip ] ) . although the origin of the unknown psf component is not clear , the above result provides us with an empirical estimate of _ the `` best performance '' _ of the psf correction scheme that we adopted in this paper . here , we examine the effects of residual psf anisotropies on suprime - cam cosmic shear data . to do so , we measure the e- and b - mode shear aperture mass variance from 5.6-degree@xmath0 suprime - cam deep imaging data , and compare the b - mode shear variance with the residual psf ellipticities measured from the dense star field discussed in section [ sec : psf ] . we use the same data as those used in hamana et al . ( 2012 ) in which basics of data and data analyses are described . therefore here we only describe points specific to this work . suprime - cam @xmath1-band data were corrected from the data archive , smoka , under the following three conditions : data are contiguous with at least four pointings , the total exposure time for each pointing is longer than 1800 sec , and the seeing full width at half - maximum ( fwhm ) is better than 0.65 arcsec . four data sets ( named by sxds , cosmos , lockman - hole and elais - n1 ) meet these requirements . the exposure times of individual exposures range from 120 and 420 seconds . data reduction , mosaic stacking , and object detection were done using the same procedure described in section [ sec : data ] . the effective area after masking regions affected by bright stars is 5.57 degree@xmath0 . the depth of the coadded images varies among four fields , and it is found that the number counts of faint galaxies are saturated at ab magnitude of @xmath62 . for weak lensing measurements , we adopt the so - called ksb method described in kaiser et al . ( 2005 ) , luppino & kaiser , and hoekstra et al . ( 1998 ) with some modifications being made following recent developments ( heymans et al . 2006a ) , which we describe below . stars are selected in a standard way by identifying the appropriate branch in the magnitude half - light radius ( @xmath63 ) plane , along with the detection significance cut @xmath64 . the number density of stars is found to be @xmath54 arcmin@xmath7 for the four fields . we only use galaxies that meet the following three conditions : ( i ) a detection significance of @xmath65 and @xmath66 , where @xmath67 is an estimate of the peak significance given by _ hfindpeaks _ , ( ii ) @xmath63 is larger than the stellar branch , and ( iii ) the ab magnitude is in the range of @xmath68 , where galaxy number counts of four fields are almost the same . the number density of resulting galaxy catalog is @xmath69 arcmin@xmath7 and is quite uniform among four fields . we measure the shape parameters of objects by _ getshapes _ of imcat . the psf correction was done on a sub - field basis , where the coadded data that were divided into sub - fields of about @xmath70 arcmin@xmath0 ( approximately one fourth of the suprime - cam s field - of - view ) . anisotropic psf correction was done following the implementations by mh , ch and ts of heymans et al . ( 2006a ) . the spatial variation of psf ellipticites measured from stars were modeled with the 4th order bi - polynomial function ; also , the psf model combined with the smear polarizability tensor ( @xmath71 ) of each galaxy was used to correct the psf ellipticities of the galaxies . in ksb formalism , the shear ( we denote by @xmath72 ) is related to the observed ellipticity through the shear polarizability tensor , @xmath73 , which is evaluated by a smoothing and weighting method developed by van waerbeke et al . ( 2000 ; and see section a5 of heymans et al . 2006a ) . but for data used for the cosmic shear analysis . the optical model fitting is applied to individual exposures . the exposure time of each exposure is @xmath74 seconds . different symbols are for different fields . [ fig : eastig_emisali.scana],width=302 ] before presenting the shear aperture mass variance measured from coadded data , it is worth looking into psf ellipticities of individual exposures . to do so , we generated mosaicked data of individual exposures , and selected stars in the same procedures as described above . the spatial variation of psf ellipticities measured from the stars were fitted to the optical model , equation([eq : psfaberrat ] ) , and the amplitude of the symmetric and asymmetric components of the model were evaluated ( see subsection [ sec : aberration ] and appendix [ appendix : aberration ] for details ) . the results are plotted in figure [ fig : eastig_emisali.scana ] , in which one may see `` v - shaped '' trend similar to the results obtained from the dense star field data ( figure [ fig : gamt_prof ] ) , though the scatter is somewhat larger for this case . this similarity implies that knowledge of the psf anisotropy learned from the dense star field analysis ( section [ sec : psf ] ) is applicable to cosmic shear analyses . it should , however , be noticed that there is one component that is not taken into consideration in the dense star field analysis , namely the stacking of multiple dithered exposures . it is obvious that the stacking of dithered exposures makes the spatial variation of psfs complicated , which may result in a poorer anisotropic psf correction . we return to this point later . ) measured from the g5 group of the dense star field data ( see section [ sec : psf ] and figure [ fig : mapvar_abbgroup_eb.g5cor ] ) ; the upper and lower curves are for @xmath75 and @xmath76-second exposure , respectively . the dotted curve shows the @xmath77cdm prediction based on the 7-year wmap cosmological model with the source galaxy redshift distribution inferred from cosmos photometric redshift catalog . [ fig : mapvar_scana_g5cor],width=332 ] we compute the e- and b - mode aperture mass variance of galaxy shears through the two - point correlation functions using equations ( [ eq : mape])-([eq : xi+- ] ) with the summations in equation([eq : xi+- ] ) being replaced with a weighted summation , @xmath78 , where @xmath79 is the weight for @xmath80-th galaxy . in addition to the shear , we compute the _ uncorrected - shear _ , that is , the galaxy shear to which the correction for the psf anisotropy is not applied . the results are presented in figure [ fig : mapvar_scana_g5cor ] . the error bars indicate the root - mean - square among 100 randomized realizations , in which the orientations of galaxies are randomized , and presumably represent the statistical error coming from the galaxy shape noise . it is found from figure [ fig : mapvar_scana_g5cor ] that the b - mode shear variance is non - zero on the aperture scale of @xmath81 arcmin . it should be noted that adjacent bins are correlated . in order to compare the shear variances with those of the residual psf ellipticites , we transform the star ellipticity to the galaxy shear by multiplying a factor @xmath82 , which is found to be @xmath83 ( thus @xmath84 ) . the results from the g5 group of the dense star field data with the @xmath85 sub - region basis correction ( see section [ sec : psf ] and figure [ fig : mapvar_abbgroup_eb.g5cor ] ) are also shown in figure [ fig : mapvar_scana_g5cor ] . the upper and lower dashed - curves are for @xmath86 seconds and @xmath76 seconds exposure , respectively . since the total exposure times of four cosmic shear fields are @xmath87 seconds , the lower curve could be considered as a `` _ best performance _ '' of our psf ellipticity correction scheme . the b - mode shear signals at smaller scales ( @xmath88 arcmin ) are larger than this expectation . a possible reason for this excess is that the stacking of dithered multiple exposures which is not involved in the dense star field analysis . this issue will be investigated in detail in a future study in which we will adopt a psf correction scheme on an individual exposure basis . on the other hand , the b - mode variance drops at larger scales ( @xmath89 arcmin ) , which is in nice agreement with that expected from the dense star field analysis . it is however noticed that in the case of the dense star field data , the turnaround aperture scale is about 5 arcmin , which is naturally set by the mean separation of stars ( @xmath54 arcmin ) ( see section [ sec : psfcorrection ] ) , whereas it is about 10 arcmin for the cosmic shear data . the reason for this is unclear , though it could be related to an excess b - mode uncorrected shear variance observed at @xmath90 arcmin . with the parameters denoted in the panel . the mean redshift of the model is 1.07 which is in a very good agreement with the value derived from the photometric redshift data ( @xmath91 ) . [ fig : nsw],height=302 ] finally , we evaluate the influence of the residual psf anisotropies on the cosmic shear e - mode measurement . we plot in the left panel of figure [ fig : mapvar_scana_g5cor ] the @xmath77cdm prediction based on the 7-year wmap cosmological model ( komatsu et al . 2011 ) as a basis for comparison . in computing the @xmath77cdm prediction , we used the halofit model by smith et al . ( 2003 ) , and we used the redshift distribution of source galaxies inferred by the following manner : we utilized the cosmos field data . we merged the galaxy catalog used for the cosmic shear measurement with the cosmos photometric redshift catalog ( ilbert et al 2009 ) , and compute the redshift distribution by adopting the photometric redshift , which is presented in figure [ fig : nsw ] . we fit the redshift distribution with the following function ( fu et al . 2008 @xmath92 where the normalization , @xmath93 , is determined by imposing @xmath94 within the integration range @xmath95 . we found that a parameter set @xmath96 gives a reasonably good fit to the data , as shown in figure [ fig : nsw ] . it is found from a comparison between the @xmath77cdm prediction and the residual psf shear variance evaluated from the dens star field analysis ( figure [ fig : mapvar_scana_g5cor ] ) that on scales below 10 arcmin , the residual shear variance is more than 10 percents of the expected cosmic shear variance . the comparison between the measured e- and b - mode shear variances on those scales results in even worse figures . on the other hand , on larger scales ( @xmath97 arcmin ) , where the anisotropic psf correction works well , the residual variance is well less than 10 percents of the measured and expected e - mode variance . this is one of important findings of this paper . before closing , we comment on an effect of the noise bias ( refregier et al . 2012 ; melchior & viola 2012 ; okura & futamase 2012 ) on the above analysis . the noise bias introduces a systematically lower ellipticity value with its amplitude depending on the sn ( a smaller bias for a higher sn object ) . refregier et al . ( 2012 ) found that the amplitude of noise bias is a few percent for sn@xmath98 objects . since the threshold sns adopted in the above analysis are sn@xmath99 for the cosmic shear galaxy selection and sn@xmath9 for the dens star field analysis , the expected amplitude of the noise bias on the above shear variance measurements might be less than 5 percent . therefore , conclusions from the above analysis are , at least qualitatively , not affected by the noise bias . we analyzed the anisotropic psf of suprime - cam data utilizing the dense star field data . we decomposed the psf ellipticities into three components ( the optical aberration , atmospheric turbulence , and the chip - misalignment ) in an empirical manner , and assessed the amplitude of each component . we then tested a standard method for correcting the psf ellipticities against a mock simulation based on models of the psf ellipticities obtained from the dense star field data analysis . finally we examined the impact of the residual psf ellipticities on the cosmic shear measurement . we found that the optical aberration has the largest contribution to the psf ellipticities of long - exposure ( say , longer than 10 minutes ) data . we also found that the spatial variation of the optical psfs can be modeled by a simple three - component model , which is based on the lowest - order aberration theory described in appendix [ appendix : aberration ] . it was also found that the optical psf ellipticities vary smoothly on the focal plane , and thus are well corrected by the standard correction method in which the spatial variation is modeled by a polynomial function . the psf ellipticities after subtracting the optical psf model ellipticities show discontinuities between ccd chips , which indicates that the misalignment of chips on the focal plane induces an additional psf ellipticity . in order to make a crude estimation of the amplitude of the chip - misalignment psf ellipticities , we fitted the residual psf ellipticities to the 2nd order bi - polynomial function on a chip - by - chip basis . it turns out that the amplitude of the chip - misalignment psf ellipticities is lower than the optical component , though it is not negligible . it was found from the mock simulation that the chip - misalignment component combined with the optical component puts a lower limit on the capability of the psf correction , which can be , in principle , avoided by employing a chip - basis correction scheme . we investigated the properties of psf ellipticities resulting from the atmospheric turbulence using a numerical simulation of wave propagation through atmospheric turbulence under a typical weather condition of the subaru telescope at mauna - kea ( appendix [ appendix : atmos ] ) . from the simulation results , we evaluated the power spectrum and aperture mass variance of the atmospheric psf ellipticities , and derived power - law fitting functions of them . as was already pointed out in literature ( wittman 2005 ; de vries et al 2007 ; jee & tyson 20011 ; chang et al . 2012 ; 2013 heymans et al . 2012 ) , the rms amplitude of the atmospheric psf ellipticities decreases as @xmath100 . we computed the aperture mass variance of the dense star field data for various exposure time ( from 30-second to @xmath76-second ) . the amplitude of the atmospheric psf ellipticities was evaluated by assuming that the deepest data represents the amplitude of ( quasi-)static psf components , and the difference from that is due to the atmospheric psfs ellipticities . the results are found to be in reasonable agreement with the simulation results . therefore , we may conclude that the aperture mass variance of the atmospheric psf ellipticies for a long - exposure data ( say , longer than 10 minutes ) , is at least one order of magnitude smaller than that of the optical psf ellipticities . since the atmospheric psfs are not smoothly varying component , its correction is affected by the common limitation that the psf correction on scales smaller than the mean star separation works very poorly because of the poor sampling of the spatial variation of psfs on those scales . the above findings provide us a clue to develop an optimal psf interpolation scheme , it is found that the spatial variation of psf ellipticities consists of two components : one is a smooth and parameterizable component arising from the optical aberration and chip - misalignment ; the other is a non - smooth and stochastic component arising from the atmospheric psfs . the former can be modeled with a parametric model , as shown in this paper . also it has been argued that an interpolation scheme based on the principle - component analysis is effective for such a case ( jarvis & jain 2004 ; jee & tyson 2011 ; see also miyatake et al 2013 and lupton et al 2001 for an actual implementation for suprime - cam data reduction pipeline ) . on the other hand , it is shown in berg , et al ( 2012 ) and gentile et al . ( 2013 ) that local - type correction schemes , such like the radial basis functions and kriging work well for atmospheric psfs . apparently , a hybrid interpolation scheme , in which the above two types of interpolation schemes are optimally incorporated , is a strong candidate for achieving a better psf correction . we examined the effects of the residual psf anisotropies on suprime - cam cosmic shear data . we also compared the b - mode shear variance measured from 5.6-degree@xmath0 @xmath1-band data with the residual psf ellipticities ( but being properly transformed into shear ) measured from the dense star field data , which can be considered as the `` best performance '' of our psf ellipticity correction scheme . it is found that the shape and amplitude of the b - mode shear variance are broadly consistent with those of the residual psf ellipticities . this indicates that most of the sources of residual systematic are understood , which is an important step for cosmic shear statistics to be a practical tool of precision cosmology . however , it is also found that the b - mode shear amplitude at scales @xmath101 arcmin are systematically larger than the residual psf . the reason for this excess is unclear ; one possible reason is thestacking of dithered multiple exposures , which is not involved in the dense star field analysis . such stacking - related issues may be avoided by employing a weak - lensing shape measurement scheme on the basis of individual exposures ( e.g. , miller et al . 2007 ; 2013 ; and miyatake et al . 2013 ) , which combined with a hybrid interpolation scheme will be addressed in a future work . we thank m. takada , m. oguri and h. miyatake for useful discussions . we thank y. utsumi for help with the photometric calibration of suprime - cam data . we also thank l. van waerbeke , h. hoekstra j. d. rhodes and r. mandelbaum for valuable comments on an earlier manuscript , which improved the paper . th greatly thanks y. mellier for many fruitful discussions on various aspects of this work . we would like to thank m. britton for making the arroyo software available . numerical computations in this paper were in part carried out on the general - purpose pc farm at center for computational astrophysics , cfca , of national astronomical observatory of japan . this work is based in part on data collected at subaru telescope and obtained from the smoka , which is operated by the astronomy data center , national astronomical observatory of japan . this work is supported in part by grant - in - aid for scientific research from the jsps promotion of science ( 23540324 ) . . in both cases , the major axis of the ellipticity of the psf are radially oriented with a scaling of @xmath102 . _ bottom two panels _ : spot diagrams for a combination of astigmatism and field curvature : bottom - left panel is for a case for @xmath103 , whereas the bottom - right panel is for @xmath104 . in both cases , the ellipticity parameter is scaled as @xmath105 . [ fig : coma],title="fig:",width=151 ] . in both cases , the major axis of the ellipticity of the psf are radially oriented with a scaling of @xmath102 . _ bottom two panels _ : spot diagrams for a combination of astigmatism and field curvature : bottom - left panel is for a case for @xmath103 , whereas the bottom - right panel is for @xmath104 . in both cases , the ellipticity parameter is scaled as @xmath105 . [ fig : coma],title="fig:",width=151 ] + . in both cases , the major axis of the ellipticity of the psf are radially oriented with a scaling of @xmath102 . _ bottom two panels _ : spot diagrams for a combination of astigmatism and field curvature : bottom - left panel is for a case for @xmath103 , whereas the bottom - right panel is for @xmath104 . in both cases , the ellipticity parameter is scaled as @xmath105 . [ fig : coma],title="fig:",width=151 ] . in both cases , the major axis of the ellipticity of the psf are radially oriented with a scaling of @xmath102 . _ bottom two panels _ : spot diagrams for a combination of astigmatism and field curvature : bottom - left panel is for a case for @xmath103 , whereas the bottom - right panel is for @xmath104 . in both cases , the ellipticity parameter is scaled as @xmath105 . [ fig : coma],title="fig:",width=151 ] + we summarize the properties of ellipticity in the psf caused by third - order optical aberrations with and without rotational symmetry . we consider a system with a circular pupil . we follow the notation by thompson ( 2005 , see fig 1 of his paper ) : @xmath106 represents the position in the image field ; without loss of generality we can choose @xmath107 . @xmath108 represents the position in the exit pupil with @xmath109 . let us start with a rotationally symmetric system . in this case , the wavefront distribution is given by ( thompson 2005 ) @xmath110 among those five seidel aberrations , those that make an ellipticity parameter @xmath111 non - zero are the seidel coma ( the second term ) and astigmatism ( the third term ) . the seidel coma results in an ellipticity whose major axis oriented toward radial direction with @xmath112 where @xmath113 is the distance from the center of the image field . on the other hand , a combination of the astigmatism and the field curvature ( the fourth term ) results in a radially elongated ( if @xmath103 ) or tangentially elongated ( if @xmath104 ) ellipticity parameter with @xmath105 ( see figure [ fig : coma ] ) . the last term , the distortion , does not generate ellipticity for a point source because it only introduces a shifting of the image position on a focal plane . however , for an extended source , its differential effect causes a radially or tangentially elongated ellipticity with @xmath114 . it should , however , be noticed that unlike other aberration terms , the image deformation caused by the distortion is due to mapping , and thus this should be treated separately from the psf . . [ fig : mis],title="fig:",width=151 ] . [ fig : mis],title="fig:",width=151 ] + . [ fig : mis],title="fig:",width=151 ] . [ fig : mis],title="fig:",width=151 ] + next , we consider the case without rotational symmetry , allowing for a tilt and decenter between the axes of the exit pupil and the image field . to compute the wavefront distribution using a perturbative approach , it is customary to introduce a vector @xmath115 with @xmath116 which represents the decentration of the center of the aberration field , @xmath117 , with respect to the unperturbed field center . in this case , optical aberrations result in a psf with non - zero ellipticity parameter are the _ misalignment _ coma and the _ misalignment _ astigmatism ( schechter & levinson 2011 ) . the wavefront distribution for the misalignment coma is given by @xmath118 in this case , the psf has the same coma shape as the seidel coma . however , the psf does not depends on the position in the image field , since the above expression does not include @xmath119 , but the psf ( and thus the ellipticity ) is constant over the image field , as shown in the two top panels of figure [ fig : mis ] . the wavefront distribution for the misalignment astigmatism is given by @xmath120 in this case , the ellipticity of the psfs depends not only on the distance from the center of the image field , but also on the azimuth angle ( we shall denote it by @xmath121 ) , and is written by @xmath122 ( see figure [ fig : mis ] ) . although an actual optical aberration may result in a more complicated psf than that expected from the third - order optical aberrations , it may be reasonable to build a model of the psf ellipticity caused by an optical aberration based on the above results . we consider the following form of a model ( already introduced in equation ( [ eq : psfaberrat ] ) , but rewritten for self - contained ) that consists three terms , namely the constant , axis - symmetric and asymmetric components ; @xmath123 we determine the model parameters @xmath30 , @xmath31 and @xmath124 by the standard least square method with the observed psf ellipticities . in doing so , we take @xmath28 and @xmath29 . as shown in section [ sec : psf ] , this simple model succeeds very well in fitting the global pattern of psf ellipticities in suprime - cam data . we examined the statistical properties of psf ellipticities caused by atmospheric truculence using a numerical simulation . the aim of this study is to investigate the shape and amplitude of the power spectrum of atmospheric psf ellipticities , which is essential knowledge for understanding the statistical properties of psf ellipticities . our approach is similar to one taken by jee & tyson ( 2011 ) , that is to utilize a numerical simulation of wave propagation through atmospheric turbulence for evaluating atmospheric psfs . here , we briefly describe our numerical simulation . since we heavily use publicity open software , arroyo ( britton 2004 ) , we refer the reader to the above reference and software documents for details of models , computational methods and their implementation . the atmospheric turbulence was modeled by eleven - layer frozen kolmogorov screens with the atmospheric model at mauna - kea proposed by ellerbroek & rigaut ( 2000 ) . the strength of the total atmospheric turbulence was set by the fried length , @xmath125 , for which we adopt a value proposed by ellerbroek & rigaut ( 2000 ) : @xmath126 m at a wavelength of @xmath127 m . neither a large scale nor small scale cutoff ( the , so - called , outer and inner scale ) on the kolmogorov power spectrum is imposed . the wind model given by equation ( 3.20 ) of hardy ( 1998 ) was adopted with random wind directions . having the atmospheric model ready , an atmospherically disturbed phase function of a wavefront is computed . we consider a monochromatic electromagnetic wave with a wavelength of @xmath128 m . then , an instantaneous psf was obtained by fourier transforming the phase function within a pupil , for which we assumed a circular pupil with 8.2 m diameter for simplicity . all of the above models and computations were implemented in arroyo subroutines . finally , a sequence of instantaneous psfs were added to obtain a long - exposure psf . we computed psfs of 1 , 10 and 60 seconds exposures on @xmath129 regular grids over a @xmath130 focal plane . we generated 20 realizations . the mean fwhm of psfs among the realizations was 0.65 arcsec for a 60-seconds exposure , which is in a good agreement with the observed value at the subaru telescope ( miyazaki et al . 2002 ) . as an illustrative example , one realization is shown in figure [ fig : emap2x3.run105 ] , where the ripple like feature appears . similar features were observed in cfht data ( heymans et al . 2012 ) and suprime - cam data ( see figures [ fig : emap1 ] and [ fig : emap2 ] ) . it should , however , be noted that the appearance varies widely between realizations , depending on turbulence and wind properties . we measured the power spectrum and the aperture mass variance from the simulation data and present them in figures [ fig : powerspec_arroyo ] and [ fig : mapvar_arroyo ] , respectively . the following three major findings were derived from the figures : first , the amplitude of the power spectrum decreases as @xmath131 , as was already pointed out in previous studies ( wittman 2005 ; de vries et al 2007 ; jee & tyson 20011 ; chang et al . 2012 ; 2013 heymans et al . second , the power spectrum is approximated by a power - law model with a power index of @xmath59 . a plausible reason for the power - law shape is our assumption of the power - law turbulence power spectrum without a cutoff . in reality , a large - scale cutoff ( the , so - called , outer scale ) at around 10@xmath38100 m is expected ( e.g. , hardy 1998 ) , which may result in a cutoff in the psf ellipticity power spectrum at a scale of around 1 degree . third , the atmospheric turbulence results in e / b - mode almost equally partitioned psf ellipticities . through a closer look at figure [ fig : mapvar_arroyo ] , one may however find that the e - mode amplitude is slightly , but systematically , larger than the b - mode . the reason for this is unclear and we leave it for future work . in addition to the above findings , we also find from figures [ fig : powerspec_arroyo ] and [ fig : mapvar_arroyo ] that models @xmath132 and @xmath133 give a good fit to the 60-second result . in what follows , we discuss how those correlation amplitudes depend on other observational parameters , including the exposure time and the atmospheric seeing fwhm ( @xmath134 ) . consider the case @xmath135 , where @xmath136 denotes the diameter of a telescope , and a monochromatic radiation with a wavelength @xmath137 from a point source . the wavefront from a point source is disturbed by atmospheric turbulence , and we consider , for a simple approximation , the disturbed wavefront to be segmented @xmath125-sized patches of constant phase and random phases between the individual patches ( saha 2007 , and see e.g. , figure 5.3 of that textbook for an illustration ) . the number of patches within a pupil is approximately @xmath138 . a wavefront from each patch results in a sharp speckle psf on a focal plane , and a total psf is viewed as a superposition of randomly distributed speckles over an extent of the atmospheric seeing size @xmath139 ( hardy 1998 ; saha 2000 ; see for illustrative examples , figure 3 of kaiser , tonry & luppino 2000 ; and fig 5 of jee & tyson 2011 ) . in this picture , the psf anisotropy is understood as a natural consequence of a finite number of speckles . since the disturbed wavefront changes quickly , more different speckles are accumulated as @xmath140 , and thus the psf becomes rounder . it is found from the above numerical simulation that the rms of psf ellipticities decreases as @xmath141 , which combined with the above mentioned relationships leads to the following relationships ; @xmath142 we tested this relationship against numerical simulations with different values of @xmath125 , from which the above dependences of @xmath125 and @xmath134 on @xmath143 were verified . combined the last relationship with the power - law fitting models , we have @xmath144 and @xmath145 although the above relationships are quite crude , they are useful to quickly evaluate a magnitude of atmospheric psf ellipticities for a given set of observational parameters . note that these relationships are valid only for a telescope with a diameter of 8.2 m . it may be worth pointing out that it is found from numerical simulations with different telescope diameters that the amplitude of psf ellipticities is insensitive to @xmath136 . this differs from a scaling relationship of @xmath146 expected from the above consideration ( likewise for @xmath147 ) . this difference may be explained by the diameter dependence of the psf , which is not taken into account in the above consideration . according to the diffraction theory ( e.g. , hardy 1998 ; saha 2007 ) , the psf is the inverse fourier transform of the optical transfer function ( otf ) which is the auto correlation function of the product of the telescope pupil function with the atmospheric phase function . in that relationship , the diameter of telescope enters through the pupil function . therefore , roughly and intuitively speaking , the power spectrum of psf ellipticities relates to the otf , and the diameter acts as a large - scale cutoff . since the turbulence power spectrum has a larger power on a larger scales ( kolmogorov power spectrum has the shape of @xmath148 ) , the large - scale cutoff imposed by the telescope diameter may have a great impact on the amplitude of the psf ellipticities . lupton r. , gunn j. e. , ivezic z. , knapp g. r. , kent s. , yasuda n. , 2001 , in harnden f. r. jr , primini f. a. , payne h. e. , eds , asp conf . 238 , astronomical data analysis software and systems x. astron . soc . pac . , san francisco , p. 269 massey r. , hoekstra h. , kitching t. , rhodes j. , cropper m. , amiaux j. , harvey d. , mellier y. , meneghetti m. , miller l. , paulin - henriksson s. , pires s. , scaramella r. , schrabback t. , 2013 , mnras , 429 , 661

the most compelling evidences for the existence of stellar - mass black holes have been obtained through observations of x - ray binary systems . the application of classical methods and the development of new techniques have allowed us to increase the number of stellar - mass black holes known . i summarize here the observational signatures of the black holes , such as the mass determination , the event horizon and the spin . i also present some observational results on the galactic centre black hole . 1.0 cm

0907.3602

the contribution presented here is part of a course on compact objects and their emission given in the first la plata international school on astronomy and geophysics . the contribution is focused on the observational evidences of stellar mass black holes in our galaxy and the super - massive black hole in its centre . the four lectures that were given about the black holes in the galaxy have been distributed here in five sections . in section 2 a brief account about how a neutron star or a black hole are formed is presented , a brief black hole history is also introduced and some concepts such as the schwarzchild radius , the event horizon and kerr black hole are given . as the x - ray binaries play a fundamental role for the study of stellar mass black holes , section 3 is devoted to introduce the x - ray binaries and explain some of their most important characteristics . in section 4 , the first black hole candidates are introduced , how were they discovered and how many of them are currently known . section 5 is devoted specifically to introduce the observational signatures of the black holes . in particular , some methods to estimate the mass , to find possible evidences for the event horizon and to measure the black hole spin are explained . finally , section 6 is devoted to the super - massive black hole in the centre of our galaxy . during their life stars evolve through different states that are linked to the nuclear burning . when the core of the star runs out of nuclear fuel , it collapses until some other form of pressure support enables a new equilibrium configuration to be attained . the possible equilibrium configurations which can exist when the star collapses are : white dwarfs ( wd ) , neutron stars ( ns ) , black holes ( bh ) . according the initial star mass ( @xmath0 ) , there will be different types of collapse and different supernova explosions . when @xmath0 < 8m@xmath1 , there is no sn explosion , only contraction , and a wd is formed . if the wd accretes mass it will explode as a type ia sn . > 8m@xmath1 , thermonuclear sn explosions of different types are produced ( types ib , ic and no h lines ; types ii and h lines ) . as a result , a ns or a bh is formed . in wd and ns , the pressure which holds up the star is the quantum mechanical pressure associated with the fact that electrons , protons and neutrons are fermions ( only one particle is allowed to occupy any one quantum mechanical state ) . wd are held up by electron degeneracy pressure , and their mass is below the chandrasekhar limiting mass @xmath2 = 1.46m@xmath1 . with increasing density , the degenerate electron gas becomes relativistic and , when the total energy of the electron exceeds the mass difference between the neutron and the proton , the inverse @xmath3-decay process can convert protons into neutrons . it is the degeneracy pressure of this neutron gas which prevents collapse under gravity and results in the formation of a ns . the same physics for the wd is responsible for providing pressure support for the ns , the only difference being that the neutrons are about 2000 times more massive than the electrons , and consequently degeneracy sets in at a correspondingly higher density . ns , in which neutron degeneracy pressure is responsible for the pressure support , can have a mass given by the tolman - oppenheimer - volkoff mass limit . this mass is not analytically well fixed because of its dependence on the equations of state for nuclear matter . in any case , most equations of state do not allow the neutron degeneracy pressure to support more than 3m@xmath1 , indicating that dead stars more massive than 3m@xmath1must be bhs . figure [ mass ] shows an scheme of the relation between the original mass and the final mass of a dead star . john michell first described the concept of a black hole in a paper that appeared in philosophical transactions of the royal society of london in 1783 . michell discussed the possibility of a star with a gravitational force so strong that not even light could escape from it , thus preventing astronomers from observing such phenomena . pierre - simon laplace mentioned the possibility of such stars in the first few editions of his book exposition du systme du monde ( 1796 ) , although he failed to include it in later editions because the idea of a star with a gravitational force that could overpower light did not comply with the wave theory of light , which was generally accepted at that time . this theory seemed to suggest that light could not be affected by gravity , a concept that is key to the theories of michell and laplace . in the early 1800 s experiments on optical interference led to the predominance of the wave theory of light and the end of the corpuscular theory . since light waves were thought to be unaffected by gravitation , interest in the hypothetical `` dark stars '' ceased . in 1905 albert einstein published his special theory of relativity and in 1915 his general theory of relativity ( gr ) . the gr was a new theory of gravitation and one of its fundamental predictions was the effect of gravity on light . matter causes space - time to curve and therefore the paths followed by light rays or matter are determined by the curvature of the space - time . this allowed for a modern scientific proof of michell s hypothesis . a short time after einstein published his gr , the german physicist karl schwarzchild wrote a paper describing a structure called a singularity . he argued that matter could theoretically be drawn into a point with virtually no volume and a infinite density . he called this object a point mass , later dubbed a singularity . in addition , schwarzchild determined that there is a definite boundary around a singularity called the event horizon . in 1928 , through his research on wd , subrahmanyan chandrasekhar hypothesized that a dying star of a certain mass might form a point with enough gravitational pull to trap light . in 1939 , american physicist robert oppenheimer developed a possible explanation for the nature of these points of infinite density . oppenheimer theorized that the gravitational pull of a star with infinite density would cause light rays to deviate from their path and bend towards the star . eventually , the gravity of the star would become so great that the light would become trapped by it and would be unable to escape , preventing one from observing it . at this point the star is a bh . the american scientist john wheeler coined the term `` black hole '' in 1969 , although from a non astrophysical point of view , the expression `` black hole '' was coined before in calcutta to describe a small dungeon where troops of the nawab of bengal held british prisoners of war after the capture of fort william on june 20 , 1756 . performing a classical calculation , the escape velocity from the surface of a star of mass @xmath4 and radius @xmath5 is @xmath6 and setting @xmath7 , the radius of such star would be @xmath8 . as we will see later , this is just the expression for the schwarzchild radius of a bh of mass @xmath4 , and the spherical surface of radius @xmath5 plays the role of the surface of the bh . the escape velocity from the earth is 11 km s@xmath9 , from the sun is 617 km s@xmath9 , from a ns ( 2m@xmath1and radius 10 km ) is 230000 km s@xmath9 and from a bh should be greater than @xmath10 . some months after the einstein s definitive formulation of the gr in 1915 , schwarzschild discovered the solution for the metric of space - time about a point mass @xmath4 : @xmath11\ ] ] this metric , known as the schwarzschild metric , has the same meaning as the interval d@xmath12 in special relativity , which is known as the minkowski metric : @xmath13\ ] ] in the limit of large distances from the point mass , @xmath14 , the two metrics become the same . they are , however , very different for small values of @xmath5 , reflecting the influence of the mass @xmath4 upon the geometry of space - time . in the case of the schwarzschild metric , the time interval between events according to an observer who is stationary in s and located at the point @xmath5 is d@xmath15d@xmath16 . this notation makes it clear how the time interval d@xmath17 depends upon the gravitational potential in which the observer is located . d@xmath17 only reduces to d@xmath18 in the limit of very large distances from the origin , @xmath14 , at which the gravitational potential goes to zero . _ redshift of electromagnetic waves_. if the time interval @xmath19 corresponds to the period of the waves emitted at the point @xmath5 , the observed period of the wave at infinity @xmath20 is given by @xmath21 therefore , the emitted and observed frequencies , @xmath22 and @xmath23 , respectively , are related by @xmath24 this expression is the general relativistic result corresponding to michell s insight . if the radiation is emitted from radial coordinate @xmath25 , the frequency of any wave is redshifted to zero frequency , and no information can reach infinity from radii @xmath26 . here @xmath27 is the _ schwarzschild radius_. since , according to gr , no radiation can escape from within this radius , the surface @xmath28 is black . the schwarzschild radius defines the _ event horizon_. not even light can escape , once it has crossed the event horizon . it can be written as @xmath29 km . another important aspect of the schwarzschild metric is the dynamics of test masses in the gravitational field of the point mass . in a newtonian treatment , from the conservation of energy we obtain @xmath30 and from the conservation of angular momentum , @xmath31 . introducing the specific angular momentum of the particle ( angular momentum per unit mass ) @xmath32 , we can write @xmath33 as long as @xmath34 , the particle can not reach @xmath35 because the energy term associated with the centrifugal force , @xmath36 , becomes greater than the gravitational potential energy @xmath37 for small enough values of @xmath5 . in a general relativistic treatment the dynamics of the test mass is given by @xmath38 there are two main differences when comparing this relativistic equation with the newtonian equation . there is an extra term and constants and variables have different meaning in gr ( @xmath5 angular diameter distance , @xmath39 means d@xmath5/d@xmath40 , where @xmath40 is proper time , @xmath41 and @xmath42 are constants equivalent to @xmath41 and @xmath43 ) . the most important difference is the additional term @xmath44 , which has the effect of enhancing the attractive force of gravity , even when the particle has a finite specific angular momentum @xmath41 ( the greater the value of @xmath41 , the greater the enhancement ) . this result can be understood by recalling that the kinetic energy associated with the rotational motion around the point mass contributes to the inertial mass of a test particle and thus enhances the gravitational force upon it . from the analysis of the relativistic equation , some interesting results can be found : + 1 ) for sufficiently small values of @xmath5 , the general relativistic term@xmath44 becomes greater than the centrifugal potential term , implying that this purely general relativistic term increases the strength of gravity close enough to @xmath35 . it can be shown easily that if the specific angular momentum of the particle @xmath45 it will inevitably fall in to @xmath35 . + 2 ) there is a last stable circular orbit , of radius @xmath46 , around the point mass . there are no stable circular orbits with smaller radii than this value because the particles would spiral rapidly into @xmath35 . this is why the bh is called hole , matter inevitably collapses in to @xmath35 if it comes too close to the point mass . + 3 ) from the metric given in eq . [ eq : metric ] , it appears that there is a singularity at @xmath27 . it can be shown that this is not a physical singularity . however , at @xmath35 , there is a real physical singularity , and according to the gr , the infalling matter collapses to a singular point.however , these schwarzschild singularities are unobservable because no information can arrive to the external observer from within @xmath27 . for all practical purposes , the bh may be considered to have a black spherical surface at @xmath27 . from the classical point of view , physics breaks down at @xmath35 . _ kerr black holes_. in 1962 , kerr discovered the general solution for a bh with angular momentum @xmath47 . it has been shown that isolated bhs can be completely characterized by only three properties : mass @xmath4 , charge @xmath48 and angular momentum @xmath47 . the rotating bhs ( kerr bhs ) are relevant to many aspects of high energy astrophysics . the kerr metric in boyer - landquist coordinates is given by : @xmath49\end{aligned}\ ] ] where @xmath50 is the angular momentum of the bh per unit mass , @xmath51 and @xmath52 . if the bh is non - rotating , @xmath53 and the kerr metric reduces to the standard schwarzschild metric . just as in the case of schwarschild metric , the metric coefficient of d@xmath54 becomes singular at a certain radial distance , when @xmath55 . this radius correspond to the surface of infinite redshift or the horizon of the rotating bh , and is given by the solution of @xmath56 . taking the larger of the two roots , the horizon occurs at radius @xmath57^{\frac{1}{2}}\ ] ] this spherical surface has exactly the same properties as the schwarschild radius in the case of non - rotating bhs . particles and photons can fall in through this radius , but they can not emerge outwards according to the classical theory of gr . if the system has too much angular momentum j , no bh will be formed . the maximum angular momentum corresponds to @xmath58 . for a maximally rotating bh , the horizon radius is @xmath59 , just half the result for the case of schwarschild bh , @xmath60 . for maximally rotating bhs the last stable circular orbit is located at ( shapiro & teukolsky 1983 ) @xmath61 for corotating test particles and at @xmath62 for counter - rotating particles . ( recall : for non - rotating bhs , the last stable orbit is at @xmath63 ) . correspondingly , the maximum binding energies of these orbits can be found , that is , the amount of energy that has to be lost in order for the material to attain a bound stable circular orbit with radius r. in the corotating case , 42.3% of the rest mass energy of the material can be released as it spirals into the bh through a sequence of almost circular equatorial orbits . in the counter - rotating case it is 3.8% . this is the process by which energy is liberated through the accretion of matter onto bhs , and is likely to be the source of energy in some of the most extreme astrophysical objects . the energy available is much greater than that attainable from nuclear fusion processes , which at most can release about 1% of the rest mass energy of matter . a more complete treatment of the topics developed in this subsection can be found in longair ( 1994a , 1994b ) . an x - ray binary is a binary system containing a compact object , either a neutron star or a stellar mass black hole , that emits x - rays as a result of a process of accretion of matter from the companion star . several scenarios have been proposed to explain this x - ray emission , depending on the nature of the compact object , its magnetic field in the case of a neutron star , and the geometry of the accretion flow . the accreted matter is accelerated to relativistic speeds , transforming its potential energy provided by the intense gravitational field of the compact object into kinetic energy . assuming that this kinetic energy is finally radiated , the accretion luminosity can be computed , finding that this mechanism provides a very efficient source of energy , even much higher efficiency than that for nuclear reactions . on its way to the compact object , the accreted matter carries angular momentum and usually forms an accretion disk around it . the matter in the disc looses angular momentum due to viscous dissipation , which produces a heating of the disc , and falls towards the compact object in a spiral trajectory . the black body temperature of the last stable orbit in the case of a bh accreting at the eddington limit is given by : @xmath64 where @xmath65 is expressed in kelvin and @xmath4 in m@xmath1 ( rees 1984 ) . for a compact object of a few solar masses , @xmath66 k. at this temperature the energy is mainly radiated in the x - ray domain . in high mass x - ray binaries ( hmxbs ) the donor star is an o or b early type star of mass in the range @xmath67@xmath68 m@xmath1 and typical orbital periods of several days . hmxbs are conventionally divided into two subgroups : systems containing a b star with emission lines ( be stars ) , and systems containing a supergiant ( sg ) o or b star . in the first case , the be stars do not fill their roche lobe , and accretion onto the compact object is produced via mass transfer through a decretion disc . most of these systems are transient x - ray sources during periastron passage . in the second case , ob sg stars , the mass transfer is due to a strong stellar wind and/or to roche lobe overflow . the x - ray emission is persistent , and large variability is common . the most recent catalogue of hmxbs was compiled by liu et al . ( 2006 ) , and contains 242 sources . in low mass x - ray binaries ( lmxbs ) the donor has a spectral type later than b , and a mass @xmath69 m@xmath1 . although it is typically a non - degenerated star , there are some examples where the donor is a wd . the orbital periods are in the range 0.2400 hours , with typical values @xmath70 hours . the orbits are usually circular , and mass transfer is due to roche lobe overflow . most lmxbs are transients , probably as a result of an instability in the accretion disc or a mass ejection episode from the companion . the typical ratio between x - ray to optical luminosity is in the range @xmath71@xmath72 , and the optical emission is dominated by x - ray heating of the accretion disc and the companion star . some lmxbs are classified as ` z ' and ` atoll ' sources , according to the pattern traced out in the x - ray color - color diagram . ` z ' sources are thought to be weak magnetic field neutron stars of the order of @xmath73 g with accretion rates around 0.51.0 @xmath74 . ` atoll ' sources are believed to have even weaker magnetic fields of @xmath75 g and lower accretion rates of 0.010.1 @xmath74 . the most recent catalogue of lmxbs was compiled by liu et al . ( 2007 ) , and contains 188 sources . recently , grimm et al . ( 2002 ) estimated that the total number of x - ray binaries in the galaxy brighter than 2@xmath76 erg s@xmath9 is about 705 , being distributed as @xmath77325 lmxbs and @xmath77380 hmxbs . the first x - ray binary known to display radio emission was sco x-1 in the late 1960s . since then , many x - ray binaries have been detected at radio wavelengths with flux densities @xmath78@xmath79 mjy . the flux densities detected are produced in small angular scales , which rules out a thermal emission mechanism . the most efficient known mechanism for production of intense radio emission from astronomical sources is the synchrotron emission mechanism , in which highly relativistic electrons interacting with magnetic fields produce intense radio emission that tends to be linearly polarized . the observed radio emission can be explained by assuming a spatial distribution of non - thermal relativistic electrons , usually with a power - law energy distribution , interacting with magnetic fields . since some rexbs , like ss 433 , were found to display elongated or jet - like features , as in active galactic nuclei ( agn ) and quasars , it was proposed that flows of relativistic electrons were ejected perpendicular to the accretion disc , and were responsible for synchrotron radio emission in the presence of a magnetic field . models of adiabatically expanding synchrotron radiation - emitting conical jets may explain some of the characteristics of radio emission from x - ray binaries ( hjellming & johnston 1988 ) . several models have been proposed for the formation and collimation of the jets , including the presence of an accretion disc close to the compact object , a magnetic field in the accretion disc , or a high spin for the compact object . however , there is no clear agreement on what mechanism is exactly at work . there are eight radio emitting hmxbs and 35 radio emitting lmxbs . since the strong magnetic field of the x - ray pulsars disrupts the accretion disc at several thousand kilometers from the neutron star , there is no inner accretion disc to launch a jet and no synchrotron radio emission has ever been detected in any of these sources . although the division of x - ray binaries in hmxbs and lmxbs is useful for the study of binary evolution , it is probably not important for the study of the radio emission in these systems , where the only important aspect seems to be the presence of an inner accretion disc capable of producing radio jets . however , the eight radio emitting hmxbs include six persistent and two transient sources , while among the 35 radio emitting lmxbs we find 11 persistent and 24 transient sources . the difference between the persistent and transient behavior clearly depends on the mass of the donor . excluding x - ray pulsars , @xmath68 to @xmath80 of the catalogued galactic x - ray binaries have been detected at radio wavelengths regardless of the nature of the donor . the corresponding ratio of detected / observed sources is probably much higher . however , it is difficult to give reliable numbers , since observational constrains arise when considering transient sources observed in the past ( large x - ray error boxes , single dish and/or poor sensitivity radio observations , etc . ) , and likely many non - detections have not been published . a microquasar is a radio emitting x - ray binary displaying relativistic radio jets . the name was given not only because of the observed morphological similarities between these sources and the distant quasars but also because of physical similarities , since when the compact object is a black hole , some parameters scale with the mass of the central object ( mirabel & rodrguez 1999 ) . a schematic illustration comparing some parameters in quasars and microquasars is shown in figure [ qmq ] . from eq . [ eq : tlast ] , a typical temperature of the accretion disc of a microquasar containing a stellar mass black hole is @xmath81 k , while that of a quasar containing a super - massive black hole ( @xmath82@xmath83 @xmath84 ) is @xmath85 k. this explains why in microquasars the accretion luminosity is radiated in x - rays , while in quasars it is radiated in the optical / uv domain . the characteristic jet sizes appear to be proportional to the mass of the black hole . radio jets in microquasars have typical sizes of a few light years , while in quasars may reach distances of up to several million light years . the timescales are also directly scaled with the mass of the black hole following @xmath86 . therefore , phenomena that take place in timescales of years in quasars can be studied in minutes in microquasars . thus , microquasars mimic , on smaller scales , many of the phenomena seen in agns and quasars , but allow a better and faster progress in the understanding of the accretion / ejection processes that take place near compact objects . the current number of confirmed microquasars is @xmath7716 among the 43 catalogued rexbs ( paredes 2005 ) . some authors ( fender 2001 ) have proposed that all rexbs are microquasars , and would be detected as such provided that there is enough sensitivity and/or resolution in the radio observations . the theoretical models attempting to understand the jet formation and its connection with the accretion disc had a seminal contribution in the works by blandford & payne ( 1982 ) . these authors explored the possibility of extracting energy and angular momentum from the accretion disc by means of a magnetic field whose lines extend towards large distances from the disc surface . their main result was the confirmation of the theoretical possibility to generate a flow of matter outwards from the disc itself , provided that the angle between the disc and the field lines was smaller than 60@xmath87 . later on , the flow of matter is collimated at large distances from the disc by the action of a toroidal component of the magnetic field . in this way , two opposite jets could be formed flowing away perpendicularly to the accretion disc plane . to confirm observationally the link between accretion disc and the genesis of the jets is by no means an easy task . the collimated ejections in grs 1915 + 105 provide one of the best studied cases supporting the proposed disc / jet connection . in figure [ grs1915 ] , from mirabel et al . ( 1998 ) , simultaneous observations are presented at radio , infrared and x - ray wavelengths . the data show the development of a radio outburst , with a peak flux density of about 50 mjy , as a result of a bipolar ejection of plasma clouds . however , preceding the radio outburst there was a clear precursor outburst in the infrared . the simplest interpretation is that both flaring episodes , radio and infrared , were due to synchrotron radiation generated by the same relativistic electrons of the ejected plasma . the adiabatic expansion of plasma clouds in the jets causes an energy loss in the electron population and , as a result , the spectral maximum of their synchrotron radiation is progressively shifted from the infrared to the radio domain . it is also important to note the behaviour of the x - ray emission in figure [ grs1915 ] . the emergence of jet plasma clouds , that produce the infrared and radio flares , seems to be accompanied by a sharp decay and hardening of the x - ray emission ( 8.088.23 h ut in the figure ) . the x - ray fading is interpreted as the disappearance , or emptying , of the inner regions of the accretion disc ( belloni et al . part of the matter content in the disc is then ejected into the jets , perpendicularly to the disc , while the rest is finally captured by the central black hole . additionally , mirabel et al . ( 1998 ) suggest that the initial time of the ejection coincides with the isolated x - ray spike just when the hardness ratio suddenly declines ( 8.23 h ut ) . the recovery of the x - ray emission level at this point is interpreted as the progressive refilling of the inner accretion disc with a new supply of matter until reaching the last stable orbit around the black hole . this behaviour in the light curves of grs 1915 + 105 has been repeatedly observed by different authors ( e.g. fender et al . 1997 ; eikenberry et al . 1998 ) , providing thus a solid proof of the so - called disc / jet symbiosis in accretion discs . all the observed events were shorter than 1/2 h , and their equivalent in quasars , or agns , would require a much longer minimum time span of some few years . despite the complexity in the grs 1915 + 105 light curves , the episodes of x - ray emission decay with associated hardening are reminiscent of the well known low / hard state typical of persistent black hole candidates ( cygnus x-1 , 1e 1740.7@xmath882942 , grs 1758@xmath88258 and gx 339@xmath884 ) . the transitions towards this state are often accompanied by radio emission with flat spectrum , interpreted as due to the continuous creation of partially self - absorbed compact synchrotron jets . it is worth mentioning the observational work by marscher et al . ( 2002 ) , who presented evidence of the disc / jet symbiosis also in the active galaxy 3c 120 . using vlbi techniques they observed episodes of ejection of superluminal plasma just after the decay and hardening of the x - ray emission . this is precisely the same behaviour displayed by grs 1915 + 105 . the events in 3c 120 seem to be recurrent with an interval of one year , which is consistent with a mass of the compact object of @xmath89 m@xmath1 . such observations strongly support the idea of continuity between galactic microquasars and agns in the universe . one of the first optical counterparts to be identified was the 9th magnitude supergiant star hd 226868 , associated with the hmxb cyg x-1 . it showed radial velocity variations which made it a prime candidate for a stellar mass bh ( webster & murdin 1972 , bolton 1972 ) . the supergiant star was shown to move with a velocity amplitude of 64 km s@xmath9(later refined to 75 km s@xmath9 ) in a 5.6 day orbit due to the gravitational influence of an unseen companion ( see fig . [ cygx1 ] ) . the mass function of the compact object was @xmath90 ) = 0.25 m@xmath1 . the mass of the bright companion @xmath91 is large and has a wide range of uncertainty . if the optical star were a normal o9.7iab its mass would be 33 m@xmath1 which , for an edge - on orbit ( @xmath92 = 90@xmath87 ) would imply a compact object of @xmath77 7 m@xmath1 . however , the optical star is likely to be undermassive for its spectral type as a result of mass transfer and binary evolution , as has been shown to be the case in several ns binaries ( e.g. rappaport & joss 1983 ) . it could be undermassive by as much as a factor of 3 given the uncertainty in distance , log @xmath93 and @xmath94 . a plausible lower limit of 10 m@xmath1 , combined with an upper limit to the inclination of 60@xmath87 , based on the absence of x - ray eclipses , leads to a compact object of @xmath95 4 m@xmath1(bolton 1975 ) . in 1975 , the x - ray satellite ariel v detected a0620 - 00 . this source is a x - ray transient ( xrt ) , a subclass of lmxbs which undergo dramatic episodes of enhanced mass - transfer or `` outbursts '' triggered by viscous - thermal instabilities in the disc . during outburst , the companion remains undetected because it is totally overwhelmed by the intense optical light from the x - ray heated disc . after a few months of activity the x - rays switch off , the reprocessed flux drops several magnitudes into quiescence and the companion star becomes the dominant source of optical light . this offers the opportunity to perform radial velocity studies of the cool companion and unveil the nature of the compact star . the first detection of the companion in a0620 - 00 revealed a mid - k star moving in a 7.8 hr period with velocity amplitude of 457 km s@xmath9 . the implied mass function was 3.2 @xmath96 0.2 m@xmath1 , the largest ever measured ( mcclintock & remillard 1986 ) . a lower limit to the mass of the compact star of 3.2 m@xmath1was established by assuming a very conservative low - mass companion of 0.25 m@xmath1and @xmath92 @xmath97 85@xmath87 , based on the lack of x - ray eclipses . this exceeds the maximum mass allowed for a stable ns and hence it also became a very compelling case for a bh . in the 1980s there was a hot debate about the real existence of bhs . there were three firm candidates , two hmxbs ( cyg x-1 and lmc x-3 ) and a transient lmxb ( a0620 - 00 ) , all with lower limits to @xmath98 very close to the maximum mass for ns stability . alternative scenarios were proposed to avoid the need for bhs such as multiple star systems ( bahcall et al . 1974 ) or non - standard models invoking exotic equations of state ( eos ) for condensed matter . it was proposed that the `` holy grail in the search for bhs is a system with a mass function that is plainly 5 m@xmath1 or greater '' ( mcclintock 1986 ) . in 1989 , ginga discovered a new xrt in outburst named gs2023 + 338 (= v404 cyg ) . its x - ray properties drew considerable attention because of the exhibition of a possible luminosity saturation at l@xmath99 = @xmath100 erg s@xmath9 and dramatic variability ( zycki et al . spectroscopic analysis during quiescence revealed a k0 star moving with a velocity amplitude of 211 km s@xmath9 in a 6.5 day orbit ( see fig . [ v404 ] ) . the obtained mass function , @xmath90 ) = 6.08 @xmath96 0.06 m@xmath1 , implies the compact object must be more massive than 6 m@xmath1(casares et al . 1992 ) . the mass function of v404 cyg was the highest yet measured and , very clearly , placed an accreting compact object comfortably above the upper limit of maximally rotating nss for any `` standard '' eos assumed . this work on v404 cygni revolutionized the field of bh searches and this source is widely considered as the best evidence for a bh , where no additional assumptions on @xmath92 nor @xmath101 have to be invoked . this remarkable result obtained by casares and collaborators established v404 cyg as the `` holy grail '' bh for almost a decade . since then , many other bhs have been unveiled through dynamical studies of xrts in quiescence , some of them having mass functions also in excess of 5m@xmath1 . this has been possible thanks to the improvement of the instrumentation and the new generation of 10-m class telescopes . the list of the confirmed stellar bhs has increased recently with the addition of two stellar bhs . one of them , m33 x-7 , is a 15.65 @xmath96 1.45 m@xmath1 bh in an eclipsing binary in the nearby spiral galaxy m33 ( orosz et al . the other is ic 10 x-1 , located in the local group starburst galaxy ic 10 , and with a mass of 32.7 @xmath96 2.6 m@xmath1 , being indeed the most massive known stellar mass bh ( silverman & filippenko 2008 ) . in table [ tabbh ] is presented a compilation of the all confirmed stellar bhs made by casares ( 2007 ) , with the addition of these two new bhs . .confirmed black holes and mass determinations . table from casares ( 2007 ) updated with m33 x-7 and ic 10 x-1 . [ cols="<,^,^,^,^,^ " , ] @xmath102 orosz et al . + @xmath103 silverman & filippenko ( 2008 ) . a complete determination of the component masses in a binary system requires the radial velocity curves of both stars and a knowledge of the inclination angle . the determination of masses in lmxbs is usually reduced to a single line spectroscopic binary problem where all the information has to be extracted from the optical star . a complete solution to the system parameters can be obtained from three observational experiments involving high resolution optical spectroscopy and infrared photometry ( the mass function , rotational broadening and ellipsoidal modulation ) . this is a fundamental equation for the determination of binary system parameters which can be derived directly from kepler s 2nd and 3rd laws . it relates the masses of the two stars @xmath104 and the inclination angle ( @xmath92 ) through two observable quantities which are readily measured from the radial velocity curve of the optical star : the orbital period @xmath105 and the radial velocity semi - amplitude @xmath106 . @xmath107 where @xmath108 is the mass ratio . @xmath109 represents an absolute lower limit to @xmath110 , i.e. at the extreme case when the mass of the companion star is neglected ( i.e. @xmath111 ) and the binary is seen edge on ( @xmath112 . however , @xmath113 implies thar the mass function is extremely sensitive to uncertainties in the radial velocity amplitude . any non - uniform brightness distribution across the surface of the companion star will modify the radial velocity curve and affect our system parameter determination . a source of systematic error in the determination of @xmath106 is produced by the x - ray irradiation , which tends to suppress the absorption lines near the l1 point . the consequence of the x - ray irradiation is the displacement of the centre of light away from the companion s centre of mass and the radial velocity curve will be significantly eccentric . a sine - wave fit to it will give an observed @xmath114 which will be larger than the true @xmath106 . irradiation is considered to be negligible in quiescent x - ray transients ( where @xmath115 erg s@xmath116 but it can be important in x - ray active states . an example of the importance of this effect was given by gro j1655@xmath8840 ( = n sco 94 ) . during the decay of the 1994 outburst ( @xmath117 erg s@xmath9 ) the radial velocity curve was fitted with a simple sine - wave , obtaining @xmath118 km s@xmath119 . this , combined with @xmath120 d gave @xmath121 0.09m@xmath1 and hence was a strong case for a bh ( orosz & bailyn 1997 ) . the same data was subsequently fitted by other authors using an irradiation model , obtaining @xmath122 192 214 km s@xmath9 which reduces the mass function to @xmath123 1.93 2.67 m@xmath1 ( phillips et al . this latter result was finally confirmed by observations in true quiescence which give @xmath124 km s@xmath9 and @xmath125 0.09m@xmath1 ( shahbaz et al . the new mass function is 16% lower than the first reported value , enough to disclaim gro j1655 - 40 as a secure bh candidate . the companion stars transfer matter onto their compact objects and hence they must be filling their roche lobes . in addition , the short orbital periods ( @xmath77 hr ) and old ages ( > 10@xmath126 yr ) of these close binaries suggest that the companion stars must be synchronised , i.e. @xmath127 where @xmath128 and @xmath129 are the stellar and orbital angular velocities respectively . the projected linear velocities would then be @xmath130 where @xmath131 , @xmath132 is the equivalent radius of the companion s roche lobe and @xmath133 is the binary separation . combining with @xmath134 one obtains @xmath135 . therefore , the mass ratio @xmath136 can be measured directly from the radial velocity curve and the observed rotational broadening ( @xmath137 ) of the secondary s absorption lines ( gies & bolton 1986 ) . the rotational velocity is determined by comparing our target with broadened versions of spectral type templates ( e.g. through a @xmath138 minimization technique ) ( marsh et al . an example of rotational broadening analysis , applied to v404 cygni , can be found in casares & charles ( 1994 ) . the lightcurves of companion stars in contact binaries display the characteristic double - humped variation on the orbital period , with amplitudes @xmath139 . this modulation is a consequence of the tidal distortion of the secondary star and the non - uniform distribution of the surface brightness , due to a combination of limb and gravity darkening . the dependence of @xmath94 on the local surface gravity @xmath93 implies that the inner hemisphere of the secondary ( facing the compact object ) will be cooler and , therefore , the phase 0.5 minimum becomes deeper than the minimum at phase 0 . model calculations show that the shape and amplitude of the ellipsoidal modulation are functions of @xmath136 and @xmath92 . in particular , the amplitude is a strong ( increasing ) function of @xmath92 , and is insensitive to @xmath136 if @xmath140 . this is normally the case in sxts and therefore model fits to the ellipsoidal modulation can be used to determine @xmath92 directly . there are 20 xrbs that lack radial velocity data . most of them even lack an optical counterpart , and only three have known orbital periods . nevertheless , they are considered black - hole candidates because they closely resemble black hole binaries in their x - ray spectral and temporal behavior ( remillard & mcclintock 2006 ) . unfortunately , they have never been seen in quiescence , or they simply become too faint for an optical detection of the companion star . fortunately , a new technique to extract dynamical information during their x - ray active states has been applied recently . it utilises narrow high - excitation emission lines powered by irradiation on the companion star , in particular the strong ciii and fluorescence niii lines from the bowen blend at @xmath141463040 . this technique was first applied to the ns lmxb sco x-1 and the doppler shift of the ciii / niii lines enabled the motion of the donor star to be traced for the first time ( steeghs & casares 2002 ) . this was also attempted during the 2002 outburst of the bh candidate gx 339 - 4 , using high - resolution spectroscopy to resolve the sharp niii / ciii lines . the orbital solution yields a velocity semi - amplitude of 317 km s@xmath9 which defines a strict lower limit to the velocity amplitude of the companion star because these lines arise from the irradiated hemisphere and not the centre of mass of the donor . a solid lower bound to the mass function is 5.8 m@xmath1 which provides compelling evidence for a bh in gx 339@xmath884 ( hynes et al . this technique opens an avenue to extract dynamical information from new xrts in outburst and x - ray persistent lmxbs , which hopefully will help increase the number of bh discoveries . in order to show that a dynamical bh candidate ( i.e. , a massive compact object ) is a genuine bh , one would hope to demonstrate that the candidate has an event horizon - the defining characteristic of a bh . the first test for the presence of the event horizon was carried out by narayan et al . ( 1997 ) using archival x - ray data . one had to wait for chandra before one could claim definitive evidence . strong evidence has been obtained for the reality of the event horizon from observations that compare bhbs with very similar ns binaries . ns binaries show signatures of the hard surface of a ns that are absent for the bh systems . + bh systems are about 100 times fainter than the nearly identical ns binaries . in figure [ chandra ] , there are plotted the eddington - scaled luminosities of bh ( circles ) and ns ( stars ) x - ray transients in quiescence ( mcclintock et al . 2007 ) . at every orbital period , bh candidates have eddington - scaled luminosities lower than ns by two to three orders of magnitude . the orbital period is a predictor of @xmath142 , according to binary mass transfer models ( menou et al . 1999 ) . by comparing quiescent bhs and nss with similar orbital periods , the uncertainty in the mass accretion rate is eliminated ( garcia et al . 2001 ) . the explanation for this result is that accretion in these quiescent systems occurs via a radiatively inefficient mode ( advection - dominated accretion ) , as confirmed through spectral studies ( e.g , narayan et al . 2002 ) . therefore , the disc luminosity @xmath143 and most of the binding energy that is released as gas falls into the potential well is retained in the gas as thermal energy . in the case of an accreting ns , this energy is eventually radiated from the stellar surface , and so an external observer still sees the full accretion luminosity @xmath144 . in the case of a bh , however , the superheated gas falls through the event horizon , carrying all its thermal energy with it . the observer therefore sees only the disc luminosity @xmath145 which is extremely small . bhs lack a soft thermal component of emission that is very prevalent in the spectra of nss and can be ascribed to surface emission ( mcclintock et al . 2004 ) . this has been quantified using observations of dynamical bhs over 9 years of rxte data . the probability that the non - detection of bursts were consistent with a solid surface is found to be @xmath146 ( remillard et al . 2006 ) the study of the fourier power spectra at high frequencies offers a way to distinguish a ns from a bh ( sunyaev & revnivtsev 2000 ) . after the analysis of the power density spectra ( pds ) of a sample of 9 ns and 9 bh binaries in the low / hard spectral state , it is concluded that in the accreting ns with a weak magnetic field , significant power is contained at frequencies close to one khz . in the case of the accreting bh , the pds show strong decline at frequencies higher than 10 50 hz . these empirical phenomenology could be used to help to distinguish the accreting ns from bh in x - ray transients . the xrt with noise in their x - ray flux at frequencies higher than 500 hz should be considered ns . sunyaev & revnivtsev propose to explain the observed difference as a result of the existence of a radiation dominated spreading layer on the ns surface . the classic colour - colour diagram of xrbs at high accretion rates shows a clear separation in the evolution of ns and bh binaries . this has been ascribed to the presence of a boundary layer in ns which gives rise to an additional thermal component in the spectrum and drags ns outside the bh region ( done & gierlinski 2003 ) . + all approaches to this subject can provide only indirect evidence for the event horizon because it is quite impossible to detect directly any radiation from this immaterial surface of infinite redshift . nevertheless , barring appeals to very exotic physics , the body of evidence just considered makes a strong case that dynamical bh candidates possess an event horizon . an astrophysical bh is described by two parameters , its mass @xmath4 and its dimensionless spin parameter @xmath147 . because the masses of 21 bhs have already been measured or constrained , the next obvious goal is to measure spin . several methods have been used to measure the bh spin . the first method , based on x - ray polarimetry , appears very promising but thus far has not been incorporated into any contemporary x - ray mission . the second method , based on x - ray continuum fitting , is already producing useful results . the third method , based on the fe k line profile , has also yielded results , although the method is hampered by significant uncertainties . the fourth method , based on high - frequency qpos , offer the most reliable measurement of spin once a model is established . here follows some details of these methods . the polarization features of bh disc radiation can be affected strongly by gr effects . conventional disc models predict that higher energy photons come from smaller disc radii . then as the photon energy increases from 1 kev to 30 kev , the plane of linear polarization swings smoothly through an angle of about 40@xmath87 for a 9m@xmath1 schwarzschild bh and 70@xmath87 for an extreme kerr bh ( connors et al . the effect is due to the strong gravitational bending of light rays . in the newtonian approximation , the polarization angle does not vary with energy . a gradual change of the plane of polarization with energy is a purely relativistic effect , and the magnitude of the change can give a direct measure of @xmath147 . polarimetry data do not even require @xmath4 , although knowledge of @xmath92 is useful in order to avoid having to include that parameter in the fit . this technique , determines the radius @xmath148 of the inner edge of the accretion disc and assumes that this radius corresponds to @xmath149 ( innermost stable circular orbit ) . because @xmath150 is a monotonic function of @xmath147 , a measurement of @xmath148 and @xmath4 directly gives @xmath147 . @xmath148 can be estimated provided that : ( a ) @xmath92 and @xmath151 are sufficiently well known , ( b ) the x - ray flux and spectral temperature are measured from well - calibrated x - ray data in the thermal state and , ( c ) the disc radiates as a blackbody . the continuum fitting is the best current method for measuring spin , although its application requires accurate estimates of bh mass ( @xmath4 ) , disc inclination ( @xmath92 ) , and distance . the first broad fe k@xmath152 line observed for either a bhb or an agn was reported in the spectrum of cyg x@xmath881 based on exosat data ( barr et al . the fe k fluorescence line is thought to be generated through the irradiation of the cold ( weakly - ionized ) disc by a source of hard x - rays ( likely an optically thin , comptonizing corona ) . in the case that the disc rotates around a bh and extends down to the isco , i.e. only a few gravitational radii away from the event horizon , relativistic beaming and gravitational redshifts in the inner disc region can serve to create an asymmetric line profile ( for a review , see reynolds & nowak 2003 ) . in 1995 , the first relativistically broadened iron k profile was observed in the seyfert1 galaxy mcg63015 with the japanese x - ray observatory asca ( tanaka et al . the sharp line profile in the rest frame of the emitter ( e.g. an accretion disc ) would be very different as observed in a distant laboratory frame . this is due to : 1 ) the doppler effect produces a redshift at the part that is receding and a blueshift at the part of the disc that is approaching to the observer . the result is a broadened line maybe with two doppler peaks if the disc is sufficiently inclined towards the observer ( this already happens in newtonian physics ) . 2 ) for intermediate to large inclinations there is a special relativistic beaming effect at work . the keplerian velocities are indeed comparable to the speed of light so that radiation is bent towards the observer . therefore , the observed radiation intensity is amplified as compared to the rest frame . 3 ) the bh captures line photons by strong space time curvature . this gravitational redshift effect is two fold : the observed photon energy is redshifted relative to the rest frame energy and the spectral line flux is lower in the observer s frame . all these three effects produce an asymmetric and skewed relativistic emission line profile ( see figure [ muller03 ] ) . in addition , relativistic emission line profiles also indicate bh spin . it is known from accretion theory that standard discs extend down to the isco . the isco depends on bh spin and is closer to the event horizon for rapidly spinning kerr bhs , @xmath153 , as compared to schwarzschild bhs , @xmath154 . as a consequence the line emission comes from regions that are deeper in the gravitational potential if the bh rotates . therefore , the line profiles from the vicinity of kerr bhs are more influenced by gravitational redshift ( see figure [ muller05 ] ) . the line has been modeled in the spectra of several bhbs . in figure [ fourbh ] , relativistic x - ray lines from the inner accretion discs around four bhs are shown ( miller 2007 ) . grs 1915 + 105 is clearly not as skewed as the others , and does not strongly require bh spin . in this source there has been some controversy about the value of the spin , being considered a rapidly rotating kerr bh in one case , @xmath155 ( mcclintock et al . 2006 ) and an intermediate spin , @xmath156 in the other case ( middleton et al . 2006 ) . in v4641 sgr , the inner disc radius deduced from the line profile is consistent with the @xmath157 radius of the isco of a schwarzschild bh , suggesting that rapid spin is not required ( miller et al . 2002 ) . in gx 339@xmath884 , the inner disc likely extends inward to ( 2 3)@xmath158 , implying @xmath159 ( miller et al . xte j1650@xmath88500 is the most extreme case with the inner edge located at @xmath160 , which suggests nearly maximal spin ( miller et al . 2002 ; miniutti et al . 2004 ) . large values of @xmath147 have also been reported for xte j1655@xmath8840 and xte j1550@xmath88564 ( miller 2007 ) . broadened fe k lines data do not even require @xmath4 to obtain the spin parameters , although knowledge of @xmath92 is useful in order to avoid having to include that parameter in the fit . high - frequency quasi periodic oscillations ( hfqpo ) are likely to offer the most reliable measurement of spin once the correct model is known . typical frequencies of these fast qpos , e.g. , 150 450 hz , correspond respectively to the frequency at the isco for schwarzschild bhs with masses of 15 5@xmath161 , which in turn closely matches the range of observed masses . these qpo frequencies ( single or pairs ) do not vary significantly despite sizable changes in the x - ray luminosity . this suggests that the frequencies are primarily dependent on the mass and spin of the bh . those bhs that show hfqpos and have well - constrained masses are the best prospects for constraining the value of the bh spin ( @xmath147 ) . theoretical work aimed at explaining hfqpos is motivated by three sources : gro j1655@xmath8840 , xte j1550@xmath88564 and grs 1915 + 105 . these sources are presently the only ones that both exhibit harmonic ( 3:2 ) hfqpos and have measured bh masses . if these hfqpos are indeed gr oscillations , then it would be possible that the three bhs have similar values of the spin parameter @xmath147 . it is relevant to try to confirm this result by taking frequency and mass measurements for more sources . assuming we have a well - tested model , qpo observations only require knowledge of @xmath4 to provide a spin estimate . bh with masses of @xmath162 to few @xmath83m@xmath1 are believed to be the engines that power nuclear activity in galaxies . agn range from faint , compact radio sources like that in m31 to quasars like 3c 273 that are brighter than the whole galaxy in which they live . some nuclei fire jets of energetic particles millions of lightyears into space . it is believed that this enormous flow of energy comes from the stars and gas that are falling into the central bh . in the past years , the search for super - massive bhs has been done by measuring rotation and random velocities of stars and gas near galactic centres . if the velocities are large enough , then they imply more mass than we see in stars , being a bh the most probable explanation . more than 50 have been found , with masses in the range expected for nuclear engines , and consistent with predictions based on the energy output of quasars . the powerful gravitational force exerted by a super - massive bh in a galactic nucleus on nearby gas and stars , causes them to move at high speeds . this is difficult to see in quasars , because they are far away and because the dazzling light of the active nucleus swamps the light from the host galaxy . however , in radio galaxies with fainter nucleus , the stars and gas are more visible . the giant elliptical galaxy messier 87 , one of the two brightest objects in the virgo cluster of galaxies , is a radio galaxy with a bright jet emerging from its nucleus . it has long been thought to contain a bh . observations of messier 87 with the hst revealed a disc of gas 500 lightyears in diameter , whose orbital speeds imply a central mass of @xmath163m@xmath1 . the ratio of this mass to the central light output is more than 100 times the solar value . no normal population of stars has such a high mass - to - light ratio . this is consistent with the presence of a bh , but it does not rule out some other concentration of underluminous matter . fortunatelly , the masers offer the possibility to have more convincing arguments . the seyfert galaxy ngc 4258 is one of the few nearby agn known to possess nuclear water masers ( the microwave equivalent of lasers ) . the enormous surface brightnesses ( @xmath164 k ) , small sizes ( @xmath165 cm ) , and narrow linewidths ( a few km s@xmath9 ) of these masers make them ideal probes of the structure and dynamics of the molecular gas in which they reside . vlbi observations ( angular resolution 100 times better than that of hst ) of the ngc 4258 maser have provided the first direct images of an agn accretion disc , revealing a thin , subparsec - scale , differentially rotating warped disc in the nucleus of this relatively weak seyfert 2 agn . the measurements imply that @xmath166m@xmath1 lie within half a lightyear of the centre ( herrnstein et al . this material can hardly be a cluster of dead stars . in the case there were failed stars ( brown dwarfs ) that never get hot enough to ignite the nuclear reactions that power stars because their low mass ( @xmath167m@xmath1 ) , there would have to be many of them to explain the dark mass in ngc 4258 . this would imply they would have to live very close together and most of them would collide with other brown dwarfs and the dark cluster would light up . in the case there were dead stars ( wds stars , nss , or stellar mass bhs ) , these are more massive than brown dwarfs , so there would be fewer of them . it is well known that during the gravitational evolution of clusters of stars , individual stars get ejected from the cluster , the remaining cluster contracts , and the evolution speeds up . calculations show that a cluster of dead stars in ngc 4258 would evaporate completely in about 100 million years . this time is much less than the age of the galaxy . therefore the most astrophysically plausible alternatives to a bh can be excluded . it seems unavoidable the conclusion that ngc 4258 contains a super - massive bh . the centre of our galaxy is only 25000 lightyears away . although its visible light is completely absorbed by intervening dust , its ir light penetrates the dust . the super - massive bh in the centre of the milky way was discovered as a bright non - thermal radio source in the 1970s and termed sagittarius a@xmath168 ( sgr a@xmath168 ) . the radio emission of sgra@xmath168 only varies slowly on time scales of several days to a few hundred days and generally with an amplitude @xmath169 . potential x - ray radiation by sgr a@xmath168 was detected with the x - ray observatory rosat in the 1990s . a reliable identification of x - rays from sgr a@xmath168 was finally possible with the new x - ray satellites chandra and xmm . a general review of sgr a@xmath168 can be found in melia & falcke ( 2001 ) . the rapid motions show that there is a mass of @xmath170m@xmath1 centered on sgr a@xmath168 . the mass enclosed inside a particular distance from sgr a@xmath168 stops dropping toward the centre at a distance of about 1 pc ( see figure [ massdis ] ) . this means that the mass in stars inside 3 lightyears has become negligible compared to the dark mass at the centre ( schdel et al . 2003 ) . as in ngc 4258 , the implied density of matter is too high to allow a cluster of dark stars or stellar remnants . ngc 4258 and the milky way give us important proof that bh exist . the velocity dispersion increases to 400 km s@xmath9 at a distance of 0.03 lightyears from sgr a@xmath168 . here stars have such small orbits that they revolve around the galactic centre in a few decades . motions in the plane of the sky have been measured for these stars for several years ( eisenhauer et al . 2005 ; ghez et al . 2005 ) . from a global fit to the positions and radial velocities of the six best stars ( see figure [ orbits ] ) within 0.5"(=0.02 pc ) of sgr a@xmath168 improved three dimensional stellar orbits were derived and their orbital parameters updated . the updated estimate for the distance to the galactic centre from the s2 orbit fit is @xmath171 kpc , resulting in a central mass value of @xmath172m@xmath1 ( eisenhauer et al . 2005 ) . vlbi observations of sgr a@xmath168 have revealed an east west elongated structure whose apparent angular size at longer wavelengths is dominated by the interstellar scattering angle , @xmath173 , where @xmath174 is the observed size , in mas , at wavelength @xmath175 in cm . thus , vlbi observations at shorter millimetre wavelengths , where the intrinsic structure of sgr a@xmath168 could become comparable to the pure scattering size , are expected to show deviations of the observed size from the scattering law . this has been demonstrated by the detection of the intrinsic size at 7 mm ( bower et al . more recently , zhi - qiang shen et al . ( 2005 ) , carried out observations with the vlba at 3.5 mm of sgr a@xmath168 . they report a radio image of sgr a@xmath168 , demonstrating that its size is 1au . this high - resolution image of sgr a@xmath168 , made at 3.5 mm , also exhibits an elongated structure . when combined with the lower limit on its mass , the lower limit on the mass density is @xmath176m@xmath1 pc@xmath177 , which provides strong evidence that sgr a@xmath168 is a super - massive bh . detection of variable infrared ( 3.8 micron ) light coincident to within 18 mas ( 1 @xmath178 ) of the super - massive bh at the centre of the milky way galaxy and the radio source sgr a@xmath168 was reported by ghez et al . the brightness variations , which occur on timescales as short as 40 minutes , reveal that the emission arises quite close to the bh , within 5 au , or 80@xmath27 . two k - band flares observed on the 15th and 16th of june 2003 show a quasi - periodicity of about 17 min ( genzel et al . the only reasonable explanation for this so short period is that the oscillations are produced by doppler boosting of hot gas near the last stable orbit of a spinning kerr bh . the spin of the bh will allow for a last stable orbit closer to the event horizon and thus with a shorter orbital frequency . from the observed 17 min quasi - periodicity it is estimated that the super - massive bh sgr a@xmath168 has a spin that is half of the maximum possible spin of such an object ( genzel et al . additional observations of flares and their quasi - periodicity will be needed in order to confirm this result . in the x - ray regime , sgra@xmath168 was found to exhibit two different states . in the quiescent state , weak x - ray emission appears to come from a slightly extended area around the bh that appears to be evidence of hot accreting gas in the environment of sgra@xmath168 . sgra@xmath168 shows x - ray flares with a recurrence of about one per day . during these flares , the emission rises by factors up to 100 during several tens of minutes and a distinctive point source becomes visible at the location of sgra@xmath168 . the short rise - and - decay times of the flares suggest that the radiation must originate from a region within less than 10@xmath27 of a @xmath179m@xmath1 bh ( boganoff et al . 2001 ) . the frame dragging that causes matter and light to co rotate with the space time has effects in the orbits around bhs . mueller ( 2007 ) shows relativistically distorted images of tight orbits around bhs for different inclination angles and for schwarzchild and kerr bhs . when viewed by a distant observer at infinity , in nearly edge on situations the observer would detect different apparent orbit shapes depending on bh spin . these orbit shapes become asymmetric in the case of rotating bhs . the maximum relative spatial offset predicted is 0.08@xmath180 . this means that it might be possible to detect it by observing the galactic centre bh but not by observing an stellar bh . the reason is that the angular size corresponding to an schwarzschild radius of a super - massive bh is about three orders of magnitude larger than the an stellar mass bh . the angular size corresponding to an schwarzschild radius is given by : i would like to thank g.e . romero and p. benaglia , for their invitation and the excellent organisation of the international school . i am indebted to j. mart and v. zabalza for a careful reading of the manuscript and their valuable comments . the author acknowledges support of the spanish ministerio de educacin y ciencia ( mec ) under grant aya2007 - 68034-c03 - 01 and feder funds . narayan , r. , garcia , m.r . , & mcclintock , j.e . 2002 , in : the ninth marcel grossmann meeting . proceedings of the mgixmm meeting held at the university of rome `` la sapienza '' , 2 - 8 july 2000 , eds . : vahe g. gurzadyan , robert t. jantzen , remo ruffini . singapore : world scientific publishing , 2002 , part a , p. 405

a graph is called a unit - distance graph in the plane if there is an injective embedding of in the plane such that every pair of adjacent vertices are at unit distance apart . if additionally the corresponding edges are non - crossing and all vertices have the same degree we talk of a regular matchstick graph . due to euler s polyhedron formula we have . the smallest known-regular matchstick graph is the so called harborth graph consisting of vertices . in this article we prove that no finite-regular matchstick graph exists . unit - distance graphs

1401.1793

one of the possibly best known problems in combinatorial geometry asks how often the same distance can occur among @xmath7 points in the plane . via scaling we can assume that the most frequent distance has length @xmath8 . given any set @xmath9 of points in the plane , we can define the so called unit - distance graph in the plane , connecting two elements of @xmath9 by an edge if their distance is one . the known bounds for the maximum number @xmath10 of edges of a unit - distance graph in the plane , see e. g. @xcite , are given by @xmath11 for @xmath12 the exact numbers of @xmath10 were determined in @xcite , see also @xcite . if we additionally require that the edges are non - crossing , then we obtain another class of geometrical and combinatorial objects : a * matchstick graph * @xmath13 consists of a graph @xmath0 and an injective embedding @xmath14 in the plane which fulfill the following conditions : * @xmath15 is a planar ( simple ) graph , * for all edges @xmath16 we have @xmath17 , where @xmath18 denotes the euclidean distance between the points @xmath19 and @xmath20 , * @xmath21 for @xmath22 , and * if @xmath23 for pairwise different vertices @xmath24 then the line segments @xmath25 and @xmath26 do not have a common point . for matchstick graphs the known bounds for the maximum number @xmath27 of edges , see e. g. @xcite , are given by @xmath28 where the lower bound is conjectured to be exact . we call a matchstick graph @xmath2-regular if every vertex has degree @xmath2 . in @xcite the authors consider @xmath2-regular matchstick graphs with the minimum number @xmath29 of vertices . obviously we have @xmath30 , @xmath31 , and @xmath32 , corresponding to a single vertex , a single edge , and a triangle , respectively . the determination of @xmath33 is an entertaining amusement . at first we observe that @xmath33 must be even since every graph contains an even number of vertices with odd degree . for @xmath34 vertices the set of @xmath35-regular planar graphs is fairly assessable , consisting of one graph of order , i. e. the number of vertices , @xmath4 , one graph of order @xmath36 , and three graphs of order @xmath37 . utilizing area arguments rules out all but one graph of order @xmath37 , so that we have @xmath38 . for degree @xmath39 the exact determination of @xmath40 is unsettled so far . the smallest known example is the so called harborth graph , see e. g. @xcite , yielding @xmath41 . due to the eulerian polyhedron formula every finite planar graph contains a vertex of degree at most five so that we have @xmath42 for @xmath43 . we would like to remark that the regular triangular lattice is the unique example of an infinite matchstick graph with degree at least @xmath36 . for degree @xmath6 it is announced at several places that no finite @xmath6-regular matchstick graph does exist , see e. g. ivars peterson s mathtrek `` match sticks in the summer '' @xcite most likely referring to personal communication with heiko harborth , the discoverer of the @xmath4-regular matchstick graph . indeed heiko harborth posed the question whether there exists a @xmath6-regular matchstick graph , or not at an oberwolfach meeting and was aware of a preprint claiming the non - existence proof ( personal communication ) . after a while he found out that there were some mistakes in the proof , so that to his knowledge there is no correct proof of the non - existence . asking the author of this preprint about the state of this problem he replied too , that the problem is still open ( at this point in time the concerning was untraceable ) . wolfram research s mathworld , see http://mathworld.wolfram.com/matchstickgraph.html , refers to erich friedman , who himself maintains a webpage @xcite stating the non - existence of a @xmath6-regular matchstick graph . asking erich friedman for a proof of this claim he responded that unfortunately he lost his reference but also thinks that such a graph can not exist . in @xcite the authors list five publications where they have mentioned an unpublished proof for the non - existence of a @xmath6-regular matchstick graph and state that up to their knowledge the problem is open . after the submission of this paper one of the referees came up with a very short proof for the non - existence @xcite . in this paper we would like to give a different , admittedly more complicated , proof using a topological equation that needs some explanation . so in some parts this ( unpublished ) paper coincides with @xcite . in the meantime heiko harborth could recover the lost preprint and send it to the original author and myself . curiously enough , after rectifying some annoying minor mistakes , the preprint turns out to be basically correct . so the author of this article took the opportunity to retype and slightly modify the original manuscript , see @xcite . it even turned out that the underlying method can be easily adopted to prove the following stronger result : no finite matchstick graph with minimum degree @xmath6 does exist . so up to now there are at least three proofs of theorem [ thm_main ] whose pairwise intersections are non - empty . we would like to mention the recent proof of the higuchi conjecture @xcite a result of similar flavor where related techniques are applied . [ thm_main ] no finite @xmath6-regular matchstick graph does exist . in order to prove theorem [ thm_main ] we denote the number of @xmath44-gons , i. e. a face consisting of @xmath44 vertices , of a given matchstick graph @xmath13 by @xmath45 , where we also count the outer face . in the example in figure [ fig_matchstick_graph ] we have @xmath46 , @xmath47 , @xmath48 , and @xmath49 for all other @xmath44 . [ lemma_euler_sum ] for a finite @xmath6-regular matchstick graph we have @xmath50 due to the eulerian polyhedral formula we have @xmath51 , where @xmath1 denotes the number of vertices , @xmath52 the number of faces , and @xmath53 the number of edges . the number of faces is given by @xmath54 since every edge is part of two faces and every vertex is part of @xmath6 faces we have @xmath55 inserting yields the proposed formula . the * face set * @xmath56 of a vertex @xmath57 contained in a planar graph is a multiset containing the number of corners of the adjacent faces . in the example in figure [ fig_matchstick_graph ] we have @xmath58 . for a @xmath6-regular matchstick graph the face sets @xmath56 of the vertices have cardinality @xmath6 , i. e. there exist integers @xmath59 with @xmath60 . due to the angle sum of @xmath61 at a vertex we have @xmath62 for at least one index @xmath44 for every vertex @xmath63 . for an integer @xmath64 we set @xmath65 and use this to define the * contribution * @xmath66 of a vertex with face set @xmath60 as @xmath67 for @xmath68 we set @xmath69 . now we can relate this definition to lemma [ lemma_euler_sum ] and state two easy lemmas . [ lemma_20 ] if @xmath13 is a ( finite ) @xmath6-regular matchstick graph we have @xmath70 . for @xmath71 we have @xmath72 . applying this to the definition of @xmath73 yields the left hand side of the formula of lemma [ lemma_euler_sum ] . we would like to remark that for an infinite @xmath6-regular matchstick graph we would have @xmath74 . [ lemma_non_negative ] the face sets with non - negative contribution are given by @xmath75 our strategy for proving theorem [ thm_main ] will be to partition the vertex set @xmath1 into subsets each having non - positive contribution , which yields a contradiction to lemma [ lemma_20 ] . to determine a suitable partition of @xmath1 into subsets for a given matchstick graph @xmath13 we consider the set @xmath76 of triangles and quadrangles with internal angles in @xmath77 and define an equivalence relation @xmath78 on @xmath76 . whenever there exist @xmath79 sharing an edge we require @xmath80 . we complete this relation to an equivalence relation by taking the transitive closure . since we prove that no finite @xmath6-regular matchstick graphs exists it is impossible to give examples to illustrate our definitions . therefore we define them ( mostly ) for matchstick graphs @xmath13 where the degrees of its vertices are at most @xmath6 . for a given ( possible infinite ) matchstick graph @xmath13 with vertex degrees at most @xmath6 we call an equivalence class @xmath81 of the above defined equivalence relation a @xmath76-class . so a @xmath76-class is an edge - connected union of triangles and quadrangles whose vertices are situated on a suitable common regular triangular lattice . in figure [ fig_honeycomb_components ] we have depicted an example , where we have marked the @xmath76-classes by different face colors . -classes of a matchstick graph . ] in the next lemma we summarize some easy facts on @xmath76-classes . let @xmath82 be a @xmath76-class , then the following holds : * the faces of @xmath82 are edge - to - edge connected , meaning that the dual graph is connected . * the vertices of @xmath82 are situated on a suitable regular triangular lattice . for brevity we associate with a @xmath76-class @xmath82 the set of its vertices @xmath83 and the set of its edges @xmath84 so that we can utilize the notation @xmath85 for the contribution of the vertex set of @xmath82 . now our aim is to show that the contribution @xmath86 for every @xmath76-class @xmath82 of a @xmath6-regular matchstick graph is at most zero . since some vertices may be contained in more than one @xmath76-class the corresponding vertex sets can not be used directly to partition the vertex set @xmath1 . instead we choose the union @xmath87 of all @xmath76-classes . here every vertex @xmath63 occurs at most once in @xmath88 , nevertheless it may be contained in several @xmath76-classes @xmath89 . in order to prove @xmath90 we have to perform some bookkeeping during the proof of @xmath91 , which will be the topic of the next section once we haven proven this we can state : * proof of the main theorem . * let @xmath13 be a finite @xmath6-regular matchstick graph and @xmath88 be the union of all @xmath76-classes of @xmath13 . using corollary [ cor_non_negativ_all ] and the fact that all vertices with positive contribution are contained in a @xmath76-class , see lemma [ lemma_non_negative ] , we conclude @xmath92 this contradicts lemma [ lemma_20 ] . in this section we want to study the contribution @xmath85 of a finite @xmath76-class @xmath82 in a @xmath6-regular matchstick graph @xmath13 . since we show that no finite @xmath6-regular matchstick graph exists @xmath13 has to be infinite in this context . nevertheless there may be some finite @xmath76-classes in @xmath13 . but since we are only interested in finite matchstick graphs we introduce another concept and consider incomplete finite parts of @xmath6-regular matchstick graphs . let @xmath13 be a finite matchstick graph with maximum vertex degree at most @xmath6 and @xmath82 a @xmath76-class of @xmath13 which induces a planar graph @xmath93 containing the vertices , edges , and faces of @xmath82 . we call a vertex @xmath57 in @xmath93 ( or @xmath82 ) an * inner vertex * if all faces being adjacent to @xmath57 in @xmath93 are contained in @xmath82 . all other vertices @xmath57 in @xmath93 ( or @xmath82 ) are called * outer vertices*. now we can call the @xmath76-class @xmath82 * prospective @xmath94-regular * exactly if all inner vertices of @xmath93 have degree @xmath6 and all outer vertices of @xmath93 have degree at most @xmath6 . -regular @xmath76-class @xmath82 of a matchstick graph @xmath13 and its induced planar graph @xmath93 . ] in figure [ fig_prospective ] we have depicted a matchstick graph with maximum vertex degree at most @xmath6 on the left hand side . the faces of the , in this case uniquely , contained @xmath76-class @xmath82 are filled with green color . if we build up a planar graph out of the vertices , edges , and faces of @xmath82 we obtain the right hand side of figure [ fig_prospective ] . in @xmath93 vertex @xmath57 is the only vertex which is not adjacent to the outer face . thus @xmath57 is an inner vertex in @xmath82 and the remaining @xmath95 vertices of @xmath82 are outer vertices . since every @xmath76-class @xmath82 of a given finite @xmath6-regular matchstick graph @xmath13 is prospective @xmath6-regular , we now study prospective @xmath6-regular @xmath76-classes . therefore we want to specify them by some parameters @xmath96 , @xmath97 , @xmath98 , @xmath99 and @xmath100 . ( for brevity we forego to use notations as @xmath101 , whenever the corresponding @xmath76-class is evident from the context . ) the parameter @xmath96 stands for the contribution of the faces of @xmath82 . to become more precisely we have to introduce a new technical notation . a given vertex @xmath57 is adjacent to five faces denoted as @xmath102 , where face @xmath103 is an @xmath104-gon . since there exist pairs of vertices @xmath105 and indices @xmath106 with @xmath107 we introduce the notation @xmath108 $ ] addressing the arc of face @xmath103 at vertex @xmath57 . the contribution of such a face arc @xmath108 $ ] is defined to be @xmath109 . if @xmath110 is the set of all face arcs @xmath108 $ ] where the face @xmath103 is contained in @xmath82 , then @xmath96 is given by @xmath111 . let us look at the four @xmath76-classes of figure [ fig_honeycomb_components ] ( it is easy to check that they are all prospective @xmath6-regular ) . the blue @xmath76-class consists of a single triangle . so it has three _ triangle_-arcs and a contribution of @xmath112 . it is easy to figure out that in general a triangle contributes @xmath8 and a quadrangle contributes @xmath113 to @xmath96 so that we have @xmath114 , @xmath115 , and @xmath116 . we have already remarked that every @xmath117 edge of a planar graph is contained in two faces . whenever both faces are contained in a given @xmath76-class @xmath82 we say that @xmath118 is an inner edge . if only one face is contained in @xmath82 we call @xmath118 an outer edge . by @xmath97 we count the outer edges of a @xmath76-class @xmath82 . in our example we have @xmath119 , @xmath120 , @xmath121 , and @xmath122 . if we consider the vertices and edges of a prospective @xmath6-regular @xmath76-class as a subgraph , then every vertex @xmath57 has a degree @xmath123 at most @xmath6 . by @xmath124 we denote the gap @xmath125 ( free valencies ) and by @xmath126 the sum over all @xmath124 where @xmath57 is in @xmath82 . so in our example we have @xmath127 , @xmath128 , @xmath129 , and @xmath130 . in a @xmath6-regular planar graph every vertex @xmath57 is contained in exactly five different faces @xmath103 . whenever all five faces are contained in a given @xmath76-class @xmath82 we say that @xmath57 in @xmath82 is an inner vertex . if the number of faces @xmath103 which are contained in @xmath82 is between one and four we call @xmath57 an outer vertex . this coincides with our previous definition of inner and outer vertices of a prospective @xmath6-regular @xmath76-class . clearly we have @xmath131 for all inner vertices if @xmath132 is an outer edge in @xmath82 then @xmath57 and @xmath133 are outer vertices of @xmath82 . for the other direction we have that for an outer vertex @xmath57 in @xmath82 there exist at least two outer edges @xmath134 , @xmath135 in @xmath82 being adjacent to @xmath57 . more precisely the number of outer edges being adjacent to @xmath57 is either @xmath136 or @xmath4 . for a given vertex @xmath57 in @xmath82 we consider the faces @xmath103 which are contained in a given @xmath76-class @xmath82 as vertices of a graph @xmath137 . two vertices of @xmath137 are connected via an edge whenever the corresponding faces have a common edge in @xmath13 . if the graph @xmath137 is connected we call @xmath57 of normal type . otherwise we say that vertex @xmath57 is of special type . in figure [ fig_special_situation ] we depict all cases of vertices of special type up to symmetry . here vertex @xmath57 is marked blue and the faces of @xmath82 are marked green . alternatively we could also define a vertex of special type as an outer vertex of a prospective @xmath6-regular @xmath76-class which is adjacent to exactly @xmath4 outer edges of @xmath82 . we would like to remark that from a local point of view at a vertex of special type it seems that the faces of the @xmath76-class are not edge connected . by @xmath99 we count the number of vertices @xmath57 in @xmath82 of special type . in our example of figure [ fig_honeycomb_components ] no vertex of special type exists and we have @xmath138 . in figure [ fig_honeycomb_components_2 ] we have depicted another example of a @xmath76-class , where we have @xmath139 the blue vertex . -class of a matchstick graph . ] if a face @xmath140 of a @xmath76-class is a quadrangle then there are two inner angles of @xmath141 and two inner angles of @xmath142 . by @xmath100 we count the number of inner angles of @xmath141 where the corresponding vertex @xmath57 is an outer vertex . it may happen that an outer vertex @xmath57 is part of two quadrangles in @xmath82 each having an inner angle of @xmath143 at @xmath57 . thus for a prospective @xmath6-regular @xmath76-class @xmath82 we may write @xmath144 for the number of inner angles of @xmath141 at an outer vertex @xmath57 . for inner vertices we set @xmath145 so that we can set @xmath146 . in the example of figure [ fig_honeycomb_components ] we have @xmath147 in the green class , @xmath148 in the brown class , and @xmath149 in the red and blue class . we remark that due to an angle sum of @xmath61 there is exactly one face with an inner angle of @xmath141 at every inner vertex of @xmath82 . now we want to prove the equation @xmath150 relating the parameters @xmath96 , @xmath97 , @xmath98 , @xmath99 and @xmath100 . this equation even holds for more general objects than prospective @xmath6-regular @xmath76-classes . a @xmath151-class @xmath82 is a union of triangles and quadrangles with inner angles in @xmath152 on a regular triangular grid such that each inner vertex has degree @xmath6 and each outer vertex has degree at most @xmath6 . clearly we can transfer the definitions of the parameters @xmath96 , @xmath97 , @xmath98 , @xmath99 and @xmath100 from ( prospective @xmath6-regular ) @xmath76-classes to @xmath151-classes . the crucial difference between @xmath76-classes and @xmath151-classes is that the dual graph of a @xmath151-class ( considered as a planar graph ) is not connected in general . -classes . ] in figure [ fig_tr_classes ] we depict some extraordinary examples of @xmath151-classes to demonstrate the whole variety being covered by the definition of a @xmath151-class . here the faces of a @xmath151-class @xmath82 are filled in green . those faces which are not contained in @xmath82 are left in white . in the left example of figure [ fig_tr_classes ] we have @xmath153 , @xmath154 , @xmath155 , @xmath156 , and @xmath149 . in the second example we have @xmath157 , @xmath158 , @xmath159 , @xmath160 , and @xmath149 . in the third example of figure [ fig_tr_classes ] we have @xmath161 , @xmath162 , @xmath163 , @xmath156 , and @xmath149 . for the central vertex @xmath57 of the right example of figure [ fig_tr_classes ] we have @xmath164 . additionally @xmath57 is of special type . the parameters are given by @xmath165 , @xmath166 , @xmath167 , @xmath139 , and @xmath147 . thus in all four cases equation ( [ eqn_paramater ] ) is valid . if we consider those four examples as a single example on the same triangular grid again equation ( [ eqn_paramater ] ) is valid . [ lemma_parameter ] for a @xmath151-class @xmath82 equation ( [ eqn_paramater ] ) is valid . we prove by induction on the number of faces of @xmath82 . for a single triangle we have @xmath168 , @xmath169 , @xmath170 , and @xmath171 . thus the left hand side of ( [ eqn_paramater ] ) sums to zero . for a quadrangle we have @xmath172 , @xmath173 , @xmath155 , @xmath138 , and @xmath148 , again summing up to zero in ( [ eqn_paramater ] ) . we remark that equation ( [ eqn_paramater ] ) is also valid for an empty @xmath151-class . for the induction step we now assume that equality ( [ eqn_paramater ] ) is valid and we show that it remains valid after adding another single triangle or quadrangle to @xmath82 . here we only consider those additions which do not alter faces of @xmath82 ( one might think of inserting an edge into a quadrangle of @xmath82 resulting in two triangles and destroying the initial quadrangle ) . in general it might happen that adding the edges of one new face produces a second new face in one step . an example arises by adding an edge into a quadrangle which is not contained in @xmath82 , see the second and the third example of figure [ fig_tr_classes ] . here , in one step we only add one of the resulting triangles to @xmath82 . if the other triangle should also be added to @xmath82 we do this in a second step where we only have a new face and no new vertices or edges . adding a triangle results in eight ( main ) cases , see figure [ fig_adding_a_triangle ] , and adding a quadrangle results in nineteen ( main ) cases , see figure [ fig_adding_a_quadrangle ] . in the different ( main ) cases we depict the new edges by dotted lines , the new vertices by empty circles , and the old edges and vertices in black . we have a closer look on the change of the parameters . if @xmath174 is the parameter of @xmath82 and @xmath175 the corresponding parameter after adding a triangle ( or a quadrangle ) , then we denote the change by @xmath176 . in order to prove the induction step it suffices to verify @xmath177 we remark that all black edges in the ( main ) cases of figure [ fig_adding_a_triangle ] and figure [ fig_adding_a_quadrangle ] are outer edges before adding the new triangle or quadrangle . after adding the new face the black edges become inner edges and the dotted edges become outer edges . .@xmath178-values for the eight ( main ) cases of figure [ fig_adding_a_triangle ] . [ cols=">,>,>,>,>",options="header " , ] next we can deal with the nineteen ( main ) cases of figure [ fig_adding_a_quadrangle ] . in every ( main ) case the situation of vertex @xmath179 and vertex @xmath180 has an equivalent in figure [ fig_adding_a_triangle ] . so up to symmetry we have to consider the situation at vertex @xmath57 . again there are four possibilities to consider showing up in ( main ) cases ( 9 ) , ( 14 ) , ( 21 ) , and ( 27 ) of figure [ fig_adding_a_quadrangle ] . we denote the corresponding ( vertex ) cases by ( v ) , ( vi ) , ( vii ) , and ( viii ) , respectively . similarly as before we denote the vertex @xmath57 before the addition of the quadrangle by @xmath181 and afterwards by @xmath182 . in ( vertex ) case ( v ) , occurring in ( main ) case ( 9 ) , @xmath181 does not exist and @xmath182 is of normal type and we have @xmath183 so that the sum of the @xmath99- and the @xmath100-count increases by one . in ( vertex ) case ( vi ) both @xmath181 and @xmath182 are outer vertices . vertex @xmath182 is of special type and @xmath181 has to be of normal type . for the @xmath100-value we have @xmath184 so that the sum of the @xmath99- and the @xmath100-count increases by two . in ( vertex ) case ( vii ) both @xmath181 and @xmath182 are outer vertices . vertex @xmath181 is of special type if and only if @xmath182 is of special type . since @xmath184 the sum of the @xmath99- and the @xmath100-count increases by one . in ( vertex ) case ( viii ) if @xmath182 is an inner vertex than @xmath181 , @xmath182 are of normal type , @xmath185 , and @xmath186 . if @xmath182 is an outer vertex then @xmath181 is of special type , @xmath182 is of normal type , and @xmath184 . thus in both cases the sum of the @xmath99- and the @xmath100-count does not change . with the above we are able to give the @xmath178-values for the ( main ) cases of figure [ fig_adding_a_quadrangle ] in table [ table_delta_quadrangles ] . it is easy to check that in all nineteen ( main ) cases equation ( [ eqn_paramater_delta ] ) is valid . [ cor_parameter ] for a ( finite ) prospective @xmath6-regular @xmath76-class of a matchstick graph we have @xmath187 now we are ready to prove @xmath91 for every ( finite ) prospective @xmath6-regular @xmath76-class of a matchstick graph with maximum vertex degree at most @xmath6 . to be more precisely we need to consider finite matchstick graphs @xmath13 with maximum vertex degree at most @xmath6 and a prospective @xmath6-regular @xmath76-class @xmath82 of @xmath13 . in @xmath13 all vertices of @xmath82 must have vertex degree exactly @xmath6 . in order to be able to speak of a contribution @xmath85 for every vertex @xmath57 in @xmath82 there must be five faces @xmath103 , unequal to the outer face , completely being contained in @xmath13 . if a finite @xmath6-regular matchstick graph would exist , it would certainly be such a graph . otherwise we can only look at local parts of such a ( possible infinite ) graph . as mentioned in the previous section we have to perform some bookkeeping in order to prove @xmath188 for a set of prospective @xmath6-regular @xmath76-classes . to every face arc @xmath108 $ ] where @xmath57 is contained in a given prospective @xmath6-regular @xmath76-class @xmath82 we will assign a real weight @xmath189)\ge c([v , i])$ ] which fulfills @xmath190)\le 0.\ ] ] we start with the face arcs @xmath108 $ ] where the face @xmath103 is contained in @xmath82 . we call those arcs inner arcs and set @xmath189)=c([v , i])$ ] . with this the sum @xmath191\text { inner arc of } \mathcal{b}}\omega([v , i])$ ] over the inner arcs equals the parameter @xmath96 from corollary [ cor_parameter ] . -regular matchstick graph . ] now we consider the remaining face arcs @xmath108 $ ] which are given by a vertex @xmath57 and two edges @xmath118 , @xmath192 being adjacent to @xmath57 . we say that the edges @xmath118 and @xmath192 are associated to @xmath108 $ ] . let @xmath57 be an outer vertex of @xmath82 and @xmath118 be an edge ( in @xmath13 ) being adjacent to @xmath57 . if @xmath118 is not an outer edge of @xmath82 we call @xmath118 a _ leaving edge_. this is only possible if @xmath118 is contained in @xmath13 but not in @xmath82 . to emphasize that we consider edge @xmath118 being rooted at vertex @xmath57 we also speak of _ leaving half edges_. it may happen that a leaving edge @xmath193 corresponds to two leaving half edges being rooted at vertex @xmath133 and @xmath57 , respectively . we remark that the number of outer edges equals @xmath97 and that the number of leaving half edges equals @xmath98 in corollary [ cor_parameter ] . in figure [ fig_honeycomb_components_3 ] we have depicted the outer edges in blue and red . the leaving half edges are depicted by dashed lines . if @xmath108 $ ] is a face arc where both associated edges @xmath118 and @xmath192 correspond to leaving half edges then we set @xmath189)=\frac{1}{3}$ ] being greater or equal to @xmath194)$ ] . for the remaining face arcs we consider the outer edges of @xmath82 . they form a union of simple cycles , i. e. cycles without repeated vertices . in the example of figure [ fig_honeycomb_components_2 ] we have a cycle of length @xmath36 and a cycle of length @xmath195 . we will treat each cycle @xmath196 separately . such a cycle @xmath196 divides the plane into two parts . we call the part containing the faces of @xmath82 the interior of @xmath196 and the other part the exterior of @xmath196 . when we speak of leaving half edges then we only want to address those which go into the exterior of @xmath196 . so the blue cycle of figure [ fig_honeycomb_components_3 ] does not contain a leaving half edge whereas the red cycle contains @xmath197 leaving half edges . it may happen that such a cycle @xmath196 does not contain any leaving half edges at all , like the blue cycle of figure [ fig_honeycomb_components_3 ] . in this case @xmath196 consists of the single face @xmath103 , where @xmath108 $ ] is an arbitrary face arc being associated to an edge of @xmath196 . since @xmath103 is not contained in @xmath82 it is neither a triangle nor a quadrangle . if @xmath103 is a pentagon we set @xmath189)=c([v , i])=-1 $ ] for all associated face arcs @xmath108 $ ] otherwise we set @xmath189)=-\frac{4}{3}\ge c([v , i])$ ] . if @xmath196 does contain leaving half edges the face which is adjacent to a given outer edge @xmath118 and not contained in @xmath82 is bordered by two leaving half edges . in figure [ fig_weights ] we have depicted the possible cases . in principle it would be possible that @xmath196 contains only one leaving half edge @xmath118 . in such a situation we consider the two half edges the cases of figure [ fig_weights ] as being identified to @xmath118 . in case ( 1 ) a single outer edge is bordered by two leaving half edges . since the outer face is not part of @xmath82 it can not be a triangle . so it is at least a quadrangle and we can set @xmath189)=-\frac{1}{2}\ge c([v , i])$ ] . in cases ( 2a ) and ( 2b ) we consider a path @xmath9 of length two which is bordered by two leaving half edges . since the outer face is not part of @xmath82 it can not be a triangle or a quadrangle . if it is a pentagon we set @xmath189)=-\frac{2}{3}\ge c([v , i])$ ] for the ends of @xmath9 and @xmath189)=-1\ge c([v , i])$ ] for the central vertex . if the outer face has at least @xmath36 edges we set @xmath189)=-\frac{1}{2}\ge c([v , i])$ ] for the ends of @xmath9 and @xmath189)=-\frac{4}{3}\ge c([v , i])$ ] for the central vertex . in the remaining cases we consider paths of length at least @xmath35 in case ( 3a ) , ( 3b ) , paths of length @xmath4 in case ( 4a ) , ( 4b ) , and paths of length at least @xmath6 in case ( 5 ) . again the outer face can not be a triangle or a quadrangle . in case ( 5 ) it even can not be a pentagon . here we do the assignments of the weights as depicted in figure [ fig_weights ] . we can easily check that we have @xmath189)\ge c([v , i])$ ] . let @xmath82 be a given prospective @xmath6-regular @xmath76-class of matchstick graph @xmath13 with vertex degree at most @xmath6 . if @xmath85 can be evaluated within @xmath13 then the weight function @xmath198 constructed above fulfills @xmath199)\le 0.\ ] ] and @xmath189)\ge c([v , i])$ ] for all vertices @xmath200 , @xmath201 . by construction we have @xmath189)\ge c([v , i])$ ] for all @xmath200 , @xmath201 . to prove equation ( [ eqn_non_negativ ] ) we utilize a booking technique . we want to book face arcs @xmath108 $ ] , outer edges , leaving half edges and vertices of special type . by @xmath202 we denote the sum of weights @xmath189)$ ] over the arcs @xmath108 $ ] which are booked . if we book an arc @xmath108 $ ] we also want to book half of its both associated edges and half edges each . therefore we define @xmath203 if edge or half edge @xmath118 was never booked , @xmath204 if @xmath118 was booked only one time , and @xmath205 if @xmath118 was booked two times . by @xmath206 we denote the sum over @xmath207 , where the edges @xmath118 are outer edges of @xmath82 and by @xmath208 we denote the sum over all leaving half edges . by @xmath209 we count the number of vertices of special type which are not booked so far . as done in the construction of @xmath198 we start with those face arcs @xmath108 $ ] , where the corresponding face @xmath103 is contained in @xmath82 . here we set @xmath189)=c([v , i])$ ] . after this initialization we have @xmath210 , @xmath211 , @xmath212 , and @xmath213 using the notation of lemma [ lemma_parameter ] . with this we have @xmath214 since @xmath215 . now we book the remaining face arcs and keep track that inequality ( [ eqn_bookkeeping ] ) endures . if @xmath108 $ ] is a face arc between two leaving half edges @xmath118 and @xmath192 , then we book these edges and assign @xmath189)=\frac{1}{3}$ ] . by this booking step @xmath202 increases by @xmath216 and @xmath208 decreases by @xmath8 . thus inequality ( [ eqn_bookkeeping ] ) remains valid . next we consider the simple cycles @xmath196 of the outer edges . we start with the cases where @xmath196 does not contain any leaving half edges . in this case @xmath196 consists of the single face @xmath103 , where @xmath108 $ ] is an arbitrary face arc being associated to an edge of @xmath196 . if @xmath103 is a pentagon both @xmath202 and @xmath206 decrease by @xmath6 . we can easily check that on a regular triangular grid there does not exist an equilateral pentagon where all inner angles are at most @xmath141 . thus face @xmath103 does contain an inner angle of at least @xmath217 at a vertex @xmath133 . this vertex @xmath133 must be of special type due to its angle sum of @xmath61 . thus we can book the special type and decrease @xmath209 by one so that inequality ( [ eqn_bookkeeping ] ) remains valid . if @xmath103 contains @xmath218 edges we set @xmath189)=-\frac{4}{3}$ ] at the corresponding face arcs . thus @xmath206 decreases by @xmath219 and @xmath202 decreases by @xmath220 so that inequality ( [ eqn_bookkeeping ] ) remains valid . next we consider the cases of figure [ fig_weights ] . in all cases @xmath208 decreases by one . the decreases of @xmath206 and @xmath202 vary in the different cases , but we can easily check that inequality ( [ eqn_bookkeeping ] ) remains valid . a the end of this procedure we have @xmath221 since all edges are booked properly . since all vertices of special type are booked at most once we have @xmath222 . thus we can conclude @xmath190)=\omega\le 0\ ] ] from inequality ( [ eqn_bookkeeping ] ) in the end . for a prospective @xmath6-regular @xmath76-class @xmath82 of a matchstick graph with maximum vertex degree @xmath6 we have @xmath223 as depicted in figure [ fig_honeycomb_components ] it may happen that a vertex @xmath57 is contained in more than one prospective @xmath6-regular @xmath76-class @xmath82 . in this cases it belongs to exactly two prospective @xmath6-regular @xmath76-classes due to an vertex degree of five . in figure [ fig_two_tq_classes ] we have depicted the possible cases . here one prospective @xmath6-regular @xmath76-class is depicted in green and the other one is depicted in blue . in case ( 1 ) the angles of face arcs @xmath224 $ ] and @xmath225 $ ] need not to be multiples of @xmath142 . similarly also in case ( 2 ) the angles of the face arcs @xmath226 $ ] , @xmath224 $ ] , and @xmath225 $ ] need not to be multiples of @xmath142 . in the next lemma we show that the assignment of our weight functions do not underestimate the contribution of vertex @xmath57 . -regular @xmath76-classes with a common vertex . ] let @xmath227 and @xmath228 be two different prospective @xmath6-regular @xmath76-classes of a matchstick graph with maximum vertex degree at most @xmath6 and weight functions @xmath229 and @xmath230 , respectively . if a vertex @xmath57 is contained in both @xmath227 and @xmath228 then we have @xmath231)+\omega_2([v , i])\ge \sum_{i=1}^5 c([v , i]).\ ] ] w.l.o.g . the prospective @xmath6-regular @xmath76-class @xmath227 is depicted green in figure [ fig_two_tq_classes ] and @xmath228 is depicted blue . at first we consider case ( i ) . here we have @xmath232)\ge c[(v,1])$ ] , @xmath233)\ge c[(v,2])$ ] , and @xmath234)\ge c([v,4])$ ] . next we look at the face arcs which are bordered by two leaving half edges . here we have @xmath235)=\frac{1}{3}$ ] , @xmath236)=\frac{1}{3}$ ] , and @xmath237)=\frac{1}{3}$ ] . due to symmetry it suffices to show @xmath238)+\omega_2([v,3])\ge c([v,3])-\frac{1}{2}.\ ] ] since at vertex @xmath57 there are leaving half edges both in @xmath227 and @xmath228 we have to be in one of the cases of figure [ fig_weights ] for the determination of the weight of face arc @xmath224 $ ] . if for @xmath227 face arc @xmath224 $ ] is in one of the cases ( 1 ) , ( 2b ) , ( 3b ) , ( 4b ) , or ( 5 ) then we have @xmath239)=-\frac{1}{2}$ ] and inequality ( [ eq_overcount ] ) is valid . due to symmetry the same holds if for @xmath228 face arc @xmath224 $ ] is in one of the cases ( 1 ) , ( 2b ) , ( 3b ) , ( 4b ) , or ( 5 ) . in all other cases @xmath103 is a pentagon so that @xmath194)= -1 $ ] . we remark that @xmath224 $ ] can not be in case ( 3a ) both in @xmath227 and @xmath228 . thus we have @xmath239)+\omega_2([v,3])\ge -\frac{5}{6}-\frac{2}{3}=-1-\frac{1}{2}\ge c([v,3])-\frac{1}{2}$ ] . to finish the proof we consider case ( ii ) . here we have @xmath232)\ge c[(v,1])$ ] , @xmath233)\ge c[(v,2])$ ] , @xmath234)\ge c([v,4])$ ] , @xmath240)\ge c([v,3])$ ] , @xmath239)=\frac{1}{3}$ ] , @xmath237)=\frac{1}{3}$ ] , @xmath235)=\frac{1}{3}$ ] , and @xmath236)=\frac{1}{3}$ ] . thus it suffices to show @xmath241)+\omega_2([v,5])\ge c([v,5])-\frac{4}{3}.\ ] ] since @xmath242),\omega_2([v,5])\ge-\frac{5}{6}$ ] and @xmath243)\le -\frac{1}{2}$ ] this inequality is true . [ cor_non_negativ_all ] if @xmath227 , @xmath228 , @xmath244 are all prospective @xmath6-regular@xmath76-classes of a ( finite ) matchstick graph with maximum vertex degree at most @xmath6 we have @xmath245 ( 13,5 ) ( 0.5,0)(1,0)4 ( 0,0.5)(1,0)1 ( 2,0.5)(1,0)1 ( 4,0.5)(1,0)1 ( 0.5,1)(1,0)4 ( 0.5,2)(1,0)4 ( 0,2.5)(1,0)1 ( 2,2.5)(1,0)1 ( 4,2.5)(1,0)1 ( 0.5,3)(1,0)4 ( 0.5,0)(-1,1)0.5 ( 0.5,0)(1,1)0.5 ( 1.5,0)(-1,1)0.5 ( 1.5,0)(1,1)0.5 ( 2.5,0)(-1,1)0.5 ( 2.5,0)(1,1)0.5 ( 3.5,0)(-1,1)0.5 ( 3.5,0)(1,1)0.5 ( 4.5,0)(-1,1)0.5 ( 4.5,0)(1,1)0.5 ( 0.5,1)(-1,-1)0.5 ( 0.5,1)(1,-1)0.5 ( 1.5,1)(-1,-1)0.5 ( 1.5,1)(1,-1)0.5 ( 2.5,1)(-1,-1)0.5 ( 2.5,1)(1,-1)0.5 ( 3.5,1)(-1,-1)0.5 ( 3.5,1)(1,-1)0.5 ( 4.5,1)(-1,-1)0.5 ( 4.5,1)(1,-1)0.5 ( 0.5,2)(-1,1)0.5 ( 0.5,2)(1,1)0.5 ( 1.5,2)(-1,1)0.5 ( 1.5,2)(1,1)0.5 ( 2.5,2)(-1,1)0.5 ( 2.5,2)(1,1)0.5 ( 3.5,2)(-1,1)0.5 ( 3.5,2)(1,1)0.5 ( 4.5,2)(-1,1)0.5 ( 4.5,2)(1,1)0.5 ( 0.5,3)(-1,-1)0.5 ( 0.5,3)(1,-1)0.5 ( 1.5,3)(-1,-1)0.5 ( 1.5,3)(1,-1)0.5 ( 2.5,3)(-1,-1)0.5 ( 2.5,3)(1,-1)0.5 ( 3.5,3)(-1,-1)0.5 ( 3.5,3)(1,-1)0.5 ( 4.5,3)(-1,-1)0.5 ( 4.5,3)(1,-1)0.5 ( 0.5,1)(0,1)1 ( 1.5,1)(0,1)1 ( 2.5,1)(0,1)1 ( 3.5,1)(0,1)1 ( 4.5,1)(0,1)1 ( 0.5,0 ) ( 1.5,0 ) ( 2.5,0 ) ( 3.5,0 ) ( 4.5,0 ) ( 0.5,1 ) ( 1.5,1 ) ( 2.5,1 ) ( 3.5,1 ) ( 4.5,1 ) ( 0.5,2 ) ( 1.5,2 ) ( 2.5,2 ) ( 3.5,2 ) ( 4.5,2 ) ( 0.5,3 ) ( 1.5,3 ) ( 2.5,3 ) ( 3.5,3 ) ( 4.5,3 ) ( 0,0.5 ) ( 1,0.5 ) ( 2,0.5 ) ( 3,0.5 ) ( 4,0.5 ) ( 5,0.5 ) ( 0,2.5 ) ( 1,2.5 ) ( 2,2.5 ) ( 3,2.5 ) ( 4,2.5 ) ( 5,2.5 ) ( 4.5,0)(1,0)4 ( 4,0.5)(1,0)1 ( 6,0.5)(1,0)1 ( 8,0.5)(1,0)1 ( 4.5,1)(1,0)4 ( 4.5,2)(1,0)4 ( 4,2.5)(1,0)1 ( 6,2.5)(1,0)1 ( 8,2.5)(1,0)1 ( 4.5,3)(1,0)4 ( 4.5,0)(-1,1)0.5 ( 4.5,0)(1,1)0.5 ( 5.5,0)(-1,1)0.5 ( 5.5,0)(1,1)0.5 ( 6.5,0)(-1,1)0.5 ( 6.5,0)(1,1)0.5 ( 7.5,0)(-1,1)0.5 ( 7.5,0)(1,1)0.5 ( 8.5,0)(-1,1)0.5 ( 8.5,0)(1,1)0.5 ( 4.5,1)(-1,-1)0.5 ( 4.5,1)(1,-1)0.5 ( 5.5,1)(-1,-1)0.5 ( 5.5,1)(1,-1)0.5 ( 6.5,1)(-1,-1)0.5 ( 6.5,1)(1,-1)0.5 ( 7.5,1)(-1,-1)0.5 ( 7.5,1)(1,-1)0.5 ( 8.5,1)(-1,-1)0.5 ( 8.5,1)(1,-1)0.5 ( 4.5,2)(-1,1)0.5 ( 4.5,2)(1,1)0.5 ( 5.5,2)(-1,1)0.5 ( 5.5,2)(1,1)0.5 ( 6.5,2)(-1,1)0.5 ( 6.5,2)(1,1)0.5 ( 7.5,2)(-1,1)0.5 ( 7.5,2)(1,1)0.5 ( 8.5,2)(-1,1)0.5 ( 8.5,2)(1,1)0.5 ( 4.5,3)(-1,-1)0.5 ( 4.5,3)(1,-1)0.5 ( 5.5,3)(-1,-1)0.5 ( 5.5,3)(1,-1)0.5 ( 6.5,3)(-1,-1)0.5 ( 6.5,3)(1,-1)0.5 ( 7.5,3)(-1,-1)0.5 ( 7.5,3)(1,-1)0.5 ( 8.5,3)(-1,-1)0.5 ( 8.5,3)(1,-1)0.5 ( 4.5,1)(0,1)1 ( 5.5,1)(0,1)1 ( 6.5,1)(0,1)1 ( 7.5,1)(0,1)1 ( 8.5,1)(0,1)1 ( 4.5,0 ) ( 5.5,0 ) ( 6.5,0 ) ( 7.5,0 ) ( 8.5,0 ) ( 4.5,1 ) ( 5.5,1 ) ( 6.5,1 ) ( 7.5,1 ) ( 8.5,1 ) ( 4.5,2 ) ( 5.5,2 ) ( 6.5,2 ) ( 7.5,2 ) ( 8.5,2 ) ( 4.5,3 ) ( 5.5,3 ) ( 6.5,3 ) ( 7.5,3 ) ( 8.5,3 ) ( 4,0.5 ) ( 5,0.5 ) ( 6,0.5 ) ( 7,0.5 ) ( 8,0.5 ) ( 9,0.5 ) ( 4,2.5 ) ( 5,2.5 ) ( 6,2.5 ) ( 7,2.5 ) ( 8,2.5 ) ( 9,2.5 ) ( 8.5,0)(1,0)4 ( 8,0.5)(1,0)1 ( 10,0.5)(1,0)1 ( 12,0.5)(1,0)1 ( 8.5,1)(1,0)4 ( 8.5,2)(1,0)4 ( 8,2.5)(1,0)1 ( 10,2.5)(1,0)1 ( 12,2.5)(1,0)1 ( 8.5,3)(1,0)4 ( 8.5,0)(-1,1)0.5 ( 8.5,0)(1,1)0.5 ( 9.5,0)(-1,1)0.5 ( 9.5,0)(1,1)0.5 ( 10.5,0)(-1,1)0.5 ( 10.5,0)(1,1)0.5 ( 11.5,0)(-1,1)0.5 ( 11.5,0)(1,1)0.5 ( 12.5,0)(-1,1)0.5 ( 12.5,0)(1,1)0.5 ( 8.5,1)(-1,-1)0.5 ( 8.5,1)(1,-1)0.5 ( 9.5,1)(-1,-1)0.5 ( 9.5,1)(1,-1)0.5 ( 10.5,1)(-1,-1)0.5 ( 10.5,1)(1,-1)0.5 ( 11.5,1)(-1,-1)0.5 ( 11.5,1)(1,-1)0.5 ( 12.5,1)(-1,-1)0.5 ( 12.5,1)(1,-1)0.5 ( 8.5,2)(-1,1)0.5 ( 8.5,2)(1,1)0.5 ( 9.5,2)(-1,1)0.5 ( 9.5,2)(1,1)0.5 ( 10.5,2)(-1,1)0.5 ( 10.5,2)(1,1)0.5 ( 11.5,2)(-1,1)0.5 ( 11.5,2)(1,1)0.5 ( 12.5,2)(-1,1)0.5 ( 12.5,2)(1,1)0.5 ( 8.5,3)(-1,-1)0.5 ( 8.5,3)(1,-1)0.5 ( 9.5,3)(-1,-1)0.5 ( 9.5,3)(1,-1)0.5 ( 10.5,3)(-1,-1)0.5 ( 10.5,3)(1,-1)0.5 ( 11.5,3)(-1,-1)0.5 ( 11.5,3)(1,-1)0.5 ( 12.5,3)(-1,-1)0.5 ( 12.5,3)(1,-1)0.5 ( 8.5,1)(0,1)1 ( 9.5,1)(0,1)1 ( 10.5,1)(0,1)1 ( 11.5,1)(0,1)1 ( 12.5,1)(0,1)1 ( 8.5,0 ) ( 9.5,0 ) ( 10.5,0 ) ( 11.5,0 ) ( 12.5,0 ) ( 8.5,1 ) ( 9.5,1 ) ( 10.5,1 ) ( 11.5,1 ) ( 12.5,1 ) ( 8.5,2 ) ( 9.5,2 ) ( 10.5,2 ) ( 11.5,2 ) ( 12.5,2 ) ( 8.5,3 ) ( 9.5,3 ) ( 10.5,3 ) ( 11.5,3 ) ( 12.5,3 ) ( 8,0.5 ) ( 9,0.5 ) ( 10,0.5 ) ( 11,0.5 ) ( 12,0.5 ) ( 13,0.5 ) ( 8,2.5 ) ( 9,2.5 ) ( 10,2.5 ) ( 11,2.5 ) ( 12,2.5 ) ( 13,2.5 ) ( 0.5,3)(1,0)4 ( 0.5,4)(1,0)4 ( 0,4.5)(1,0)1 ( 2,4.5)(1,0)1 ( 4,4.5)(1,0)1 ( 0.5,5)(1,0)4 ( 0.5,3)(-1,-1)0.5 ( 0.5,3)(1,-1)0.5 ( 1.5,3)(-1,-1)0.5 ( 1.5,3)(1,-1)0.5 ( 2.5,3)(-1,-1)0.5 ( 2.5,3)(1,-1)0.5 ( 3.5,3)(-1,-1)0.5 ( 3.5,3)(1,-1)0.5 ( 4.5,3)(-1,-1)0.5 ( 4.5,3)(1,-1)0.5 ( 0.5,4)(-1,1)0.5 ( 0.5,4)(1,1)0.5 ( 1.5,4)(-1,1)0.5 ( 1.5,4)(1,1)0.5 ( 2.5,4)(-1,1)0.5 ( 2.5,4)(1,1)0.5 ( 3.5,4)(-1,1)0.5 ( 3.5,4)(1,1)0.5 ( 4.5,4)(-1,1)0.5 ( 4.5,4)(1,1)0.5 ( 0.5,5)(-1,-1)0.5 ( 0.5,5)(1,-1)0.5 ( 1.5,5)(-1,-1)0.5 ( 1.5,5)(1,-1)0.5 ( 2.5,5)(-1,-1)0.5 ( 2.5,5)(1,-1)0.5 ( 3.5,5)(-1,-1)0.5 ( 3.5,5)(1,-1)0.5 ( 4.5,5)(-1,-1)0.5 ( 4.5,5)(1,-1)0.5 ( 0.5,3)(0,1)1 ( 1.5,3)(0,1)1 ( 2.5,3)(0,1)1 ( 3.5,3)(0,1)1 ( 4.5,3)(0,1)1 ( 0.5,4 ) ( 1.5,4 ) ( 2.5,4 ) ( 3.5,4 ) ( 4.5,4 ) ( 0.5,5 ) ( 1.5,5 ) ( 2.5,5 ) ( 3.5,5 ) ( 4.5,5 ) ( 0,4.5 ) ( 1,4.5 ) ( 2,4.5 ) ( 3,4.5 ) ( 4,4.5 ) ( 5,4.5 ) ( 4.5,3)(1,0)4 ( 4.5,4)(1,0)4 ( 4,4.5)(1,0)1 ( 6,4.5)(1,0)1 ( 8,4.5)(1,0)1 ( 4.5,5)(1,0)4 ( 4.5,3)(-1,-1)0.5 ( 4.5,3)(1,-1)0.5 ( 5.5,3)(-1,-1)0.5 ( 5.5,3)(1,-1)0.5 ( 6.5,3)(-1,-1)0.5 ( 6.5,3)(1,-1)0.5 ( 7.5,3)(-1,-1)0.5 ( 7.5,3)(1,-1)0.5 ( 8.5,3)(-1,-1)0.5 ( 8.5,3)(1,-1)0.5 ( 4.5,4)(-1,1)0.5 ( 4.5,4)(1,1)0.5 ( 5.5,4)(-1,1)0.5 ( 5.5,4)(1,1)0.5 ( 6.5,4)(-1,1)0.5 ( 6.5,4)(1,1)0.5 ( 7.5,4)(-1,1)0.5 ( 7.5,4)(1,1)0.5 ( 8.5,4)(-1,1)0.5 ( 8.5,4)(1,1)0.5 ( 4.5,5)(-1,-1)0.5 ( 4.5,5)(1,-1)0.5 ( 5.5,5)(-1,-1)0.5 ( 5.5,5)(1,-1)0.5 ( 6.5,5)(-1,-1)0.5 ( 6.5,5)(1,-1)0.5 ( 7.5,5)(-1,-1)0.5 ( 7.5,5)(1,-1)0.5 ( 8.5,5)(-1,-1)0.5 ( 8.5,5)(1,-1)0.5 ( 4.5,3)(0,1)1 ( 5.5,3)(0,1)1 ( 6.5,3)(0,1)1 ( 7.5,3)(0,1)1 ( 8.5,3)(0,1)1 ( 4.5,4 ) ( 5.5,4 ) ( 6.5,4 ) ( 7.5,4 ) ( 8.5,4 ) ( 4.5,5 ) ( 5.5,5 ) ( 6.5,5 ) ( 7.5,5 ) ( 8.5,5 ) ( 4,4.5 ) ( 5,4.5 ) ( 6,4.5 ) ( 7,4.5 ) ( 8,4.5 ) ( 9,4.5 ) ( 8.5,3)(1,0)4 ( 8.5,4)(1,0)4 ( 8,4.5)(1,0)1 ( 10,4.5)(1,0)1 ( 12,4.5)(1,0)1 ( 8.5,5)(1,0)4 ( 8.5,3)(-1,-1)0.5 ( 8.5,3)(1,-1)0.5 ( 9.5,3)(-1,-1)0.5 ( 9.5,3)(1,-1)0.5 ( 10.5,3)(-1,-1)0.5 ( 10.5,3)(1,-1)0.5 ( 11.5,3)(-1,-1)0.5 ( 11.5,3)(1,-1)0.5 ( 12.5,3)(-1,-1)0.5 ( 12.5,3)(1,-1)0.5 ( 8.5,4)(-1,1)0.5 ( 8.5,4)(1,1)0.5 ( 9.5,4)(-1,1)0.5 ( 9.5,4)(1,1)0.5 ( 10.5,4)(-1,1)0.5 ( 10.5,4)(1,1)0.5 ( 11.5,4)(-1,1)0.5 ( 11.5,4)(1,1)0.5 ( 12.5,4)(-1,1)0.5 ( 12.5,4)(1,1)0.5 ( 8.5,5)(-1,-1)0.5 ( 8.5,5)(1,-1)0.5 ( 9.5,5)(-1,-1)0.5 ( 9.5,5)(1,-1)0.5 ( 10.5,5)(-1,-1)0.5 ( 10.5,5)(1,-1)0.5 ( 11.5,5)(-1,-1)0.5 ( 11.5,5)(1,-1)0.5 ( 12.5,5)(-1,-1)0.5 ( 12.5,5)(1,-1)0.5 ( 8.5,3)(0,1)1 ( 9.5,3)(0,1)1 ( 10.5,3)(0,1)1 ( 11.5,3)(0,1)1 ( 12.5,3)(0,1)1 ( 8.5,4 ) ( 9.5,4 ) ( 10.5,4 ) ( 11.5,4 ) ( 12.5,4 ) ( 8.5,5 ) ( 9.5,5 ) ( 10.5,5 ) ( 11.5,5 ) ( 12.5,5 ) ( 8,4.5 ) ( 9,4.5 ) ( 10,4.5 ) ( 11,4.5 ) ( 12,4.5 ) ( 13,4.5 ) in this paper we have proven that no finite @xmath6-regular matchstick graph exists . in figure [ fig_infinite ] we have depicted a fraction of an infinite @xmath6-regular matchstick graph . since we can shift single rows of such a construction there exists an uncountable number of these graphs ( even in the combinatorial sense ) . as mentioned by bojan mohar there are several further examples . starting with the infinite @xmath36-regular triangulation we can delete several vertices at different rows and obtain an example by suitably removing horizontal edges to enforce degree @xmath6 for all vertices . taking an arbitrary @xmath76-class and continueing the boundary vertices with trees gives another set of examples . we strongly believe that there exists a more general topological interpretation of equation ( [ eqn_paramater ] ) . we have simply discovered it going along the concept of a potential function in order to prove theorem [ thm_main ] . a. blokhuis , _ regular finite planar maps with equal edges _ , tech . report , university of technology , eindhoven , the netherlands , 1982 , retyped in 2009 - available at http://www.wm.uni-bayreuth.de/fileadmin/sascha/publikationen/aartb.pdf .

we investigate the states generated in continuous variable ( cv ) optical fields on operating them with a number - conserving operator of the type , formed by the generalised superposition of products of field annihilation ( ) and creation ( ) operators , with . such an operator is experimentally realizable and can be suitably manipulated to generate nonclassical optical states when applied on single- and two - mode coherent , thermal , and squeezed input states . at low intensities , these nonclassical states can interact with a secondary mode via a linear optical device to generate two - mode discrete entangled states , which can serve as a resource in quantum information protocols . the advantage of these operations are tested by applying the generated entangled states as quantum channels in cv quantum teleportation , under the braunstein and kimble protocol . we observe that , under these operations , while the average fidelity of cv teleportation is enhanced for the nonclassical channel formed using input squeezed states , it remains at the classical threshold for input coherent and thermal states . this is due to the fact that though these operations can introduce discrete entanglement in all input states , it enhances the einstein - podolosky - rosen ( epr ) correlations only in the nonclassical squeezed state inputs , leading to an advantage in cv teleportation . this shows that nonclassical optical states generated using the above operations on classical coherent and thermal state inputs are not resourceful for cv teleportation . this investigation could prove useful in efficient implementation of noisy non - gaussian channels , formed by linear operations , in future teleportation protocols . ( some figures may appear in colour only in the online journal ) _ keywords _ : continuous variable system , non - gaussian operation , entanglement , quantum teleportation , epr correlation

1312.6226

the wide use of gaussian and non - gaussian electromagnetic field states for various quantum tasks and protocols renders the generation and investigation of these quantum states an essential aspect of modern research . entangled states generated using both linear and nonlinear quantum optical operations have been extensively used in quantum information @xcite and communications @xcite applications . the accessibility of continuous variable ( cv ) processing in quantum information theory has led to successful implementation of novel quantum tasks such as quantum teleportation @xcite , quantum cryptography @xcite and quantum memory @xcite ( for reviews , see @xcite ) . a recent method of generating nonclassicality in electromagnetic field states is by the operation of photon addition ( @xmath2 ) @xcite and subtraction ( @xmath1 ) @xcite . a remarkable aspect of the non - gaussian photon addition and subtraction operations is the relative efficiency with which these operations can be experimentally realised using linear optical devices and parametric down - converters @xcite . these non - gaussian operations are known to enhance entanglement @xcite and quantum information protocols , such as quantum teleportation @xcite . an extension of these non - gaussian operations , such as photon addition followed by photon subtraction or vice - versa , has also been implemented to show improvements in quantum correlations and other quantum tasks @xcite . the study of nonclassical optical states generated by such non - gaussian operations has received considerable theoretical attention in quantum optics @xcite . further , the experimental implementation of these operations has been used to prove the canonical commutation relation @xcite . a useful operation based on the non - gaussian photon - addition and subtraction protocols is the number - conserving generalised superposition of products ( gsp ) of field annihilation ( @xmath1 ) and creation ( @xmath2 ) operators of the form @xmath0 , with @xmath3 . such an operation is a generalization of the experimental scheme proposed to study the bosonic canonical commutation in optical fields @xcite . using an analysis of quasi - probability functions , quadrature squeezing and sub - poissonian statistics , we have recently shown that such gsp operations introduces nonclassicality in single - mode coherent and thermal states @xcite . the experimental scheme to implement the gsp operation using linear optical devices and parametric down - converters has been illustrated and mathematically explained in ref . however , the feasibility of such a scheme is reliant on the efficiency with which photon addition and subtraction can be experimentally achieved . in particular , it is limited by the low efficiency with which a single photon can be detected in the relevant modes @xcite ( cf . @xcite ) . it is known that gsp operations can be suitably manipulated to generate nonclassical output states when applied on single- and two - mode coherent , thermal and squeezed vacuum states @xcite . hence , the primary motivation of the present article is to analyze the nonclassicality introduced by the gsp operations from a quantum information perspective and investigate if the generated nonclassicality can enhance the performance of specific quantum information tasks . these nonclassical optical states , in mode _ a _ , can then be converted into two - mode entangled states by interacting with a secondary mode _ b _ , via a linear optical device such as a 50:50 beam - splitter . for a single - mode input state , the secondary mode is the vacuum , while for a two - mode input state , one mode serves as _ a _ and the other mode as _ b _ ( see figure [ fig1 ] ) . we note that the generated entanglement is characteristic of the nonclassicality of the gsp operated state , as beam - splitters do not generate entanglement for classical inputs @xcite . thus the generated entanglement is an indicator of the resource obtained due to the gsp operations . the suitability of the two - mode entangled state is investigated for potential use as a quantum channel in a cv quantum teleportation , under the braunstein and kimble ( bk ) protocol @xcite . the success of the protocol is evaluated in terms of the average fidelity obtained in teleporting a single - mode optical state . the efficiency of the quantum channel for cv teleportation can be justified in terms of the phase - quadrature or the einstein - podolsky - rosen ( epr ) correlations present in the quantum channels . the main results obtained in our investigation can be summarized in the following points : 1 ) at low field intensities , the gsp operation can be suitably tuned to generate highly entangled two - mode discrete level optical states for both single- and two - mode coherent , thermal and squeezed input states . this is a signature of the nonclassicality generated by the gsp operation on these input states . 2 ) these two - mode entangled states can be used as quantum channels in cv quantum teleportation of single - mode coherent and squeezed states , under the bk protocol . 3 ) we observe that while the average fidelity of cv teleportation is enhanced for the squeezed input states under gsp operations , the fidelity does not rise above the classical threshold for coherent and thermal inputs . this shows that the generated nonclassicality in coherent and thermal states , under gsp operations , is not resourceful for cv teleportation . 4 ) investigating further , we observe that the gsp operations do not enhance the nonclassicality of coherent and thermal input states in terms its phase - quadrature or the epr correlations , which is crucial for gaining an advantage in cv teleportation . interestingly , the above results show that though gsp operations introduce nonclassicality and discrete level entanglement for all input states , the injected nonclassicality is not resourceful for cv teleportation when the input state is coherent or thermal , as it fails to enhance the epr correlations . the real advantage is observed only when the input state is already nonclassical , such as the squeezed state . this result is important from the consideration of noisy quantum channels , where classical noise can be modelled using coherent or thermal states . the paper is organised as follows . the introduction in section [ ss1 ] is followed by a discussion on the characterization of nonclassicality and two - mode entanglement in section [ ent ] . we then discuss the state generation protocol for single - mode ( section [ ss2 ] ) and two - mode ( section [ ss3 ] ) coherent , thermal and squeezed state inputs . we investigate the entanglement properties in section [ ss5 ] followed by the investigation of average teleportation fidelities in section [ ss6 ] . in section [ sec - epr ] , we look at the epr correlations of the quantum channels , and end with a discussion of the results in section [ discuss ] . the action of number - conserving operators formed by the gsp of field annihilation ( @xmath1 ) and creation ( @xmath2 ) operators of the type @xmath0 , with @xmath3 , on field modes introduces distinct nonclassicality in the operated states @xcite . the nonclassicality in these cases is characterized by the indicators such as negative quasiprobability distributions , sub - poissonian statistics or quadrature squeezing . under such a description , coherent and thermal states can be deemed as classical , wheras squeezed vacuum states are highly nonclassical . from the perspective of quantum information theory , the nonclassicality of the gsp operated state can be investigated in terms of the two - mode entanglement that develops upon interaction of the modes via some linear optical device . nonclassical states upon interacting with a secondary mode via a beam splitter will develop finite entanglement @xcite . for example , non - gaussian operations based on photon addition and subtraction are known to enhance entanglement in squeezed vacuum states @xcite . the protocol adopted to measure the nonclassicality using entanglement is different for the case of single - mode and two - mode inputs . for a single - mode gsp operated state , in mode _ a _ , we interact the single - mode output with a vacuum in mode _ b _ , via a 50:50 beam splitter to generate a two - mode output state . a schematic for the gsp protocol and the beam - splitter interaction for single - mode input states is given in the upper row of figure [ fig1 ] . the entanglement in this two - mode output state is measured using logarithmic negativity ( ln ) @xcite , which gives us a measure of the entanglement capacity or potential @xcite of the gsp operated state . we call this measure entanglement capacity ( ec ) , as it measures the capacity of the nonclassical gsp operated state to create entanglement via a linear optical device , and is thus a measure of the nonclassicality of the gsp operated state . the ec is zero in cases where a classical input state interacts with the vacuum mode via a beam - splitter @xcite . for two - mode input states which are initially `` classical '' , such as the coherent and thermal states , the gsp operation is applied separately on each mode . this operation does not introduce any entanglement in the two - mode output since the gsp operations are local . however , the two gsp operated modes are individually nonclassical and this can again be converted into two - mode entanglement using a linear optical device , which introduces a nonlocal interaction . to calculate the ec for the two - mode gsp operated state , we allow the two modes to interact via a 50:50 beam splitter to generate entangled two - mode output states . the ec is then calculated using ln . for two - mode input states which are initially entangled , such as a two - mode squeezed state , the gsp operation is globally applied to the two modes without further interaction between the modes . a schematic for the complete protocol for two - mode input states is given in the lower row of figure [ fig1 ] . we briefly describe the entanglement measure called logarithmic negativity that is used to characterize the entanglement capacity . ln is defined @xcite on the known premise that the negativity of the partial transpose is a sufficient condition for a bipartite quantum state to be entangled . this is the peres - horodecki separability criterion @xcite . for low - dimensional bipartite systems ( qubit - qubit and qubit - qutrit ) the negativity of partial transpose is not only sufficient but also a necessary condition for bipartite entanglement @xcite . to calculate ln , we evaluate a quantity called `` negativity '' , @xmath4 , where @xmath5 is the trace norm of the partial transpose @xmath6 , which is positive for all non - entangled states . hence @xmath7 is zero for separable states . ln of @xmath8 is defined as @xmath9 . \label{logneg}\ ] ] we use ln as the measure of two - mode entanglement to calculate ec for all gsp operated states . in this section , we consider the number - conserving generalized superposition operation of the form @xmath0 , with @xmath3 , acting on single- and two - mode coherent , thermal and squeezed vacuum states . we mathematically derive the relations to convert the generated nonclassical state into entangled two - mode states , by using a linear optical device , such as a 50:50 beam - splitter , as per the protocol formalised in section [ ent ] . the final output state obtained is then varied with the system parameter @xmath10 , to analyze the entanglement capacity of these states and their utility in cv quantum teleportation in the following sections . _ single - mode coherent state _ : let us consider the gsp operation acting on a single - mode coherent state @xmath11 [ figure [ fig1](a ) , top row ] . the single - mode ( in mode @xmath12 ) gsp operated coherent state is given by @xmath13 where @xmath14 is the normalization constant . in terms of the displacement operator , @xmath15 , a coherent state can be written as @xmath16 @xcite . hence we can write relation ( [ 1 ] ) as @xmath17 ) , \label{2}\end{aligned}\ ] ] where we have used the relation @xmath18 . the single - mode gsp operated coherent state is then interacted with a vacuum mode via a 50:50 beam splitter . the two - mode output state after the interaction has the form @xmath19 , where @xmath20 is the beam - splitter operator with input ports @xmath12 and @xmath21 [ figure [ fig1](a ) , top row ] . using relation ( [ 2 ] ) , we obtain @xmath22 , \label{5}\end{aligned}\ ] ] where @xmath23 and @xmath24 . using the fact that @xmath25 = @xmath26 , and the beam - splitter relations , @xmath27 = @xmath28 and @xmath29 = @xmath30 , we obtain @xmath31|0\rangle_a|0\rangle_b \nonumber \\ & = & \mathcal{a}_{ab } \left [ p_1 |00\rangle_{ab } + \frac{p_2}{\sqrt{2}}(|10\rangle_{ab } - |01\rangle_{ab } ) \right ] , \label{6}\end{aligned}\ ] ] where @xmath32=@xmath33 , @xmath34=@xmath35 . the density matrix for the two - mode state given by relation ( [ 6 ] ) is @xmath36 , where @xmath37 is given by @xmath38 the operator @xmath32 acts locally on the modes @xmath12 and @xmath21 , and hence can not change the entanglement properties of the two - qubit density matrix ( [ 7 ] ) . this reduces an infinite - dimensional two - mode interaction ( in an infinite dimensional fock state basis ) to a discrete two - qubit problem ( in the \{@xmath39 } basis ) . hence , the entanglement capacity can simply be calculated by computing the ln of the smaller two - qubit state . hence , the gsp operation generates a nonclassical two - mode output state that can be used in possible quantum protocols . + _ single - mode thermal state _ : a similar approach can be taken if the single - mode input is a classical gaussian thermal state . the thermal state is a maximally mixed state in the fock state basis , and the thermal density matrix can be represented as @xmath40 where @xmath41 is the average photon number . the action of gsp on the thermal field density matrix gives us the following operated state @xcite : @xmath42 where @xmath43 is the normalization constant . the gsp operated thermal state ( [ 9 ] ) is thus a mixed state in the infinite dimensional basis . it is known that the gsp operated single - mode thermal state has nonclassicality introduced by the operation and is non - gaussian @xcite . the gsp operated state ( in mode @xmath12 ) is interacted with a vacuum state in mode @xmath21 via a 50:50 beam splitter [ figure [ fig1](b ) , top row ] to generate a nonclassical two - mode output state . the operated two - mode state is of the form @xmath44 where @xmath45 . the infinite - dimensional interaction between the two modes can be reduced by considering a truncated thermal state input with a low average photon number . for @xmath46 0.1 , the two - mode density matrix ( [ 10 ] ) can be written as @xmath47 where only the first three terms ( @xmath48 0 , 1 , 2 ) in the summation have been retained after truncation . applying the unitary operation @xmath49 , we get the final two - mode output state as @xmath50 the final density matrix can be written as @xmath51 the low average photon number ensures that the density matrix is effectively truncated to a discrete @xmath52-dimensional two - mode system . hence , the gsp operation generates a discrete two - mode nonclassical state . + _ single - mode squeezed state _ : the gsp operation and subsequent conversion to a two - mode entangled state for a single - mode squeezed state are similar to the approach adopted for coherent and thermal input states . however , the squeezed state is not a classical input state unlike the cases of coherent and thermal inputs . the single - mode ( in mode @xmath12 ) squeezed state can be written as @xmath53|0\rangle_a \nonumber \\ & = & ( 1-\lambda^2)^{1/4}\sum_{n=0}^\infty \frac{\sqrt{(2n)!}}{n!}\left(-\frac{\lambda}{2}\right)^n|2n\rangle_a , \label{sq1.1}\end{aligned}\ ] ] where @xmath54 is the squeezing operator , @xmath55 is the squeezing parameter , and @xmath56 . the gsp operated squeezed state [ figure [ fig1](c ) , top row ] is given by @xmath57 where @xmath14 is the normalization constant . the gsp operated single - mode squeezed state ( in mode @xmath12 ) is interacted with a vacuum state ( in mode @xmath21 ) via a 50:50 beam splitter . the resulting two - mode state is given by @xmath58 where @xmath59 is the vacuum mode , @xmath49 is the beam - splitter operator , and @xmath60 . the infinite - dimensional two - mode output can be truncated by considering regimes where @xmath61 1 . in the low @xmath62 regime , the gsp operated single - mode squeezed state can be written as @xmath63 where @xmath64 , and @xmath65 . applying the beam - splitter operation , we get @xmath66 the output two - mode density matrix is given by @xmath67 . the gsp operated two - mode output can be used as a quantum channel in cv teleportation . _ two - mode coherent state _ : for a classical coherent state input , the gsp operation is applied separately on the two modes @xmath12 and @xmath21 [ figure [ fig1](a ) , bottom row ] . each mode operation is similar to the case for a single - mode input . the two modes than interact via the beam - splitter . the single - mode gsp operated input coherent state is given by relation ( [ 2 ] ) . the beam - splitter interaction for the two modes can then be written as @xmath68 , where @xmath69 and @xmath70 are the two input coherent states . the output state can then be written as @xmath71 , \nonumber\\ & = & \mathcal{a}_{ab}^{local}\left [ p'_1 |0\rangle_{a}|0\rangle_{b } + p'_2 ~\hat{\mathcal{b}}_{ab}\hat{b}^\dag \hat{\mathcal{b}}_{ab}^\dag |0 \rangle_a |0 \rangle_b \right . + p'_3 \hat{\mathcal{b}}_{ab}\hat{a}^\dag \hat{\mathcal{b}}_{ab}^\dag |0 \rangle_a |0 \rangle_b \nonumber\\ & & \left . + p'_4 ~\hat{\mathcal{b}}_{ab}\hat{a}^\dag\hat{b}^\dag \hat{\mathcal{b}}_{ab}^\dag |0 \rangle_a |0 \rangle_b \right ] , \label{14}\end{aligned}\ ] ] where @xmath72 and @xmath73 hence , applying the beam - splitter operation on equation ( [ 14 ] ) , we obtain @xmath74|00\rangle_{ab } \nonumber\\ \nonumber\\ & = & \mathcal{a}_{ab}^{local}\left[p''_1|00\rangle_{ab } + p''_2 |01\rangle_{ab } + p''_3|10\rangle_{ab } \right . \nonumber\\ \nonumber\\ & & - \left . p''_4 |02\rangle_{ab } + p''_4|20\rangle_{ab}\right ] , \label{16}\end{aligned}\ ] ] where @xmath75 , @xmath76 , @xmath77 and @xmath78 . the two - mode output density matrix can be written in the form @xmath79 . hence , the entanglement between the two modes of the output is encoded in the density matrix @xmath37 since the operation @xmath80 only acts locally on the two modes . + _ two - mode thermal state _ : the two - mode gsp operated thermal input state can be calculated using an approach similar to that for the two - mode input coherent state . the two modes are operated separately and then interact via a 50:50 beam - splitter [ figure [ fig1](b ) , bottom row ] . the single - mode gsp operated thermal state can be written in a form similar to relations ( [ 9 ] ) , ( [ 10 ] ) . the action of the beam - splitter can be represented as the following : @xmath81 where @xmath82 corresponds to the two input thermal modes @xmath83 . we have used the low average photon number approximation and retained terms up to @xmath84 2 . since the most of the terms in the gsp operated density matrix are product of the probabilities from the single - mode gsp operated thermal state , the smaller probabilities can be neglected for low average fields . the two - mode output state generated , using ( [ 17 ] ) , is of the form @xmath85 expanding the above relation ( [ 18 ] ) by applying the beam - splitter operation , we obtain @xmath86 on solving , we get @xmath87 where @xmath88 , @xmath89 , @xmath90 , and @xmath91 . + the nonclassical properties of the two - mode output gsp operated thermal state can be obtained using the density matrix ( @xmath92 ) in relation ( [ 20 ] ) . + _ two - mode squeezed state _ : the two - mode squeezed state is a well - known gaussian state that can be experimentally prepared using any pure gaussian state upon unitary gaussian operations . the initial squeezed state is an entangled two - mode state and hence it is not necessary to interact the modes via any active linear optical media as in the case of coherent or thermal input states . the two - mode squeezed state is an infinite dimensional superposition state of the two modes . the two - mode squeezed state can be written as , @xmath93 where @xmath94 is the squeezing parameter and @xmath62 = @xmath95 . applying the gsp operation on the two - mode squeezed state [ figure [ fig1](c ) , bottom row ] , we obtain @xmath96 , \label{22}\end{aligned}\ ] ] where @xmath14 is the normalization constant . the infinite dimensional two - mode squeezed state can be reduced by considering the low squeezing region . for @xmath61 1 , the gsp operated squeezed state reduces to @xmath97|1\rangle_a|1\rangle_b \right ] . \label{23}\end{aligned}\ ] ] the gsp operated two - mode squeezed state in relation ( [ 23 ] ) gives us the two - mode output pure state density matrix ( @xmath98 ) that can be used to study its nonclassical properties . [ ent1 ] [ ent2 ] as discussed in sections [ ss2 ] and [ ss3 ] , using a low average field - intensity regime , the infinite - dimensional output optical state can be mapped to a discrete - level optical state . the entanglement properties of the single- and two - mode gsp operated input states is characterized using the entanglement capacity discussed in section [ ent ] , where logarithmic negativity ( [ logneg ] ) is used as the measure of two - mode entanglement . the parameters that control the ec of the final two - mode states are the parameter @xmath10 of the gsp operator and the initial field parameters of the input states , such as amplitude of the coherent state , average photon number of the thermal state and squeezing parameter of the squeezed state . + _ single - mode states _ : the variation of ec with gsp operation parameter @xmath10 and field parameters for the single - mode input states is given in figure [ ent1 ] . in figure [ ent1](a ) , we observe that high ec is observed for lower values of @xmath10 and the amplitude ( @xmath99 ) of the coherent state input . the operation generates maximal entanglement at @xmath100 . for greater @xmath99 , the states have lower values of ec but are not very sensitive to variation in @xmath10 . we also observe that for the truncated input thermal state [ figure [ ent1](b ) ] , the ec is maximum at @xmath100 and steadily decreases with increase in @xmath10 . the maximum ec corresponds to the lowest average photon number ( @xmath101 ) . this is due to the fact that the classicality of coherent and thermal states increases with increase in average photon number . figure [ ent1](c ) gives us the variation of ec with @xmath102 , where @xmath55 is the squeezing parameter of the input single - mode squeezed state . the maximum ec is obtained at @xmath103 , and the ec decreases with increasing @xmath10 . the state is seen to be more entangled for higher values of @xmath62 for all values of @xmath10 away from @xmath10 = 0 . hence , the squeezed output with high @xmath62 is nonclassical for all ranges of the gsp operation . + _ two - mode states _ : the ec of the gsp operated two - mode coherent , thermal and squeezed input states is shown in figure [ ent2 ] . the variation of the ec is with respect to the two - mode gsp operation parameters @xmath104 and @xmath105 for fixed values of field parameters such as amplitude , average photon number and squeezing parameter for coherent , thermal and squeezed states , respectively . for coherent and thermal fields , we consider inputs with equal field parameters for both the input modes . figure [ ent2](a ) gives us the variation of ec with the two - mode gsp operator parameters @xmath104 and @xmath105 for the two - mode coherent state input with equal amplitudes , @xmath99 = @xmath106 = 0.5 . we observe that low values of either of the mode operators @xmath104 and @xmath105 are sufficient for obtaining entangled states . the only region that produces low entanglement is the region corresponding to simultaneous high values of both @xmath104 and @xmath105 marked by the dark region in the colour scheme of figure [ ent2](a ) . in contrast , the plot for the ec of two - mode thermal state input in figure [ ent2](b ) , with fixed photon number averages @xmath107 = @xmath108 = 0.05 , shows that regions of relatively high entanglement is restricted to a small region along the two axes that correspond to the small values of @xmath104 ( @xmath109 ) or @xmath105 ( @xmath109 ) . for other values of @xmath104 and @xmath105 , the output state has very low entanglement . the behaviour of ec for the two - mode squeezed state ( @xmath62 = 0.05 ) is completely opposite of that of the two - mode thermal state . figure [ ent2](c ) shows that the ec remains high at most values of @xmath104 and @xmath105 , dropping in the region of @xmath110 . the two - mode input squeezed state , unlike the two - mode coherent and thermal state , is entangled prior to the gsp operation . however , the gsp operation enhances the amount of entanglement @xcite over the range of parameters @xmath104 and @xmath105 . hence , it is observed that the gsp operation introduces nonclassicality in single- and two - mode coherent , thermal and squeezed states , which in turn can be characterized by high entanglement capacity . this clearly shows that gsp operation , in the low field - intensity limit and in suitable parameter regimes , can be used to generate highly entangled two - mode optical states , for all the three inputs considered in our investigation . to investigate the utility of the generated nonclassical and entangled two - mode states , under gsp operations , we use these states as quantum resources in specific quantum information protocols . in particular , we investigate the advantage obtained by using these two - mode entangled states as quantum channels in continuous variable quantum teleportation . quantum teleportation was originally devised as a quantum information protocol that involved sending an unknown qubit , from one party to another , across an entangled epr channel shared between the two parties @xcite . in the cv regime , the earliest example of teleportation was formulated for a one - dimensional phase - space particle @xcite . a more practical version of the cv quantum teleportation was devised by braunstein and kimble @xcite , who proposed the teleportation of quadrature components of an electromagnetic field using finite degrees of correlation . such cv quantum teleportation protocol is experimentally realizable @xcite . in this section , we investigate the resourcefulness of the two - mode entangled output states generated in sections [ ss2 ] and [ ss3 ] under gsp operations , in cv quantum teleportation . in particular , we explore whether the nonclassicality generated under the gsp operations is useful in providing a quantum advantage in specific information protocol such as teleportation . the success of the cv quantum teleportation , under the bk protocol , is indicated by the average fidelity ( @xmath111 ) between the final output and the initial input state being teleported . a genuine quantum advantage under the gsp operation will be observed if it can enhance the average fidelity of the cv teleportation protocol . under the bk protocol , perfect teleportation fidelity ( @xmath112 ) is obtained with an infinitely nonclassical channel such as the ideal epr entangled state @xcite . for non - ideal channels , the average fidelity is less than 1 and the maximum value attainable using a classical channel is @xmath113 @xcite . this is the maximum permissible fidelity without the use of any entangled quantum channels . an important figure of merit in cv teleportation is the no - cloning limit , which is equal to @xmath114 @xcite . to ensure that the teleported state is the best copy of the state remaining after the protocol and the nonclassical features of the input state have been teleported @xcite , the average fidelity must be greater than the no - cloning limit . hence , one can define a fidelity of cloning @xmath115 , which is an important benchmark in the success of the protocol . in the cv quantum teleportation formulated by bk , the two - mode quantum channel and the input state to be teleported are described by a joint wigner function @xmath116 . the wigner function description can be more clearly stated in terms of the symmetrically ordered characteristic function @xcite . the average fidelity of the quantum teleportation is @xcite @xmath117 where ( from @xcite ) @xmath118 @xmath119 and @xmath120 being the characteristic functions of the input state to be teleported and the two - mode quantum channel , respectively . hence , an important part of the calculation of the average fidelity of cv teleportation is to obtain the characteristic function of the state to be teleported and the two - mode quantum channel . in our work , we use the gsp operated , two - mode entangled states generated in sections [ ss2 ] and [ ss3 ] , as the quantum channel in the teleportation protocol . in the following , we show how the characteristic functions for the gsp operated single- and two - mode coherent , thermal and squeezed states can be derived in order to calculate the average fidelity of teleportation . + [ fid1 ] _ single - mode states _ : we have considered three different single - mode input states for the gsp operation , namely , coherent , thermal and squeezed states . these input states undergo gsp operation and then interact via a symmetric beam - splitter with a vacuum mode to generate a two - mode bipartite entangled output state , as shown in section [ ss2 ] . the generated two - mode nonclassical states are used as a quantum channel in cv teleportation . for a single - mode gsp operated coherent state coupled to a vacuum mode , the final two - mode nonclassical state can be written as @xmath121 the characteristic function of the state ( [ 27 ] ) is given by @xmath122 \exp(\alpha^*x-\alpha x ) , \label{28}\end{aligned}\ ] ] where @xmath123 and @xmath124 . @xmath14 is the normalization constant given by @xmath125 $ ] . the average fidelity can be calculated using relations ( [ 24 ] ) and ( [ 25 ] ) provided the characteristic function of the state to be teleported is also known . for the gsp operated single - mode thermal state , the characteristic function can be calculated using the density matrix expression for the truncated gsp operated two - mode density matrix given by relation ( [ 13 ] ) . the characteristic function of any two - mode state with a density matrix @xmath126 is given by @xmath127 where @xmath128 is the displacement operator acting on mode @xmath12 . the detailed derivation of the characteristic function for any two - mode density matrix in the number state basis is shown in [ app1 ] . the exact analytic expression for the characteristic function of the gsp operated single - mode thermal state can be calculated using ( [ 29 ] ) . the average fidelity is then calculated using expressions ( [ 24 ] ) and ( [ 25 ] ) . the characteristic function of a single - mode gsp operated squeezed state interacting with a beam - splitter can be written as @xmath129 , \label{sqz1ch}\end{aligned}\ ] ] where @xmath130 and @xmath131 . @xmath132 . @xmath133 and @xmath134 . the normalization constant is @xmath135 . the symbols @xmath10 , @xmath136 and @xmath55 have the usual meanings as in section [ ss2 ] . the average fidelity can be calculated using the characteristic function ( [ sqz1ch ] ) . + _ two - mode states _ : now we consider three two - mode input states in our analysis . the first two are the two - mode gsp operated coherent and thermal input states interacting via a symmetric beam splitter to generate a bipartite entangled two - mode nonclassical output . the third is the gsp operated two - mode squeezed state . the generation of the nonclassical gsp operated states have been shown in section [ ss3 ] . the calculation of the characteristic function for the two - mode gsp operated input is more complicated than for the case of single - mode input interacting with vacuum mode . for a two - mode coherent input state , the derivation of the characteristic function of the gsp operated final bipartite state ( [ 16 ] ) is given in [ app2 ] . the average fidelity of the teleportation of a state across the nonclassical two - mode gsp operated coherent input can be calculated using relations ( [ 24 ] ) and ( [ 25 ] ) . the characteristic function of the two - mode gsp operated thermal state can be derived using the relation ( [ 29 ] ) . the complete steps for the derivation of the characteristic function is given in [ app1 ] . the density matrix elements in the number state representation required for calculating the characteristic function in ( [ 29 ] ) are given by the gsp operated two - mode output state given by relation ( [ 20 ] ) . the characteristic function of the final bipartite state ( [ 22 ] ) obtained by gsp operated two - mode squeezed state input is given in [ app3 ] . note that the truncated state ( valid for low field intensity ) is needed for the calculation of characteristic function only in the case of the thermal input state . the characteristic function of the nonclassical output states of the gsp operated single and two - mode input is used for calculating the average fidelity in teleporting a single - mode state , using these gsp generated states as quantum channels . in the following subsections , we consider the teleportation of two single - mode states : ( a ) single - mode coherent state , and ( b ) single - mode squeezed state . we consider the teleportation of a single - mode coherent state using the gsp operated two - mode states as a quantum teleportation channel , generated using single- and two - mode coherent , thermal and squeezed state , as shown in the previous sections . the teleportation is conducted and optimized using the bk protocol . the symmetrically ordered characteristic function of the input coherent state @xmath137 is @xmath138 \exp({\alpha^*\gamma-\alpha\gamma^ * } ) . \label{26}\ ] ] the suitability of a two - mode entangled quantum channel to successfully transport a single - mode state under the bk teleportation protocol can be measured by calculating the average fidelity using eqs.([24 ] ) and ( [ 25 ] ) . the fidelity of teleportation of a coherent state input under the bk protocol is independent of the amplitude of the state to be teleported . [ fid2cohthm ] let us consider the average fidelity of teleporting a single - mode coherent state via a two - mode output state generated from gsp operations on single - mode states . the characteristic function of the gsp operated single - mode coherent , thermal and squeezed input interacting with vacuum is given by eqs.([28 ] ) , ( [ 29 ] ) and ( [ sqz1ch ] ) , respectively . figure [ fid1 ] shows the variation of the average fidelity ( @xmath111 ) with changing field parameters of the quantum channel at different values of the gsp operation parameter @xmath10 . we observe in figure [ fid1](a ) , that for the gsp operated single - mode coherent state coupled with vacuum channel the maximum average fidelity is , @xmath139 0.5 . this value is the classical upper limit of teleportation , @xmath140 , with an unentangled source and is well below the threshold of cloning fidelity . the maximum fidelity is achieved at very low values of the channel field amplitude @xmath99 and is independent of the gsp operator parameter @xmath10 . interestingly , the decaying @xmath111 with increasing @xmath99 for the gsp operated state is always bounded above by that for the non - operated two - mode state . hence , the gsp operations reduce the teleportation fidelity under optimizations in the bk protocol . figure [ fid1](b ) gives us the fidelity for the gsp operated single - mode thermal state coupled with vacuum channel for varying average photon number . the maximum fidelity , @xmath139 0.5 , which , again , is the classical threshold . hence , the gsp operated single - mode coherent and thermal channels are not resourceful for quantum teleportation , even though they are entangled , as shown in section [ ss5 ] . similarly , the fidelity of the gsp operated thermal state is always bounded above by the fidelity of the non - operated thermal state . hence , the operation reduces the average fidelity in gsp operated thermal channels . the enhancement of average fidelity can be observed for the quantum channel generated from the gsp operated single - mode squeezed state . from figure [ fid1](c ) , we can observe that a high fidelity , @xmath139 0.816 , can be achieved for highly squeezed states . this value breaches the threshold for quantum cloning . in the low squeezing regime , gsp operation can highly enhance the average fidelity . for low squeezing parameters , @xmath141 0.2 , the gsp operated states at @xmath142 0.5 offer enhanced fidelity values , @xmath143 0.74 , which is again higher than the cloning fidelity , @xmath114 and also higher than the non - operated squeezed channel . hence , one observes that the nonclassical states generated using single - mode coherent and thermal states do not offer any quantum advantage in cv quantum teleportation , even though they are discrete level two - mode entangled . however , nonclassical states formed by gsp operation on initially nonclassical squeezed states offer a distinct advantage in enhancing average fidelity . let us now consider the average fidelity of teleporting a single - mode coherent state via a bipartite two - mode output generated from a gsp operated two - mode input state . the average fidelity ( @xmath111 ) for the two - mode gsp operated coherent and thermal state channel can be observed in figure [ fid2cohthm ] as a function of the field parameters @xmath99 and @xmath41 . we have considered symmetric field parameters for the two modes . the behaviour of the average fidelity for the two - mode case is similar to the results for the single - mode fidelity . for the two - mode gsp operated coherent and thermal input the maximum average fidelity is again @xmath139 0.5 . the two - mode gsp operated coherent and thermal state has a maximum fidelity attainable by a classical channel . however , the average fidelity is always bounded by the non - operated state . figure [ fid2sqz ] gives us the average fidelity ( @xmath111 ) for a two - mode gsp operated squeezed state channel with varying field parameter @xmath62 . @xmath62 = @xmath95 , where @xmath94 is the two - mode squeezing parameter . @xmath111 is calculated for different values of the gsp operator parameter @xmath10 . for @xmath144 0 and @xmath144 1 , the fidelity , @xmath111 , has an analytical solution : @xmath145 the maximum fidelity , @xmath146 = 1 is achieved for all values of @xmath10 at high squeezing ( @xmath62 = 1 ) . in the low - squeezing regime , for @xmath141 0.2 , the highest fidelity , @xmath147 0.76 is achieved at @xmath10 = 0.9 . at slightly higher squeezing , @xmath148 , the highest fidelity , @xmath147 0.91 is achieved at @xmath10 = 1.0 . figure [ fid2sqz3d ] shows the variation of average fidelity @xmath111 with the two - mode gsp operator parameters @xmath104 and @xmath105 applied on a two - mode squeezed state , with @xmath62 = 0.6 . the average fidelity is independent of the amplitude of the single - mode coherent state to be teleported . @xmath111 varies over a small range of values ( 0.790.85 ) for different values of @xmath104 and @xmath105 , with the higher range corresponding to higher values of the gsp operator parameters . hence , for a fixed squeezing , @xmath62 , the fidelity is not very sensitive to the operator parameters . hence , we once again observe that gsp operated coherent and thermal states offer no advantage in cv quantum teleportation , in contrast to the performance of the gsp operated squeezed states . we now consider the teleportation of single - mode squeezed vacuum state using the gsp operated two - mode output states discussed in section [ ss2 ] and [ ss3 ] as the teleportation channel . as for the single - mode coherent input , the teleportation is conducted using the bk protocol . the symmetrically ordered characteristic function of the input squeezed vacuum state , @xmath149|0\rangle$ ] is @xmath150 .\ ] ] for the teleportation of the single - mode squeezed state , we consider the single- and two - mode gsp operated coherent and squeezed state quantum channels using the bk protocol . the average fidelity of the bk protocol is not independent of the squeezing parameter @xmath151 of the squeezed state to be teleported . the fidelity for the coherent and thermal channel is very similar to the fidelity obtained for teleporting a single - mode coherent state . [ fidsqz1 ] figure [ fidsqz1 ] shows the average fidelity ( @xmath111 ) of teleporting a single - mode squeezed state , with squeezing parameter @xmath151=0.1 , using a gsp operated single - mode coherent and squeezed state input coupled with a vacuum mode . for a single - mode gsp operated coherent state , as shown in figure [ fidsqz1](a ) , the maximum average fidelity is @xmath139 0.5 , which is the classical threshold . the gsp operation again reduces the average fidelity of teleportation . figure [ fidsqz1](b ) shows the average fidelity of teleportation using a single - mode gsp operated squeezed state channel . the maximum average fidelity is @xmath146 = 0.842 , at @xmath62 = 1 , for highly squeezed channels . at low squeezing limits , the gsp operations enhance the average fidelity over the non - operated state . for @xmath141 0.2 , the highest fidelity is 0.752 for @xmath152 0.5 . figure [ fidsqz2 ] shows the average fidelity ( @xmath111 ) of teleporting a single - mode squeezed state , with squeezing parameter @xmath151 = 0.5 , using a gsp operated two - mode squeezed state input . since @xmath111 is dependent on the squeezing parameter @xmath151 of the state to be teleported , there is no analytical solutions for the fidelity . the maximum fidelity is @xmath146 = 1 , achieved for highly squeezed channels with @xmath153 1.0 . at the mid - squeezing range , in the vicinity of @xmath153 0.5 , we observe that all gsp operated states have enhanced average fidelity over the non - operated state , with the highest fidelity , @xmath111 = 0.785 , achieved at @xmath10 = 1.0 . the behaviour of the average fidelity for a two - mode gsp operated coherent channel is similar to the single - mode coherent channel . to summarize , we find that the gsp operation , while enhancing the teleportation fidelity for the nonclassical two - mode squeezed input state channel , reduces the fidelity for channels formed by gsp operated classical thermal and coherent input states , under the bk protocol . hence , the nonclassicality generated under gsp operations does not directly lead to any genuine advantage in quantum information protocols for all input states . in particular , nonclassical states generated from classical inputs , such as coherent and thermal , perform no better in teleportation than classical channels . this is a significant result showing that the nonclassicality under non - gaussian operations , such as the gsp , does not necessarily contribute to advantages in teleportation . another significant ramification is for gsp operations on noisy quantum channels , which can be modelled using a linear characteristic function of a low intensity thermal and an entangled quantum channel @xcite . using gsp operations , under the bk protocol , the average fidelity of teleportation under the quantum channel is enhanced while the noise is reduced . to analyze the results obtained for cv teleportation using the gsp operated nonclassical channels , we consider the epr correlations in these two - mode states . the nonclassicality of the state in terms of epr correlations will give us a better insight into the performance of these channels in the teleportation protocol , and in particular , it will clarify why certain gsp operated nonclassical states do not provide any quantum advantage in teleportation . the success of the cv teleportation protocol is dependent on the correlations developed between the phase - quadrature components of the gsp operated two - mode output states . these correlations also known as epr ( einstein - podolosky - rosen ) correlations can be quantified in terms of the variance of the differences of the quadrature components of the two modes . the quadrature operators are defined as @xmath154 for the two modes @xmath12 and @xmath21 , respectively . the epr correlation is then defined as the total variance of the operators @xmath155 and @xmath156 . the correlation for the epr state @xcite , which is a maximally entangled state , is @xmath157 = @xmath158 = 0 . for a two - mode vacuum state , the epr correlation is @xmath157 + @xmath158 = 2 . for states with classical epr correlation the variance difference is always greater than 2 . [ epr ] figure [ epr ] gives us the variance of the epr operator for the two - mode gsp operated coherent and squeezed state inputs . from the values of the variance , with changing field parameters , it is evident that the two - mode gsp operated coherent state is not epr correlated at any field value . the variance for the two - mode non - operated coherent state is constant at 2.0 . the epr correlation patterns throw light into the teleportation fidelity plots of the two - mode coherent state [ figure [ fid2cohthm](a ) and figure [ fidsqz1](a ) ] . the maximum fidelity achieved for low @xmath99 , @xmath146 = 0.5 , which is at the upper limit of teleportation using unentangled quantum channels ( @xmath140=@xmath159 ) . this is explained by the fact that the two - mode state is not epr correlated for all @xmath99 and has no quadrature entanglement . hence , the gsp operation is unable to introduce phase - quadrature entanglement in the gsp operated coherent state even though low intensity states are two - mode entangled in the number state basis . in comparison , figure [ epr](b ) shows that the two - mode squeezed state is strongly epr correlated , with the gsp operation enhancing the correlation . this results in high teleportation fidelity above the cloning fidelity limit even for relatively low squeezing values ( @xmath62 ) . the variances in the epr quadrature operators are equal and behave as @xmath160 $ ] , where @xmath94 is the squeezing parameter . hence , at maximum squeezing the gsp operated state is maximally epr correlated . hence , we observe that non - gaussian operations , such as the gsp operations , are able to introduce nonclassicality in optical states and thus create two - mode entanglement in the number state basis , in the low intensity regime , even when the initial states are coherent and thermal . however , these operations do not inject any entanglement in the phase quadrature or generate epr correlations , which are crucial to gain advantage in cv quantum information protocols . interestingly , for initial two - mode squeezed state inputs , which already contain significant quadrature correlations , the gsp operations are able to enhance epr correlations to offer advantage in information protocols such as cv teleportation . hence , the role of non - gaussian operations in enhancing quantum resources is subject to the class of input states considered in the protocol . we further note that the two - mode entanglement generated in low intensity fields , under gsp operations , may play a significant role in discrete level information protocols . in this paper , we have used a generalised superposition of products of annihilation and creation operators to obtain nonclassical continuous variable field states . we have applied the gsp operator along with linear optical interaction on single- and two - mode cv states such as coherent , thermal and squeezed states to obtain highly entangled two - mode output states . the primary findings of the paper can be efficiently summarised in two logical parts . the first is the generation of highly entangled , two - mode discrete level output states using the gsp operation . upon application of the gsp protocol , the classical cv inputs , such as coherent and thermal states , can be converted to entangled two - mode states , in the low field - intensity limit . manipulating the gsp parameter @xmath10 along with the field parameters , one can obtain highly entangled discrete photonic systems . entangled photons are useful in a wide range of applications in quantum optics and are of fundamental importance in practical implementation of discrete level quantum information theory . the second aspect of the study is the application of the gsp operated states as quantum channels in cv quantum teleportation . we observe that the gsp operations enhance average teleportation fidelity of teleporting single - mode coherent and squeezed states using two - mode entangled channels obtained from single- and two - mode squeezed state inputs . the average fidelity obtained is higher than the fidelity of cloning even at low squeezing regime and betters the performance of the non - operated two - mode squeezed channel . the optimised gsp operation outperforms most non - gaussian operations with regard to fidelity enhancement in cv teleportation . interestingly , the gsp operation does not enhance the classical fidelity when the two - mode gsp operated thermal and coherent channels are used . the maximum fidelity possible using the coherent and thermal channels is the classical fidelity that is achieved without any entangled resources . hence , the nonclassical states introduced by gsp operations on coherent and thermal states are not resourceful for cv teleportation . this proves that not all two - mode entangled states generated under gsp operations are useful for cv information protocols . we show that this is due to the fact that gsp operations do not inject any nonclassicality in quadrature or epr correlations , thus rendering these states useless for cv information protocols . the results obtained may have interesting ramifications on noisy quantum channels . such channels can be modelled by a linear characteristic function of a low intensity thermal state and an entangled two - mode squeezed channel . for the gsp operations considered in our study , and restrictive optimizations under the bk protocol , the average fidelity of teleportation with a squeezed channel is enhanced while that with a noisy thermal state is reduced . this could lead to interesting interplay between nonclassicality and noise in determining the overall teleportation fidelity . the experimental advances in the generation of photon - added and photon - subtracted field states ensure that the advantages of non - gaussian operations will be put to further applications in the future . in the light of such developments the importance of theoretical studies on non - gaussian states is highly relevant . hsd thanks university grants commission , government of india for support under the senior research fellowship scheme . ac thanks serb , department of science and technology , government of india . the general expression of a symmetrically ordered characteristic function for a density matrix ( @xmath126 ) in terms of its number state matrix elements , for a single - mode state , is given by @xcite : @xmath161 generalizing the above relation for a two - mode density matrix , we obtain @xmath162 to obtain the characteristic function , one needs to derive the inner product @xmath163 . the general expression can be written as @xmath164\langle n , m|\exp(\xi\hat{a}^\dag)\exp(-\xi^*\hat{a})\exp(\eta\hat{b}^\dag)\exp(-\eta^*\hat{b})|n - l , m - l'\rangle \nonumber\\ & = & \exp[-\frac{1}{2}(|\xi|^2+|\eta|^2)]\langle n , m| \exp(\xi\hat{a}^\dag)\exp(-\xi^*\hat{a})\exp(\eta\hat{b}^\dag)\exp(-\eta^*\hat{b})\hat{a}^l\hat{b}^{l'}|n , m \rangle \nonumber\\ & & \times \sqrt{\frac{(n - l)!(m - l')!}{n!~m ! } } \nonumber\\ & = & \exp[-\frac{1}{2}(|\xi|^2+|\eta|^2)]\sqrt{\frac{(n - l)!(m - l')!}{n!~m!}}\left(-\frac{\partial}{\partial\xi^*}\right)^l\left(-\frac{\partial}{\partial\eta^*}\right)^{l ' } \nonumber\\ & & \times\langle n , m|\exp(\xi\hat{a}^\dag)\exp(-\xi^*\hat{a})\exp(\eta\hat{b}^\dag)\exp(-\eta^*\hat{b})|n , m\rangle \nonumber\\ & = & ~ \exp[-\frac{1}{2}(|\xi|^2+|\eta|^2)]\sqrt{\frac{(n - l)!(m - l')!}{n!~m!}}(-\xi)^l ( -\eta)^{l'}\left(-\frac{\partial}{\partial|\xi|^2}\right)^l\left(-\frac{\partial}{\partial|\eta|^2}\right)^{l'}\nonumber\\ & & \times \langle n , m|\exp(\xi\hat{a}^\dag)\exp(-\xi^*\hat{a})\exp(\eta\hat{b}^\dag)\exp(-\eta^*\hat{b})|n , m\rangle.\ \label{a3}\end{aligned}\ ] ] we now need to evaluate the term @xmath165 for all @xmath166 , @xmath167 . @xmath168 where @xmath169 are the laguerre polynomials . using relations ( [ a3 ] ) and ( [ a4 ] ) , we get @xmath170\sqrt{\frac{(n - l)!(m - l')!}{n!~m ! } } \nonumber\\ & \times & ( -\xi)^l ( -\eta)^{l'}\left(-\frac{\partial}{\partial|\xi|^2}\right)^l\left(-\frac{\partial}{\partial|\eta|^2}\right)^{l'}l_n(|\xi|^2 ) l_m(|\eta|^2 ) . \label{a5}\end{aligned}\ ] ] further , @xmath171 where @xmath172 is the generalised or the associated laguerre polynomials . hence , using relation ( [ a6 ] ) in ( [ a2 ] ) , we can obtain the characteristic function of a two - mode state , provided its number state density matrix elements are known . this relation can be useful in cases where truncated number states are considered to represent the field . the gsp operated two - mode coherent state interacting via a beam splitter can be written in the following way : @xmath173 the characteristic function can be written as @xmath174 an important step in the calculation is the evaluation of the following term : @xmath175 where we have used the properties of the beam - splitter and the displacement operator . the relation can be further simplified using the following identities : @xmath176 , \nonumber\\ d_b\left(\frac{\xi}{2}\right)d_b\left(\frac{\eta}{2}\right ) & = & d_b\left(\frac{\xi+\eta}{2}\right)\exp[\frac{1}{4}(\xi\eta^*-\xi^*\eta ) ] .\end{aligned}\ ] ] using ( [ b4 ] ) in relation ( [ b3 ] ) , we get @xmath177\exp[-\frac{1}{4}((\xi+\eta)^*\beta-(\xi+\eta)\beta^ * ) ] \nonumber\\ & & \times \langle \alpha,\beta|\frac{\xi-\eta}{2}+\alpha,\frac{\xi+\eta}{2}+\beta\rangle \nonumber\\ & = & \exp[-\frac{1}{2}(x^*\alpha - x\alpha^*)]\exp[-\frac{1}{2}(y^*\beta - y\beta^*)]\langle \alpha,\beta|x+\alpha , y+\beta\rangle \nonumber\\ & = & \exp[-\frac{1}{2}(|x|^2+|y|^2)]\exp(\alpha^*x-\alpha x^*)\exp(\beta^*y-\beta y^ * ) , \label{b5}\end{aligned}\ ] ] where @xmath178 , and @xmath179 . let @xmath180= @xmath181 . proceeding as in ( [ b5 ] ) , we can derive the following relations : @xmath182 let @xmath183 ( @xmath184 = 1 , 2 , 3 ) be the inner products corresponding to @xmath185 ( similar to @xmath186 in ( [ b6 ] ) ) : @xmath187 using relations ( [ b5 ] ) , ( [ b6 ] ) , and ( [ b7 ] ) , we can write the normalised characteristic function for a gsp operated two - mode coherent state interacting via a symmetric beam - splitter as @xmath188 where @xmath189 is the normalization constant given by + + @xmath190 the gsp operated two - mode squeezed state can be written in the following way : @xmath191 where @xmath192 $ ] is the two - mode squeezing operator , and @xmath94 is the squeezing parameter . the characteristic function can be written as @xmath193 we first calculate the following term : @xmath194\exp[\eta\hat{b}^\dag-\eta^*\hat{b } ] \hat{s}(r)|00\rangle , \label{c3}\end{aligned}\ ] ] where we have dropped the mode indices @xmath12 , @xmath21 from the states , and used the definition of the displacement operator . using the fact that @xmath195 and expanding relation ( [ c3 ] ) , we get @xmath196\exp[\eta\hat{b}^\dag-\eta^*\hat{b } ] \hat{s}(r)|00\rangle \nonumber\\ & = & \langle 00| \exp [ \xi(\cosh r~\hat{a}^\dag+\sinh r~\hat{b})-\xi^*(\cosh r~\hat{a}+\sinh r~\hat{b}^\dag ) ] \nonumber\\ & & \times \exp[\eta(\cosh r~\hat{b}^\dag+\sinh r~\hat{a})-\eta^*(\cosh r~\hat{b}+\sinh r~\hat{a}^\dag)]|00\rangle \nonumber\\ & = & \langle 00| d_a(\xi')d_b(\eta')|00\rangle\nonumber\\ & = & \exp[-\frac{1}{2}(|\xi'|^2+|\eta'|^2 ) ] , \label{c5}\end{aligned}\ ] ] where @xmath197 , and @xmath198 . let @xmath199 . the characteristic function of the gsp operated two - mode squeezed state can be calculated using the results ( [ c4 ] ) and ( [ c5 ] ) . let us consider some general terms to be calculated to obtain the characteristic function . @xmath200 in terms of the relations ( [ c6 ] ) , the characteristic function can be written as @xmath201 where @xmath189 is the normalization constant given by @xmath202 the terms @xmath203 and @xmath204 have been defined in relation ( [ c6 ] ) . 99 m. hillery 2000 _ phys . rev . a _ * 61 * 022309 d. gottesman and j. preskill 2001 _ phys . a _ * 63 * 022309 n. j. cerf , m. lvy and g. van assche 2001 _ phys . a _ * 63 * 052311 f. grosshans and p. grangier 2002 _ phys . lett . _ * 88 * 057902 c. silberhorn , t. c. ralph , n. ltkenhaus and g. leuchs 2002 _ phys . lett . _ * 89 * 167901 c. weedbrook , a. m. lance , w. p. bowen , t. symul , t. c. ralph and p. k. lam 2004 _ phys . lett . _ * 93 * 170504 m. navascus , f. grosshans and antonio acn 2006 _ phys . lett . _ * 97 * 190502 a. i. lvovsky , b. c. sanders and w. tittel 2009 _ nature photon . _ * 3 * 706 k. jensen , w. wasilewski , h. krauter , t. fernholz , b. m. nielsen , m. owari , m. b. plenio , a. serani , m. m. wolf and e. s. polzik 2011 _ nature phys . _ * 7 * 13 a. ourjoumtsev , a. dantan , r. tualle - brouri and p. grangier 2007 _ phys . lett . _ * 98 * 030502 m. takahashi , j. s. neergaard - nielsen , m. takeuchi , m. takeoka , k. hayasaka , a. furusawa and m. sasaki 2010 _ nature photon . _ * 4 * 178 s. y. lee , s. w. ji , h. j. kim and h. nha 2011 _ phys . rev . a _ * 84 * 012302 t. opatrny , g. kurizki and d. g. welsch 2000 _ phys . rev . a _ * 61 * 032302 p. t. cochrane , t. c. ralph and g. j. milburn 2002 _ phys . rev . a _ * 65 * 062306 s. olivares , m. g. a. paris and r. bonifacio 2003 _ phys . a _ * 67 * 032314 a. kitagawa , m. takeoka , m. sasaki and a. chefles 2006 _ phys a _ * 73 * 042310 g. s. agarwal and k. tara 1991 _ phys . rev . a _ * 43 * 492 s. s. mizrahi and v. v. dodonov 2002 _ j. phys . a : math . theor . _ * 35 * 8847 m. s. kim , e. park , p. l. knight and h. jeong 2005 _ phys . rev . a _ * 71 * 043805 a. r. usha devi , r. prabhu and m. s. uma 2006 _ eur j. d. _ * 40 * 133 l. y. hu , x. x. xu , z. s. wang and x. f. xu 2010 _ phys . rev . a _ * 82 * 043842 s. y. lee and h. nha 2010 _ phys . a _ * 82 * 053812

we study minimal energy problems for strongly singular riesz kernels , where and , considered for compact-dimensional-manifolds immersed into . based on the spatial energy of harmonic double layer potentials , we are motivated to formulate the natural regularization of such minimization problems by switching to hadamard s partie finie integral operator which defines a strongly elliptic pseudodifferential operator of order on . the measures with finite energy are shown to be elements from the sobolev space , , and the corresponding minimal energy problem admits a unique solution . we relate our continuous approach also to the discrete one , which has been worked out earlier by d.p . hardin and e.b . saff .

1602.08544

the classical gauss problem of minimizing the coulomb energy to solve the problem of thomson and its generalization to riesz potentials together with the discretization is the basic problem of many applications ( in @xcite are listed coding theory , cubature formulas , tight frames and packing problems ) . in the works @xcite , @xcite , @xcite , @xcite , and @xcite , the discretization is obtained by approximating the minimizing charges by a distribution of finitely many dirac measures on the given manifold . if the number of dirac points tends to infinity , then the minimizing densities approach distributions in the form of sobolev space elements . therefore , in @xcite , @xcite , @xcite , the minimizing measures are considered as distributions in hilbert spaces of finite riesz energy . this continuous setting is simpler and more efficient from the numerical point compared to the discrete approach in @xcite , @xcite , @xcite , @xcite . for potentials with riesz kernel @xmath11 , where @xmath12 , and borel measures supported on a given @xmath3-dimensional manifold @xmath5 immersed into @xmath6 , a surface potential is generated , which on @xmath5 defines a boundary integral operator with weakly singular kernel . this boundary integral operator is a pseudodifferential operator of _ negative _ order @xmath13 if @xmath14 . the energy space of this pseudodifferential operator on @xmath5 is thus the sobolev space @xmath15 of distributions and the minimizing measure of finite energy is an element of this sobolev space . hence , the determination of the minimizer is reduced to an optimization problem with a quadratic functional which is defined in terms of the single layer riesz potential on @xmath5 . the strong ellipticity of the corresponding pseudodifferential operator in @xmath16 and its trace on @xmath5 then provides the coerciveness of the associated quadratic functional . for @xmath17 , which corresponds to the newtonian kernel , the riesz energy of the single layer potential is just its dirichlet integral over @xmath18 . in this paper , however , we consider the riesz kernels with @xmath19 . for @xmath20 , in classical potential theory , the energy of the harmonic double layer potential in @xmath18 now equals the riesz energy if we define the latter as to be hadamard s partie finie integral of the hypersingular potential which is the natural distributional regularization ( see section [ s : double ] where @xmath5 is a @xmath3-dimensional planar bounded domain in @xmath6 ) . let @xmath21 where @xmath22 , @xmath23 , are finitely many compact , connected @xmath3-dimensional @xmath4-manifolds immersed into @xmath6 . in section [ s : manifold ] , we then consider the riesz potential as a pseudodifferential operator just on @xmath8 since we can not use its extension to @xmath24 ( for @xmath25 , the transmission conditions ( * ? ? ? * theorem 8.3.11 ) are not satisfied ) . we call the bilinear form with the strongly singular partie finie integral of the riesz kernel the _ energy of the riesz potential_. the partie finie integral operator with the hypersingular riesz kernel defines now a strongly elliptic pseudodifferential operator @xmath26 of _ positive _ order @xmath27 on @xmath5 . in contrast to the analysis of weakly singular riesz kernels provided earlier by the authors in @xcite , @xcite , in the case under consideration , the trace theorem in @xmath28 is not valid anymore , because of the negativity of the order @xmath29 , cf . nevertheless , we have succeeded in overcoming this difficulty , and we have shown that all the borel measures on @xmath8 with finite riesz energy whose restriction on any @xmath22 takes sign either @xmath30 or @xmath31 form a certain cone in the sobolev space @xmath32 , @xmath33 . _ this is our main result in section [ s : manifold ] , theorems [ th:3.3 ] and [ th:3.4]_. in this framework , the corresponding gauss variational problem admits a unique solution which belongs to @xmath32 , which is a compact subspace of @xmath34 . these results have again a potential theoretic meaning in the particular situation @xmath35 in relation to the harmonic double layer potential as explained in section [ s : null ] . in the fundamental work @xcite by d.p . hardin and e.b . saff , discrete minimal energy problems have been investigated . there , the discrete riesz energies are obtained by distributing a finite number ( @xmath36 ) of evenly weighted dirac measures on a compact @xmath3-dimensional manifold @xmath37 where the set @xmath38 is excluded . then , the discrete minimal riesz energy determines an optimal geometric arrangement of the @xmath36 distinct dirac points on @xmath37 . in @xcite , three cases are distinguished : ( i ) the riesz kernel is weakly singular , ( ii ) the case @xmath39 ( see @xcite ) , and ( iii ) the hypersingular case @xmath40 . for all these three cases , the behavior of the discrete minimal energies for @xmath36 tending to infinity is explicitly determined ( see section [ s : mini ] below for details ) . in the works @xcite , @xcite , and @xcite , these results are generalized to more general riesz kernels with weights . during a miniworkshop in august 2012 in stuttgart with e.b . saff , d.p . hardin , and p.d . dragnev , we have learned from them that in the hypersingular case the discretized minimal energies tend to infinity if the number of dirac basic points approaches infinity and at the same time those minimizing charges tend to a charge with a constant density . this discussion inspired us to pick up this topic gratefully in our paper and to analyze also this approach by cutting out the set @xmath41 of @xmath42 where @xmath43 . we first figure out the idea in section [ s : mini ] by studying a perturbed riesz energy problem . then , in section [ s : comp ] , we perform the computations in detail for the punched riesz energy problem and give an asymptotic expansion of the solution in the corresponding family of finite energy spaces for @xmath44 . in particular cases ( see corollary [ c:4.6 ] for details ) , the minimizers tend to a constant distribution on @xmath5 while the corresponding minimal energies tend to infinity . we shall motivate our approach by an example from potential theory where @xmath45 , i.e. @xmath46 . to this end , let @xmath47 be a planar bounded domain in @xmath6 and @xmath48 with @xmath49 when @xmath50 , see figure [ fig : plane ] for an illustration . illustration of the geometrical setting.,width=302,height=151 ] the double layer potential of the laplacian with given dipole charge density @xmath51 is given by @xmath52 for @xmath53 with @xmath54 and @xmath55 . the fundamental solution for the laplacian is given by @xmath56 the vector @xmath57 is the @xmath58-th basis vector of @xmath6 and the unit normal vector on @xmath5 . if @xmath59 is continuous at @xmath60 , then there holds the jump relation @xmath61 since for @xmath49 and @xmath62 the scalar product @xmath63 and , hence , the integral vanishes . consequently , the harmonic potential @xmath64 solves the transmission problem in @xmath18 @xmath65_\gam : = u_{\bvfi}(\bx\p,-0)-u_{\bvfi}(\bx\p,+0)=\bvfi(\bx\p)\ ] ] where @xmath59 is a given element of @xmath66 , the closure of @xmath67 in @xmath68 . the energy of the harmonic field @xmath69 is given by its dirichlet integral , and green s theorem yields @xmath70 we can interchange differentiation and integration in this expression by means of hadamard s finite part integral . namely , due to @xmath71 with the constant @xmath72 , it holds @xmath73 where @xmath74 is hadamard s finite part integral with @xmath75 , @xmath76 and with the hypersingular kernel function @xmath77 , that is @xmath78 ( for the definition of hadamard s partie finie integral operators , see @xcite and ( * ? ? ? * chapter 3.2 ) . ) hence , the finite part integral on the right of which has the riesz kernel @xmath79 for @xmath80 defines the energy of the harmonic double layer potential in @xmath18 given by the dirichlet integral on the left of . since the riesz kernel is a homogeneous function of degree @xmath81 , it defines on @xmath82 a strongly elliptic pseudodifferential operator @xmath83 of order @xmath84 ( see ( * ? ? ? * section 7.1.2 ) ) . in all that follows , without stated otherwise , we fix @xmath85 and @xmath86 , and write @xmath87 . in @xmath24 , consider a strongly singular riesz kernel @xmath88 and a manifold @xmath89 , where @xmath90 are finitely many compact , connected , mutually disjoint , boundaryless , @xmath3-dimensional , oriented @xmath91-manifolds , immersed into @xmath16 . then , the surface measure @xmath92 on @xmath8 is well defined . in what follows , @xmath93 will stand for the extension of the @xmath94-scalar product to dualities as @xmath95 and @xmath96 and also to the applications of distributions @xmath97 on @xmath5 operating on @xmath98 . we call the strongly singular partie finie integral of the riesz kernel @xmath99 with respect to @xmath100 , @xmath101 , operating on @xmath102 , the _ energy _ of the riesz potential @xmath103 generated by the surface charge @xmath59 ( see e.g. @xcite ) . for @xmath104 , the hadamard partie finie integral operator @xmath105 which underlies , is for @xmath106 given by @xmath107 where @xmath108 where @xmath109 stands for the calderon mikhlin principal value integral ( see @xcite ) . ( see appendix [ a : a ] for the explicit computation of the partie finie integrals @xmath110 and @xmath111 . ) [ th:2.1 ] the partie finie integral operator @xmath112 is a strongly elliptic pseudodifferential operator of order @xmath113 on @xmath5 . the principal symbol of this operator is given by the equivalence class associated with the homogeneous function @xmath114 @xmath112 defines the linear and continuous mapping @xmath115 for every @xmath116 . in particular , for @xmath117 , @xmath112 maps @xmath32 into @xmath118 and there exist @xmath119 and @xmath120 such that the inequalities @xmath121 are satisfied for any @xmath122 . for justifying the inequalities , recall that for each of the components of the @xmath91-manifolds @xmath90 , @xmath123 , immersed into @xmath24 , we may associate a family of finite - dimensional atlases @xmath124 ( see @xcite ) . each atlas @xmath124 is a family of local charts @xmath125 , where @xmath126 ranges through a finite set @xmath127 . the open sets @xmath128 define an open covering of @xmath129 , while @xmath130 is a @xmath91-diffeomorphism of @xmath131 onto @xmath132 . let @xmath133 be a @xmath91-partition of unity of @xmath129 which is subordinate to the atlas @xmath124 . in addition to the partition of unity , let @xmath134 be a second system of functions @xmath135 with the properties @xmath136 thus , it holds that @xmath137 with respect to the atlas @xmath124 , let @xmath138 denote the corresponding pushforwards and @xmath139 the pullbacks . then @xmath140 . without loss of generality , the local parametric representations can always be chosen in such a way that at one point @xmath141 where @xmath142 we have @xmath143 and , moreover , at this point the tangent bundle @xmath144 forms a positively oriented system of @xmath145 mutually orthogonal unit vectors . this implies that the riemannian tensor of @xmath129 in the local coordinates at the point @xmath146 is the unity matrix . hence , the surface measure satisfies @xmath147 given an atlas @xmath148 on @xmath149 , define @xmath150 thus , one can choose @xmath151 so that for any given @xmath152 there exists a finite - dimensional atlas @xmath153 satisfying all the above formulated properties and @xmath154 . hence , we have a whole family of finite atlases @xmath153 with @xmath152 which will be under consideration . ( we then shall omit the index @xmath155 in the notation . ) note that the jacobians @xmath156 depend on the geometric properties of @xmath129 only , and @xmath156 together with their derivatives are uniformly continuous relative to @xmath157 . corresponding to the partition of unity , the pseudodifferential operator @xmath112 on @xmath8 can be decomposed as @xmath158 and @xmath159 herein , @xmath160 are smoothing operators of order @xmath161 in view of @xmath162 for @xmath163 and @xmath164 if @xmath165 ( see also ( * ? ? ? * chapter 8) ) . moreover , @xmath166 are the localized operators in the parametric domains @xmath167 , defined by the operator @xmath112 . the inequalities now follow locally on each chart of the atlas @xmath148 for the localized operators @xmath168 in local coordinates in @xmath169 . with martensen s surface polar coordinates ( , in appendix [ a : b ] ) , the kernel of @xmath168 admits a pseudohomogeneous asymptotic expansion of the form @xmath170 where @xmath171 since @xmath172 satisfies the expansion ( b9 ) . correspondingly , the symbol @xmath173 of @xmath168 has the asymptotic expansion @xmath174 where @xmath175 with @xmath176 , @xmath177 . then , fourier transform and parseval s theorem yield @xmath178 where @xmath179 is a pseudodifferential operator of order @xmath180 : @xmath181 here , the function @xmath182 is arbitrary but fixed such that @xmath183 and @xmath184 the remainder operator @xmath185 is a smoothing operator with the smooth kernel function @xmath186 . hence , there exists a constant @xmath187 such that @xmath188 which implies with some constant @xmath189 that @xmath190 vice versa , from latexmath:[\[(v_\beta\boldsymbol\varphi,\boldsymbol\varphi)_{l_2(\cu_{\ell r } ) } \geq \int\limits_{\mathbb{r}^{n-1 } } ( 1+|\boldsymbol{\xi}^\prime|)^\beta |\widehat{\boldsymbol\varphi } ( \boldsymbol{\xi}^\prime|^2\d\boldsymbol{\xi}^\prime -c^\prime_{\ell r}\int\limits_{\mathbb{r}^{n-1 } } ( 1+|\boldsymbol{\xi}^\prime|)^{\beta-2 } after summation over @xmath123 , we obtain the grding inequality @xmath192 the embedding @xmath193 is compact since @xmath10 . for @xmath194 the tricomi condition needs to be satisfied for @xmath26 being a pseudodifferential operator and reads here as @xmath195 ( see ( * ? ? ? * theorem 7.1.7 ) ) . in addition , we have @xmath196 . so , @xmath26 is a classical pseudodifferential operator on @xmath8 of order @xmath197 . now , we introduce a set of so - called _ admissible measures _ or charges located on @xmath5 . recall that @xmath198 , where the finitely many @xmath149 are compact , nonintersecting , boundaryless , connected , @xmath3-dimensional , orientable @xmath4-manifolds , immersed into @xmath6 . with each @xmath90 we associate a prescribed sign @xmath199 where @xmath200 for @xmath201 and @xmath202 for @xmath203 . then @xmath204 and @xmath205 , @xmath206 is admitted . let @xmath207 denote the @xmath208-algebra of signed borel measures @xmath209 on @xmath5 equipped with the topology of pointwise convergence on @xmath210 , the class of all real - valued continuous functions on @xmath5 . next , consider the manifold @xmath5 being loaded by charges of the form @xmath211 where , for every @xmath123 , @xmath212 is a nonnegative borel measure on @xmath90 . the convex cone of all signed measures @xmath213 of the form will be denoted by @xmath214 . the following theorem deals with absolutely continuous @xmath215 , i.e.@xmath216 , with densities @xmath217 , @xmath218 for brevity , we shall often identify an absolutely continuous borel measure @xmath215 with @xmath219 , its density . likewise , the cone of all @xmath215 with @xmath217 will be denoted by @xmath220 , provided that this will not cause any misunderstanding . similar to as it has been done in ( [ eq:2.1 ] ) , we define _ the riesz energy _ of @xmath221 by @xmath222 [ th : x ] for any @xmath223 , the riesz energy is finite and satisfies the inequalities @xmath224 the constants @xmath225 and @xmath226 being strictly positive and independent of @xmath227 . this means that @xmath26 is continuously invertible on @xmath228 . write @xmath229 let @xmath230 be a @xmath231-function with @xmath232 for @xmath233 , @xmath234 and @xmath235 for @xmath236 . having the atlas @xmath237 at hand , we use the decomposition and the representation in local coordinates to arrive at @xmath238 where @xmath239 and @xmath240 are the pushforwards of @xmath241 and @xmath219 , respectively . summation gives @xmath242 since the localized operators @xmath168 are all pseudodifferential operators of order @xmath243 with positive definite principal symbol , hence strongly elliptic , one finds with fourier transform and parseval s theorem the estimate @xmath244 with @xmath245 since @xmath193 compactly and if @xmath246 is chosen sufficiently small . the remaining quadratic form @xmath247 has a strictly positive @xmath91-kernel . hence the left inequality in is satisfied for @xmath248 , @xmath249 . since @xmath250 is dense in @xmath251 , the left inequality in also holds for @xmath252 , @xmath249 by completion . the right inequality in was already shown in for @xmath219 in place of @xmath253 . this completes the proof . we next proceed by defining the notion of the riesz energy for arbitrary ( not necessarily absolutely continuous ) measures @xmath254 . since @xmath26 is a classical pseudodifferential operator on @xmath8 , it maps the distribution given by the radon measure @xmath255 to @xmath256 which is a distribution again , and this linear mapping is continuous in the weak topology of distributions ( see theorem ii.1.5 in @xcite ) . therefore , the action of the measure @xmath256 on functions @xmath102 is well defined and @xmath257 since @xmath26 is symmetric . let @xmath258 consist of all @xmath259 which the property @xmath260 hence , for @xmath261 , we can identify @xmath256 with an associated element @xmath262 satisfying @xmath263 and @xmath264 this leads us to the following theorem . [ th:3.3 ] for any @xmath261 there exists a unique element @xmath265 such that @xmath266 and @xmath267 moreover , @xmath268 , and the riesz energy @xmath269 of any @xmath261 is equivalent to the @xmath9-norm of the corresponding @xmath270 in the sense of ( [ eq : stp ] ) . choose an arbitrary @xmath261 . as has been observed just above , @xmath271 is a linear functional on @xmath272 , and it is bounded on @xmath9 because of @xmath273 since @xmath274 is dense in @xmath32 , according to the fischer riesz lemma on the representation of bounded linear functionals there exists a unique @xmath275 with @xmath276 where @xmath277 . if @xmath59 traces @xmath274 , so does @xmath278 ; hence , holds for @xmath279 by replacing @xmath278 by @xmath59 . since @xmath280 , we actually have @xmath281 , and the inclusion @xmath282 follows . now , let @xmath283 . then @xmath284 and @xmath285 due to the duality @xmath286 of @xmath287 , @xmath26 being a pseudodifferential operator on @xmath8 . hence , @xmath288 , which completes the proof . although we have shown that the distributions in @xmath289 all have finite riesz energy @xmath290 , it is not clear yet whether there are no other measures in @xmath214 whose riesz energy is finite . to elaborate on this problem , we employ an idea by j. deny @xcite . a measure on @xmath8 can be considered as a distribution on @xmath8 and , hence , can be fourier transformed . in connection with the localization of @xmath26 on one chart of the atlas on @xmath8 , we have relation where the pseudodifferential operator @xmath168 is defined via fourier transform . if @xmath255 is given , then it becomes via the pushforward @xmath291 the localized distribution @xmath292 with compact support in @xmath293 ( see e.g. ) , which can be fourier transformed to @xmath294 on @xmath295 . the measures in @xmath214 for which @xmath296 are precisely all those having finite riesz energy ( cf . ( [ eq:3.8a ] ) ; observe that the first summand on the left - hand side of ( [ eq:3.8a ] ) is the dominant one ) . let @xmath297 consist of all @xmath255 satisfying ( [ hryu ] ) . [ th:3.4 ] there holds @xmath298 let @xmath299 . then , by theorem [ th : x ] , @xmath300 with and parseval s identity , together with @xmath301 , we obtain further @xmath302 since @xmath303 . consequently , it holds @xmath304 and thus @xmath305 . next , suppose @xmath306 ; then by localization inequality holds . since @xmath307 we find @xmath308 with pullback to @xmath8 this implies that @xmath309 application of theorem [ th:3.3 ] then gives @xmath310 , which in view of the arbitrary choice of @xmath306 finally yields @xmath311 . [ t:2.3 ] let @xmath312 with @xmath313 . then there exists a sequence of absolutely continuous measures @xmath314 , where @xmath315 with @xmath316 and @xmath317 for @xmath318 , such that @xmath319 converges weakly and strongly in the hilbert space @xmath15 to @xmath227 , i.e. , @xmath320 for @xmath227 we have @xmath321 due to theorem [ th:3.3 ] . hence , since @xmath322 densely , there exists a sequence @xmath323 with @xmath324 for all @xmath325 . we define @xmath326 by @xmath327 where @xmath328 . then , @xmath329 since @xmath330 is piecewise smooth and @xmath331 herein , it holds @xmath332 for all @xmath333 , particularly @xmath334 . from @xmath335 , it immediately follows that @xmath336 for all @xmath23 and therefore @xmath337 as desired . the gauss variational problem is the problem of minimizing the riesz energy for particularly signed borel measures on the given @xmath3-dimensional manifold @xmath338 , in the presence of an external field . let @xmath339 be a given continuous , positive function on @xmath5 and @xmath340 a given vector with @xmath341 , @xmath23 . then , the set of _ admissible charges _ for the gauss problem is defined as @xmath342 where we set @xmath343 . note that the set @xmath344 is convex and weakly and strongly closed in @xmath251 . the _ gauss minimal energy problem _ reads as follows ( see @xcite and @xcite ) : to given @xmath345 , @xmath346 and @xmath347 such that @xmath348 , find the borel measure @xmath349 which is the minimizer of @xmath350 where the gauss functional is given by @xmath351 since @xmath352 is on @xmath353 strictly convex and weakly and strongly continuous , the gauss problem admits a unique solution @xmath354 . based on theorem [ th : x ] , the minimization problem can also be formulated as a variational problem in @xmath32 . namely , minimize the functional @xmath355 over the affine cone @xmath356 where @xmath357 , @xmath347 and @xmath358 are given . this minimization problem will be called the _ dual gauss problem_. [ th:2.4 ] to the unique solution @xmath354 of the gauss problem , there corresponds a unique element @xmath359 with the properties @xmath360 and @xmath361 the element @xmath362 is the minimizer of the functional @xmath363 over @xmath364 , i.e. , @xmath365 by theorems [ th : x ] and [ th:3.3 ] , to any borel measure @xmath366 , there corresponds a unique element @xmath367 satisfying both and . moreover , since @xmath272 is dense in @xmath368 , from we get @xmath369 hence , @xmath370 . applying , for these @xmath213 and @xmath371 we also obtain @xmath372 thus , the correspondence @xmath373 between @xmath374 and @xmath364 is one - to - one and satisfies , which immediately implies @xmath375 if now @xmath376 is the ( unique ) solution of the gauss problem , then @xmath362 , the image of @xmath376 under this correspondence , is the unique solution of the minimizing problem , and vice versa . in the following , we will focus on the particular situation @xmath35 from the potential theoretic point of view . the double layer energy @xmath378 of a function @xmath379 , @xmath380 , which is harmonic in @xmath381 ( see ( * ? ? ? * equation ( 1.2.17 ) ) ) is given by @xmath382 with the hypersingular integral operator @xmath383 : @xmath384 the hadamard partie finie integral operator is given by the finite part with respect to @xmath385 of @xmath386 if @xmath104 . the limit @xmath387 exists ( as a cauchy principal value integral ) , whereas , from @xmath388 , we have to take the _ finite part _ @xmath389 ( for the evaluation of @xmath110 , see . ) hence , we finally arrive at @xmath390 instead of considering the continuous minimal riesz energy problem , d.p . hardin and e.b . saff investigate in @xcite the discrete riesz energy problem of minimizing @xmath391 , @xmath392 , of the sum of a finite number @xmath36 of dirac measures @xmath393 , @xmath394 being a set on a @xmath395-rectifiable manifold @xmath37 . the energy @xmath391 is defined by removing the self - interactions . for the sake of simplicity , assume that @xmath37 is compact and has the positive hausdorff measure @xmath396 . then , the infimum of @xmath397 over all point sets @xmath398 is attained at some @xmath399 . in particular , it is shown in ( * ? ? ? * theorem 2.4 ) that @xmath400 where @xmath401 is a constant independent of @xmath37 and defined explicitly by the unit cube . furthermore , in the weak - star topology of measures it holds @xmath402 if @xmath37 is a bi - lipschitz image of a single compact set in @xmath403 , then the separation estimate ( * ? ? ? * eq . ( 16 ) ) has also been established for an optimal @xmath36 point @xmath404-energy configuration @xmath405 for @xmath37 . now , for our @xmath3-dimensional manifold @xmath21 , every compact smooth @xmath22 immersed into @xmath16 , satisfies all the assumptions on @xmath37 with @xmath406 , @xmath407 , @xmath408 . following the inspiration of these results , in our continuous setting , this corresponds to integrating for a small @xmath246 over @xmath409 , i.e. , by cutting out a set with @xmath41 near the singularity . in order to explain the computations in the next section , we shall focus first on the following perturbation problem which , for @xmath410 , is essentially the minimization problem we will finally get . [ th:4.1 ] for @xmath411 sufficiently small , consider the minimization problem : @xmath412 subject to @xmath413 let the given data satisfy the additional conditions : @xmath414 then , the minimizer @xmath415 admits the asymptotic expansion @xmath416 with a constant @xmath417 independent of @xmath418 , and where @xmath419 in particular , with @xmath420 , it holds @xmath421 the quadratic form in induces for @xmath411 the @xmath418-dependent family of hilbert spaces @xmath422 . let us denote by @xmath423 the dual space to @xmath422 whose norm is then defined by @xmath424 satisfying the estimate @xmath425 since @xmath426 densely and the unit ball in @xmath422 is contained in the ball @xmath427 . the problem can also be written as to minimize @xmath428 subject to . it possesses the lagrangian @xmath429 where @xmath430 and @xmath431 . thus , the necessary conditions at the minimum read as @xmath432 or @xmath433 here , @xmath434 since the constraints are always active as it follows from and will also be seen below . for @xmath435 and @xmath436 , we insert the expansion into and obtain the system @xmath437 equating equal order terms in @xmath418 yields with @xmath438 : we find @xmath439 thus , it follows @xmath440 the assumptions imply the properties @xmath441 . moreover , @xmath442 it holds @xmath443 and @xmath444 this yields with @xmath445 : @xmath446 \si_1^k = g_k(g_k , g_k)^{-1}_{l_2(\gam_k)}(v_\bet\si_0^k - f_k , g_k)_{l_2(\gam_k)}+f_k -v_\bet \si_0^k\in h^{\bet/2}(\gam_k).\end{array}\ ] ] hence , @xmath447 we derive the identities @xmath448 \vep(v_\bet\si_2^k , g_k)_{l_2(\gam_k)}+(v_\bet \si_1^k , g_k)_{l_2(\gam_k ) } = \lam_2^k ( g_k , g_k)_{l_2(\gam_k ) } \end{array}\ ] ] since @xmath449 . therefore , we conclude @xmath450 for every fixed @xmath411 sufficiently small , the mapping @xmath451 defines an isomorphism due to and the lax milgram lemma . therefore , amounts to the estimate @xmath452 with a constant @xmath453 depending on @xmath5 and @xmath339 due to but _ not _ on @xmath418 . with @xmath454 and @xmath112 being a pseudodifferential operator of order @xmath455 on @xmath5 , we further have @xmath456 implying that @xmath457 consequently , since the constants do not depend on @xmath418 , there exists an @xmath101 such that @xmath458 and @xmath459 independent of @xmath418 . with @xmath460 , @xmath461 , we find @xmath462 and @xmath463 . hence , is justified which completes the proof of theorem [ th:4.1 ] . with the help of the previous theorem , we immediately find the following asymptotic behaviour of the minimizer @xmath464 if @xmath418 tends to zero . [ th:4.2 ] under the assumptions in theorem [ th:4.1 ] , we find that @xmath465 with some constant @xmath466 , independent of @xmath467 and @xmath411 , where @xmath464 is the minimizer of for @xmath411 . since @xmath468 , with we find @xmath469 as proposed . we consider next the punched hypersingular riesz potential which is defined by integrating for a small @xmath470 only over @xmath409 , i.e. , by cutting out a set with @xmath41 near the singularity . thus , the corresponding riesz energy is defined as @xmath471 in view of theorems [ th:3.3 ] , [ t:2.3 ] and [ th:2.4 ] , the associated minimal riesz energy problem is then equivalent to minimizing the punched functional @xmath472 where @xmath473 and @xmath474 , over the affine cone @xmath475 . then , the measures satisfy @xmath476 with @xmath92 being the surface measure on @xmath5 . for @xmath477 one has the following monotonicity property . [ l:4.2 ] let @xmath478 and @xmath479 be the minimizers of @xmath480 and @xmath481 , respectively . then , it holds that @xmath482 since @xmath483 , the minimizer @xmath484 with @xmath485 is an admissible element for minimizing @xmath481 . in particular , it holds @xmath486 we further find @xmath487 as proposed in . in order to see the relation between @xmath488 and @xmath489 in , let us introduce the compensating quadratic functional @xmath490 then , we obtain @xmath491 and thus @xmath492 for the corresponding functional @xmath493 , there holds [ l:4.3 ] let @xmath494 . then @xmath495 where @xmath496 satisfies @xmath497 with a constant @xmath466 independent of @xmath498 . moreover @xmath499 using martensen s coordinates of @xmath5 in the vicinity of @xmath49 ( see theorem [ t:3.3 ] in appendix [ a : b ] ) , for every @xmath500 , we have @xmath501 since @xmath502 and @xmath503 in is symmetric , we find @xmath504 herein , @xmath505 is given by @xmath506 we rewrite @xmath507 according to @xmath508 where @xmath509 and @xmath510 are kernels which posess pseudohomogeneous expansions of degree @xmath511 . this means that @xmath512 with classical pseudodifferential operators @xmath513 and @xmath514 of degree @xmath515 on @xmath5 . since the constant charge @xmath84 is a smooth function on @xmath5 and @xmath516 for @xmath122 , we finally arrive at @xmath517 as proposed . , @xmath518 is a singular mikhlin calderon integral operator with principal part @xmath519 which satisfies the mikhlin condition @xmath520 . ] in order to show , consider first @xmath104 and use taylor s expansion about @xmath49 in . then , all the integrals on the right hand side are weakly singular tending to zero with @xmath44 . for @xmath122 approximate @xmath59 by @xmath521 , satisfying @xmath522 . then @xmath523 with @xmath524 for @xmath525 , one has @xmath526 then @xmath527 for any @xmath528 which implies . since @xmath529 where @xmath530 for @xmath44 , lemma [ l:4.2 ] implies [ c:4.4 ] let @xmath531 and @xmath532 be given . then , there exist positive constants @xmath533 such that the estimate @xmath534 holds uniformly for @xmath535 , @xmath536 and @xmath537 . the functional coincides with the functional from except for the perturbation term @xmath507 . hence , we can proceed in complete analogy to the proof of theorem [ th:4.1 ] . the lagrangian to the punched energy functional reads as @xmath538 where the first order necessary optimality condition is given by @xmath539 again , @xmath540 since the constraints are active . here , @xmath541 with @xmath542 and @xmath513 and @xmath543 being bounded linear operators : @xmath544 under the assumptions of theorem [ th:4.1 ] and as in the proof of theorem [ th:4.1 ] , we finally obtain the asymptotic expansion of the minimizer @xmath464 as well as of the lagrangian multipliers : @xmath545 and @xmath546 where @xmath547 it turns out that @xmath548 and @xmath549 are exactly the same as in . moreover , we have to replace @xmath112 by @xmath550 in the equations and , . note that @xmath551 and , hence , @xmath552 for @xmath44 tends to zero in @xmath32 due to . ( if @xmath553 for @xmath554 or @xmath555 for @xmath556 , then @xmath557 and @xmath558 , respectively @xmath559 . ) collecting these results , we have for the punched energy riesz minimum problem the following result : [ th:4.5 ] under the same assumptions as for theorem [ th:4.1 ] , the minimization problem @xmath560 subject to , has for every @xmath561 a unique solution @xmath464 . it admits for @xmath562 the asymptotic expansion @xmath563 with a constant @xmath466 independent of @xmath418 . in particular , with @xmath564 there holds @xmath565 where @xmath566 and @xmath567 with @xmath568 [ c:4.6 ] for @xmath44 and @xmath569 one finds @xmath570 with some constant @xmath466 , independent of @xmath467 and @xmath411 , where @xmath464 is the minimizer . moreover , it holds @xmath571 due to , we obtain @xmath572 the conditions imply that @xmath573 and , hence , @xmath574 for all @xmath418 with @xmath575 with some @xmath576 . thus , there holds @xmath577 with uniformly bounded @xmath578 . hence , @xmath44 implies . for the torus @xmath579 in @xmath580 , considered in @xcite and @xcite , where @xmath460 , @xmath581 , @xmath582 , the minimizers @xmath464 of the punched minimization problem tend to the constant charge @xmath583 . in this appendix , we shall compute the partie finie integrals which define the functions @xmath110 and @xmath111 from and , respectively . [ l : a.1 ] ( i ) let @xmath86 , @xmath14 and @xmath104 . then , one has @xmath584 where @xmath585 denotes the mikhlin calderon principal value integral . \(ii ) the function @xmath110 from is given by @xmath586 with any @xmath417 sufficiently small . \(iii ) for the function @xmath111 from , one obtains @xmath587 with any @xmath417 sufficiently small . _ ( i ) _ locally on @xmath5 one has near @xmath49 : @xmath588 @xmath589 describes the @xmath590-dimensional unit sphere @xmath591 and @xmath592 is its polar coordinate with @xmath593 its @xmath590-dimensional surface measure . consequently , with an appropriate constant @xmath417 depending on @xmath5 , one has @xmath594 the first integral on the right is zero because of @xmath595 . whereas , the integrand of the remaining integral has at @xmath596 a weak singularity @xmath597 with @xmath598 whose principal value integral exists . this follows by using e.g. martensen s surface polar coordinates , cf . theorem [ t : b1 ] in appendix [ a : b ] . _ ( ii ) _ for the function latexmath:[\[h(\bx ) = \lim_{\de\to 0}\pf \intli_{0<\de<|\bx-\by|\wedge \by\in\gam } h(\bx ) & = \lim_{\de\to 0}\pf \intli_{0<\de<|\bx-\by|\le c } \big\{|\bx-\by|^{-\bet-(n-1)}-r^{-\bet-(n-1)}\big\}\d s_\by\\ & \qquad+\lim_{\de\to 0}\pf \intli_{0<\de}^c r^{-\bet-(n-1 ) } r^{n-2}\d r\intli_{|\btheta|=1}\d\bom\\ & \qquad-\lim_{\de\to 0}\pf\intli_{0<\de}^c\ \intli_{|\btheta|=1 } r^{-\bet-(n-1)}\big\{r^{n-2}\d r\d\bom-\d s_\by\big\}\\ & \qquad + \intli_{c\le |\bx-\by|\wedge\by\in\gam } martensen s surface polar coordinates on @xmath5 , then the integrand of the first integral on the right is identical zero due to @xmath600 . for the second integral on the right , we have with @xmath601 that @xmath602 for the third integral , theorem [ t:3.3 ] in appendix [ a : b ] implies @xmath603 hence , the integrand is of order @xmath604 and the integral exists as a principal value integral . collecting these properties proves . _ ( iii ) _ we proceed with respect to the vector - valued function latexmath:[\[\bh(\bx)=\pf\lim_{\de\to 0}\intli_{\by\in\gam\wedge 0<\de<|\bx-\by| } @xmath110 . inserting @xmath606 for martensen s surface polar coordinates , we have : @xmath607 . the third integral contains an integrand of order @xmath608 . hence , the integrand is bounded and the integral exists . consequently , also holds . for @xmath609 , surface polar coordinates have been introduced by martensen in ( * ? ? ? * chapter 2.1 ) . a graphical illustration of these coordinates are found in figure [ fig : sphere ] . we generalize this approach to arbitrary spatial dimension @xmath610 . the given surface has locally the parametric representation @xmath611 , @xmath612 . for @xmath613 , define the family of @xmath590-dimensional manifolds ( level sets ) as @xmath614 given by @xmath615 , where @xmath616 denotes the euclidian distance in @xmath6 and @xmath617 is the radial parameter . let @xmath618 where @xmath619 denotes the euclidian scalar product in @xmath6 . in the local neighborhood of @xmath5 , define @xmath620 with @xmath621 , where @xmath622 denotes the ( exterior ) unit normal vector of @xmath5 at @xmath623 . then , on @xmath5 , the vector @xmath624 is tangential to @xmath5 at @xmath625 , @xmath626 for @xmath627 chosen , the surface polar coordinates then locally are given by the family of closed surfaces @xmath628 , i.e. , the level sets of the function @xmath629 , and curved radial rays on @xmath5 through @xmath630 which are perpendicular to @xmath628 for constant @xmath617 . let us denote such a radial ray curve by @xmath631 , given by @xmath632 , @xmath633 , where the parametric representation @xmath634 at @xmath635 starts in @xmath636 in the direction of the unit vector @xmath637 , @xmath638 , @xmath639 @xmath640 , @xmath641 . hence , for the curve @xmath642 , we require that @xmath643 ( note that for @xmath609 one usually uses @xmath644 and @xmath645 . ) without restriction of generality , we assume in what follows that at @xmath646 we have @xmath647 ( the kronecker tensor ) . since the radial ray curves @xmath631 are perpendicular to the level sets , which implicitly are given by @xmath648 const , they satisfy the ordinary differential equations for fixed @xmath649 : @xmath650 where @xmath651 is the surface gradient and @xmath652 . since implies on @xmath5 that @xmath653 we find @xmath654 and the differential equations take the form @xmath655 after multiplication with @xmath656 , we thus arrive at @xmath657 [ t : b1 ] to every @xmath627 there exists a neighborhood of @xmath630 on @xmath5 where the system admits a unique solution @xmath658 for @xmath659 which is the solution of the volterra integral equations @xmath660 the transformation @xmath661 to surface polar coordinates about @xmath630 is given by @xmath662 , i.e. , @xmath663 . since @xmath664 the expansion about @xmath665 gives on the one hand @xmath666 on the other hand , with the gaussian equations @xmath667 and the weingarten relations @xmath668 we obtain @xmath669 where @xmath670 are the christoffel symbols of the second kind of @xmath5 at @xmath671 , @xmath672 , and @xmath673 with @xmath674 the second fundamental form of @xmath5 at @xmath671 ( see e.g. @xcite ) . here and in what follows , we abbreviate @xmath675 we get thus from @xmath676 \os{\circ}{\bx}_{|t}+ \big[\os{\circ}{l}_{jk|\ell}+\os{\circ}{\gam}_{jk}^m\os{\circ}{l}_{m\ell}\big ] \os{\circ}{\bn}\big)\theta^j\theta^k\theta^\ell\\ & \quad\quad+3\big(\os{\circ}{\gam}_{k\ell}^t\os{\circ}{\bx}_{|t}+\os{\circ}{l}_{k\ell } \os{\circ}{\bn}\big)\theta^\ell \f{\d^2\os{\circ}{u}^k}{\d s^2}+\os{\circ}{\bx}_{|t } \f{\d^3\os{\circ}{u}^t}{\d s^3}\bigg\ } s^3+\co(s^4).\no\end{aligned}\ ] ] by combining this expansion with @xmath677 we arrive at @xmath678\theta^k \theta^\ell+\overset{\circ}{\bx}_{|k}\f{\d^2\overset{\circ}{u}^k}{\d s^2}\big ) + \ha \overset{\circ}{l}_{k\ell}\theta^k\theta^\ell\overset{\circ}{\bn}\bigg\}s^2+\co(s^3),\\ a & = s+\co(s^2).\end{aligned}\ ] ] hence , we conclude @xmath679 for any @xmath680-curve @xmath634 through @xmath630 . consequently , the kernel function @xmath681 of the volterra operator in is continuous and there exists a solution @xmath682 for fixed given @xmath649 in some vicinity of @xmath630 on @xmath5 . this solution is in @xmath683)$ ] for some @xmath684 and is as many times continuously differentiable for @xmath659 as is the manifold @xmath5 . by bootstrapping , it follows from that @xmath662 is higher order differentiable up to @xmath688 . to see this , consider the taylor expansions of the left and the right hand sides of the equations about @xmath689 up to the order two : ( by using ) @xmath690\overset{\circ}{\bx}_{|m } + \big[\overset{\circ}{\gam}_{\ell j}^m \overset{\circ}{l}_{mk } + \overset{\circ}{l}_{\ell j|k}\big]\overset{\circ}{\bn}\big ) \theta^\ell\theta^j\theta^k\\ & \qquad+3\big(\overset{\circ}{\gam}_{\ell j}^m\overset{\circ}{\bx}_{|m } + \overset{\circ}{l}_{\ell j}\overset{\circ}{\bn}\big ) \theta^j\f{\d^2\overset{\circ}{u}}{\d\vr^2}^\ell\bigg\}\vr^2+\co(\vr^3).\end{aligned}\ ] ] ( see ) @xmath691\overset{\circ}{\bx}_{|m } + \big[\overset{\circ}{\gam}_{\ell j}^m \overset{\circ}{l}_{mk } + \overset{\circ}{l}_{\ell j|k}\big]\overset{\circ}{\bn}\big ) \theta^\ell\theta^j\theta^k\\ & \quad+3\big(\overset{\circ}{\gam}_{\ell j}^m\overset{\circ}{\bx}_{|m } + \overset{\circ}{l}_{\ell j}\overset{\circ}{\bn}\big)\theta^j \f{\d^2\overset{\circ}{u}}{\d\vr^2}^\ell \bigg\}\vr^2+\co(\vr^3 ) , \\ & \bn(\bu ) = \overset{\circ}{\bn}+\overset{\circ}{\bn}_{|j } \f{\d\overset{\circ}{u}}{\d\vr}^j\vr+\ha \bigg\{\overset{\circ}{\bn}_{|j|k}\f{\d\overset{\circ}{u}^j}{\d\vr } \f{\d\overset{\circ}{u}}{\d\vr}^k+\overset{\circ}{\bn}_{|j } \f{\d^2\overset{\circ}{u}}{\d\vr^2}^j\bigg\}\vr^2 + \co(\vr^3)\\ & \quad= \overset{\circ}{\bn}-\overset{\circ}{l}_j^m \overset{\circ}{\bx}_{|m}\theta^j\vr\\ & \qquad-\ha\bigg\{\big(\overset{\circ}{l}_{j|k}^m\overset{\circ}{\bx}_{|m } + \overset{\circ}{l}_j^\ell\overset{\circ}{\gam}_{\ell k}^m \overset{\circ}{\bx}_{|m}+\overset{\circ}{l}_j^m \overset{\circ}{l}_{mk}\overset{\circ}{\bn}\big ) \theta^j\theta^k + \overset{\circ}{l}_j^m \f{\d^2\overset{\circ}{u}}{\d\vr^2}^j \overset{\circ}{\bx}_{|m}\bigg\}\vr^2+\co(\vr^3),\\ & \f{1}{\vr}g = \ha\overset{\circ}{l}_{k\ell}\theta^k\theta^\ell \vr -\ha \overset{\circ}{l}_{j\ell}\f{\d^2 \overset{\circ}{u}}{\d\vr^2}^j\theta^\ell\\ & \quad + \bigg\{\f{1}{6 } \overset{\circ}{l}_{\ell j|k}-\f{1}{3 } \overset{\circ}{\gam}_{\ell j}^m \overset{\circ}{l}_{mk } -\ha \overset{\circ}{g}_{m\ell } \overset{\circ}{l}_{j|k}^m-\ha \overset{\circ}{g}_{m\ell}\overset{\circ}{l}_j^t\overset{\circ } { \gam}_{tk}^m\bigg\}\theta^j\theta^k\theta^\ell\vr^2+\co(\vr^3),\\ & \big(\f{1}{\vr}g\big)\bn(\bu ) = \big(\f{1}{\vr}g\big)\big(\overset{\circ}{\bn}- \overset{\circ}{l}_j^m\theta^j\overset{\circ}{\bx}_{|m}\vr\big)+\co(\vr^3)\\ & \quad= \overset{\circ}{\bn}\ha\overset{\circ}{l}_{k\ell}\theta^k\theta^\ell\vr + \overset{\circ}{\bn}\vr^2 -\ha\overset{\circ}{l}_j^m \overset{\circ}{l}_{k\ell}\overset{\circ}{\bx}_{|m } \theta^j\theta^k\theta^\ell\vr^2 -\ha\overset{\circ}{l}_{j\ell } \f{\d^2\overset{\circ}{u}}{\d\vr^2}^j\theta^\ell\overset{\circ}{\bn}\vr^2 + \co(\vr^3),\\ & \f{g^2}{a^2 } = \big(\f{1}{\vr}g\big)^2= \f{1}{4 } \big(\overset{\circ}{l}_{k\ell}\theta^k\theta^\ell\big)^2\vr^2+\co(\vr^3 ) , \\ & \big(1-\f{g^2}{a^2}\big)^{-1 } = 1+\f{1}{4}\big(\overset{\circ}{l}_{k\ell}\theta^k\theta^\ell\big)^2\vr^2+\co(\vr^3).\end{aligned}\ ] ] with the expressions for @xmath699 and @xmath700 , we recollect the relations for @xmath701 and find instead of : @xmath702 \os{\circ}{\bn}\big)\theta^m\theta^j\theta^k + \f{4}{3}\big(\os{\circ}{l}_{kj}\theta^k\theta^j\big)^2\btheta\bigg\}+\co(\vr^4).\end{aligned}\ ] ] collecting the first three derivatives of @xmath703 at @xmath689 implies that the first terms of the transform @xmath687 read as @xmath704 for all @xmath705 . whereas , the inverse mapping @xmath661 can be obtained from @xmath706 and the nonlinear equations for @xmath707 for all @xmath705 can , for @xmath708 sufficiently small , be solved via successive iteration . [ t:3.3 ] let @xmath709 . then , the surface measure @xmath710 of @xmath5 in martensen s surface polar coordinates satisfies @xmath711 where @xmath712 and @xmath713,\ \sumli_{j=1}^{n-1}\theta^j\theta^j=1.\ ] ] here , @xmath593 is the surface measure of the unit sphere @xmath591 in @xmath714 . for the surface measure of @xmath5 , we use the exterior pfaffian products , defining the exterior normal vector s components times @xmath710 ( see ( * ? ? ? * chapter 11.4 ) . then , we multiply scalarly with the exterior unit normal @xmath715 which yields @xmath716n^j(\bx).\ ] ] expansions about @xmath646 up to the order @xmath717 give @xmath718 and @xmath719 with @xmath720 for @xmath721 and @xmath722 inserting this into , yields @xmath723 \big(1-\ha\vr^2\os{\circ}{l}_j^m\os{\circ}{l}_{mk}\theta^j\theta^k + \co(\vr^3)\big)\no\\ & \quad + \sumli_{j=1}^{n-1}\big[\big[\d x^1\wedge\dots\wedge\kreuzx\wedge\d x^{n-1}\big ] \wedge\d x^n\big]\no\\ & \qquad \x\bigg\{-\vr\os{\circ}{l}_k^j\theta^k -\ha\big(\os{\circ}{l}_{m|k}^j+\os{\circ}{l}_m^\ell \os{\circ}{\gam}_{\ell k}^j-\os{\circ}{\gam}_{mk}^t \os{\circ}{l}_t^j\big)\theta^m\theta^k\vr^2+\co(\vr^3)\bigg\}.\end{aligned}\ ] ] for the first term in , we obtain ( modulo @xmath724 terms ) with the relations for @xmath725 : @xmath726 = \big[\d\vr\theta^1+\vr\d\theta^1+\vr^2\d\vr c^1(\btheta ) \wedge \d\vr \theta^2+\vr\d\theta^2+\vr^2\d\vr c^2(\btheta)\\ & \wedge\dots\wedge \d\vr\theta^{n-1}+\vr\d\theta^{n-1}+\vr^2\d\vr c^{n-1}(\btheta)\big ] = \vr^{n-1}\big[\d\theta^1\wedge\dots\wedge\d\theta^{n-1}\big]\\ & + \vr^{n-2}\bigg[\d\vr\wedge\sumli_{j=1}^{n-1}(-1)^j \big\{\theta^j+\vr^2 c^j(\btheta)\big\ } \big[\d\theta^1\wedge\dots\wedge\kreuz\wedge\dots\wedge\d\theta^{n-1}\big ] \bigg].\end{aligned}\ ] ] since the variables @xmath727 vary on the @xmath590-dimensional sphere @xmath591 , where @xmath728 we have @xmath729 on @xmath591 . hence , the differentials @xmath730 with @xmath721 are linearly dependent and @xmath731=0.\ ] ] moreover , on @xmath591 , we have that the exterior unit normal vector @xmath209 to @xmath591 satisfies @xmath732 therefore , it holds @xmath733 = \theta^j\d\bom,\ ] ] and with one obtains @xmath734.\ ] ] for the first term in , we find therefore @xmath735 for the remaining terms in , we have ( modulo @xmath724 terms ) that @xmath736\wedge \d x^n\big]\big(-\vr \os{\circ}{l}_t^j\theta^t -\ha\vr^2\{\dots\}\theta^m\theta^k\big)\\ & = \big[\big[\d\vr\theta^1+\vr\d\theta^1+\vr^2\d\vr c^1(\btheta)\wedge \d\vr\theta^2+\vr\d\theta^2+\vr^2\d\vr c^2(\btheta)\\ & \qquad\qquad\wedge\dots\wedge\kreuzx\wedge\dots\wedge\d\vr^2\theta^{n-1}+ \vr\d\theta^{n-1}+\vr^2\d\vr c^{n-1}(\btheta)\big]\\ & \qquad\qquad\wedge\big(\vr\d\vr\os{\circ}{l}_{jk}\theta^j\theta^k + \ha \vr^2\d\vr \{\dots\}\theta^m\theta^j\theta^k\big)\big ] \big(-\vr \os{\circ}{l}_t^k \theta^t-\ha\{\dots\}\theta^m\theta^k\big)\\ & = \sumli_{j=1}^{n-1}(-1)^{n+j}\big(\vr^n+\co(\vr^{n+1})\big)\d\vr\wedge \big[\d\theta^1\wedge\dots\wedge\kreuz\wedge\dots\wedge\d\theta^{n-1}\big ] \os{\circ}{l}_{mk}\theta^m\theta^k\os{\circ}{l}_t^j\theta^t\\ & = ( -1)^{n+1}\sumli_{j=1}^{n-1}\os{\circ}{l}_{mk}\theta^m\theta^k \os{\circ}{l}_t^j\theta^t\theta^j\big(\vr^n+\co(\vr^{n+1})\big ) \d\vr\wedge\d\bom.\end{aligned}\ ] ] consequently , we finally get in @xmath737 with @xmath738 this is the proposed relation for the surface measure . s.v . borodachov , d.p . hardin , e.b . saff : asymptotics for discrete weighted minimal riesz energy problems on rectifiable sets . _ transactions american math . soc . _ ( 2008 ) 15591580 . borodachov , d.p . hardin , e.b . saff : low complexity methods for discretizing manifolds via riesz energy minimization . math . _ ( 2014 ) 11731208 . brauchart , d.p . hardin , e.b . saff : riesz energy and sets of revolution in @xmath580 . _ contemporary math . functional analysis and complex analysis _ ams ( 2009 ) 4757 . j. deny : les potentiels dnergy finie , _ acta math . _ * 82 * ( 1950 ) 107183 . dragnev , e.b . saff : riesz spherical potentials with external fields and minimal energy points separation . _ potential analysis _ ( 2007 ) 139162 . j. hadamard : le problme de cauchy et les quations aux drives partielles linaires hyperboliques . hermann , paris 1932 . h. harbrecht , w.l . wendland , n. zorii : on riesz minimal energy problems . _ j. math . ( 2012 ) 397412 . h. harbrecht , w.l . wendland , n. zorii : riesz minimal energy problems on @xmath739manifolds . ( 2014 ) 4869 . hardin , e.b . saff : discretizing manifolds via minimum energy points . _ notices ams _ ( 2004 ) 11861194 . hardin , e.b . saff : minimal riesz energy point configurations for rectifiable @xmath395-dimensional manifolds . _ advances in mathematics _ ( 2005 ) 174204 . hsiao , w.l . wendland : boundary integral equations . springer , berlin 2008 . w. khnel : differential geometry : curves surfaces manifolds , 2nd edition . american mathematical society 2006 . kuijlaars and e.b . saff : asymptotics for minimal discrete energy on the sphere _ trans . amer . _ * 350 * ( 1998 ) 523538 . landkof : foundations of modern potential theory . springer , berlin 1992 . e. martensen : potentialtheorie . teubner , stuttgart 1968 . mikhlin , s. prssdorf : singular integral operators . verlag berlin 1986 . g. of , w.l . wendland , n. zorii : on the numerical solution of minimal energy problems . _ complex variables and elliptic equations _ ( 2010 ) 9911012 . m. ohtsuka : on potentials in locally compact spaces , _ j. sci . hiroshima univ . a1 _ * 25 * ( 1961 ) 135352 . taylor : pseudodifferential operators . princeton university press , princeton 1981 . wendland , o. steinbach : analysis . teubner , wiesbaden 2005 .

we develop the shape derivative analysis of solutions to the problem of scattering of time - harmonic electromagnetic waves by a penetrable bounded obstacle . since boundary integral equations are a classical tool to solve electromagnetic scattering problems , we study the shape differentiability properties of the standard electromagnetic boundary integral operators . the latter are typically bounded on the space of tangential vector fields of mixed regularity . using helmholtz decomposition , we can base their analysis on the study of pseudo - differential integral operators in standard sobolev spaces , but we then have to study the gteaux differentiability of surface differential operators . we prove that the electromagnetic boundary integral operators are infinitely differentiable without loss of regularity . we also give a characterization of the first shape derivative of the solution of the dielectric scattering problem as a solution of a new electromagnetic scattering problem .

1105.2479

consider the scattering of time - harmonic electromagnetic waves by a bounded obstacle @xmath1 in @xmath2 with a smooth and simply connected boundary @xmath3 filled with an homogeneous dielectric material . this problem is described by the system of maxwell s equations with piecewise constant electric permittivity and magnetic permeability , valid in the sense of distributions , which implies two transmission conditions on the boundary of the obstacle guaranteeing the continuity of the tangential components of the electric and magnetic fields across the interface . the transmission problem is completed by the silver mller radiation condition at infinity ( see @xcite and @xcite ) . boundary integral equations are an efficient method to solve such problems for low and high frequencies . the dielectric scattering problem is usually reduced to a system of two boundary integral equations for two unknown tangential vector fields on the interface ( see @xcite and @xcite ) . we refer to @xcite for methods developed by the authors to solve this problem using a single boundary integral equation . optimal shape design with a goal function involving the modulus of the far field pattern of the dielectric scattering problem has important applications , such as antenna design for telecommunication systems and radars . the analysis of shape optimization methods is based on the analysis of the dependency of the solution on the shape of the dielectric scatterer , and a local analysis involves the study of derivatives with respect to the shape . an explicit form of the shape derivatives is desirable in view of their implementation in shape optimization algorithms such as gradient methods or newton s method . in this paper , we present a complete analysis of the shape differentiability of the solution of the dielectric scattering problem and of its far field pattern , using integral representations . even if numerous works exist on the calculus of shape derivatives of various shape functionals @xcite , in the framework of boundary integral equations the scientific literature is not extensive . however , one can cite the papers @xcite , @xcite and @xcite , where r. potthast has considered the question , starting with his phd thesis @xcite , for the helmholtz equation with dirichlet or neumann boundary conditions and the perfect conductor problem , in spaces of continuous and hlder continuous functions . using the integral representation of the solution , one is lead to study the gteaux differentiability of boundary integral operators and potential operators with weakly singular and hypersingular kernels . the natural space of distributions ( energy space ) which occurs in the electromagnetic potential theory is @xmath0 , the set of tangential vector fields whose components are in the sobolev space @xmath4 and whose surface divergence is in @xmath4 . we face two main difficulties : on one hand , the solution of the scattering problem is given in terms of products of boundary integral operators and their inverses . in order to be able to construct shape derivatives of such products , it is not sufficient to find shape derivatives of the boundary integral operators , but it is imperative to prove that the derivatives are bounded operators between the same spaces as the boundary integral operators themselves . on the other hand , the very definition of shape differentiability of operators defined on the shape - dependent space @xmath0 poses non - trivial problems . our strategy consists in using the helmholtz decomposition of this hilbert space which gives a representation of a tangential vector field in terms of ( tangential derivatives of ) two scalar potentials . in this way , we split the analysis into two steps : first the gteaux differentiability analysis of scalar boundary integral operators and potential operators with strongly and weakly singular kernels , and second the study of shape derivatives of surface differential operators . this work contains results from the thesis @xcite where this analysis has been used to develop a shape optimization algorithm of dielectric lenses in order to obtain a prescribed radiation pattern . this is the second of two papers on shape derivatives of boundary integral operators , the first one @xcite being aimed at a general theory of shape derivatives of singular integral operators appearing in boundary integral equation methods . the paper is organized as follows : in section [ boundintop ] we recall some standard results about trace mappings and regularity properties of the boundary integral operators in electromagnetism . in section [ scatprob ] we define the scattering problem for time - harmonic electromagnetic waves at a dielectric interface . we then give an integral representation of the solution and of the quantity of interest , namely the far field of the dielectric scattering problem following the single source integral equation method developed in @xcite . the remaining parts of the paper are dedicated to the shape differentiability analysis of the solution of the dielectric scattering problem . we use the results of our first paper @xcite on the gteaux differentiability of boundary integral operators with pseudo - homogeneous kernels . we refer to this paper for a discussion of the notion of gteaux derivatives in frchet spaces and of some of their basic properties . in section [ helmholtzdec ] we discuss the difficulties posed by the shape dependency of the function space @xmath0 on which the integral operators are defined , and we present a strategy for dealing with this difficulty , namely using the well - known tool @xcite of helmholtz decomposition . in our approach , we map the variable spaces @xmath5 to a fixed reference space with a transformation that preserves the hodge structure . this technique involves the analysis of surface differential operators that have to be considered in suitable sobolev spaces . therefore in section we recall and extend the results on the differentiability properties of surface differential operators established in ( * ? ? ? * section 5 ) . using the rules on derivatives of composite and inverse functions , we obtain in section [ shapesol ] the shape differentiability properties of the solution of the scattering problem . more precisely , we prove that the boundary integral operators are infinitely gteaux differentiable without loss of regularity , whereas previous results allowed such a loss @xcite , and we prove that the shape derivatives of the potentials are smooth away from the boundary but they lose regularity in the neighborhood of the boundary . this implies that the far field is infinitely gteaux differentiable , whereas the shape derivatives of the solution of the scattering problem lose regularity . these new results generalize existing results : in the acoustic case , using a variational formulation , a characterization of the first gteaux derivative was given by a. kirsch @xcite for the dirichlet problem and then by hettlich @xcite for the impedance problem and the transmission problem . an alternative technique was introduced by kress and pivrinta in @xcite to investigate frchet differentiability in acoustic scattering by the use of a factorization of the difference of the far - field pattern of the scattered wave for two different obstacles . in the electromagnetic case , potthast used the integral equation method to obtain a characterization of the first shape derivative of the solution of the perfect conductor scattering problem . in @xcite , kress improved this result by using a far - field identity and in @xcite kress and haddar extended this technique to acoustic and electromagnetic impedance boundary value problems . at the end of section [ shapesol ] we obtain a characterization of the first shape derivative of the solution of the dielectric scattering problem as the solution of a new electromagnetic transmission problem . we show by deriving the integral representation of the solution that the first derivative satisfies the homogeneous maxwell equations , and by directly deriving the boundary values of the solution itself we see that the first derivative satisfies two new transmission conditions on the boundary . in the end we will have obtained two different algorithms for computing the shape derivative of the solution of the dielectric scattering problem and of the far field pattern : a first one by differentiating the integral representations and a second one by solving the new transmission problem associated with the first derivative . the characterization of the derivatives as solutions to boundary value problems has been obtained in the acoustic case by kress @xcite , kirsch @xcite , hettlich and rundell @xcite and hohage @xcite and has been used for the construction of newton - type or second degree iterative methods in acoustic inverse obstacle scattering . whereas the use of these characterizations requires high order regularity assumption for the boundary , we expect that the differentiation of the boundary integral operators does not require much regularity . although in this paper we treat the case of a smooth boundary , in the last section we give some ideas on possible extensions of the results of this paper to non - smooth domains . a final remark on the terminology of derivatives with respect to the variation of the domain : for the solutions of boundary value problems with a moving boundary , in the context of continuum mechanics one distinguishes frequently between _ material derivatives _ and _ shape derivatives_. the former correspond to the gteaux derivative with respect to the deformation , which is then interpreted as a flow associated with a velocity field , when the solution is pulled back to a fixed undeformed reference domain ( lagrangian coordinates ) . depending on the interpretation of the flow , other names are used , such as `` lagrangian derivative '' , or `` substantial derivative '' . the shape derivatives , on the other hand , correspond to the gteaux derivative of the solution in the deformed domain ( eulerian coordinates ) . one also talks about `` eulerian derivative '' . the difference between the two has the form of a convection term , which can lead to a loss of one order of regularity of the shape derivative on the support of the velocity field @xcite . in the context of this terminology , the derivatives of the boundary integral operators that we studied in part i of this paper @xcite for general pseudohomogeneous kernels and in this paper for the boundary integral operators of electromagnetism , correspond to the lagrangian point of view , because they are obtained by pull - back to a fixed reference boundary . however , the derivative of the solution of dielectric scattering problem that we construct in section [ shapesol ] with the help of the derivatives of the boundary integral operators is the eulerian shape derivative . we also observe the loss of one order of regularity of the shape derivative with respect to the solution of the transmission problem . finally , for the far field , the two notions of derivative coincide , because the deformation has compact support . let @xmath1 be a bounded domain in @xmath6 and let @xmath7 denote the exterior domain @xmath8 . throughout this paper , we will for simplicity assume that the boundary @xmath3 of @xmath1 is a smooth and simply connected closed surface , so that @xmath1 is diffeomorphic to a ball . we use standard notation for surface differential operators and boundary traces . more details can be found in @xcite . for a vector function @xmath9 with @xmath10 , we denote by @xmath11 $ ] the matrix the @xmath12-th column of which is the gradient of the @xmath12-th component of @xmath13 , and we set @xmath14={{[\nabla{\boldsymbol { v}}]}^{\mathsf t}}$ ] . let @xmath15 denote the outer unit normal vector on the boundary @xmath3 . the tangential gradient of a complex - valued scalar function @xmath16 is defined by @xmath17 and the tangential vector curl is defined by @xmath18 where @xmath19 is a smooth extension of @xmath20 to the whole space @xmath2 . for a complex - valued vector function @xmath21 , we denote @xmath22 $ ] the matrix the @xmath12-th column of which is the tangential gradient of the @xmath12-th component of @xmath23 and we set @xmath24={{[\nabla_{\gamma}{\boldsymbol { u}}]}^{\mathsf t}}$ ] . the surface divergence of @xmath21 is defined by @xmath25{\boldsymbol { n}}\cdot{\boldsymbol { n}}\right),\ ] ] and the surface scalar curl is defined by @xmath26 these definitions do not depend on the choice of the extension @xmath27 . [ 2.1 ] for a vector function @xmath28 a scalar function @xmath29 and @xmath30 , we define the traces : @xmath31 @xmath32 @xmath33 @xmath34 we define in the same way the exterior traces @xmath35 , @xmath36 , @xmath37 and @xmath38 . for a domain @xmath39 we denote by @xmath40 the usual @xmath41-based sobolev space of order @xmath42 , and by @xmath43 the space of functions whose restrictions to any bounded subdomain @xmath44 of @xmath45 belong to @xmath46 . spaces of vector functions will be denoted by boldface letters , thus @xmath47 if @xmath48 is a differential operator , we write : @xmath49 the space @xmath50 is endowed with the natural graph norm . when @xmath51 , this defines in particular the hilbert spaces @xmath52 and @xmath53 . we introduce the hilbert spaces @xmath54 and @xmath55 . for @xmath56 , the trace mappings @xmath57 @xmath58 @xmath59 are then continuous . the dual of @xmath60 and @xmath61 with respect to the @xmath41 ( or @xmath62 ) scalar product is denoted by @xmath63 and @xmath64 , respectively . the surface differential operators defined above can be extended to sobolev spaces : for @xmath42 the tangential gradient and the tangential vector curl are obviously linear and continuous operators from @xmath65 to @xmath66 . the surface divergence and the surface scalar curl can then be defined on tangential vector fields by duality , extending duality relations valid for smooth functions @xmath67 @xmath68 we have the following equalities : @xmath69 @xmath70 let @xmath42 . we define the hilbert space @xmath71 endowed with the norm @xmath72 [ 2.3 ] the operators @xmath73 and @xmath74 are linear continuous from @xmath75 to @xmath76 and they can be extended to continuous linear operators from @xmath52 and @xmath77 , respectively , to @xmath78 . for @xmath79 and @xmath80 we define @xmath81 and @xmath82 in the same way and the same mapping properties hold true . recall that we assume that the boundary @xmath3 is smooth and topologically trivial . for a proof of the following result , we refer to @xcite . [ lapbel ] let @xmath42 . the laplace beltrami operator defined by @xmath83 is linear and continuous from @xmath84 to @xmath60 . for @xmath85 and @xmath86 , the equation @xmath87 has the equivalent formulation @xmath88 the operator @xmath89 is fredholm of index zero , its kernel and cokernel consisting of constant functions , so that @xmath90 is an isomorphism . here we define the space @xmath91 by @xmath92 for @xmath93 we denote the unique solution @xmath94 of by @xmath95 . this result is due to the injectivity of the operator @xmath96 from @xmath97 to @xmath98 , the lax - milgram lemma applied to for @xmath99 , and standard elliptic regularity theory . note that @xmath100 is also injective from @xmath97 to @xmath98 , and by duality both @xmath101 and @xmath102 are surjective from @xmath98 to @xmath91 notice that @xmath102 is defined in a natural way on all of @xmath103 and maps to @xmath91 , because we have @xmath104 for any scalar function @xmath105 . thus is still valid for a not necessarily tangential density @xmath106 . an analogous property for @xmath101 defined by is not available . we now recall some well known results about electromagnetic potentials . details can be found in @xcite . let @xmath107 be a positive real number and let @xmath108 be the fundamental solution of the helmholtz equation @xmath109 the single layer potential @xmath110 is given by @xmath111 , and its trace by @xmath112 as discussed in the first part of this paper @xcite , the fundamental solution is pseudo - homogeneous of class @xmath113 . the single layer potential @xmath114 is continuous across the boundary @xmath3 . as a consequence we have the following result : [ 3.1 ] let @xmath42 . the operators @xmath115 v_{\kappa } & : h^{s-\frac{1}{2}}(\gamma)\rightarrow h^{s+\frac{1}{2}}(\gamma ) \end{array}\ ] ] are continuous . the electric potential operator @xmath116 is defined for @xmath117 by @xmath118 in @xmath119 , this can be written as @xmath120 because of the helmholtz equation and the identity @xmath121 . the magnetic potential operator @xmath122 is defined for @xmath123 by @xmath124 we denote the identity operator by @xmath125 . [ 3.2 ] the potentials operators @xmath116 and @xmath122 are continuous from @xmath78 to @xmath126 for @xmath127 we have @xmath128 and @xmath129 and @xmath130 satisfy the silver - mller condition . we define the electric and the magnetic far field operators for a density @xmath131 and an element @xmath132 of the unit sphere @xmath133 of @xmath2 by @xmath134 these operators are bounded from @xmath135 to + , for all @xmath42 . we can now define the main boundary integral operators : @xmath136 the operators @xmath137 and @xmath138 are bounded operators from @xmath78 to itself . we consider the scattering of time - harmonic waves at a fixed frequency @xmath139 by a three - dimensional bounded and non - conducting homogeneous dielectric obstacle represented by the domain @xmath1 . the electric permittivity @xmath140 and the magnetic permeability @xmath141 are assumed to take constant positive real values in @xmath1 and @xmath7 . thus they will be discontinuous across the interface @xmath3 , in general . the wave number is given by @xmath142 . we distinguish the dielectric quantities related to the interior domain @xmath1 through the index @xmath12 and to the exterior domain @xmath7 through the index @xmath143 . the time - harmonic maxwell system can be reduced to second order equations for the electric field only . the time - harmonic dielectric scattering problem is then formulated as follows . * the solution of the dielectric scattering problem :* consider the scattering of a given incident electric wave @xmath144 that satisfies @xmath145 in a neighborhood of @xmath146 . the interior electric field @xmath147 and the exterior electric scattered field @xmath148 satisfy the time - harmonic maxwell equations @xmath149 the two transmission conditions , @xmath150 and the silver - mller radiation condition : @xmath151 it is well known that the problem - admits a unique solution for any positive real values of the exterior wave number @xmath152 . we refer the reader to @xcite for a proof via boundary integral equation methods . to analyze the dependency of the solution on the shape of the scatterer @xmath1 , we will use an integral representation of the solution , obtained by the single boundary integral equation method developped by the autors in @xcite . it is based on the layer ansatz for the exterior electric field @xmath153 : @xmath154 where @xmath155 is a positive real number , @xmath127 and the operator @xmath156 is defined for @xmath127 by @xmath157 thanks to the transmission conditions and the stratton - chu formula , we have the integral representation of the interior field @xmath158 where @xmath159 and @xmath160 @xmath161 the exterior dirichlet trace applied to the right - hand side vanishes . the density @xmath131 then solves the following boundary integral equation @xmath162 [ tht ] the operator @xmath163 from @xmath78 to itself is linear , bounded and invertible . moreover , given the electric incident field @xmath164 , the integral representations , of @xmath165 and @xmath166 give the unique solution of the dielectric scattering problem for all positive real values of the dielectric constants @xmath167 , @xmath168 , @xmath169 and @xmath170 . an important quantity that is of interest in many shape optimization problems is the far field pattern of the electric field , defined on the unit sphere of @xmath2 by @xmath171 we have @xmath172 . to obtain the integral representation of the far field @xmath173 of the solution , it suffices to replace in the potential operators @xmath174 and @xmath175 by the far field operators @xmath176 and @xmath177 defined in , respectively . in the method we have described , the solution @xmath178 and the far field @xmath173 are constructed from operators defined by integrals on the boundary @xmath3 and the incident field . for a fixed incident field and fixed constants @xmath179 , @xmath152 , @xmath167 , @xmath168 , these quantities therefore depend on the geometry of the boundary @xmath3 of the scatterer @xmath1 only . in the sequel we analyze the @xmath3-dependence of the solution following the definition of shape derivatives and the notations of section 4 of the paper @xcite . let us fix a reference domain @xmath1 . we consider variations of @xmath1 generated by transformations of the form @xmath180 of points @xmath181 , where @xmath182 is a smooth vector function defined on @xmath3 . this transformation deforms the domain @xmath1 in a domain @xmath183 of boundary @xmath184 . the functions @xmath182 are assumed to belong to the frchet space @xmath185 . for @xmath186 and some metric @xmath187 on @xmath185 , we set @xmath188 in the following , we choose @xmath189 small enough so that for any @xmath190 , @xmath191 is a diffeomorphism from @xmath3 to @xmath192 . the aim of this paper is to study the shape differentiability , that is , the gteaux differentiability with respect to @xmath182 , of the functionals mapping @xmath182 to the solution @xmath193 of the dielectric scattering problem with obstacle @xmath183 , and to the far field @xmath194 . in the following , we use the superscript @xmath182 for integral operators and trace mappings pertaining to @xmath184 , while functions defined on @xmath184 will often have a subscript @xmath182 . according to the boundary integral equation method described in the previous section , we have @xmath195 where @xmath196 solves the integral equation @xmath197 and @xmath198 with @xmath199 the far field pattern of the dielectric scattering problem by the interface @xmath184 is @xmath200 as defined in , the operator @xmath201 is composed of the operators @xmath202 , @xmath203 , @xmath204 et @xmath205 , which are all bounded operators on the space @xmath206 . therefore we have to study the gteaux differentiability of the following mappings on @xmath207 @xmath208 finally , the differentiability properties of the mapping @xmath209 can be deduced from those of the mapping @xmath210 . in this approach , several difficulties have to be overcome . the first one is that if we want to find the derivatives of the solution of the scattering problem , which is given as a product of operators and of their inverses , all defined on the same space @xmath78 ( for the derivative at @xmath211 ) , it is necessary to prove that the derivatives themselves are defined as bounded operators on the same space , too . on the other hand , the very definition of the differentiability of operators defined on @xmath0 raises non - trivial questions . for reducing the variable space @xmath206 to a fixed reference space , it is not sufficient , as we did in the scalar case studied in the first part @xcite , to use a change of variables . let us discuss this question : _ how to define the shape derivative of operators defined on the variable space @xmath206 ? _ in detail . we recall the notation @xmath212 for the `` pullback '' induced by the change of variables . it maps a function @xmath213 defined on @xmath184 to the function @xmath214 defined on @xmath3 . for @xmath190 , the transformation @xmath212 is an isomorphism from @xmath215 to @xmath216 . we have @xmath217 the natural idea to use this for a product of operators , proposed by potthast in @xcite in the acoustic case , is to insert the identity @xmath218 between the factors . this allows to consider integral operators on the fixed boundary @xmath3 only and to would require study the differentiability of the mappings @xmath219 but as has been already pointed out in @xcite , difficulties remain . the main cause for this is that @xmath212 does not map vector fields tangential to @xmath184 to vector fields tangential to @xmath3 , and in particular , @xmath220 this will lead to a loss of regularity if we simply try to differentiate the mappings in . let us explain this for the operator @xmath137 . the operator @xmath221 , when acting on vector fields tangential to @xmath184 , has additional regularity like what is known for the scalar double layer potential , namely it has a pseudo - homogeneous kernel of class @xmath113 , whereas it is of class @xmath222 when considered on all vector fields ( see ) . if we differentiate the kernel of @xmath223 , we will not obtain a pseudo - homogeneous kernel of class @xmath113 on the set of vector fields tangential to @xmath3 , so that we find a loss of regularity for the gteaux derivative of @xmath223 . for mapping tangent vector fields to tangent vector fields , the idea of potthast was to use projectors from one tangent plane to the other . let us denote by @xmath224 the pullback @xmath212 followed by orthogonal projection to the tangent plane to @xmath3 . this maps any vector function on @xmath184 to a tangential vector function on @xmath3 , and we have @xmath225 the restriction of @xmath224 to tangential functions on @xmath184 admits an inverse , denoted by @xmath226 , if @xmath182 is sufficiently small . the mapping @xmath226 is defined by @xmath227 and it is easy to see that @xmath224 is an isomorphism between he space of continuous tangential vector functions on @xmath184 and on @xmath3 , and for any @xmath228 between @xmath229 and @xmath230 . in the framework of continuous tangential functions it suffices to insert the product @xmath231 between factors in the integral representation of the solution to reduce the analysis to the study of boundary integral operators defined on @xmath232 , which does not depend on @xmath182 . in our case , we would obtain operators defined on the space @xmath233 which still depends on @xmath182 and is , in general , different from @xmath78 . we propose a different approach , using the helmholtz decomposition of the space @xmath206 to introduce a new pullback operator @xmath234 that defines an isomorphism between @xmath206 and @xmath78 . recall that we assume that the boundary @xmath3 is smooth and simply connected . we have the following decomposition . we refer to @xcite for the proof . the hilbert space @xmath78 admits the following helmholtz decomposition:@xmath235 since @xmath189 is chosen such that for all @xmath190 the surfaces @xmath184 are still regular and simply connected , the spaces @xmath206 admit similar decompositions . the operator of change of variables @xmath212 is an isomorphism from @xmath236 to @xmath237 and from @xmath238 to @xmath239 , and it maps constant functions to constant functions . let @xmath240 and let @xmath241 be its helmholtz decomposition . the scalar functions @xmath242 and @xmath243 are determined uniquely up to additive constants . the following operator : @xmath244 is therefore well defined , linear , continuous and invertible . its inverse @xmath245 is given by @xmath246 obviously for @xmath211 we have @xmath247 . we can now insert the identity @xmath248 between factors in the integral representation of the solution @xmath249 , and we are finally led to study the gteaux differentiability properties of the following mappings , defined on @xmath182-independent spaces . @xmath250 where @xmath251 is a compact subset of @xmath252 . these mappings are composed of scalar singular integral operators , the shape derivatives of which we studied in the first part @xcite , of surface differential operators , and of the inverse of the laplace beltrami operator , which appears in the construction of the helmholtz decomposition . let us look at the representation of the operators in in terms of the helmholtz decomposition . * helmholtz representation of @xmath253 * + the operator @xmath254 is defined for @xmath255 and @xmath256 by : @xmath257 * helmholtz representation of @xmath258 * + the operator @xmath259 is defined for @xmath255 and @xmath256 by : @xmath260 * helmholtz representation of @xmath261 * + recall that for @xmath262 , the operator @xmath263 is defined by @xmath264 we want to write @xmath265 in the form @xmath266 . using formulas , we find @xmath267 as a consequence we have for @xmath268 @xmath269 and @xmath270 the operator @xmath261 is defined for @xmath255 by @xmath271 with @xmath272 and @xmath273 * helmholtz representation of @xmath274 * + recall that for all @xmath275 , the operator @xmath221 is defined by @xmath276 using the equalities and the identity @xmath277 , we have @xmath278 proceeding in the same way as with the operator @xmath261 , we obtain that the operator @xmath274 is defined for @xmath255 by : @xmath279 with @xmath280 and @xmath281 these operators are composed of boundary integral operators with weakly singular kernel and of the surface differential operators defined in section [ boundintop ] . each of these weakly singular boundary integral operators has a pseudo - homogeneous kernel of class -1 . the @xmath282-gteaux differentiability properties of such boundary integral operators has been established in the preceding paper @xcite . it remains now to show that the surface differential operators , more precisely @xmath283 , as well as @xmath284 and its inverse , preserve their mapping properties by differentiation with respect to @xmath182 . the analysis of the surface differential operators requires the differentiability properties of some auxiliary functions , such as the outer unit normal vector @xmath285 and the jacobian @xmath286 of the change of variable @xmath180 . we recall some results established in the first part @xcite . for the definition of gteaux derivatives and the corresponding analysis , see @xcite . [ n ] the mapping @xmath287 is @xmath282-gteaux - differentiable and its first derivative in the direction of @xmath288 is given by @xmath289=-\left[\tau_{r}\nabla_{\gamma_{r}}(\tau_{r}^{-1}\xi)\right]\mathcal{n}(r).\ ] ] [ j ] the mapping @xmath290 from @xmath190 to the surface jacobian @xmath291 is @xmath282-gteaux differentiable and its first derivative in the direction of @xmath288 is given by @xmath292=j_{r_{0}}\cdot\big(\tau_{r_{0}}{\operatorname{\mathrm{div}}}_{\gamma_{r_{0}}}(\tau^{-1}_{r_{0}}\xi)\big).\ ] ] the differentiability properties of the tangential gradient and of the surface divergence in the framework of classical sobolev spaces is established in @xcite . [ nabla ] the mapping @xmath293 is @xmath282-gteaux differentiable and its first derivative for @xmath288 is given by @xmath294u=-[\mathcal{g}(r)\xi]\mathcal{g}(r)u+\big(\mathcal{g}(r)u\cdot[\mathcal{g}(r)\xi]\mathcal{n}(r)\big)\,\mathcal{n}(r).\ ] ] [ d ] the mapping @xmath295 is @xmath282-gteaux differentiable and its first derivative for @xmath288 is given by @xmath296{\boldsymbol { u}}=-{\operatorname{trace}}([\mathcal{g}(r)\xi][\mathcal{g}(r){\boldsymbol { u}}])+\left([\mathcal{g}(r){\boldsymbol { u}}]\mathcal{n}(r)\cdot[\mathcal{g}(r)\xi]\mathcal{n}(r)\right).\ ] ] similar results can now be obtained for the tangential vector curl by composition of the tangential gradient with the normal vector . [ rot2 ] the mapping @xmath297 is @xmath282-gteaux differentiable and its first derivative for @xmath288 is given by @xmath298u={{[\mathcal{g}(r)\xi]}^{\mathsf t}}\boldsymbol{\mathcal{r}}(r)u-\mathcal{d}(r)\xi\cdot\boldsymbol{\mathcal{r}}(r)u.\ ] ] let @xmath299 . by definition , we have @xmath300 . by lemmas [ n ] and [ nabla ] this application is @xmath282-gteaux differentiable . for the derivative in the direction @xmath288 we find @xmath301u=-{[\mathcal{g}(r)\xi]}\mathcal{g}(r)u\wedge \mathcal{n}(r)-\mathcal{g}(r)u\wedge[\mathcal{g}(r)\xi]\mathcal{n}(r).\ ] ] for any @xmath302 matrix @xmath303 and vectors @xmath304 and @xmath305 there holds @xmath306 we obtain the expression of the first derivative with the choice @xmath307 $ ] , @xmath308 and @xmath309 . [ rot1 ] the mapping @xmath310 is @xmath282-gteaux differentiable and its first derivative for @xmath288 is given by @xmath311{\boldsymbol { u}}=-\sum_{i=1}^3\left(\mathcal{g}(r)\xi_{i}\cdot\boldsymbol{\mathcal{r}}(r)u_{i}\right)- \mathcal{d}(r)\xi\cdot \mathcal{r}(r){\boldsymbol { u}}\ ] ] where @xmath312 and @xmath313 . let @xmath314 . notice that we have @xmath315\big)$ ] . we can therefore write @xmath316 the @xmath282-differentiability of @xmath317 results from the @xmath282-differentiability of @xmath318 . the first derivative in the direction @xmath319 is @xmath320{\boldsymbol { u}}=&-{\operatorname{trace}}\left(d \boldsymbol{\mathcal{r}}[r,\xi]{\boldsymbol { u}}\right ) \\ = & -{\operatorname{trace}}\left({{[\mathcal{g}(r)\xi]}^{\mathsf t}}[\boldsymbol{\mathcal{r}}(r){\boldsymbol { u}}]\right)-\mathcal{d}(r)\xi\cdot { \operatorname{trace}}\left(-\boldsymbol{\mathcal{r}}(r){\boldsymbol { u}}]\right ) \\ = & -\sum_{i=1}^3\left(\mathcal{g}(r)\xi_{i}\cdot\boldsymbol{\mathcal{r}}(r){\boldsymbol { u}}_{i}\right)- \mathcal{d}(r)\xi\cdot \mathcal{r}(r){\boldsymbol { u}}. \end{split}\ ] ] higher order derivatives of the tangential vector curl operator and of the surface scalar curl operator can be obtained by applying these results recursively . in view of the integral representations of the operators @xmath321 and @xmath322 , we have to study the gteaux differentiability of the mappings @xmath323 we have seen that for @xmath190 the operator @xmath324 is linear and continuous from @xmath325 to @xmath326 , that the operator @xmath327 is linear and continuous from @xmath328 to @xmath329 and that @xmath330 is defined from @xmath331 to @xmath332 . to use the chain rules , it is necessary to prove that the derivatives at @xmath211 act between the spaces @xmath103 and @xmath91 for the scalar curl operator , between the spaces @xmath98 and @xmath91 for the divergence operator and between the spaces @xmath91 and @xmath333 for the laplace beltrami operator . an important observation is @xmath334 using the duality on the boundary @xmath184 we can write for any vector density @xmath106 and any scalar density @xmath335 @xmath336 taking @xmath337 ( i.e. @xmath338 is a constant function ) then the right - hand side vanishes . this means that @xmath339 is of vanishing mean value . the mapping @xmath340 is @xmath282-gteaux differentiable and we have in any direction @xmath341 @xmath342&=&-\mathcal{j}(r)\cdot\sum\limits_{i=1}^3\left(\mathcal{g}(r)\xi_{i}\cdot\boldsymbol{\mathcal{r}}(r){\boldsymbol { u}}_{i}\right),\vspace{2mm}\\ \dfrac{\partial^m \mathcal{r}^{*}}{\partial r^m}[r,\xi]&=&0,\;\text { for all } m\ge2.\end{array}\right.\ ] ] looking at the expression of the derivatives of the tangential gradient and of the tangential vector curl in lemmas [ nabla ] and [ rot2 ] we prove iteratively that all the derivatives of @xmath343 and of @xmath344 are composed of @xmath343 and @xmath344 , so that for @xmath337 the derivatives of the right - hand side of vanishes . we have for all @xmath345 and @xmath314 : @xmath346=\int_{\gamma}\frac{\partial^m}{\partial r^m}\left\{\mathcal{r}^{*}\right\}[r,\xi]{\boldsymbol { j}}\,ds=0.\ ] ] it can also be obtained by directly deriving the expression of @xmath347 using the formulas obtained in the lemmas [ n ] to [ rot1 ] . the first derivative of @xmath348 is given by @xmath349{\boldsymbol { u}}&=-\mathcal{j}(r)\cdot\sum\limits_{i=1}^3\left(\mathcal{g}(r)\xi_{i}\cdot\boldsymbol{\mathcal{r}}(r){\boldsymbol { u}}_{i}\right)\\ & = -j_{r}.\tau_{r}\left(\sum\limits_{i=1}^3\nabla_{\gamma_{r}}(\tau_{r}^{-1}\xi_{i})\cdot{\operatorname{\mathbf{curl}}}_{\gamma_{r}}(\tau_{r}^{-1}{\boldsymbol { u}}_{i})\right ) \end{array}\ ] ] the right - hand side is of vanishing mean value since the space @xmath350 is orthogonal to @xmath351 for the @xmath352 duality product . for the second order derivative we derive @xmath353{\boldsymbol { j}}$ ] in the direction @xmath354 and we obtain @xmath355{\boldsymbol { u}}&=&\mathcal{j}(r)\cdot\sum\limits_{i=1}^3\left([\mathcal{g}(r)\eta]\mathcal{g}(r)\xi_{i}\cdot\boldsymbol{\mathcal{r}}(r){\boldsymbol { u}}_{i}\right)\\&&-\mathcal{j}(r)\cdot\sum\limits_{i=1}^3\left(\mathcal{g}(r)\xi_{i}\cdot{{[\mathcal{g}(r)\eta]}^{\mathsf t}}\boldsymbol{\mathcal{r}}(r){\boldsymbol { u}}_{i}\right)\\&=&0.\end{array}\ ] ] higher order derivatives of @xmath348 vanish . for the surface divergence , similar arguments can be applied . using the duality , we can write for @xmath356 : @xmath357 this shows that for constant @xmath337 the right - hand side vanishes and therefore @xmath358 is of vanishing mean value . we obtain the following result . the mapping @xmath359 is @xmath282-gteaux differentiable . now it remains to analyze the inverse of the laplace beltrami operator @xmath360 . we apply the following abstract result on the gteaux derivative of the inverse in a banach algebra . we leave its proof to the reader . [ d-1 ] let @xmath361 be an open subset of a frchet space @xmath362 and let @xmath363 be a banach algebra . assume that @xmath364 is gteaux differentiable at @xmath365 and that @xmath366 is invertible in @xmath363 for all @xmath367 . then @xmath368 is gteaux differentiable at @xmath369 and its first derivative in the direction @xmath370 is @xmath371 = -f(r_{0})^{-1 } \circ d f[r_{0},\xi ] \circ f(r_{0})^{-1}.\ ] ] moreover if @xmath372 is @xmath373-gteaux differentiable then @xmath368 is , too . from the preceding results we deduce the @xmath282-gteaux differentiability of the mapping @xmath374 let us note that @xmath212 induces an isomorphism between the quotient spaces @xmath375 and @xmath376 . [ delta-1 ] the mapping @xmath377 is @xmath282-gteaux differentiable and we have in any direction @xmath288 @xmath378=\;-\delta_{\gamma}^{-1}\circ d\mathcal{l}^{*}[0,\xi]\circ\delta_{\gamma}^{-1}.\ ] ] we have seen in section [ boundintop ] that the laplace beltrami operator is invertible from @xmath332 to @xmath326 . as a consequence @xmath379 is invertible from @xmath97 to @xmath91 . we conclude by using lemma [ d-1 ] . let us give another formulation of . for any @xmath86 and @xmath335 we have @xmath380 it is more convenient to differentiate the right - hand side than the left hand side . for @xmath381 , the element @xmath382 is the solution @xmath20 of @xmath383 the formula means that the first derivative at @xmath211 in the direction @xmath288 of @xmath384 is the solution @xmath385 of @xmath386ds=-\int_{\gamma}\nabla_{\gamma}v\cdot\nabla_{\gamma}\varphi\,ds,\;\text { for all } \varphi\in h^s(\gamma),\ ] ] with @xmath387 . now we have all the tools to establish the differentiability properties of the electromagnetic boundary integral operators and then of the solution to the dielectric scattering problem . for the shape - dependent integral operators we now use the following simplified notation @xmath388 in the following we use the results of the preceding paper @xcite about the gteaux differentiability of potentials and boundary integral operators with pseudo - homogeneous kernels . [ thpsi ] the mappings @xmath389 are infinitely gteaux differentiable . the derivatives can be written in explicit form by differentiating the kernels of the operators @xmath390 and @xmath391 , see ( * ? ? ? * theorem 4.7 ) , and by using the formulas for the derivatives of the surface differential operators given in section [ surfdiffop ] . the first derivatives at @xmath211 can be extended to bounded linear operators from @xmath392 to @xmath52 and to @xmath393 . given @xmath394 , the potentials @xmath395{\boldsymbol { j}}$ ] and @xmath396{\boldsymbol { j}}$ ] satisfy the maxwell equations @xmath397 in @xmath1 and @xmath7 , and the silver - mller radiation condition . let @xmath398 and let @xmath399 be its helmholtz decomposition . recall that @xmath400 and @xmath401 can be written as : @xmath402 by composition of differentiable mappings , we deduce that @xmath403 and @xmath404 are infinitely gteaux differentiable far from the boundary and that their first derivatives are continuous from @xmath392 to @xmath405 . recall that we have @xmath406 far from the boundary we can invert the differentiation with respect to @xmath407 and the derivation with respect to @xmath182 , which gives @xmath408{\boldsymbol { j}}=\kappa\ , d\psi_{m_{\kappa}}[0,\xi]{\boldsymbol { j}}\text { and } { \operatorname{\mathbf{curl}}}d\psi_{m_{\kappa}}[0,\xi]{\boldsymbol { j}}=\kappa\ , d\psi_{e_{\kappa}}[0,\xi]{\boldsymbol { j}}.\ ] ] it follows that @xmath395{\boldsymbol { j}}$ ] and @xmath409{\boldsymbol { j}}$ ] are in @xmath410 and that they satisfy the maxwell equations and the silver - mller condition . we recall from section [ helmholtzdec ] that with the notation of section [ surfdiffop ] the operator @xmath411 admits the following representation @xmath412 where @xmath413\end{array}\ ] ] and @xmath414 \vspace{3mm}\\ & & + \dfrac{1}{\kappa}\left(\tau_{r}v^{r}_{\kappa}\tau_{r}^{-1}\right)\left(\tau_{r}\delta_{\gamma_{r}}(\tau_{r}^{-1}p)\right).\end{array}\ ] ] let @xmath415 and let @xmath416 be its helmholtz decomposition . we want to derive @xmath417 we find @xmath418{\boldsymbol { j}}=\nabla_{\gamma}dp[0,\xi]+{\operatorname{\mathbf{curl}}}_{\gamma}dq[0,\xi].\ ] ] thus the derivative with respect to @xmath182 of @xmath419 is given by the derivatives of the functions @xmath420 and of @xmath421.we also note that for an @xmath182-dependent vector function @xmath366 on @xmath3 there holds @xmath422=\pi(0 ) df[0,\xi].\ ] ] by composition of infinitely differentiable mappings we obtain the following theorem . the mapping : @xmath423 is infinitely gteaux differentiable . the derivatives can be written in explicit form by differentiating the kernel of the operator @xmath263 , see ( * ? ? ? * corollary 4.5 ) , and by using the formulas for the derivatives of the surface differential operators given in section [ surfdiffop ] . similarly , recall that the operator @xmath424 admits the following representation : @xmath425 where @xmath426\vspace{2mm}\\ & + ( \mathcal{l}^{*}(r))^{-1}(j_{r}\cdot\tau_{r}d_{\kappa}^{r}\tau_{r}^{-1})(\tau_{r}\delta_{\gamma_{r}}(\tau_{r}^{-1}p ) ) \end{array}\ ] ] and @xmath427 \end{array}\ ] ] with @xmath428 the mapping : @xmath429 is infinitely gteaux differentiable . the gteaux derivatives have the same regularity as @xmath137 , so that they are compact operators in @xmath78 . the derivatives can be written in explicit form by differentiating the kernel of the operators @xmath221 , see ( * ? ? ? * corollary 4.5 ) , and by using the formulas for the derivatives of the surface differential operators given in section [ surfdiffop ] . the differentiability of the double layer boundary integral operator is established in ( * ? ? ? * example 4.10 ) . it remains to prove the infinite gteaux differentiability of the mapping @xmath430 the function @xmath431 is pseudo - homogeneous of class 0 . we then have to prove that for any fixed @xmath432 and any function @xmath433 the gteaux derivatives of @xmath434 behave as @xmath435 when @xmath436 tends to zero . to do so , either we write @xmath437 or we use lemmas [ n ] and [ nabla ] and straighforward computations . [ shapederiv ] assume that @xmath438 and that the mappings @xmath439 are gteaux differentiable at @xmath211 . then the mapping from @xmath190 to the solution @xmath440 of the scattering problem by the obstacle @xmath183 is gteaux differentiable at @xmath211 . we use the integral equation method described in theorem [ tht ] . let @xmath131 be the solution of the integral equation . by composition of infinitely differentiable mappings we see that @xmath441 is @xmath282-gteaux differentiable . then with and we get for the exterior field @xmath442 @xmath443=\;&\left(-d\psi_{e_{\kappa_{e}}}[0,\xi]-i\eta d\psi_{m_{\kappa_{e}}}[0,\xi]c^{*}_{0}-i\eta\psi_{m_{\kappa_{e}}}d c^{*}_{0}[0,\xi]\right){\boldsymbol { j}}\\ & + ( -\psi_{e_{\kappa_{e}}}-i\eta\psi_{m_{\kappa_{e}}}c^*_{0}){\boldsymbol{\mathsf s}}^{-1}\big(-d { \boldsymbol{\mathsf s}}[0,\xi]\,{\boldsymbol { j}}\big ) \\ + ( -&\psi_{e_{\kappa_{e}}}\!\!\!-i\eta\psi_{m_{\kappa_{e}}}c^*_{0}){\boldsymbol{\mathsf s}}^{-1}\left(-\rho d m_{\kappa_{i}}[0,\xi]\gamma_{d}{\boldsymbol{\mathsf e}}^{inc}- d c_{\kappa_{i}}[0,\xi]\gamma_{n_{\kappa_{e}}}{\boldsymbol{\mathsf e}}^{inc}\right ) \\ & \!\!\!+(-\psi_{e_{\kappa_{e}}}-i\eta\psi_{m_{\kappa_{e}}}c^*_{0}){\boldsymbol{\mathsf s}}^{-1}\left(-\rho\big(\tfrac{1}{2}+m_{\kappa_{i}}\right)d\left\ { { \boldsymbol{\mathsf p}}_{r}\gamma_{d}^{r}{\boldsymbol{\mathsf e}}^{inc}\right\}[0,\xi]\big ) \\ & + ( -\psi_{e_{\kappa_{e}}}-i\eta\psi_{m_{\kappa_{e}}}c^*_{0}){\boldsymbol{\mathsf s}}^{-1}\left(-c_{\kappa_{i}}d\left\ { { \boldsymbol{\mathsf p}}_{r}\gamma_{n_{\kappa_{e}}}^{r}{\boldsymbol{\mathsf e}}^{inc}\right\}[0,\xi]\right ) . \end{split}\ ] ] we know that @xmath394 , so that the first terms on the right - hand side are in @xmath444 , and the hypotheses guarantee that the last two terms are in @xmath444 . for the interior field we write @xmath445=\;&-\frac{1}{\rho}d \psi_{e_{\kappa_{i}}}[0,\xi]\gamma_{n_{\kappa_{e}}}^c\left({\boldsymbol{\mathsf e}}^{s}+{\boldsymbol{\mathsf e}}^{inc}\right)-d\psi_{m_{\kappa_{i}}}[0,\xi]\gamma_{d}^c\left({\boldsymbol{\mathsf e}}^{s}+{\boldsymbol{\mathsf e}}^{inc}\right ) \\ & -\frac{1}{\rho}\psi_{e_{\kappa_{i}}}d\hspace{-.5mm}\left\{{\boldsymbol{\mathsf p}}_{r}\gamma_{n_{\kappa_{e}}}^{c , r}\left(\mathscr{e}^s(r)+{\boldsymbol{\mathsf e}}^{inc}\right)\right\}[0,\xi]\\ & \qquad\qquad-\psi_{m_{\kappa_{i}}}d\hspace{-.5mm}\left\{{\boldsymbol{\mathsf p}}_{r}\gamma_{d}^{c , r}\left(\mathscr{e}^s(r)+{\boldsymbol{\mathsf e}}^{inc}\right)\right\}\hspace{-1mm}[0,\xi ] . \end{split}\ ] ] the hypotheses guarantee that @xmath446 and @xmath447 are in @xmath392 , which implies that the first two terms are in @xmath52 , and that the last two terms are in @xmath52 . the mapping sending @xmath190 to the far field pattern @xmath448 of the solution to the scattering problem by the obstacle @xmath183 is @xmath282-gteaux differentiable . the mapping @xmath449 is @xmath282-gteaux differentiable and the derivatives define smooth kernels . by the linearity of the integral we deduce that the boundary to far field operators @xmath450 are @xmath282-gteaux differentiable . for @xmath451 we have : @xmath452{\boldsymbol { j}}(\hat{x})=i\kappa\hat{x}\wedge\left(\int_{\gamma}e^{-i\kappa\hat{x}\cdot y}\big({\operatorname{\mathrm{div}}}_{\gamma}\xi(y)-i\kappa\hat{x}\cdot\xi(y)\big){\boldsymbol { j}}(y)ds(y)\right)\wedge\hat{x},\ ] ] and @xmath453{\boldsymbol { j}}(\hat{x})=\kappa\hat{x}\wedge\left(\int_{\gamma}e^{-i\kappa\hat{x}\cdot y}\big({\operatorname{\mathrm{div}}}_{\gamma}\xi(y)-i\kappa\hat{x}\cdot\xi(y)\big){\boldsymbol { j}}(y)ds(y)\right).\ ] ] we conclude by using the integral representation of @xmath454 and previous theorems . the following theorem gives a caracterization of the first gteaux derivative of @xmath455 at @xmath211 . [ t : firstder ] under the hypotheses of theorem [ shapederiv ] , the first derivative of the solution @xmath456 of the dielectric scattering problem at @xmath211 in the direction @xmath288 solves the following transmission problem : @xmath457-\kappa_{i}^2 d\mathscr{e}^{i}[0,\xi]=0 \\ { \operatorname{\mathbf{curl}}}{\operatorname{\mathbf{curl}}}d \mathscr{e}^{s}[0,\xi]-\kappa_{e}^2 d\mathscr{e}^{s}[0,\xi]=0 \end{aligned}\right.\ ] ] with the interface conditions @xmath458-{\boldsymbol { n}}\wedge d\mathscr{e}^{s}[0,\xi]=g_{d } \\ \mu_{i}^{-1}{\boldsymbol { n}}\wedge{\operatorname{\mathbf{curl}}}d\mathscr{e}^{i}[0,\xi]-\mu_{e}^{-1}{\boldsymbol { n}}\wedge{\operatorname{\mathbf{curl}}}d\mathscr{e}^{s}[0,\xi]=g_{n } , \end{aligned}\right.\ ] ] where with the solution @xmath459 of the scattering problem , @xmath460 and @xmath461 and @xmath462 $ ] satisfies the silver - mller radiation condition . we have shown in the previous paragraph that the potential operators and their gteaux derivatives satisfy the maxwell equations and the silver - mller radiation condition . it remains to compute the boundary conditions . they can be obtained from the integral representation , but this is rather tedious . a simpler way consists in deriving for a fixed @xmath463 the expression @xmath464 this gives @xmath465(x)\wedge\left({\boldsymbol{\mathsf e}}^{i}(x)-{\boldsymbol{\mathsf e}}^s(x)-{\boldsymbol{\mathsf e}}^{inc}(x)\right ) \\ & + { \boldsymbol { n}}(x)\wedge\left(d\mathscr{e}^i[0,\xi](x)-d\mathscr{e}^s[0,\xi](x)\right ) \\ & + { \boldsymbol { n}}\wedge\left(\xi(x)\cdot\nabla\left({\boldsymbol{\mathsf e}}^{i}-{\boldsymbol{\mathsf e}}^s-{\boldsymbol{\mathsf e}}^{inc}\right)\right ) . \end{aligned}\ ] ] we now use the explicit form of the shape derivatives of the normal vector given in lemma [ n ] : @xmath466=-\left[\nabla_{\gamma}\xi\right]{\boldsymbol { n}}$ ] , and the formula @xmath467 we obtain @xmath468(x)-d\mathscr{e}^s[0,\xi](x)\right)=\\ \left[\nabla_{\gamma}\xi\right]{\boldsymbol { n}}\wedge\left({\boldsymbol{\mathsf e}}^{i}(x)-{\boldsymbol{\mathsf e}}^s(x)-{\boldsymbol{\mathsf e}}^{inc}(x)\right ) \\ -{\boldsymbol { n}}\wedge\left(\xi(x)\cdot\nabla_{\gamma}\left({\boldsymbol{\mathsf e}}^{i}(x)-{\boldsymbol{\mathsf e}}^s(x)-{\boldsymbol{\mathsf e}}^{inc}(x)\right)\right ) \\ -(\xi\cdot{\boldsymbol { n}}){\boldsymbol { n}}\wedge\dfrac{\partial}{\partial{\boldsymbol { n}}}\left({\boldsymbol{\mathsf e}}^{i}(x)-{\boldsymbol{\mathsf e}}^s(x)-{\boldsymbol{\mathsf e}}^{inc}(x)\right).\end{gathered}\ ] ] since the tangential component of @xmath469 vanishes , we have @xmath470\xi\right)\end{gathered}\ ] ] and @xmath471{\boldsymbol { n}}\right)\wedge\left({\boldsymbol{\mathsf e}}^{i}(x)-{\boldsymbol{\mathsf e}}^s(x)-{\boldsymbol{\mathsf e}}^{inc}(x)\right)\\ = \left(\left[\nabla_{\gamma}\xi\right]{\boldsymbol { n}}\right)\wedge{\boldsymbol { n}}\left({\boldsymbol{\mathsf e}}^{i}(x)-{\boldsymbol{\mathsf e}}^s(x)-{\boldsymbol{\mathsf e}}^{inc}(x)\right)\cdot{\boldsymbol { n}}.\end{gathered}\ ] ] since we are on a regular surface , we have @xmath472 and @xmath473{\boldsymbol { n}}\right)\wedge{\boldsymbol { n}}-{\boldsymbol { n}}\wedge\left(\left[{{\nabla_{\gamma}{\boldsymbol { n}}}^{\mathsf t}}\right]\xi\right)={\operatorname{\mathbf{curl}}}_{\gamma}\left(\xi\cdot{\boldsymbol { n}}\right).\ ] ] using the expansion ( see @xcite ) @xmath474{\boldsymbol { u}}\right)\wedge{\boldsymbol { n}}-\left(\frac{\partial{\boldsymbol { u}}}{\partial{\boldsymbol { n}}}\right)\wedge{\boldsymbol { n}}\ ] ] we obtain that @xmath475 here we used that the curvature operator @xmath476 $ ] acts on the tangential component of vector fields , so that @xmath477\left(\gamma{\boldsymbol{\mathsf e}}^{i}-\gamma^c\big({\boldsymbol{\mathsf e}}^s+{\boldsymbol{\mathsf e}}^{inc}\big)\right)=0.\ ] ] thus we have @xmath478 to obtain the second transmission condition , we use similar computations with the electric field @xmath479 replaced by the magnetic field @xmath480 . we have presented a complete differentiability analysis of the electromagnetic integral operators with respect to smooth deformations of a smooth boundary in the framework of sobolev spaces . using the boundary integral equation approach we have established that the far - field pattern of the dielectric scattering problem is infinitely differentiable with respect to the deformations and we gave a characterization of the first derivative as the far - field pattern of a new transmission problem . in the case of a non - smooth boundary a polyhedral or more generally a lipschitz boundary the formulas determining the first derivative given in theorem [ t : firstder ] are problematic . the normal vector field @xmath15 will have discontinuities , and the factor @xmath481 and vector product with @xmath15 that appear in the right - hand side of may not be well defined in the energy trace spaces . it is , however , known for the acoustic case that the far field is infinitely shape differentiable for non - smooth boundaries , too , see @xcite for a proof via the implicit function theorem . our procedure of using a boundary integral representation gives an alternative way of characterizing the shape derivatives of the solution of the dielectric scattering problem and of its far field . we do not require the computation of the boundary traces of the solution and taking tangential derivatives and multiplication by possibly discontinuous factors . instead we determine the shape derivatives of the boundary integral operators . while the study of gteaux differentiability of boundary integral operators , for the case of smooth deformations of a lipschitz domain , is still an open problem that will require further work , our approach via helmholtz decompositions seems to be a promising starting point for tackling this question . let us briefly indicate why we think this is so . we consider the case where the boundary @xmath3 is merely lipschitz , but the deformation is defined by a vector field @xmath319 that is smooth ( at least @xmath482 ) in a neighborhood of @xmath3 . note that the reduction to purely normal displacements that is often used for studying shape optimization problems for smooth boundaries does not make sense here , as soon as there are corners present . in this situation , many of the ingredients of our toolbox are still available . here are some of them . first , the change of variables mapping @xmath212 still defines an isomorphism between @xmath238 and @xmath239 . by duality , we see that the mapping @xmath483 defines an isomorphism between @xmath484 and @xmath485 . when we want to transport the energy space @xmath206 , we can still use helmholtz decomposition . namely , the following result is known , see @xcite . a natural idea for the transport of the energy trace space is then , instead of , to define @xmath489 this is justified by the sequence of isomorphisms @xmath490 the inverse of the transformation @xmath234 is given by @xmath491 in this situation it seems to be more convenient to rewrite the operators @xmath261 and @xmath274 as operators acting on the the scalar fields @xmath492 and @xmath493 instead of @xmath494 and @xmath495 . for example , the operator @xmath261 is defined for @xmath496 by @xmath497 with @xmath498\end{array}\ ] ] and @xmath499 \vspace{3mm}\\ & & + \dfrac{1}{\kappa}\left(\tau_{r}v^{r}_{\kappa}\tau_{r}^{-1}\right)\left(j_{r}^{-1}p^{*}\right ) . \end{array}\ ] ] here we have used the same notation for the surface differential operators as introduced in section [ surfdiffop ] . these formulas together with similar ones for the operator @xmath500 can now be the starting point for generalization of the analysis of shape differentiability of the maxwell boundary integral operators to lipschitz domains . we expect that the results for the differentiability of the surface differential operators and then also of the boundary integral operators will be similar to what we have obtained for the case of smooth domains . this is , however , far from trivial and will require further work . height 2pt depth -1.6pt width 23pt , _ on traces for functional spaces related to maxwell s equations . hodge decompositions on the boundary of lipschitz polyhedra and applications _ , math . methods appl . , 24 ( 2001 ) , pp . 3148 . height 2pt depth -1.6pt width 23pt , _ shape derivatives of boundary integral operators in electromagnetic scattering . part i : shape differentiability of pseudo - homogeneous boundary integral operators _ , ( 2011 ) . height 2pt depth -1.6pt width 23pt , _ erratum : `` frechet derivatives in inverse obstacle scattering '' [ inverse problems * 11 * ( 1995 ) , no . 2 , 371382 ; mr1324650 ( 95k:35217 ) ] _ , inverse problems , 14 ( 1998 ) , pp . 209210 . height 2pt depth -1.6pt width 23pt , _ frchet - differenzierbarkeit von randintegraloperatoren und randwertproblemen zur helmholtzgleichung und den zeitharmonischen maxwellgleichungen _ , phd in numerical analysis , gttingen , 1996 . , _ nonlinear functional analysis _ , gordon and breach science publishers , new york , 1969 . notes by h. fattorini , r. nirenberg and h. porta , with an additional chapter by hermann karcher , notes on mathematics and its applications .

we analyze the position and momentum uncertainties of the energy eigenstates of the harmonic oscillator in the context of a deformed quantum mechanics , namely , that in which the commutator between the position and momentum operators is given by=i\hbar(1+\beta \hat{p}^2)$ ] . this deformed commutation relation leads to the minimal length uncertainty relation , which implies that at small while at large . we find that the uncertainties of the energy eigenstates of the normal harmonic oscillator ( ) , derived in ref . , only populate the branch . the other branch , , is found to be populated by the energy eigenstates of the ` inverted ' harmonic oscillator ( ) . the hilbert space in the inverted case admits an infinite ladder of positive energy eigenstates provided that^{1/4}$ ] . correspondence with the classical limit is also discussed .

1109.2680

the consequences of deforming the canonical commutation relation between the position and momentum operators to @xmath9 \;=\ ; i\hbar ( 1 + \beta \hat{p}^2 ) \label{commutationrelation}\ ] ] have been studied in various contexts by many authors @xcite . the main motivation behind this was to use such deformed quantum mechanical systems as models which obey the minimal length uncertainty relation ( mlur ) @xcite : @xmath10 this relation is expected on fairly generic grounds in quantum gravity @xcite , and has been observed in perturbative string theory @xcite . the mlur implies the existence of a minimal length @xmath11 below which the uncertainty in position , @xmath12 , can not be reduced . for quantum gravity , @xmath13 would be the planck scale , @xmath14 , while for string theory this would be the string length scale , @xmath15 . the investigation of said model systems could be expected to shed some light on the nature of these , and other , theories which possess a minimal length scale . as a function of @xmath3 when they obey the minimal length uncertainty relation , eq . ( [ mlur ] ) , is shown in red . the blue line indicates @xmath16 while the green line indicates @xmath17 . , width=302 ] note that the uncertainty in position which saturates the mlur bound behaves as @xmath18 for @xmath19 , while @xmath4 for @xmath20 , as illustrated in fig . [ mlurfig ] . while we are familiar with the @xmath18 behavior from canonical quantum mechanics , the @xmath4 behavior is quite novel . it behooves us to understand how it can come about , and how it can coexist with the canonical @xmath18 behavior within a single quantum mechanical system . to this end , we calculate @xmath12 and @xmath3 for the energy eigenstates of the harmonic oscillator , @xmath21 the wave - functions of which were derived explicitly in ref . we find that all of the eigenstates of this hamiltonian inhabit the @xmath2 branch , and not the other branch , as long as both the spring constant @xmath22 and the mass @xmath23 remain positive . the cross - over happens when the inverse of the mass , @xmath24 , is allowed to decrease through zero into the negative , in which case all the energy eigenstates will move smoothly over to the @xmath6 branch . the objective of this paper is to provide a detailed account of this result . in the following , we solve the schrdinger equation for the above hamiltonian without assuming a specific sign for the mass @xmath23 . the spring constant @xmath22 is kept positive throughout . we find that the ` inverted ' harmonic oscillator with @xmath25 and @xmath7 admits an infinite ladder of normalizable positive energy eigenstates provided that @xmath26 where @xmath27^{1/4 } \label{adef}\ ] ] is the characteristic length scale of the harmonic oscillator . the uncertainties @xmath12 and @xmath3 are calculated for the energy eigenstates , and the above mentioned cross - over through @xmath28 from the @xmath2 branch to the @xmath29 branch is demonstrated . we then take the classical limit of our deformed commutation relation and work out the evolution of the classical harmonic oscillator for both the positive and negative mass cases . it is found that for the ` inverted ' @xmath7 case , the time it takes for the particle to travel from @xmath30 to @xmath31 is finite , demanding the compactification of @xmath32-space , and also rendering the classical probability of finding the particle near the origin finite . this provides a classical explanation of why ` bound ' states are possible for the ` inverted ' harmonic oscillator in this modified mechanics . the position and momentum operators obeying eq . ( [ commutationrelation ] ) can be represented in momentum space by @xcite @xmath33 the inner product between two states is @xmath34 this definition ensures the symmetricity of the operator @xmath35 . the schrdinger equation for the harmonic oscillator in this representation is thus @xmath36\psi(p ) \;=\ ; e\,\psi(p)\;. \label{schrodinger - p}\ ] ] here , we do not assume @xmath5 as usual , so the kinetic energy term can contribute with either sign . a change of variable from @xmath37 to @xmath38 maps the region @xmath39 to @xmath40 and casts the @xmath35 and @xmath41 operators into the forms @xmath42 with inner product given by @xmath43 note that @xmath35 is the wave - number operator in @xmath44-space , so the fourier coefficients of the wave - function in @xmath44-space will provide the probability amplitudes for a discretized @xmath32-space . ( [ schrodinger - p ] ) is thus transformed into : @xmath45\psi(\rho ) \;=\ ; e\,\psi(\rho)\;. \label{schrodinger - rho}\ ] ] in @xmath44-space , the potential energy term @xmath46 effectively becomes the kinetic energy , and the kinetic energy term @xmath47 effectively becomes a tangent - squared potential which is ` inverted ' when @xmath48 . we next introduce dimensionless parameters and a dimensionless variable by @xmath49^{1/4 } \;=\ ; \dfrac{\delta x_{\min}}{a } \;,\cr \varepsilon & \equiv & \frac{2|m|e\beta}{\kappa^2 } \;,\cr \xi & \equiv & \frac{\sqrt{\beta}\rho}{\kappa } \;,\end{aligned}\ ] ] where the length - scale @xmath50 was introduced in eq . ( [ adef ] ) . the dimensionless variable @xmath51 is in the range @xmath52 and the inner product is @xmath53 the dimension of the inner product has all been absorbed into the prefactor @xmath54 . the schrdinger equation becomes @xmath55\psi(\xi ) = 0\ ; , \label{schrodinger - xi}\end{aligned}\ ] ] where the minus sign in front of the tangent - squared potential is for the case @xmath5 , and the plus sign for the case @xmath7 . let @xmath56 , where @xmath57 , @xmath58 , and @xmath59 is a constant to be determined . the variable @xmath60 is in the range @xmath61 with inner product given by @xmath62 the equation for @xmath63 is @xmath64 f = 0\;. \cr & & \label{schrodinger - s1}\end{aligned}\ ] ] we fix @xmath59 by requiring the coefficient of the tangent squared term to vanish : @xmath65 the solutions are @xmath66 where we have chosen the branches for which @xmath67 to prevent the inner - product , eq . ( [ inner - s ] ) , from blowing up at the domain boundaries . note that @xmath68 while @xmath69 , and that @xmath70 in the limit @xmath71 . the dependence of @xmath72 on @xmath73 is shown in fig . [ lambdaplot ] . the @xmath74 branch does not extend below @xmath75 . ( blue ) and @xmath74 ( red ) , eq . ( [ lambdaplusminus ] ) , as functions of @xmath76.,width=302 ] setting @xmath77 for @xmath5 , and @xmath78 for @xmath7 simplifies eq . ( [ schrodinger - s1 ] ) to @xmath79 the sign of @xmath23 being encoded in the value of @xmath59 . since @xmath63 should be non - singular at @xmath80 , we demand a polynomial solution to eq . ( [ schrodinger - s2 ] ) . this requirement imposes the following condition on the coefficient of @xmath81 : @xmath82 where @xmath83 is a non negative integer @xcite . ( [ schrodinger - s2 ] ) becomes @xmath84 the solution of which is given by the gegenbauer polynomial : @xmath85 the gegenbauer polynomials satisfy the following orthogonality relation : @xmath86 ^ 2\,n!\,(n+\lambda)\ , } \,\delta_{nm}\;. \label{gegenbauerorthogonalityrelation}\ ] ] the energy eigenvalues follow from the condition eq . ( [ eigenpm1 ] ) . replacing @xmath59 with @xmath72 , we find @xmath87 \phantom{\bigg| } \cr & = & \dfrac{n^2+(2n+1)\lambda_\pm}{\sqrt{\lambda_\pm\bigl|\lambda_\pm-1\bigr| } } \cr & = & \kappa^2 \left ( n^2 + n + \frac{1}{2 } \right ) + \left ( 2n+1 \right ) \sqrt { \frac{\kappa^4}{4 } \pm 1 } \ ; , \cr & & \label{epsilonpmn}\end{aligned}\ ] ] or in the original dimensionful units , @xmath88 \cr & = & \dfrac{k}{2 } \biggl [ \left ( n + \frac{1}{2 } \right ) \sqrt { ( \delta x_{\min})^4 \pm 4 a^4 } \cr & & \qquad\qquad + \left ( n^2 + n + \frac{1}{2 } \right ) ( \delta x_{\min})^2 \biggr ] \ ; , \cr & & \label{enpm}\end{aligned}\ ] ] where @xmath89 . for the @xmath5 case , we can take the limit @xmath90 , and we recover @xmath91 for the @xmath7 case , it is clear that we must have @xmath92 for the square - root in eq . ( [ enpm ] ) to remain real . this is the condition we cited in eq . ( [ negativemasscondition ] ) . therefore , the limit @xmath90 can not be taken in this case for non - zero @xmath50 . the two cases converge when @xmath93 , at which @xmath94 , and we find that the energy levels in that limit are @xmath95 thus , the @xmath96 and @xmath48 cases connect smoothly at @xmath28 . the normalized energy eigenfunctions are thus given by : @xmath97 where @xmath98{\beta}\left [ 2^{\lambda}\gamma(\lambda ) \sqrt { \dfrac { n!\,(n+\lambda ) } { 2\pi\,\gamma(n+2\lambda ) } } \right ] \;,\cr c & = & \cos\sqrt{\beta}\rho \;=\ ; \dfrac { 1 } { \sqrt{1+\beta p^2 } } \;,\cr s & = & \sin\sqrt{\beta}\rho \;=\ ; \dfrac { \sqrt{\beta } p } { \sqrt{1+\beta p^2 } } \;.\end{aligned}\ ] ] the wave - functions for the first few energy eigenstates for several representative values of @xmath59 are shown in figs . [ wavefunctions ] . -dependence of the wave - functions of the first few energy eigenstates . the @xmath99 values correspond to @xmath5 , while the @xmath100 values correspond to @xmath7 . the @xmath101 case corresponds to the limit @xmath93 . , title="fig:",width=302 ] -dependence of the wave - functions of the first few energy eigenstates . the @xmath99 values correspond to @xmath5 , while the @xmath100 values correspond to @xmath7 . the @xmath101 case corresponds to the limit @xmath93 . , title="fig:",width=302 ] -dependence of the wave - functions of the first few energy eigenstates . the @xmath99 values correspond to @xmath5 , while the @xmath100 values correspond to @xmath7 . the @xmath101 case corresponds to the limit @xmath93 . , title="fig:",width=302 ] versus @xmath12 for the lowest six energy eigenstates of the harmonic oscillator . @xmath12 is in units of @xmath102 , while @xmath3 is in units of @xmath103 . the location of the state along the curves shown is determined by the value of @xmath59 defined in eq . ( [ lambdaplusminus ] ) . the @xmath101 points correspond to the case @xmath28 . as @xmath24 is increased to the positive side , the value of @xmath59 will increase away from one and the state will move toward the left along the @xmath18 branch . if @xmath24 is decreased into the negative , the value of @xmath59 will decrease toward @xmath104 , and the state will move toward the right along the @xmath29 branch . the @xmath105 curve ( shown in red ) corresponds to the minimal length uncertainty bound , eq . ( [ mlur ] ) . , width=302 ] using the wave - functions derived above , and the formula provided in the appendix , the expectation values of @xmath35 , @xmath41 , @xmath106 , and @xmath107 for the energy eigenstates are found to be @xmath108}{(2\lambda-1)}\;,\cr { \left\langle n,\lambda \right|}\hat{p}^2{\left| n,\lambda \right\rangle } & = & \dfrac{1}{\beta}\left(\dfrac{2n+1}{2\lambda-1}\right)\;,\end{aligned}\ ] ] giving the uncertainties in @xmath109 and @xmath110 as @xmath111}{(2\lambda-1)}}\;,\cr \delta p_n & = & \dfrac{1}{\sqrt{\beta}}\sqrt{\dfrac{2n+1}{2\lambda-1}}\;. \label{dxndpn}\end{aligned}\ ] ] for fixed @xmath83 and fixed @xmath112 , @xmath113 is a monotonically decreasing function of @xmath59 in the range @xmath114 , and a monotonically increasing one in the range @xmath115 . @xmath116 on the other hand is a monotonically decreasing function of @xmath59 throughout . eliminating @xmath59 from the above expressions , we find @xmath117 \bigl [ ( 2n+1)^2 + \beta\delta p_n^2 \bigr ] } \cr & \ge & \dfrac{1}{2 } \left ( \dfrac{1}{\sqrt{\beta}\delta p_n } + \sqrt{\beta}\delta p_n \right ) \ ; , \label{dpdxcurve}\end{aligned}\ ] ] where the equality in the second line is saturated for the @xmath105 case only the first line gives the curve on the @xmath3-@xmath12 plane that the point @xmath118 follows as @xmath59 is varied . differentiating with respect to @xmath116 we find @xmath119 } \cr & = & \dfrac { \beta^2\delta p_n^4 - ( 2n+1)^2 } { 2\sqrt{\beta}\delta p_n^2\sqrt { \bigl [ 1 + \beta\delta p_n^2 \bigr ] \bigl [ ( 2n+1)^2 + \beta\delta p_n^2 \bigr ] } } \;,\end{aligned}\ ] ] indicating that the curve is flat at the @xmath101 point where @xmath120 and @xmath113 reaches its minimum of @xmath121 . therefore , the @xmath101 point is the turn - around point where the uncertainties switch from the @xmath2 behavior to the @xmath6 behavior . to go from one branch to another one must flip the sign of the mass @xmath23 . eliminating @xmath83 from eq . ( [ dxndpn ] ) , we find @xmath122 } \;.\ ] ] this gives the curve on which the points @xmath118 fall on for constant @xmath59 . in particular , for @xmath101 this reduces to @xmath123 and gives the @xmath28 boundary between the @xmath96 and @xmath48 regions in @xmath12-@xmath3 space . these properties of the uncertainties have been plotted in fig . [ ho - dpdx ] . as shown in fig . [ ho - dpdx ] , the value of @xmath59 determines where the uncertainties @xmath118 are along their trajectories given by eq . ( [ dpdxcurve ] ) , with @xmath99 keeping the uncertainties on the @xmath2 branch of the trajectory , while @xmath114 keeping them on the @xmath6 branch . let us consider a few limiting values of @xmath59 to see the behavior of the solutions there . the @xmath125 limit only exist for the @xmath5 case where @xmath77 . as @xmath125 , the parameter @xmath126 diverges to infinity as @xmath127 where @xmath128 . in that limit , the gegenbauer polynomials become hermite polynomials : @xmath129 noting that as @xmath125 , we have @xmath130 we can conclude that @xmath131 similarly , @xmath132 using stirling s formula @xmath133 the normalization constant can be shown to converge to @xmath134{m\hbar\omega\pi } } \;.\end{aligned}\ ] ] therefore , @xmath135{m\hbar\omega\pi } } \exp\left(-\dfrac{p^2}{2m\hbar\omega}\right ) h_n\!\left(\dfrac{p}{\sqrt{m\hbar\omega}}\right)\ ; , \cr & & \end{aligned}\ ] ] which are just the usual harmonic oscillator wave - functions in momentum space . the energy eigenvalues reduce to the usual ones given in eq . ( [ betazerolimit ] ) . the uncertainties reduce to the usual ones as well @xmath136 which satisfy @xmath137 the limit @xmath101 is reached when @xmath112 and @xmath22 are kept constant while @xmath139 is taken to infinity . when @xmath101 , the gegenbauer polynomials become the chebycheff polynomials of the second kind : @xmath140 where @xmath141 while the normalization constant reduces to @xmath142{\beta } \sqrt{\dfrac{2}{\pi}}\;.\ ] ] therefore , @xmath143{\beta } } \;=\ ; \sqrt{\dfrac{2}{\pi}}\,c\,u_n(s)\;.\ ] ] the orthonormality relation for the chebycheff polynomials is @xmath144 and we can see that the correct normalization constant is obtained . since the argument in our case is @xmath145 , it is more convenient to express the chebycheff polynomials as @xmath146}{\cos\theta}\;,\cr \cr u_{2n+1}(\sin\theta ) & = & ( -1)^n\dfrac{\sin[(2n+2)\theta]}{\cos\theta}\ ; , \label{chebycheff2}\end{aligned}\ ] ] for @xmath147 . this will allow us to write @xmath148{\beta } } & = \ ; ( -1)^n \sqrt{\dfrac{2}{\pi}}\cos\bigl[(2n+1)\sqrt{\beta}\rho\bigr ] \;,\\ \dfrac{\psi_{2n+1}^{(1)}(p)}{\sqrt[4]{\beta } } & = \ ; ( -1)^n \sqrt{\dfrac{2}{\pi}}\sin\bigl[(2n+2)\sqrt{\beta}\rho\bigr ] \;. \end{array}\ ] ] the energy eigenvalues in this limit were given in eq . ( [ minfinitelimit ] ) . here , our procedure of keeping the spring constant @xmath22 fixed while taking @xmath139 to infinity maintains the finiteness of @xmath149 , while taking the kinetic energy contribution to @xmath149 to zero . from the @xmath83-dependence of the energies , we can see that , in this limit , the problem reduces to that of an infinite square well potential , of width @xmath150 , in @xmath44-space . indeed , the effective potential in @xmath44-space was @xmath151 this can also be seen from the form of the energy eigenfunctions , which have reduced to simple sines and cosines . we will see in the next section that the classical solution also behaves as that of a particle in an infinite square well potential in @xmath44-space in the same limit . the uncertainties become @xmath152 as was shown in fig . [ ho - dpdx ] . note that @xmath153 & = & \delta x_\mathrm{min}\;\dfrac{n+1}{\sqrt{2n+1 } } \cr & \le & \delta x_\mathrm{min}\bigl(n+1\bigr ) \;=\ ; \delta x_n\ ; , \cr & & \end{aligned}\ ] ] the bound being saturated only for the ground state @xmath105 . the @xmath155 limit is reached as @xmath156 when @xmath7 . in this limit , the gegenbauer polynomials become the legendre polynomials , @xmath157 , while the normalization constant reduces to @xmath158{\beta}\sqrt{\dfrac{2n+1}{2}}\;.\ ] ] the wavefunctions are @xmath159{\beta } } \;=\ ; \sqrt{\dfrac{2n+1}{2}}\sqrt{c}\,p_n(s)\;.\ ] ] note that the orthonormality relation for the legendre polynomials is @xmath160 so these wave - functions are properly normalized . the integrals for @xmath161 and @xmath162 diverge in this limit , so both @xmath113 and @xmath116 are divergent for all @xmath83 . however , the energy , which is the difference between @xmath163 and @xmath164 , stays finite : @xmath165 as we have seen above , for values of @xmath112 which maintain the inequality @xmath166 , the harmonic oscillator hamiltonian admits an infinite ladder of positive energy eigenstates even when @xmath7 . furthermore , these are states with finite @xmath12 and @xmath3 , implying that the particle is ` bound ' close to the phase space origin , just as in the @xmath5 case . but how can a particle be ` bound ' for an ` inverted ' harmonic oscillator ? to gain insight into this question , we solve the corresponding classical equation of motion . we assume that the classical limit of our commutation relation , eq . ( [ commutationrelation ] ) , is obtained by the usual correspondence between commutators and poisson brackets : @xmath167\;\rightarrow\;\{a,\,b\,\}\;.\ ] ] therefore , we assume @xmath168 then , the equations of motion for the harmonic oscillator with hamiltonian given by @xmath169 are @xmath170 we allow @xmath23 to take on either sign : if @xmath5 , then @xmath171 and @xmath37 will have the same sign ; if @xmath7 they will have opposite sign . note that , even though the equations of motion of @xmath32 and @xmath37 have changed , the total energy will still be conserved . consequently , the time - evolution of @xmath32 and @xmath37 in phase space will be along the trajectory given by @xmath172 . for the @xmath5 case this will be an ellipse , while for the @xmath7 case this will be a hyperbola . to solve these equations , we change the variable @xmath37 to @xmath44 , which was introduced in eq . ( [ rhodef ] ) for the quantum case . then , the equations become @xmath173 \;=\ ; \dfrac{1}{2m\beta}\dfrac{d}{d\rho}\left[\tan^2(\sqrt{\beta}\rho)\right]\ ; , \cr \dot{\rho } & = & -k x\;.\phantom{\dfrac{x}{x}}\end{aligned}\ ] ] therefore , @xmath174\;,\ ] ] which integrates to @xmath175\ ; , \label{rhodotsquared}\ ] ] where @xmath176 is the integration constant . since we must have @xmath177 , the range of allowed values of @xmath176 will depend on whether @xmath5 or @xmath7 . we will consider the two cases separately . , where @xmath178 is the particle s energy , and @xmath179 is the amplitude of the oscillation in @xmath32 . the undeformed @xmath180 case is shown in red . the other four cases are @xmath181 ( orange ) , @xmath182 ( green ) , @xmath183 ( blue ) , and @xmath184 ( purple ) . , title="fig:",width=296 ] , where @xmath178 is the particle s energy , and @xmath179 is the amplitude of the oscillation in @xmath32 . the undeformed @xmath180 case is shown in red . the other four cases are @xmath181 ( orange ) , @xmath182 ( green ) , @xmath183 ( blue ) , and @xmath184 ( purple ) . , title="fig:",width=302 ] , where @xmath178 is the particle s energy , and @xmath179 is the amplitude of the oscillation in @xmath32 . the undeformed @xmath180 case is shown in red . the other four cases are @xmath181 ( orange ) , @xmath182 ( green ) , @xmath183 ( blue ) , and @xmath184 ( purple ) . , title="fig:",width=302 ] ) , in the limit @xmath185 with @xmath186 where @xmath178 and @xmath112 are kept fixed . @xmath187 and @xmath188 take on the behavior of the position and momentum of a particle in an infinite square well . , title="fig:",width=302 ] ) , in the limit @xmath185 with @xmath186 where @xmath178 and @xmath112 are kept fixed . @xmath187 and @xmath188 take on the behavior of the position and momentum of a particle in an infinite square well . , title="fig:",width=302 ] ) , in the limit @xmath185 with @xmath186 where @xmath178 and @xmath112 are kept fixed . @xmath187 and @xmath188 take on the behavior of the position and momentum of a particle in an infinite square well . , title="fig:",width=307 ] when @xmath5 , we introduce the angular frequency @xmath190 as usual . then , eq . ( [ rhodotsquared ] ) becomes @xmath191\ ; , \label{rhodotsquared2}\ ] ] and taking the square - root , we obtain @xmath192 in this case , we must have @xmath193 for the content of the square - root to be positive . separating variables , we obtain @xmath194 the left - hand side integrates to @xmath195 \;.\end{aligned}\ ] ] therefore , @xmath196 \cr & & \end{aligned}\ ] ] where @xmath197 is the integration constant . without loss of generality , we can choose the sign inside the curly brackets to be plus . setting the clock so that @xmath198 , we obtain : @xmath199 and the energy is given by @xmath200 ^ 2 + \dfrac{1}{2\beta m}\left[\sqrt{\beta}\,p(t)\right]^2 \cr & = & \dfrac{c}{2\beta m } \;>\ ; 0\;.\end{aligned}\ ] ] the period of oscillation @xmath201 is no longer equal to @xmath202 when @xmath203 . it is now @xmath204 let @xmath205 then @xmath206 and we can identify @xmath179 as the oscillation amplitude in @xmath32 . if we take the limit @xmath125 while keeping @xmath179 fixed , we find : @xmath207 which shows that the canonical behavior is recovered in this limit . the behavior of the solution when @xmath203 is compared with the @xmath180 limit for several representative values of @xmath176 in fig . [ xplotpplotnorm1 ] . another interesting limit is obtained by setting @xmath186 and letting @xmath185 while keeping @xmath178 and @xmath112 fixed . in that limit , @xmath208 and we find @xmath209 \;,\cr \sqrt{\beta}\,p(t ) & \;\xrightarrow{m\rightarrow\infty}\ ; & + \dfrac{\sin(\omega_0 t)}{|\cos(\omega_0 t)| } \;,\cr x(t)/a & \;\xrightarrow{m\rightarrow\infty}\ ; & -\dfrac{\cos(\omega_0 t)}{|\cos(\omega_0 t)| } \;. \label{mplusinfinitylimit}\end{aligned}\ ] ] the behavior of the solution in this limit is shown in fig . [ rhoplotxplotpplot ] . the motion of a particle in an infinite square well potential ( in @xmath44-space ) is reproduced , in correspondence to the quantum @xmath138 limit . for the @xmath7 case , by an abuse of notation , let us set @xmath211 then , eq . ( [ rhodotsquared ] ) becomes @xmath212\;. \label{rhodotsquared3}\ ] ] the integration constant @xmath176 can have either sign in this case . we will consider the three cases @xmath213 , @xmath214 , and @xmath215 separately . , where @xmath178 is the particle s energy , and @xmath216 is the distance of closest approach to the origin in @xmath32-space . note that due to the negative mass , @xmath217 is negative when @xmath218 is positive , and vice versa . the undeformed @xmath180 case is shown in red . the other four cases are @xmath219 ( orange ) , @xmath220 ( green ) , @xmath221 ( blue ) , and @xmath222 ( purple ) . , title="fig:",width=302 ] , where @xmath178 is the particle s energy , and @xmath216 is the distance of closest approach to the origin in @xmath32-space . note that due to the negative mass , @xmath217 is negative when @xmath218 is positive , and vice versa . the undeformed @xmath180 case is shown in red . the other four cases are @xmath219 ( orange ) , @xmath220 ( green ) , @xmath221 ( blue ) , and @xmath222 ( purple ) . , title="fig:",width=302 ] , where @xmath178 is the particle s energy , and @xmath216 is the distance of closest approach to the origin in @xmath32-space . note that due to the negative mass , @xmath217 is negative when @xmath218 is positive , and vice versa . the undeformed @xmath180 case is shown in red . the other four cases are @xmath219 ( orange ) , @xmath220 ( green ) , @xmath221 ( blue ) , and @xmath222 ( purple ) . , title="fig:",width=302 ] for the @xmath223 case , the square - root of eq . ( [ rhodotsquared3 ] ) gives us @xmath224 therefore , @xmath225 the left - hand side integrates to @xmath226 & \bigl(|c|>1\bigr ) \\ \;\sin\bigl(\sqrt{\beta}\rho\bigr)\phantom{\bigg| } & \bigl(|c|=1\bigr ) \\ \dfrac{1}{\sqrt{1-|c|}}\,{\mathrm{arcsinh}}\left[\!\sqrt{\dfrac{1-|c|}{|c|}}\sin\bigl(\sqrt{\beta}\rho\bigr)\!\right ] & \bigl(|c|<1\bigr ) \\ \end{array } \right . \cr & & \end{aligned}\ ] ] therefore , @xmath227 & \bigl(|c|>1\bigr ) \\ \pm\;{\omega}(t - t_0 ) \phantom{\bigg| } & \bigl(|c|=1\bigr ) \\ \sqrt{\dfrac{|c|}{1-|c|}}\ ; \sinh\bigl[\pm\sqrt{1-|c|}\;{\omega}(t - t_0)\bigr ] & \bigl(|c|<1\bigr ) \\ \end{array } \right . \cr & & \end{aligned}\ ] ] where @xmath197 is the integration constant , which we will set to zero in the following . from this , we find : @xmath228 in all three cases , we have @xmath229 ^ 2 -\dfrac{1}{2\beta|m|}\left[\sqrt{\beta}\,p(t)\right]^2 \cr & = & \dfrac{|c|}{2\beta|m| } \;>\ ; 0\;. \label{eclassicalcneg}\end{aligned}\ ] ] let @xmath230 then @xmath231 and we can identify @xmath216 as the distance of closest approach to the origin ( aka impact parameter ) . taking the limit @xmath125 while keeping @xmath216 fixed , we find : @xmath232 which recovers the canonical solution . this behavior of @xmath188 and @xmath217 for the @xmath180 case is compared with that in the @xmath203 case for several representative values of @xmath176 in fig . [ xplotpplotinvertedcminus ] . it should be noted that for any finite value of @xmath213 , it only takes a finite amount of time for the particle to get from @xmath233 to @xmath234 , or equivalently , for @xmath235 to evolve from @xmath236 to @xmath237 . we will call this time @xmath238 for reasons that will become clear later . @xmath201 is given by : @xmath239 this dependence on @xmath213 is shown in fig . [ omegatplotfig ] . , where @xmath178 is the particle s energy , and @xmath240 is the momentum of the particle at the origin @xmath241 . the undeformed @xmath180 case is shown in red . the other four cases are @xmath242 ( orange ) , @xmath243 ( green ) , @xmath244 ( blue ) , and @xmath245 ( purple ) . , title="fig:",width=302 ] , where @xmath178 is the particle s energy , and @xmath240 is the momentum of the particle at the origin @xmath241 . the undeformed @xmath180 case is shown in red . the other four cases are @xmath242 ( orange ) , @xmath243 ( green ) , @xmath244 ( blue ) , and @xmath245 ( purple ) . , title="fig:",width=302 ] , where @xmath178 is the particle s energy , and @xmath240 is the momentum of the particle at the origin @xmath241 . the undeformed @xmath180 case is shown in red . the other four cases are @xmath242 ( orange ) , @xmath243 ( green ) , @xmath244 ( blue ) , and @xmath245 ( purple ) . , title="fig:",width=302 ] as a function of @xmath246 , where @xmath247 , and @xmath178 is the particle s energy . @xmath238 is the time it takes for the particle to traverse the entire trajectory . , width=302 ] for the @xmath248 case , taking the square - root of eq . ( [ rhodotsquared3 ] ) yields @xmath249 therefore , @xmath250 the left - hand side integrates to @xmath251 therefore , @xmath252\end{aligned}\ ] ] where @xmath197 is the integration constant , which we will set to zero in the following . the sign on the argument of the hyperbolic cosine is also irrelevant so we will set it to plus . from this , we find : @xmath253 and @xmath254 ^ 2 -\dfrac{1}{2\beta|m|}\left[\sqrt{\beta}\,p(t)\right]^2 \cr & = & -\dfrac{c}{2\beta|m| } \;<\ ; 0\;.\end{aligned}\ ] ] let @xmath255 then @xmath256 and we can identify @xmath240 as the magnitude of the momentum that the particle has at the origin @xmath241 . taking the limit @xmath125 while keeping @xmath240 constant , we find @xmath257 as in the @xmath213 , the time @xmath238 it takes for the particle to travel from @xmath258 to @xmath259 is finite . @xmath201 is given by : @xmath260 this dependence on @xmath214 is also shown in fig . [ omegatplotfig ] . at time @xmath261 and asymptotically approaches the origin . , title="fig:",width=302 ] at time @xmath261 and asymptotically approaches the origin . , title="fig:",width=302 ] at time @xmath261 and asymptotically approaches the origin . , title="fig:",width=312 ] for @xmath215 , eq . ( [ rhodotsquared3 ] ) leads to @xmath263 or @xmath264 which can be integrated easily to yield @xmath265 or @xmath266 with all combinations of signs allowed . set the clock so that @xmath198 . the solutions for the @xmath267 region are @xmath268 the particle starts out at @xmath233 and asymptotically approaches the origin , taking an infinite amount of time to get there . this behavior is show in fig . [ xplotpplotinvertedczero ] . the example shown is for @xmath221 . , title="fig:",width=302 ] -space . the example shown is for @xmath221 . , title="fig:",width=302 ] -space . the example shown is for @xmath221 . , title="fig:",width=302 ] as we have seen , when @xmath7 and @xmath203 , it only takes a finite amount of time for the particle to traverse the entire classical trajectory as long as the energy of the particle is non - zero . this means that we must specify what happens to the particle after it reaches infinity . for this , we could either compactify @xmath32-space so that the particle which reaches @xmath259 will return from @xmath258 , in which case the momentum of the particle will bounce back from infinite walls at @xmath269 , or we could compactify @xmath37-space so that the particle which reaches @xmath269 will return from @xmath270 , in which case the position of the particle will bounce back from infinite walls at @xmath259 . here , we choose to compactify @xmath32-space so that the @xmath93 limit of the @xmath7 , @xmath213 solution will match the @xmath185 limit of the @xmath5 , @xmath214 solution . this choice also agrees with the boundary condition we imposed in the quantum case , in which the wave - function in @xmath44 was demanded to vanish at the domain boundaries @xmath271 , which corresponds to placing infinite potential walls there . the @xmath5 , @xmath214 solution was given by eq . ( [ mplusinfinitylimit ] ) . taking the @xmath93 limit of eq . ( [ mminuseplussolutions ] ) while keeping @xmath216 fixed , we find @xmath272 and @xmath273 \ ; , \cr \sqrt{\beta}\,p(t ) & \quad\xrightarrow{|m|\rightarrow\infty}\quad & + \dfrac{\sin(\pm{\omega}_0 t)}{|\cos(\pm{\omega}_0 t)| } \ ; , \cr x(t)/b & \quad\xrightarrow{|m|\rightarrow\infty}\quad & \mp\dfrac{\cos(\pm{\omega}_0 t)}{|\cos(\pm{\omega}_0 t)| } \;. \label{mminusinfinitylimit}\end{aligned}\ ] ] which formally agrees with eq . ( [ mplusinfinitylimit ] ) , and if graphed will lead to a figure similar to fig . [ rhoplotxplotpplot ] . the one significant difference is , however , that when the particle jumps from @xmath274 to @xmath275 in the @xmath5 case it goes through @xmath241 , while when it jumps from @xmath276 to @xmath277 in the @xmath7 case , it must go through @xmath278 . by compactifying @xmath32-space , all motion when @xmath279 will become oscillatory through @xmath278 , and the @xmath201 calculated above becomes the oscillatory period . as an example , we plot the @xmath32-compactified solution for @xmath221 in fig . [ compactificationplot ] , for which the period is @xmath280 . note that the period for the @xmath7 , @xmath213 solution in the limit of @xmath93 becomes @xmath281 the arccosine providing a @xmath282 . - , @xmath37- , and @xmath32-spaces of a negative mass particle in a harmonic oscillator potential for the case @xmath283 . the time dependence of this solution was shown in fig . [ compactificationplot ] . though the trajectory of the particle is not confined to a finite region of phase space , the classical probabilities of finding the particle near the phase - space - origin is still finite due to the finiteness of the time it takes for the particle to traverse the entire trajectory . , title="fig:",width=302 ] - , @xmath37- , and @xmath32-spaces of a negative mass particle in a harmonic oscillator potential for the case @xmath283 . the time dependence of this solution was shown in fig . [ compactificationplot ] . though the trajectory of the particle is not confined to a finite region of phase space , the classical probabilities of finding the particle near the phase - space - origin is still finite due to the finiteness of the time it takes for the particle to traverse the entire trajectory . , title="fig:",width=302 ] - , @xmath37- , and @xmath32-spaces of a negative mass particle in a harmonic oscillator potential for the case @xmath283 . the time dependence of this solution was shown in fig . [ compactificationplot ] . though the trajectory of the particle is not confined to a finite region of phase space , the classical probabilities of finding the particle near the phase - space - origin is still finite due to the finiteness of the time it takes for the particle to traverse the entire trajectory . , title="fig:",width=312 ] consider the @xmath7 , @xmath284 case . @xmath235 evolves from @xmath285 to @xmath282 in time @xmath238 , that is : @xmath286 thus , we can identify @xmath287 as the classical probability density of the particle in @xmath44-space . the classical probablity in @xmath37- and @xmath32-spaces can be defined in a similar manner : @xmath288 } \;. \cr & & \end{aligned}\ ] ] these probability functions are plotted in fig . [ classicalprobabilityplot ] for the case @xmath289 . comparing the energies of the quantum and classical solutions given in eqs . ( [ enpm ] ) and ( [ eclassicalcneg ] ) , we can conclude that the correspondence is given by the relation @xmath290 we expect the quantum and classical probabilities to match for large @xmath83 . as an example , we take @xmath291 and @xmath292 , which correspond to : @xmath293{\lambda(1-\lambda ) } } \;=\ ; \dfrac{2}{\sqrt[4]{3 } } \;\approx\;1.52,\cr b & = & \dfrac{\sqrt{|c|}}{\kappa^2}\,\delta x_{\min } \;\stackrel{1\ll n}{\approx}\ ; n\,\delta x_{\min } \;=\ ; 30\,\delta x_{\min}\;. \cr & & \end{aligned}\ ] ] the comparison of the quantum and classical probabilities for this case in @xmath44- , @xmath37- , and @xmath32-spaces are shown in fig . [ probabilitycomparison ] . if we average out the bumps in the quantum case , it is clear that the distributions agree , up to the typical quantum mechanical phenomenon of seepage of the probability into energetically forbidden regions . thus , the existence of ` bound ' states with a finite @xmath12 and @xmath3 in the quantum case can be associated with the fact that the particle spends a finite amount of time near the phase space origin in the classical limit . - , @xmath37- , and @xmath32-spaces for a negative mass particle in a harmonic oscillator potential for the case @xmath291 and @xmath292 , which corresponds to the classical @xmath294 . the quantum distribution in @xmath32-space is discrete due to the existence of the minimal length . there is also some seepage of the probability into classically forbidden regions in @xmath32-space as is expected of quantum probabilities . , title="fig:",width=302 ] - , @xmath37- , and @xmath32-spaces for a negative mass particle in a harmonic oscillator potential for the case @xmath291 and @xmath292 , which corresponds to the classical @xmath294 . the quantum distribution in @xmath32-space is discrete due to the existence of the minimal length . there is also some seepage of the probability into classically forbidden regions in @xmath32-space as is expected of quantum probabilities . , title="fig:",width=302 ] + - , @xmath37- , and @xmath32-spaces for a negative mass particle in a harmonic oscillator potential for the case @xmath291 and @xmath292 , which corresponds to the classical @xmath294 . the quantum distribution in @xmath32-space is discrete due to the existence of the minimal length . there is also some seepage of the probability into classically forbidden regions in @xmath32-space as is expected of quantum probabilities . , title="fig:",width=302 ] + we have solved for the eigenstates of the harmonic oscillator hamiltonian under the assumption of the deformed commutation relation between @xmath35 and @xmath41 as given in eq . ( [ commutationrelation ] ) , with the objective of calculating their uncertainties in position and momentum . for the normal harmonic oscillator with positive mass ( @xmath96 ) , the eigenstates are found on the @xmath2 branch of the mlur , where decreasing @xmath24 leads to larger @xmath3 , and thus smaller @xmath12 . somewhat surprisingly , @xmath24 can be decreased through zero into the negative , thereby ` inverting ' the harmonic oscillator , while still maintaining an infinite ladder of positive energy eigenstates so long as the condition @xmath295^{1/4}$ ] is satisfied . there , the eigenstates are found on the @xmath29 branch of the mlur , where decreasing @xmath24 away from zero further into the negative leads to larger @xmath3 , and thus larger @xmath12 , with both diverging as @xmath50 approaches the above bound from below . the @xmath28 line separating the @xmath2 and @xmath6 regions is given by eq . ( [ maginotline ] ) . taking the classical limit by replacing our deformed commutator with a deformed poisson bracket , we solve the corresponding classical equations of motion and find that the solutions for the ` inverted ' harmonic oscillator are such that the particle only takes a finite amount of time to traverse its entire trajectory . this leads to a finite classical probability density of finding the particle near the phase space origin , and provides and explanation of why ` bound ' states with discrete energy levels are possible in the quantum case . one significant difference between the classical and quantum cases is , however , that the classical system has no restriction on the sign of the energy , whereas the quantum system only allows for positive energy eigenstates . the latter is guaranteed by the above mentioned condition on @xmath13 and @xmath50 . indeed , the condition is equivalent to @xmath296 when one assumes @xmath297 . we have found that the energy eigenstates of the harmonic oscillator only populate the @xmath2 branch of the mlur when the mass is positive , and only the @xmath29 branch when the mass is negative . a natural question to ask is whether some superposition of the energy eigenstates could cross over the @xmath28 line to the other side . for the free particle , which can be considered the @xmath298 limit of the harmonic oscillator , the answer is in the affirmative . in that case , the uncertainties of the energy eigenstates are @xmath299 and @xmath300 . superpositions of these states , with uncertainties positioned anywhere along the mlur bound , can be easily constructed , as will be discussed in detail in a subsequent paper @xcite . is a similar ` cross - over ' possible for the @xmath301 case ? another interesting question is whether it is possible to construct classically behaving ` coherent states ' in either of these mass sectors , and whether their uncertainties are contained as they evolve in time . in particular , when the mass is negative , the classical equations of motion calls for the particle to move at arbitrarily large speeds . what is the corresponding quantum phenomenon ? we are also assuming that eq . ( [ commutationrelation ] ) embodies the non - relativistic limit of some relativistic theory with a minimal length . from that perspective , the infinite speed that the negative - mass particle attains seems problematic . would such states not exist in the relativistic theory , or will they metamorphose into imaginary mass tachyons ? these , and various other related questions will be addressed in future works . we would like to thank lay nam chang , george hagedorn , and djordje minic for helpful discussions . this work is supported by the u.s . department of energy , grant de - fg05 - 92er40709 , task a. the following integral formula is necessary in calculating the expectation value of @xmath107 . for two non - negative integers @xmath23 and @xmath83 such that @xmath302 , and @xmath303 , we have @xmath304 ^ 2 m!}\ ; \qquad & \mbox{if $ n - m=\mathrm{even}$ , } \\ \phantom{xxx } & \\ 0 & \mbox{if $ n - m=\mathrm{odd}$ , } \end{array } \right . \cr & & \label{integral}\end{aligned}\ ] ] where @xmath305 . we were unable to find this result in any of the standard tables of integrals @xcite though _ mathematica _ seems to be aware of it . here we present a proof . we start from recursion relations which can be found in ref . @xcite . in section 8.933 of gradshteyn and ryzhik , we have : @xmath306 \ ; , \phantom{\bigg| } \cr & & n\,c_{n}^{\lambda}(x ) \;=\ ; 2\lambda\bigl[\ , x\,c_{n-1}^{\lambda+1}(x ) - c_{n-2}^{\lambda+1}(x ) \,\bigr ] \;.\end{aligned}\ ] ] eliminating @xmath307 and then shifting @xmath59 by one unit , we obtain @xmath308 which is equation 22.7.23 of abramowitz and stegun . iterating this relation , we deduce that @xmath309 since @xmath310 we can write @xmath311 thus , the even @xmath312 s can be expressed as a sum of the even @xmath313 s , and the odd @xmath312 s as a sum of the odd @xmath313 s . invoking the orthogonality relation , eq . ( [ gegenbauerorthogonalityrelation ] ) , which is valid when @xmath314 , it is clear that @xmath315 for @xmath303 . so for the integral of eq . ( [ integral ] ) to be non - zero , @xmath23 and @xmath83 must be both even , or both odd . for two non - negative integers @xmath22 and @xmath316 such that @xmath317 , we find @xmath318 \cr & & \qquad\quad\;\times \biggl[\ , \sum_{j=0}^{\ell}\left(\dfrac{2j+\lambda-1}{\lambda-1}\right ) c_{2j}^{\lambda-1}(s ) \,\biggr]ds \cr & = & \sum_{i=0}^{k}\sum_{j=0}^{\ell } \left(\dfrac{2i+\lambda-1}{\lambda-1}\right ) \left(\dfrac{2j+\lambda-1}{\lambda-1}\right ) \cr & & \qquad\qquad\;\times \int_{-1}^{1 } c^{2\lambda-3}\,c_{2i}^{\lambda-1}(s)\,c_{2j}^{\lambda-1}(s)\,ds \cr & = & \sum_{i=0}^{k } \left(\dfrac{2i+\lambda-1}{\lambda-1}\right)^2 \dfrac{2\pi\,\gamma(2i+2\lambda-2)}{(2i+\lambda-1)\bigl[\,2^{\lambda-1}\gamma(\lambda-1)\,\bigr]^2 ( 2i ) ! } \cr & = & \dfrac{2\pi}{\bigl[\,2^{\lambda-1}\gamma(\lambda)\,\bigr]^2 } \sum_{i=0}^{k } \dfrac{(2i+\lambda-1)\,\gamma(2i+2\lambda-2)}{(2i ) ! } \ ; , \cr & & \phantom{xxx } \cr \lefteqn{\int_{-1}^{1 } c^{2\lambda-3}\ , c_{2k+1}^\lambda(s)\,c_{2\ell+1}^\lambda(s)\,ds } \cr & = & \int_{-1}^{1 } c^{2\lambda-3 } \biggl[\ , \sum_{i=0}^{k}\left(\dfrac{2i+\lambda}{\lambda-1}\right ) c_{2i+1}^{\lambda-1}(s ) \,\biggr ] \cr & & \qquad\quad\;\times \biggl[\ , \sum_{j=0}^{\ell}\left(\dfrac{2j+\lambda}{\lambda-1}\right ) c_{2j+1}^{\lambda-1}(s ) \,\biggr]ds \cr & = & \sum_{i=0}^{k}\sum_{j=0}^{\ell } \left(\dfrac{2i+\lambda}{\lambda-1}\right ) \left(\dfrac{2j+\lambda}{\lambda-1}\right ) \cr & & \qquad\qquad\;\times \int_{-1}^{1 } c^{2\lambda-3}\,c_{2i+1}^{\lambda-1}(s)\,c_{2j+1}^{\lambda-1}(s)\,ds \cr & = & \sum_{i=0}^{k } \left(\dfrac{2i+\lambda}{\lambda-1}\right)^2 \dfrac{2\pi\,\gamma(2i+2\lambda-1)}{(2i+\lambda)\bigl[\,2^{\lambda-1}\gamma(\lambda-1)\,\bigr]^2 ( 2i+1 ) ! } \cr & = & \dfrac{2\pi}{\bigl[\,2^{\lambda-1}\gamma(\lambda)\,\bigr]^2 } \sum_{i=0}^{k } \dfrac{(2i+\lambda)\,\gamma(2i+2\lambda-1)}{(2i+1 ) ! } \;.\end{aligned}\ ] ] the sums in the above expressions are given by @xmath319 these relations can be proved by induction in @xmath22 . putting everything together , we obtain eq . ( [ integral ] ) . using this formula , we find the matrix elements of the operator @xmath107 to be : @xmath320 } & \\ & \mbox{for $ n - m=\mathrm{even}$ , $ m\le n$ , } \\ \phantom{xxx } & \\ 0 \hspace{3 cm } & \mbox{for $ n - m=\mathrm{odd}$. } \end{array } \right . \cr & & \end{aligned}\ ] ] in particular , the diagonal elements are given by @xmath321 the expectation value of @xmath106 is obtained from @xmath322 l. n. chang , d. minic , n. okamura and t. takeuchi , phys . d * 65 * , 125028 ( 2002 ) [ arxiv : hep - th/0201017 ] , s. benczik , l. n. chang , d. minic , n. okamura , s. rayyan and t. takeuchi , phys . d * 66 * , 026003 ( 2002 ) [ arxiv : hep - th/0204049 ] ; arxiv : hep - th/0209119 , s. benczik , l. n. chang , d. minic and t. takeuchi , phys . a * 72 * , 012104 ( 2005 ) [ arxiv : hep - th/0502222 ] , s. z. benczik , ph.d . thesis , virginia tech ( 2007 ) , l. n. chang , d. minic and t. takeuchi , mod . a * 25 * , 2947 ( 2010 ) [ arxiv:1004.4220 [ hep - th ] ] , l. n. chang , z. lewis , d. minic and t. takeuchi , arxiv:1106.0068 [ hep - th ] . s. hossenfelder , m. bleicher , s. hofmann , j. ruppert , s. scherer and h. stoecker , phys . b * 575 * , 85 ( 2003 ) [ arxiv : hep - th/0305262 ] , u. harbach , s. hossenfelder , m. bleicher and h. stoecker , phys . b * 584 * , 109 ( 2004 ) [ arxiv : hep - ph/0308138 ] , u. harbach and s. hossenfelder , phys . b * 632 * , 379 ( 2006 ) [ arxiv : hep - th/0502142 ] , s. hossenfelder , phys . rev . * d70 * , 105003 ( 2004 ) . [ hep - ph/0405127 ] ; class . * 23 * , 1815 ( 2006 ) [ arxiv : hep - th/0510245 ] ; phys . rev . d * 73 * , 105013 ( 2006 ) [ arxiv : hep - th/0603032 ] . s. das and e. c. vagenas , phys . * 101 * , 221301 ( 2008 ) [ arxiv:0810.5333 [ hep - th ] ] ; can . j. phys . * 87 * , 233 ( 2009 ) [ arxiv:0901.1768 [ hep - th ] ] ; phys . lett . * 104 * , 119002 ( 2010 ) [ arxiv:1003.3208 [ hep - th ] ] , a. f. ali , s. das and e. c. vagenas , phys . b * 678 * , 497 ( 2009 ) [ arxiv:0906.5396 [ hep - th ] ] , s. basilakos , s. das and e. c. vagenas , jcap * 1009 * , 027 ( 2010 ) [ arxiv:1009.0365 [ hep - th ] ] , s. das , e. c. vagenas and a. f. ali , phys . b * 690 * , 407 ( 2010 ) [ erratum - ibid . * 692 * , 342 ( 2010 ) ] [ arxiv:1005.3368 [ hep - th ] ] . b. bagchi , a. fring , phys . . * a373 * , 4307 - 4310 ( 2009 ) . [ arxiv:0907.5354 [ hep - th ] ] , a. fring , l. gouba , f. g. scholtz , j. phys . a * a43 * , 345401 ( 2010 ) . [ arxiv:1003.3025 [ hep - th ] ] , a. fring , l. gouba , b. bagchi , j. phys . a * a43 * , 425202 ( 2010 ) . [ arxiv:1006.2065 [ hep - th ] ] . p. pedram , europhys . lett . * 89 * , 50008 ( 2010 ) [ arxiv:1003.2769 [ hep - th ] ] ; int . j. mod . d * 19 * , 2003 ( 2010 ) [ arxiv:1103.3805 [ hep - th ] ] , k. nozari and p. pedram , europhys . lett . * 92 * , 50013 ( 2010 ) [ arxiv:1011.5673 [ hep - th ] ] , p. pedram , k. nozari and s. h. taheri , jhep * 1103 * , 093 ( 2011 ) [ arxiv:1103.1015 [ hep - th ] ] . d. j. gross and p. f. mende , phys . b * 197 * , 129 ( 1987 ) ; nucl . b * 303 * , 407 ( 1988 ) , + d. amati , m. ciafaloni and g. veneziano , phys . b * 197 * , 81 ( 1987 ) ; int . j. mod . phys . a * 3 * , 1615 ( 1988 ) . i. s. gradshteyn and i. m. ryzhik , `` table of integrals , series and products , '' 7th edition ( acedemic press , 2007 ) ; m. abramowitz and i. a. stegun , `` handbook of mathematical functions , with formulas , graphs , and mathematical tables '' ( dover , 1965 ) .

we consider the supergravity solution describing a configuration of intersecting d-4-branes with non - vanishing worldvolume gauge fields . the entropy of such a black hole is calculated in terms of the d - branes quantised charges . the non - extreme solution is also considered and the corresponding thermodynamical quantities are calculated in terms of a d - brane / anti - d - brane system . to perform the quantum mechanical d - brane analysis we study open strings with their ends on branes with a magnetic condensate . applying the results to our d - brane system we managed to have a perfect agreement between the d - brane entropy counting and the corresponding semi - classical result . the landau degeneracy of the open string states describing the excitations of the d - brane system enters in a crucial way . we also derive the near - extreme results which agree with the semi - classical calculations . = msbm10 at 12pt # 1 = msbm10 at 10pt # 1 damtp - r/97/59 + * landau degeneracy and black hole entropy * miguel s. costa and malcolm j. perry + _ d.a.m.t.p . + university of cambridge + cambridge cb3 9ew + uk _

hep-th9712026

it is now widely accepted that superstring theory is the most promising unifying theory . among its great achievements is the resolution of a longstanding problem in black hole physics : the statistical origin of the bekenstein - hawking entropy at least for certain extreme and near - extreme black holes @xcite . the resolution of this problem has been made possible by the realization of polchinski @xcite that d - branes , i.e. branes carrying ramond - ramond charge , are extended objects with the property that open strings can end on them . the d - brane excitations are described by such open strings . the microscopic origin of the black hole entropy arises in d - brane physics in two steps : firstly , we identify a certain classical supersymmetric black hole carrying charge in the @xmath0 sector with a corresponding excited d - brane configuration ; secondly , we count the number of open string states that reproduce the excited d - brane system . the worldvolume bosonic field content that describes the low energy excitations of a d-@xmath1-brane is given by @xmath2 scalar fields @xmath3 ( @xmath4 ) and by a 1-form gauge potential @xmath5 ( @xmath6 ) @xcite . these fields are interchanged by @xmath7-duality transformations . the scalar fields are associated to the massless excitations of the brane that describe transverse displacements , and the gauge field is associated to worldvolume massless excitations on the brane . through this paper we will be interested in non - threshold d - brane bound states . these configurations arise whenever the worldvolume gauge field associated with a d-@xmath1-brane forms a condensate . if this is the case the d-@xmath1-brane may carry the @xmath8 charge that is usually associated with lower dimensional d - branes . in fact , the d-@xmath1-brane effective action contains the correct coupling terms to @xmath0 gauge potentials with rank lower than @xmath9 @xcite . remarkably , the existence of these generalised wess - zumino terms ( and for the d-9-brane the anomalous bianchi identity and the g - s coupling ) is a consequence of @xmath7-duality and lorentz invariance @xcite . to be definite consider the @xmath10 d - brane bound state . this configuration may be obtained by performing a @xmath7-duality at an angle on a d-@xmath11-brane , i.e. p-1 : & x^1, ... ,x^p-1 + & ^p = x^p-1 & & p : & x^1, ... ,x^p + & a_p = x^p-1 [ 1.1 ] where we start with a d-@xmath11-brane parallel to the @xmath12-directions and making an angle @xmath13 with the @xmath14-direction in the @xmath15 2-torus . the @xmath7-duality transformation along the @xmath16-direction interchanges the scalar field @xmath17 with the gauge field @xmath18 forming a magnetic condensate ( through this paper we shall scale the gauge potential as @xmath19 ) . from the supergravity theory perspective the existence of these bound states has been deduce in @xcite . in the linear approximation the field theory describing the low - lying states on the d - brane bound state in ( [ 1.1 ] ) is the super yang mills theory with @xmath20 gauge group @xcite . the integer @xmath21 is the d-@xmath1-brane winding number in the @xmath14-direction . the brane carries a @xmath22 flux which induces a thooft twist in the @xmath23 part of the theory @xcite . as a consequence the gauge potential and the scalar fields will obey twisted boundary conditions . for an appropriate choice of this @xmath20 bundle multiple transition functions the only non - vanishing gauge potential is in the @xmath22 center of the group and it is given by ( [ 1.1 ] ) . had we considered @xmath24 d - branes on top of each other , the scalar and gauge fields would become @xmath25 matrices and equation ( [ 1.1 ] ) would still refer to the @xmath22 center of the group . the aim of this paper is to study the d - brane entropy counting when the d - brane system has non - vanishing worldvolume gauge fields . in particular , we will be considering a d - brane configuration presented in @xcite interpreted as the intersection of @xmath26 and @xmath27 d-@xmath28-branes with non - vanishing worldvolume gauge fields and carrying momentum along a common direction . although this configuration is @xmath7-dual to the configuration used by callan and maldacena @xcite , the origin of the statistical entropy has a remarkable new feature . namely , we have to take into account the landau degeneracy of the open string sector corresponding to open strings with each end on different branes . we begin in section two by presenting our d - brane configuration and by considering the corresponding charges quantisation . we shall then present the supergravity solution describing our d - brane system and write the integer valued form of the bekenstein - hawking entropy . the supergravity solution corresponding to the non - extreme configuration will also be presented and the corresponding thermodynamical formulae in terms of a brane / anti - brane system will be written . in section three we shall consider open strings with ends on different d - branes which have a magnetic condensate on its worldvolume theory . we will keep our discussion general and specify for our d - brane system in section four , where we shall give the d - brane entropy counting . we will also consider the near - extreme black hole solution , matching the corresponding thermodynamical quantities with the brane / anti - brane system . we shall give our conclusions in section five . let us start by considering the @xmath29 d - brane configuration . we place @xmath26 d-@xmath30-branes along the @xmath31-directions and @xmath27 d-@xmath30-branes along the @xmath32-directions . these directions are taken to be circles of radius @xmath33 . next , we rotate this configuration in the @xmath34 @xmath35-torus by an angle @xmath13 as described in figure [ torus ] . the resulting d - brane system may be represented by 3_1:&x^1,x^2,x^5 , + & ^4_(1)=- x^5 , + 3_2:&x^1,x^3,x^5 , + & ^4_(2)= x^5 , [ 2.1 ] where as explained in the introduction we are considering the @xmath22 center of the worldvolume scalar fields associated with each group of d-@xmath30-branes . wrapping a direction of the d-@xmath30-branes around a cycle of the @xmath34 @xmath35-torus yields quantisation conditions on the rotation angle @xmath13 . start by considering the first d-@xmath30-branes . if the corresponding winding number in the @xmath36-direction is @xmath37 , and in the @xmath38-direction is @xmath39 , the angle @xmath13 will obey the quantisation condition = , [ 2.2 ] where @xmath40 and @xmath39 are co - prime . similarly , if the other d-@xmath30-branes are wrapped on the @xmath41-cycle the quantisation condition on the angle @xmath13 is = . [ 2.3 ] thus , the moduli @xmath42 and @xmath43 are not independent but obey the relation ( ) ^2= . [ 2.4 ] the mass formula for @xmath24 d-@xmath1-branes wrapped on a @xmath1-torus is given by @xcite m_p = n , [ 2.5 ] where @xmath44 is the string coupling constant , @xmath45 is the inverse string tension and @xmath33 is the radius of the circle along which the d - branes are wrapped with winding number @xmath46 . it is straightforward to generalise this formula for the case where a given direction of the d - branes wraps around a @xmath35-torus . for the configuration described by ( [ 2.1 ] ) we have m_3_1=n_1 , + + m_3_2=n_2 . [ 2.6 ] we proceed by performing a @xmath7-duality transformation along the @xmath36-direction on the configuration ( [ 2.1 ] ) . the resulting d - brane system is described by @xcite 4_1:&x^1,x^2,x^4,x^5 , + & f_45^(1)= , + 4_2:&x^1,x^3,x^4,x^5 , + & f_45^(2)=- . [ 2.7 ] consider the first group of d-@xmath28-branes . they arise by performing a @xmath7-duality transformation along the @xmath36-direction on @xmath26 d-@xmath30-branes that were wrapped on the @xmath38-direction with winding number @xmath39 . therefore we end up with @xmath26 d-@xmath28-branes with winding @xmath39 in the @xmath38-direction . also , we have seen that the d-@xmath30-branes were wrapped on the dualised direction with winding @xmath37 . therefore the d-@xmath28-branes will also carry d-@xmath35-brane charge . they carry @xmath47 units of such charge . in the language of @xcite , the d-@xmath35-branes dissolve in the d-@xmath28-branes leaving flux . this fact is described by the magnetic field condensate @xmath48 . similar comments apply to the second group of d-@xmath28-branes . the resulting intersecting d-@xmath28-brane system does not obey the standard intersecting rules , but supersymmetry of this configuration is guaranteed by the presence of non - vanishing worldvolume gauge fields @xcite . in the dual theory the quantisation conditions on the rotation angle @xmath13 translate into quantisation conditions on the flux carried by the worldvolume gauge field strengths @xmath48 and @xmath49 . in terms of the dual radius @xmath50 and dual string coupling @xmath51 these read as & & f_45^(1)== , + [ 2.8 ] + & & f_45^(2)=-= - . similarly , the mass of the d-@xmath28-branes is now given by m_4_1=n_1 = p_1n_1 , + [ 2.9 ] + m_4_2=n_2 = p_2n_2 . the factor @xmath52 multiplying the usual d-@xmath28-brane mass formula is due to the presence of the magnetic field condensate . as it will be explained in the subsection 3.4 , this term also arises in the vacuum amplitude calculation for a system of two parallel d - branes with such fields . finally , we remark that all constituent d - branes in the configuration ( [ 2.7 ] ) , including the d-@xmath35-branes that are dissolved in the d-@xmath28-branes , have a common @xmath53-direction . as a consequence , the d - branes may carry momentum along this direction while leaving some unbroken supersymmetry . if we introduce @xmath24 units of ( left - moving ) momentum , the compactified system will have an extra mass term given by p = m_p= . [ 2.10 ] the d - brane system presented above has a corresponding supergravity solution . this solution may be found by applying @xmath7-duality at an angle on the @xmath29 d - brane solution or by using the @xmath54 symmetry of @xmath55 , @xmath56 supergravity @xcite . the corresponding type iia background fields are ds^2&= & ( h_1h_2)^ , + + e^2(-_)&= & _ 12 ^ -1 ( h_1h_2)^ , + [ 2.11 ] + f&=&- + + & & - , + + h&=&- d()dx_4dx_5 , where h_i=1 + , i=1,2,3 , [ 2.12 ] are one - centered harmonic function on the @xmath57 euclidean space and @xmath58 is the dual operator with respect to the euclidean metric on @xmath57 . the harmonic function @xmath59 is defined by @xmath60 . reducing the solution ( [ 2.11 ] ) to five dimensions we obtain a black hole with a regular horizon . the corresponding einstein frame metric is ds_e^2=-(h_1h_2h_3)^-dt^2 + ( h_1h_2h_3)^ds^2(^4 ) . [ 2.13 ] calculating the horizon area the bekenstein - hawking entropy is seen to be s_bh== , [ 2.14 ] where @xmath61 is the five - dimensional newton s constant and @xmath62 is the volume of the unit @xmath30-sphere . we want to express the macroscopic entropy formula ( [ 2.14 ] ) in terms of the quantities describing the d - brane system ( [ 2.7 ] ) . this may be done by matching the mass formulae ( [ 2.9 ] ) and ( [ 2.10 ] ) to the corresponding string frame five - dimensional masses @xcite . the result is m_i=2_i . [ 2.15 ] the five- and ten - dimensional newton s constants are related by g_n^10=(2)^5r_1 ... r_5g_n^5=8 ^ 6g^2^4 . after some straightforward algebra the previous relations will give the following entropy formula s_bh=2 , [ 2.16 ] where n_l = p_2q_1+p_1q_2 , [ 2.17 ] is , as we shall see , the landau degeneracy for open strings with ends on different branes . in the t - dual picture described by ( [ 2.1 ] ) the same factor arises in the entropy formula . this factor is interpreted as the d-3-branes number of intersecting points on the @xmath34 2-torus @xcite . the string theory entropy counting is then clear because on the intersecting points strings with both ends on different d-3-branes become massless . the d-@xmath28-brane charges @xmath63 may be expressed as an integral over spacial infinity of the @xmath28-form field strength . the properly normalised expression for these charges is q_i=__if = \ { _ 1 , i=1 , + + _ 2 , i=2 , . [ 2.18 ] where @xmath64 with @xmath65 the asymptotic @xmath30-sphere and @xmath66 the compactified dimension transverse to the @xmath67-th @xmath28-branes . we remark that it is also possible to write the d-@xmath35-brane charges carried by the d-@xmath28-branes as an integral of the dual of the @xmath28-form field strength . the ` charge ' associated with the ( left - moving ) momentum carried by the d-@xmath28-branes may be calculated by using equations ( [ 2.10 ] ) and ( [ 2.15 ] ) n= _ 3 . [ 2.19 ] the supergravity solution ( [ 2.11 ] ) can be easily generalised to the corresponding non - extreme solution @xcite . start with the solution ( [ 2.11 ] ) without the plane wave and perform the following redefinitions dt^2 & f dt^2 , + dr^2 & f^-1dr^2 , + q_i & q_i , i=1,2 , [ 2.20 ] where @xmath68 and @xmath69 is either the electric or magnetic charge associated with the @xmath67-th @xmath28-branes . for later convenience we will express our solution in terms of the parameters @xmath70 and @xmath71 defined by = r_0 ^ 2 , + _ i = r_0 ^ 2 ^ 2_i , i=1,2 . [ 2.21 ] in the limit @xmath72 and @xmath73 , holding @xmath74 fixed , we obtain the extreme solution . the charge associated with the momentum along the @xmath53-direction may be obtained by performing the boost transformation t & t- x^1 , + x^1&- t+ x^1 . [ 2.22 ] the resulting non - extreme solution is described by ds^2&= & ( h_1h_2)^ , + [ 2.23 ] + f&=&- + + & & - , and the dilaton field and @xmath30-form field strength are still given as in ( [ 2.11 ] ) . the charges ( [ 2.18 ] ) and ( [ 2.19 ] ) previously defined are now given by q_1&= & r_0 ^ 2 , + + q_2&= & r_0 ^ 2 , [ 2.24 ] + + n&= & r_0 ^ 2 . the adm mass , entropy and hawking temperature may be calculated by reducing our solution to five dimensions . the result is m&=&r_0 ^ 2 _ i=1 ^ 3 , + + s_bh&=&2r_0 ^ 3 _ i=1 ^ 3 , [ 2.25 ] + + t_h&= & ( 2r_0_i=1 ^ 3)^-1 = s_bh^-1 . as in the extreme case , we are interested in expressing these quantities in terms of an associated d - brane system . in other words , we want to introduce right - moving momentum and anti - d-4-branes . in the near - extremal limit the resulting system will describe the near - extremal black - hole @xcite . with that in mind , we define the quantities ( @xmath26 , @xmath75 , @xmath27 , @xmath76 , @xmath77 , @xmath78 ) by q_4_1= n_1p_1= e^2_1 , & q_|4_1= n_|1p_1= e^-2_1 , + + q_4_2= n_2p_2= e^2_2 , & q_|4_2= n_|2p_2= e^-2_2 , + + n_l= e^2_3 , & n_r= e^-2_3 . [ 2.26 ] we shall see in section 4 that in the near - extremal limit , these quantities are interpreted as the units of left ( @xmath78 ) and right ( @xmath77 ) moving momentum along the @xmath53-direction , and as the winding numbers along the @xmath53-direction of the d-@xmath28-branes ( @xmath26 , @xmath27 ) and anti - d-@xmath28-branes ( @xmath75 , @xmath76 ) both with the worldvolume gauge field turned on as in ( [ 2.7 ] ) . in terms of these quantities the adm mass , entropy and hawking temperature in ( [ 2.25 ] ) are given by m&= & p_1 ( n_1+n_|1 ) + + & & + p_2 ( n_2+n_|2 ) + ( n_r+n_l ) , + [ 2.28 ] + s_bh&= & 2(+ ) ( + ) ( + ) , + + t_h&= & s_bh^-1 ( ) , where @xmath79 is defined in ( [ 2.17 ] ) . in this section we will study open superstrings ending on d - branes which have a worldvolume gauge field strength of the form f_= ( 0&f + -f&0 ) [ 3.1 ] where @xmath80 is a constant and the indices @xmath81 and @xmath82 label two spatial worldvolume coordinates . we will just consider the superstring mode expansion for these two coordinates . this , together with well known facts , will be sufficient to explain the three open string sectors of the d - brane system ( [ 2.7 ] ) , i.e. the two sectors corresponding to open strings with both ends on the same d - branes and the sector corresponding to open strings with each end on different d - branes . in the first cases , the field strength ( [ 3.1 ] ) is equal on both ends of the string while in the second case it is not . of course there are other possibilities , for example one end of the string may be let to move freely @xcite . an important remark is that we are just considering the @xmath22 center of the d - branes worldvolume fields . in our d - brane system we have @xmath26 and @xmath27 d-4-branes on top of each other wrapping on the @xmath38-direction with winding numbers @xmath39 and @xmath83 , respectively . this means that the corresponding worldvolume fields are in the adjoint representation of @xmath84 . as explained in the introduction , the @xmath22 flux induces a twist on the fields . the gauge potential as well as the scalar fields will then obey twisted boundary conditions @xcite . we can choose a set of multiple transition functions on this @xmath84 bundle such that the only non - vanishing d - branes worldvolume fields are in the @xmath22 center of the group . this means that we can perform the calculations as if our open strings have each of its ends on a singly wrapped d - brane . let us start by writing the boundary term of the open superstring action in the presence of a @xmath22 background gauge field . using the orthonormal gauge coordinates @xmath85 @xmath86 and @xmath87 @xmath88 we have ( taking @xmath89 ) s_b=._-^+ d|_=0^= , [ 3.2 ] where @xmath90 and @xmath91 and @xmath92 are respectively the positive and negative chirality spinor components of the worldsheet majorana spinor @xmath93 @xcite . performing @xmath7-duality transformations along the directions @xmath4 we find the boundary term for the action describing open superstrings with the ends on a d-@xmath1-brane with worldvolume fields @xmath94 ( @xmath95 ) and @xmath96 s_b&=&_-^+ d|_=0^= , where we used the @xmath7-duality transformation rules @xmath97 , @xmath98 , @xmath99 . if the open string ends on different d - branes the background fields will differ for @xmath100 and @xmath101 . as explained above , we will find the superstring mode expansion for the case where the only non - vanishing worldvolume fields are f_^(1)f_(=0)= ( 0&1 + -1&0 ) , + + f_^(2)f_(=)= ( 0&1 + -1&0 ) , [ 3.4 ] where @xmath102 and @xmath103 are arbitrary constants and the indices @xmath81 and @xmath82 label only two spatial worldvolume coordinates . for the d - brane system considered in this paper , and for the open string sector corresponding to open strings ending on different branes , the indices @xmath81 and @xmath82 label the @xmath36- and @xmath38-directions and = , = - . [ 3.4a ] the variation of the @xmath104 bosonic fields in the superstring action with a boundary term like in ( [ 3.3 ] ) yields the usual wave equation together with the boundary conditions _ x^= f^ _ _ x^ , = 0 , . [ 3.5 ] similarly , to cancel the boundary term that arises on varying the fermionic fields in the action ( [ 3.3 ] ) , the right ( @xmath105 ) and left ( @xmath106 ) moving world - sheet fermions should obey the boundary conditions @xcite _ + ^-_-^=f^ _ ( _ + ^+_-^ ) , = 0 , + + _ + ^_-^=f^ _ ( _ + ^_-^ ) , = . [ 3.6 ] the upper and lower sign choices correspond to the ramond and neveu - schwarz sectors of the theory , respectively . in fact , the boundary conditions ( [ 3.6 ] ) are found by requiring the fermionic fields variations to obey @xmath107 at @xmath101 , and @xmath108 at @xmath100 . in this subsection we will study the mode expansion for the worldsheet bosonic fields satisfying the boundary conditions ( [ 3.5 ] ) . all the results presented here have been obtained in @xcite in the context of open strings in a constant magnetic field , and are included for the sake of clarity . has explained above we will just consider the excitations along the @xmath109-directions . relabelling these worldsheet fields by @xmath110 and @xmath111 we define the complex field z ( x+iy ) . [ 3.7 ] the boundary conditions ( [ 3.5 ] ) now read _ z =- i _ z , & = 0 , + _ z =- i _ z , & = . [ 3.8 ] the mode expansion is then seen to be z = z+i , [ 3.9 ] where _ n- = e^-i(n- ) , [ 3.10 ] with @xmath112 , @xmath113 an integer and @xmath114 real . note that by an appropriate redefinition of the modes in ( [ 3.9 ] ) the parameter @xmath115 may be taken to lie between @xmath116 and @xmath117 . in fact , by interchanging the @xmath118 and @xmath119 mode operators it may be taken to lie between 0 and 1/2 . since the integrals of the mode functions @xmath120 are non - zero , the constant @xmath121 can not be interpreted as the string s center - of - mass . the functions @xmath120 , together with a constant function , form a complete basis of functions on the interval @xmath122 $ ] ( the details may be found in @xcite ) . the canonical momentum is as usual given by p_z== , [ 3.11 ] where the complex gauge fields @xmath123 are defined by a^(i)= ( a_x^(i)+ia_y^(i ) ) , i=1,2 . [ 3.12 ] at this point we have to choose a gauge for the d - branes worldvolume gauge fields . a convenient choice is a^(1)_dx^= ( xdy - ydx ) , + [ 3.13 ] + a^(2)_dx^= ( xdy - ydx ) .the canonical momentum is then seen to be p_z= ( _ z^+ z^ ) . [ 3.14 ] the first quantised string is obtained by introducing the equal time canonical commutation relations . this gives the following non - vanishing commutators for the mode operators in the expansion ( [ 3.9 ] ) = _ nm , = _ nm , = . [ 3.15 ] note that these commutation relations do not depend on the gauge choice ( [ 3.13 ] ) . thus , the spectrum of string states is gauge invariant . introducing the light cone coordinates @xmath124 , the virasoro operators are defined to be l_n=_0^d ( e^int_+++ e^-int_ ) , [ 3.16 ] where @xmath125 are the @xmath126 components of the energy - momentum tensor . the corresponding contributions from the @xmath110 and @xmath111 worldsheet fields are t_=_x^_x_= ( _ z_z ) ( _ z_z)^ . [ 3.17 ] after some straightforward algebra we obtain for @xmath127 l_n^(b)&=&_p=1^a_p-^a_n+p- + + & & + _ p=0^ a_p+^a_n+p+ [ 3.18 ] + + & & + _ p=0^n-1 a_p+a_n - p- . for @xmath128 , @xmath129 is obtained by substituting @xmath113 by @xmath130 and by taking the hermitian conjugate of ( [ 3.18 ] ) . the virasoro operators satisfy the algebra = ( n - m ) l_n+m^(b)+a(n)_n+m,0 , [ 3.19 ] where the @xmath131-number @xmath132 is due to the usual normal ordering ambiguity in the definition of @xmath133 . the value of @xmath132 may be calculated for the bosonic string as it was done in @xcite . we will calculate this anomalous term in the full superstring theory . as a final remark , we note that the frequencies of the @xmath118 and @xmath119 ( @xmath134 ) oscillators will be shifted with respect to their usual values by @xmath135 and @xmath136 , respectively . also , the operators @xmath137 and @xmath138 annihilate and create quanta of frequency @xmath115 . if @xmath139 the zero mode operators disappear ( we will comment on this case in subsection 3.4 ) . if @xmath140 the frequencies of the mode operators @xmath118 and @xmath119 , and therefore the spectrum , become the same as those for open strings with nd boundary conditions . the only difference is that the zero mode @xmath121 is not absent . at the beginning of this section we wrote the boundary conditions satisfied by the worldsheet fermionic fields for the case when the d - brane worldvolume gauge field does not vanish . as in the bosonic case we will just consider the two relevant directions corresponding to the ansatz ( [ 3.4 ] ) . start by defining the complex spinor field ( ^x+i^y ) = ( _ -^x+i_-^y + _ + ^x+i_+^y ) ( _ - + _ + ) . [ 3.25 ] in terms of this field the boundary conditions ( [ 3.6 ] ) read _ + -_-=-i ( _ + + _ - ) , & = 0 , + _ -=-i ( _ + _ - ) , & = , [ 3.26 ] where the upper and lower sign choices correspond to the ramond and neveu - schwarz sectors , respectively . let us consider first the ramond sector . the mode expansion for the field @xmath141 may be written as _ = , [ 3.27 ] where _ n-^()=e^(n- ) ( ) , [ 3.28 ] with @xmath115 defined as in ( [ 3.10 ] ) and @xmath113 an integer . the canonical momentum of the complex spinor field @xmath141 is p_= = ^ . [ 3.29 ] introducing the equal time anti - commutation relations we find the following non - vanishing anti - commutators for the mode operators in the expansion ( [ 3.27 ] ) \{_n-,_m-^}=_nm , \{_n+,_m+^}=_nm . [ 3.30 ] the contribution of the worldsheet fermionic fields @xmath142 and @xmath143 to the @xmath126 components of the energy - momentum tensor is t_=_^__= ( _ _ _ ^+ _ ^ _ _ ) . [ 3.31 ] thus , using the definition ( [ 3.16 ] ) and after some algebra we find that for @xmath144 l_n^(r)&= & l_n^(b)+_p=1^(+p- ) _ p-^_n+p- + + & & + _ p=0^(+p+ ) _ p+^_n+p+ [ 3.32 ] + + & & + _ p=0^n-1(-p- ) _ n - p-_p+ , where @xmath145 is the bosonic contribution to the virasoro operators given in ( [ 3.18 ] ) . for @xmath128 we have to replace @xmath113 by @xmath130 and take the hermitian conjugate . the virasoro algebra may be seen to be = ( n - m ) l_n+m^(r)+n^3_n+m,0 . [ 3.33 ] this is exactly the standard result expected for the ramond sector . thus the zero energy is not changed in the presence of the magnetic field condensate ( [ 3.4 ] ) . this is just the consequence of the fact that both the worldsheet bosons and fermions have the same moding . as in the bosonic case the frequencies of the @xmath146 and @xmath147 ( @xmath134 ) oscillators are shifted with respect to their usual values by @xmath135 and @xmath115 , respectively . also , the operators @xmath148 and @xmath149 annihilate and create quanta of frequency @xmath115 . if @xmath139 these operators will take ground states into ground states forming the usual ramond vacua . if @xmath150 the mode operators , and therefore the spectrum , become similar to the case of open superstrings with nd boundary conditions , i.e. the expansion becomes of neveu - schwarz type . we now turn to the neveu - schwarz sector of the theory . in this case the mode expansion takes the form _ = , [ 3.34 ] where _ r-^()=e^(r- ) ( ) , [ 3.35 ] with @xmath115 defined as before and @xmath151 a half - integer . after quantisation the non - vanishing anti - commutation relations for the mode operators in the expansion ( [ 3.34 ] ) are \{_r-,_s-^}=_rs , \{_r+,_s+^}=_rs . [ 3.36 ] a similar calculation to the one in the previous case gives the virasoro operators ( @xmath144 ) l_n^(ns)&= & l_n^(b)+_r=^(+r- ) _ r-^_n+r- + + & & + _ r=^(+r+ ) _ r+^_n+r+ [ 3.37 ] + + & & + _ r=^n-(-r- ) _ n - r-_r+ . for @xmath128 we have to replace @xmath113 by @xmath130 and take the hermitian conjugate . the virasoro algebra is now given by = ( n - m ) l_n+m^(ns)+_n+m,0 . [ 3.38 ] the first term inside the square brackets is the usual term that would be obtained if the magnetic field was absent . the other term is proportional to @xmath113 and it may be absorbed by a redefinition @xmath152 , which shifts the zero energy by a term @xmath153 ( in units of @xmath154 ) . as in the previous cases the frequencies of the @xmath155 and @xmath156 oscillators are shifted with respect to their usual values by @xmath135 and @xmath115 , respectively . if @xmath157 we just obtain the standard spectrum for the worldsheet fermions in the ns sector . if @xmath158 the frequency of the mode operators @xmath159 and @xmath160 vanishes . in this case the spectrum is similar to open superstrings with nd boundary conditions , i.e. it becomes of ramond type . in the subsection 3.1 we have seen that the zero modes @xmath121 and @xmath161 are non - commuting variables . in terms of the @xmath162 and @xmath163 coordinates the corresponding commutation relation reads = i . [ 3.20 ] thus , @xmath164 may be seen as the conjugate momentum of @xmath162 . since both zero modes @xmath162 and @xmath163 commute with @xmath133 we can take any string state to be an eigenstate of either @xmath162 or @xmath163 . in the d - brane system we are interested on the @xmath162- and @xmath163-directions are compactified on a torus . as a result , the eigenstates of @xmath162 or @xmath163 will have a degeneracy , the landau degeneracy @xcite . assume that the d - branes where the open strings end are wrapped around the @xmath163-direction with winding numbers @xmath39 and @xmath83 for each d - brane . we consider first the case where @xmath39 and @xmath83 are co - prime . the open strings ending on both branes carry chan - paton factors in the fundamental representation of @xmath165 . the worldsheet fields are identified according to ( y(,))_a + 2r_y ( y(,))_a+1 , , [ 3.21 ] where @xmath166 and @xmath167 . thus , going around @xmath168 cycles in the @xmath163-direction we have the identification ( y(,))_a + 2p_1p_2r_y ( y(,))_a . [ 3.21a ] to deduce the existence of the landau levels we essentially follow the same steps as in ref . @xcite . since the d - branes are singly wrapped in the @xmath162-direction the zero mode wave function @xmath169 must be single - valued . this implies that the @xmath170 eigenvalues are quantised as @xmath171 . in terms of the zero mode @xmath172 this condition reads ( y_m)_a== l , [ 3.22 ] where @xmath173 and @xmath174 . in the last equality we have used equation ( [ 3.4a ] ) which applies to the d - brane system considered in this paper . more generally , the fact that @xmath79 has to be an integer follows from the flux quantisation condition on the d - branes gauge fields this may also be derived has a dirac quantisation condition since a constant magnetic field on a torus is a monopole field @xcite . in order to interpret equation ( [ 3.22 ] ) we note that @xmath39 and @xmath83 were taken to be co - prime . this means that for a given @xmath175 all the other zero modes @xmath176 are determined by ( [ 3.21 ] ) . since the system has periodicity @xmath177 it follows from ( [ 3.22 ] ) that there are @xmath79 independent @xmath178 states . thus , the degeneracy of any string state is @xmath79 , i.e. any string state will be found by acting with creation operators on the degenerated ground state |y_m , m=1, ... ,n_l . [ 3.23 ] to relax the assumption that @xmath39 and @xmath83 are co - prime , we just have to realise that in the general case the system has periodicity @xmath179 , where @xmath180 and @xmath181 with @xmath182 and @xmath183 co - prime and @xmath184 an integer . equation ( [ 3.22 ] ) may be written as ( y_m)_a = l , [ 3.22a ] where @xmath185 . the point now is that for a given @xmath175 not all the other zero modes @xmath176 are determined by ( [ 3.21 ] ) . a minimal set of zero modes that determines all the other zero modes is ( y_m)_a , ( y_m)_a+1 , , ... , ( y_m)_a+l-1 , . [ 3.24 ] thus , there are @xmath186 landau levels but each level is itself @xmath184 times degenerated . this means that again any string state will be found by acting with the creation operators on the @xmath79 times degenerated ground state for completeness we will comment on the @xmath10 d - brane bound state @xcite . this configuration is interpreted as a d-@xmath1-brane carrying also the charge of a d-@xmath187-brane . as it is well known the excitations of this d - brane bound state are described by open superstrings with the following boundary condition for the worldsheet bosons ( at @xmath188 and @xmath189 ) _ x^=0 , & = 0, ... ,p-2 , + _ x^=f^ _ _ x^ , & , = p-1,p , + _ x^m=0 , & m = p+1, ... ,9 , [ 3.39 ] where @xmath190 is given by ( [ 3.1 ] ) . the boundary conditions for the fermionic fields in the @xmath109-directions are as in ( [ 3.6 ] ) . the spectrum for directions other than the @xmath109-directions is just standard . for the @xmath109-directions , and using the above results , we have @xmath139 and therefore the fermionic worldsheet excitations are also standard . for the bosonic fields the only relevant difference is the corresponding zero modes which become commuting variables and no longer satisfy the commutation relation ( [ 3.20 ] ) . the commutation relation ( [ 3.20 ] ) becomes ill defined . however , in this limit the spectrum of the landau levels becomes continuum as may be seen from ( [ 3.22 ] ) . this is the expected result because the zero modes are now continuum variables . we thank gary gibbons for bringing this point to our attention . ] also , the operators @xmath137 and @xmath138 disappear and are substituted by a linear term in @xmath191 @xcite . the resulting spectrum is exactly the same as in the case for vanishing worldvolume gauge field ( up to a normalisation factor in the momentum modes that is correctly reproduced for the low - lying excitations if one considers the born - infeld action @xcite ) . a crucial difference between the open superstrings obeying the boundary conditions ( [ 3.39 ] ) and the boundary conditions corresponding to a d-@xmath1-brane with vanishing worldvolume gauge field is the vacuum amplitude for two parallel of such branes . in the bound state case there is a overall factor @xmath192 multiplying the corresponding result for two d-@xmath1-branes . this factor arises in the momentum integration as it was explained in @xcite . in the effective field theory this factor also appears as it is expected from the agreement of both calculations @xcite . consider first the attractive term in the amplitude that is due to the graviton , dilaton and antisymmetric tensor field exchanges . the coupling of a d - brane to this fields will have an extra @xmath193 factor due to the born - infeld character of the d - brane action @xcite , giving the correct result in the corresponding one - loop calculation . the repulsive rr exchange will now have two contributions as the d - branes carry the charge of the @xmath194 and @xmath195 form field potentials . to be more precise , the coupling of a d-@xmath1-brane to these fields will be given by _ pd^p+1 ( _ p+1+f_p-1 ) = + + _ pd^p+1_p+1 + _ p f ( 2r_p-1s ) ( 2r_pp)d^p-1_p-1 , [ 3.40 ] where the hat denotes the pullback to the d - brane s worldvolume , @xmath196 is the d-@xmath1-brane quantum of charge and this d - brane is wrapped around the @xmath14- and @xmath16-directions with winding numbers @xmath197 and @xmath1 , respectively . both the rr exchanges will give the desired @xmath198 factor . we remark that since just integer quanta of the @xmath195 field can be exchanged there has to be a quantisation condition on the worldvolume gauge field . in other words , the usual flux quantisation condition has to be satisfied . this gives the relation q_p-2=_p f ( 2r_p-1s ) ( 2r_pp ) , [ 3.41 ] where @xmath199 is an integer . substituting for the known values of @xmath196 and @xmath200 we find f= . [ 3.42 ] this is exactly the quantisation condition ( [ 2.8 ] ) for the case @xmath201 obtained geometrically be performing @xmath7-duality at an angle . if @xmath202 the corresponding t - dual d - brane system is given by @xmath197 parallel branes wrapped on a @xmath203-cycle on this 2-torus . in this section we will study the d - brane system describing the extreme and near - extreme black hole presented in section 2 . this discussion parallels that of @xcite with some new ingredients . we start by performing the d - brane entropy counting in the extreme case . our d - brane system consisted of @xmath26 and @xmath27 intersecting d-@xmath28-branes with the @xmath22 center of the corresponding gauge fields turned on . it was represented as 4_1:&x^1,x^2,x^4,x^5 , + & f_45^(1)= , + 4_2:&x^1,x^3,x^4,x^5 , + & f_45^(2)=- , [ 4.1 ] where the branes carry @xmath24 units of left - moving momentum in the @xmath53-direction . the d-4-branes are also wrapped on the @xmath38-direction with windings @xmath39 and @xmath83 . the excitations of this d - brane system are described by three different sectors . the first two correspond to open strings with both ends on the same group of d - branes and the third to open strings with each end on a different group of d - branes . the former cases were explained at the end of the previous section and the spectrum is similar to the case with vanishing worldvolume gauge field . the maximum number of massless excitation is @xmath204 ( and not @xmath205 because some excitations will give a mass term to others ) which is far to few to explain the entropy formula ( [ 2.16 ] ) @xcite . this excitations represent subleading contributions to the entropy formula . we expect these string states to condense ensuring supersymmetry of the excited d - brane system . in the gauge theory description this fact translates into the conditions for the vanishing of the d - terms of the theory and of the mass terms for the hypermultiplet associated with strings ending on different branes@xcite . the relevant excitations come from the sector of open strings with the ends on different branes @xcite . open strings attached to both type of branes will have the following boundary conditions x^0 , x^1 : & nn , + x^2 , x^3 : & nd , dn , + x^4 , x^5 : & ff , + x^6, ... ,x^9 : & dd , + [ 4.2 ] where the ff boundary conditions represent open strings satisfying the boundary conditions ( [ 3.5 ] ) and ( [ 3.6 ] ) with the field strength given by ( [ 3.4 ] ) and ( [ 3.4a ] ) . the only non - standard results come from these @xmath36- and @xmath38-directions . the parameters in the previous section read now in terms of our d - brane system as = , = - , = , + n_l = p_2q_1+p_1q_2 . [ 4.3 ] consider first the neveu - schwarz sector of the theory . since @xmath206 , the worldsheet bosons become half - integer moded and the worldsheet fermions integer moded . the zero moded operators @xmath207 , @xmath208 take ground states into ground states . thus , the mode expansion is just as in the nd case . the zero energy is then seen to be zero as it is the case of intersecting d - branes with four nd directions e=(6 - 2 ) ( --)+ 4 ( + ) = 0 , [ 4.4 ] where the first term is the contribution of the @xmath209-directions and the second term the contribution of the dn , nd and ff directions . with respect to the usual result for two nn or dd directions as it was explained in the previous section . ] there is however a crucial difference between the nd and ff directions since the latter admits a degenerated zero mode in the bosonic expansion . the vacuum state for the ns sector forms a representation of the @xmath28-dimensional algebra \ { _ 0^i,_0^j } = ^ij , i , j=2, ... ,5 , [ 4.5 ] and it has a further degeneracy due to the bosonic zero mode . a convenient basis for the vacuum states is @xmath210 comes from the landau degeneracy as explained in subsection 3.3 . this vacuum transforms as a spinor under the internal space group @xmath211 giving therefore bosonic states . the gso projections leaves half of the states out . we end up with @xmath212 states , and since there are @xmath213 of these strings ( the factor of @xmath35 is because the strings are oriented ) we have @xmath214 massless bosonic particles . next , consider the ramond sector . the zero energy is zero as it is usual even when we have the ff directions . since @xmath150 , the worldsheet fermions become half - integer moded as it is the case for nd directions . this vacuum forms a representation of the 6-dimensional dirac algebra \ { _ 0^,_0^ } = ^ , , = 0,1,6, ... ,9 , [ 4.7 ] with the landau degeneracy . a basis for the vacuum states is |s_1,s_2,s_3,y_m , [ 4.8 ] where @xmath215 and @xmath210 is defined as in ( [ 4.6 ] ) . this vacuum transforms as a spinor under the corresponding @xmath216 group . the gso projection plus the fact that physical states are annihilated by the zero mode of the supersymmetry generator leaves us with @xmath212 states and therefore we have @xmath214 massless fermionic excitations . the argument now is exactly as presented in @xcite . we have a gas of strings with @xmath214 fermionic and bosonic species carrying all together the left - moving momentum @xmath217 . this gives exactly the entropy formula ( [ 2.16 ] ) . the previous counting argument is valid for @xmath218 @xcite . outside this regime , we would obtain an entropy much smaller than the bekenstein - hawking entropy . the point is that the configurations of branes winding around in the @xmath53-direction become the relevant ones to explain the classical geometry limit . consider the case where @xmath26 and @xmath27 are co - prime ( this assumption may be easily relaxed @xcite ) . for the system corresponding to two intersecting d-4-branes with winding numbers @xmath26 and @xmath27 in the @xmath53-direction we will have a gas of @xmath219 bosonic and fermionic species . the momentum carried by these states is however quantised in units of @xmath220 giving the correct entropy formula for @xmath221 . in the extremal case we can extrapolate the entropy calculation from the weakly coupled d - brane phase to the strongly coupled classical black hole geometry phase because of supersymmetry . in other words , the bps nature of our configuration allow us to extrapolate between small and strong coupling regimes while leaving the number ( degeneracy ) of our bps states unchanged . for the non - extreme solution case this argument is no longer valid . while the small coupling calculation is still reliable in the non - bps case , we can not extrapolate it to the strong coupling region . however , it is seen that in the near - extreme case the d - brane picture reproduces the correct semi - classical thermodynamical analysis @xcite . as explained in @xcite , the excitations of our d - brane system above the bps state that correctly reproduce the near - extreme behaviour correspond to multiply wrapped branes . the corresponding d - brane system is defined by keeping the total momentum and d - brane charges fixed , while introducing some right - moving momentum @xmath222 , and anti - d-@xmath28-branes both with magnetic condensates as in ( [ 4.1 ] ) and with windings along the @xmath53-direction @xmath223 and @xmath224 . the calculation of the variation of the entropy due to the change in momentum is similar to the one presented in @xcite . we have @xmath219 bosonic and fermionic species with momentum quantised in units of @xmath220 . the change in the left - moving entropy is @xmath225 , while the change in the right - moving entropy is @xmath226 which dominates . the resulting change in the entropy agrees with the semi - classical value in ( [ 2.28 ] ) . we proceed by calculating the contribution to the entropy due to the anti - d-4-branes . in order to do that we shall assume that in the near - extreme regime we may perform duality transformations leaving the degeneracy of states unchanged @xcite . consider the case when we have a d-4-brane and an anti - d-4-brane along the @xmath227-directions with a magnetic condensate as in ( [ 4.1 ] ) and wrapped on the @xmath53-direction with winding numbers @xmath228 and @xmath229 , respectively . next , perform the following duality transformations t_1 s t_4 t_5 t_2 ( up ) , [ 4.9 ] where @xmath230 means a t - duality transformation along the @xmath231-direction , @xmath232 the s - duality transformation and @xmath233 the uplift of the configuration to 11-dimensions . to follow this duality orbit one has to realize that s - duality acts as electromagnetic duality on the worldvolume gauge field of the d-3-branes @xcite . the duality transformation ( [ 4.9 ] ) interchanges the @xmath24 quanta of momentum in the @xmath53-direction with a m-5-brane parallel to the @xmath234-directions with winding number @xmath24 along the @xmath235-direction . the d-4-brane with winding @xmath27 in the @xmath53-direction is transformed into a membrane singly wrapped on the @xmath236-direction and wrapping diagonally the @xmath237 2-torus on the @xmath238-cycle @xcite . this membrane makes an angle @xmath239 with the @xmath235-direction in this squared 2-torus where = . [ 4.10 ] the m-5-brane intersects the membrane along the string defined by this direction . the d-4-brane with winding @xmath240 ( anti - d-4-brane with winding @xmath229 ) is transformed under the duality operation ( [ 4.9 ] ) into left(right)-moving momentum modes propagating along the common string direction . there will be @xmath241 left - moving quanta of momentum along the @xmath235-direction and @xmath242 left - moving quanta of momentum along the @xmath243-direction . the condition that these momentum modes propagate along the common string is = . [ 4.11 ] the total left - moving momentum along this direction is then seen to be p_l= , r= . [ 4.12 ] we now note that the momentum modes along the intersection string are described by a superconformal field theory with central charge @xmath244 @xcite . in fact , reducing our m - theory configuration along the @xmath36-direction we obtain a d-4-brane intersecting a d-2-brane over the string direction described above . the open strings carrying the momentum along this string direction are described by a theory with the required central charge . thus , from equation ( [ 4.12 ] ) we conclude that the left sector of the theory is at level @xmath245 giving a change in the entropy @xmath246 . similarly , the change in the entropy arising from the right - moving sector is @xmath247 which dominates . the resulting change in the entropy again agrees with the semi - classical result ( [ 2.28 ] ) . the near - extreme entropy due to the other anti - d-4-brane may be calculated in a similar way . the calculation of the hawking temperature goes through as in @xcite . the result again agrees with ( [ 2.28 ] ) and is t_h = 2t_r = , [ 4.13 ] where @xmath248 is the right - moving temperature . in this paper we have studied a black hole described by a configuration of intersecting d-4-branes with non - vanishing worldvolume gauge fields . we have shown the agreement between the semi - classical and d - brane calculations of the entropy and hawking temperature in the extreme and near - extreme cases . as a new ingredient , we found that the landau degeneracy of open string states describing the excitations of the d - brane system enters in a fundamental way in the explanation of the microscopic structure of this black hole . it is amusing that an old result of quantum mechanics that has been brought into string theory in @xcite now has its place in the fundamental description of black holes , providing an impressive new matching with the corresponding semi - classical results . the results presented in this paper should also arise in the worldvolume description of the low - lying states of our d - brane system . the corresponding theory should be dual to an 8-dimensional theory with two vector gauge fields @xmath5 ( @xmath249 ) and two scalar fields @xmath250 ( @xmath251 ) in the adjoint representation of @xmath252 and @xmath253 , plus a hypermultiplet transforming in the fundamental representation of @xmath254 . naively one should expect from the d - brane intersection rules that such a supersymmetric theory does not exist . however , the coupling to the non - vanishing background gauge fields should render the theory supersymmetric . as another application of our work there is the possibility of considering the dual configuration of @xmath113 d - branes intersecting at @xmath255 angles @xcite . the entropy of such configuration of d-5-branes with a self - dual field strength on a compact @xmath256 also has its origin in the landau degeneracy of open strings describing the excitations of this system . further , in this case it is easy to write down the worldvolume field theory governing the dynamics of the d - brane system . it is just the super yang mills theory with twisted boundary conditions on @xmath256 . the landau levels arise in this description as torons @xcite , and the correct matching with the entropy formula arises from non - trivial results on @xmath257-functions on @xmath256 @xcite . finally , let us note that the intersection of other d - branes with worldvolume gauge fields may be studied by using the results of section 3 . an example , are the configurations found in @xcite . the corresponding worldvolume theory should also be expressed in terms of bundles with twisted boundary conditions . we would like to thank m. gutperle and g. gibbons for helpful comments . one of us ( m.s.c . ) acknowledges the financial support of jnict ( portugal ) under programme praxis xxi . 50 a. strominger and c. vafa , phys . lett . * b379 * ( 1996 ) 99 . c. callan and j.m . maldacena , nucl . phys . * b472 * ( 1996 ) 591 . breckenridge , r.c . myers , a.w . peet and c. vafa , phys . b391 * ( 1997 ) 93 . maldacena and a. strominger , phys . * 77 * ( 1996 ) 428 . johnson , r.r . khuri and r.c . myers , phys . b378 * ( 1996 ) 78 . maldacena , _ black holes in string theory _ , ph.d . thesis , hep - th/9607235 . j. polchinski , phys . * 75 * ( 1995 ) 4724 . j. polchinski , _ tasi lectures on d - branes _ , hep - th/9611050 . j. dai , r.g . leigh and j. polchinski , mod . a4 * ( 1989 ) 2073 . leigh , mod . * a4 * ( 1989 ) 2767 . m. douglas , _ branes within branes _ , hep - th/9512077 . c. bachas , phys . * b374 * ( 1996 ) 37 . j.c . breckenridge , g. michaud and r.c . myers , phys . * d55 * ( 1997 ) 6438 . costa and g. papadopoulos , _ superstring dualities and p - brane bound states _ , hep - th/9612204 ( to appear in nucl . e. witten , nucl . b460 * ( 1995 ) 335 . g. t hooft , nucl . phys . * b153 * ( 1979 ) 141 . g. t hooft , commun . math * 81 * ( 1981 ) 267 . z. guralnik and s. ramgoolam , nucl . b499 * ( 1997 ) 241 . a. hashimoto and w. taylor , nucl . b503 * ( 1997 ) 193 . costa and m. cveti , phys . rev . * d56 * ( 1997 ) 4834 . v. balasubramanian and r.g . leigh , phys . rev . * d55 * ( 1997 ) 6415 . izquierdo , n.d . lambert , g. papadopoulos and p.k . townsend , nucl . b460 * ( 1996 ) 560 . v. balasubramanian , f. larsen and r.g . leigh , _ branes at angles and black holes _ , hep - th/9704143 . duff , h. lu and c.n . pope , phys . lett . * b382 * ( 1996 ) 73 . m. cveti and a.a . tseytlin , nucl . phys . * b478 * ( 1996 ) 181 . costa , nucl . phys . * b495 * ( 1997 ) 195 . g. horowitz , j.m . maldacena and a. strominger , phys . lett . * b383 * ( 1996 ) 151 . m. li , nucl . phys . * b460 * ( 1996 ) 351 . callan , jr . and r. klebanov , nucl . phys . * b465 * ( 1996 ) 473 . green and m. gutperle , nucl . phys . * b476 * ( 1996 ) 484 . g. lifschytz , nucl . phys . * b499 * ( 1997 ) 283 . green , j.h . schwarz and e. witten , _ superstring theory _ , vol . 1 , cambridge university press , 1987 . c.g . callan , c. lovelace , c.r . nappi and s.a . yost , nucl . phys . * b308 * ( 1988 ) 221 . c. bachas and m. porrati , phys b296 * ( 1992 ) 77 . a. abouelsaood , c.g . callan , c.r . nappi and s.a . yost , nucl . * b280 * ( 1987 ) 599 . green , _ d - branes and related topics _ , lectures given at the cargse summer school on strings , branes and dualities . maldacena and l. susskind , nucl . phys . * b475 * ( 1996 ) 679 . tseytlin , nucl . phys . * b469 * ( 1996 ) 51 . green and m. gutperle , phys . b377 * ( 1996 ) 28 . russo and a.a . tseytlin , nucl . phys . * b490 * ( 1997 ) 121 . i.r . klebanov and a.a . tseytlin , nucl . phys . * b475 * ( 1996 ) 179 . v. balasubramanian and f. larsen , nucl . phys . * b478 * ( 1996 ) 199 . m. berkooz , m.r . douglas and r.g . leigh , nucl . * b480 * ( 1996 ) 265 . breckenridge , g. michaud and r.c . myers , phys . rev . * d56 * ( 1997 ) 5172 . p. van baal , commun . * 94 * ( 1984 ) 397 . costa and m.j . perry , _ torons and black hole entropy _ , to appear . costa , nucl . phys . * b490 * ( 1997 ) 202 .

the _ spectrum _ of a finite group is the set of element orders of . finite groups and are _ isospectral _ if their spectra coincide . suppose that is a simple classical group of sufficiently large dimension ( the lower bound varies for different types of groups but is at most 62 ) defined over a finite field of characteristic . it is proved that a finite group isospectral to can not have a nonabelian composition factor which is a group of lie type defined over a field of characteristic distinct from . together with a series of previous results this implies that every finite group isospectral to is ` close ' to . namely , if is a linear or unitary group , then , in particular , there are only finitely many such groups for given . if is a symplectic or orthogonal group , then has a unique nonabelian composition factor and , for given , there are at most 3 variants for ( including ) . groups of lie type , finite classical groups , element orders 20d06 , 20d20

1405.4374

the _ spectrum _ @xmath0 of a finite group @xmath1 is the set of element orders of @xmath1 . finite groups having the same spectra are said to be _ isospectral_. recently , the following general result was obtained . suppose that @xmath3 is a finite simple group , and @xmath1 is a finite group with @xmath8 and @xmath9 . then @xmath1 is isomorphic to @xmath3 . the statement was conjectured by w. shi @xcite in 1987 , while its proof was completed in @xcite ( see the latter article for background information and complete list of references ) . it is worth mentioning that theorem a together with ( * ? ? ? * corollary 5.2 ) implies that a finite simple group and a finite group with the same burnside rings are isomorphic as well . what happens if we omit the condition @xmath9 in theorem a ? then @xmath1 is not necessary isomorphic to @xmath3 . for example , there are infinitely many groups with the same spectrum as the spectrum of the alternating permutation group of degree six @xcite . on the other hand , it turns out that for a bulk of finite nonabelian simple groups @xmath3 , a finite group isospectral to @xmath3 is isomorphic to @xmath3 or , at least , to a group @xmath1 with @xmath10 . investigations on this subject have 30-year history and resulted in more than a hundred papers of numerous authors . we do not intend to give a detailed review , rather prefer to draw an overall picture . in order to do that we formulate the following conjecture attributed to v.d . mazurov : for every finite nonabelian simple group @xmath3 , apart from a finite number of sporadic , alternating and exceptional groups and apart from several series of classical groups of small dimensions , if a finite group @xmath1 is isospectral to @xmath3 then @xmath1 is an almost simple group with socle isomorphic to @xmath3 . the conjecture was proved for sporadic groups @xcite , for alternating groups @xcite , and very recently for exceptional groups of lie type @xcite . here we deal with groups isospectral to finite simple classical groups . observe that there are a lot of results on this topic concerning groups in particular characteristics or dimensions . the peculiarity of our approach is that we sacrifice groups of small dimensions in order to obtain a general result ( cf . the theorems below with recent papers @xcite ) . throughout this paper we use single - letter names for simple classical groups , following @xcite , i.e. , for example , @xmath11 means @xmath12 , as well as the standard abbreviation @xmath13 , where @xmath14 , @xmath15 , and @xmath16 . [ t : l&u ] suppose that @xmath17 is a finite simple linear or unitary group and @xmath18 . then a finite group isospectral to @xmath3 is isomorphic to a group @xmath1 with @xmath10 . in particular , there are only finitely many pairwise non - isomorphic finite groups @xmath1 with @xmath8 . [ t : s&o ] suppose that @xmath3 is a finite simple symplectic or orthogonal group , where @xmath19 for @xmath20 , @xmath21 for @xmath22 , and @xmath23 for @xmath24 . if @xmath1 is a finite group with @xmath8 , then @xmath1 has a unique nonabelian composition factor @xmath6 , and one of the following holds : _ ( i ) _ @xmath25 _ ( ii ) _ @xmath20 and @xmath26 _ ( iii ) _ @xmath27 is even , @xmath22 , and @xmath28 in particular , there exist at most @xmath29 possibilities for @xmath6 for given @xmath3 . it seems very likely that the conclusion of theorem [ t : l&u ] is valid under the hypothesis of theorem [ t : s&o ] . in fact , as shown in the last section , theorems [ t : l&u ] and [ t : s&o ] are straightforward consequences of a series of previous results @xcite and the following theorem whose proof is the main goal of this paper . [ t : main ] let @xmath3 be a simple classical group over a finite field of characteristic @xmath4 , and @xmath1 be a finite group with @xmath8 . suppose that @xmath18 for @xmath17 , @xmath19 for @xmath20 , @xmath21 for @xmath22 , and @xmath23 for @xmath24 . then @xmath1 has a unique nonabelian composition factor @xmath6 , and @xmath6 is not isomorphic to a group of lie type over a field of characteristic distinct from @xmath4 . we strongly believe that the conclusion of theorem [ t : main ] remains true for all simple classical groups except the well - known examples of isomorphic groups in different characteristics such as @xmath30 ; three groups @xmath31 , @xmath32 , and @xmath33 , which are the only simple groups of lie type isospectral to some solvable groups ( * ? ? ? * corollary 1 ) ; and a few exotic cases such as @xmath34 @xcite . moreover , for many classical groups of small dimensions and specific characteristics the conclusion of the theorem has already been proved . in the article we concentrate on groups of sufficiently large dimensions in order to achieve a generic proof covering classical groups of all types in all characteristics . a final remark . the determination of properties of a group by means of its element orders is widely applied in computational group theory , especially in development of the so - called black - box algorithms , i.e. algorithms that do not exploit specific features of a group representation . we mention here just one of the numerous results on this subject , which is nearest to our main assertion . namely , w. kantor and . seress in @xcite proved that the characteristic of a finite simple group @xmath1 of lie type can be determined if three greatest element orders of @xmath1 are known ( it is additionally assumed that the characteristic of @xmath1 is odd ) . one may observe that theorem [ t : main ] says the same thing but only for classical groups of large dimensions and involving the whole spectrum of @xmath1 . however , we do not presuppose that @xmath1 is a simple group , and this is an important distinction between our approach and that of @xcite . on the other hand , in contrast to @xcite we do not propose here any practical implementation of our results . for nonzero integers @xmath35 , let @xmath36 and @xmath37 $ ] denote their greatest common divisor and least common multiple , respectively . given a nonzero integer @xmath27 , we put @xmath38 for the euler totient function of @xmath27 , @xmath39 for the set of prime divisors of @xmath27 , and if @xmath1 is a finite group then , as usual , @xmath40 stands for @xmath41 . if @xmath42 is a set of primes , then @xmath43 denotes the @xmath42-part of @xmath27 , that is , the largest divisor @xmath44 of @xmath27 with @xmath45 ; and @xmath46 denotes the @xmath47-part of @xmath27 , that is , the ratio @xmath48 . if @xmath27 is a nonzero integer and @xmath49 is an odd prime with @xmath50 , then @xmath51 denotes the multiplicative order of @xmath27 modulo @xmath49 . given an odd integer @xmath27 , we put @xmath52 if @xmath53 , and @xmath54 otherwise . fix an integer @xmath55 with @xmath56 . a prime @xmath49 is said to be a _ primitive prime divisor _ of @xmath57 if @xmath58 . we write @xmath59 to denote some primitive prime divisor of @xmath57 , if such a prime exists , and @xmath60 to denote the set of all such divisors . zsigmondy @xcite proved that primitive prime divisors exist for almost all pairs @xmath61 . [ l : zsigmondy ] let @xmath55 be an integer and @xmath56 . for every natural number @xmath62 the set @xmath60 is nonempty , except for the pairs @xmath63 . for @xmath64 the product of all primitive divisors of @xmath57 taken with multiplicities is denoted by @xmath65 . put @xmath66 . the number @xmath65 is said to be the _ greatest primitive divisor _ of @xmath57 . it follows from the definition that @xmath67 if @xmath68 . it is easy to check that @xmath69 if @xmath70 , and @xmath71 if @xmath72 , as well as @xmath73 if @xmath74 , and @xmath75 if @xmath76 . it follows from @xcite that for @xmath77 , @xmath78 where @xmath79 is the @xmath62th cyclotomic polynomial and @xmath49 is the largest prime dividing @xmath62 ; moreover , if @xmath80 does not divide @xmath81 then @xmath82 . note that for a divisor , the property of being primitive depends on the pair @xmath61 and is not determined by the number @xmath57 . for example , @xmath83 but @xmath84 , and @xmath85 but @xmath86 . [ l : kjpu0 ] suppose that @xmath55 and @xmath62 are integers with @xmath56 and @xmath87 , and @xmath4 is a prime . then @xmath88 divides @xmath89 . furthermore , if @xmath4 divides @xmath62 then @xmath90 . let @xmath49 be odd . then @xmath91 means that the multiplicative order of @xmath55 modulo @xmath49 equals @xmath92 . hence the order of @xmath93 modulo @xmath49 equals @xmath62 , that is @xmath94 . vice versa , if the order of @xmath93 modulo @xmath49 is equal to @xmath62 , and @xmath4 divides @xmath62 , then the order of @xmath55 modulo @xmath49 is equal to @xmath92 . thus , for odd @xmath88 and @xmath89 the assertion holds by the definition of a greatest primitive divisor . assume that @xmath95 . then @xmath55 is odd , @xmath96 , and @xmath97 . therefore , @xmath98 divides @xmath99 , and so divides @xmath100 . if @xmath101 and @xmath4 divides @xmath62 , then @xmath55 is odd , @xmath96 , and @xmath102 , so @xmath103 . [ l : kipma ] let @xmath55 and @xmath62 be integers with @xmath56 and @xmath87 . if @xmath62 is odd then @xmath104 , and if @xmath62 is a multiple of @xmath105 then @xmath106 . let @xmath62 be odd . by the definition of @xmath107 , we may assume that @xmath108 , so both @xmath109 and @xmath110 are odd . the order of @xmath55 modulo an odd prime @xmath49 is equal to @xmath111 if and only if the order of @xmath112 modulo @xmath49 is equal to @xmath62 . hence @xmath113 . it follows that @xmath104 because @xmath110 divides @xmath114 . the latter assertion follows from lemma [ l : kjpu0 ] . indeed , put @xmath115 and observe that @xmath116 . [ l : ki ] suppose that @xmath55 , @xmath62 , and @xmath117 are integers with @xmath56 , @xmath87 , and @xmath118 , @xmath49 is an odd prime such that @xmath119 and @xmath49 divides @xmath120 . the following hold : _ ( i ) _ if @xmath121 , then @xmath122 divides @xmath123 _ ( ii ) _ if @xmath124 , then @xmath122 divides @xmath125 _ ( iii ) _ if @xmath126 , then @xmath122 divides @xmath127 _ ( iv ) _ if @xmath128 , then @xmath122 divides @xmath129 _ ( v ) _ if @xmath130 , then @xmath122 divides @xmath131 _ ( vi ) _ if @xmath132 , then @xmath122 divides @xmath133 in particular , @xmath134 . moreover , @xmath135 in _ ( ii ) _ and _ ( v)_. observe that , by ( [ eq : ki ] ) , for @xmath136 and @xmath137 we have @xmath138 . it is easy to verify the following equalities : if @xmath121 , then @xmath139 for even @xmath55 , and @xmath140 for odd @xmath55 ; if @xmath124 , then @xmath141 ; if @xmath126 , then @xmath142 ; if @xmath143 , then @xmath144 if @xmath145 , then @xmath146 if @xmath147 , then @xmath148 . these equalities yield the lemma . [ l : estimknbyvarphi ] let @xmath55 and @xmath62 be integers , and @xmath14 . if @xmath149 , @xmath150 , and @xmath151 , then @xmath152 . we prove that @xmath153 first . let @xmath49 be the greatest prime divisor of @xmath62 and @xmath154 where @xmath155 . it follows that @xmath156 , and if @xmath81 is not a multiple of @xmath44 , then @xmath157 . if @xmath157 , then the desired inequality holds by ( * ? ? ? * lemma 7 ) , so we assume that @xmath49 divides @xmath158 and , in particular , @xmath44 divides @xmath81 . as observed in @xcite , the inequality @xmath159 where @xmath160 , holds true . since @xmath161 , we have @xmath162 let @xmath163 and @xmath164 . then @xmath165 and , therefore , @xmath166 by ( [ eq : estimkn ] ) . let @xmath167 and @xmath168 . then @xmath169 , @xmath170 . furthermore , @xmath44 divides @xmath171 . if @xmath172 , then @xmath173 , so we may assume that @xmath174 and @xmath175 . in this case the assertion follows because @xmath176 . let @xmath177 . then @xmath172 and @xmath178 . if @xmath179 then @xmath180 . therefore , by ( [ eq : estimkn ] ) , @xmath181 thus , @xmath182 , so @xmath183 and @xmath170 . since @xmath184 , the case @xmath169 is impossible by the hypothesis . if @xmath185 then @xmath186 . let , finally , @xmath187 . then @xmath188 , as required . thus , the inequality @xmath153 is proved . now we apply lemma [ l : kipma ] . if @xmath189 then @xmath106 . if @xmath190 then @xmath191 , so @xmath192 . finally , if @xmath62 is odd then @xmath193 , hence @xmath194 . the lemma is proved . define the following function on positive integers : @xmath195 [ l : kjubyu7 ] let @xmath196 be a prime power , @xmath14 , @xmath4 be an odd prime , and @xmath197 be a natural number with @xmath198 . then @xmath199 and @xmath200 . by lemma [ l : estimknbyvarphi ] , the assertion holds for sufficiently large @xmath197 . indeed , if @xmath201 , then @xmath202 @xmath203 where the latest inequality follows because @xmath204 . therefore , we may assume that @xmath205 and , in particular , @xmath206 . let @xmath207 stand for the set of @xmath197 , satisfying the hypothesis of the lemma and the inequality @xmath205 . by brute - force attack , we obtain that @xmath208 , @xmath209 , @xmath210 , @xmath211 , @xmath212 , @xmath213 , @xmath214 , @xmath215 , @xmath216 , @xmath217 , @xmath218 . if @xmath219 , then @xmath220 by ( [ eq : ki ] ) , so it is sufficient to prove the assertion for all @xmath221 , @xmath211 , @xmath213 , @xmath215 , @xmath216 , @xmath217 , @xmath218 . we use the following inequalities on @xmath222 for @xmath223 : @xmath224 @xmath225 @xmath226 @xmath227 @xmath228 @xmath229 @xmath230 by ( [ eq : ki ] ) , for @xmath231 , @xmath211 , @xmath213 , @xmath216 the equality @xmath232 holds , so @xmath199 by the above inequalities . for the other numbers from @xmath233 we have : @xmath234 @xmath235 @xmath236 thus , the first required inequality is proved . if @xmath4 divides @xmath197 then @xmath237 by lemma [ l : kjpu0 ] , so @xmath238 . therefore , we assume that @xmath4 does not divide @xmath197 . then @xmath239 . suppose that @xmath240 . then @xmath4 is the greatest prime divisor of @xmath241 , hence , using ( [ eq : ki ] ) , we obtain @xmath242 suppose that @xmath243 . in this case the greatest prime divisor of @xmath241 is at most @xmath217 . it follows that @xmath244 unless @xmath197 equals @xmath216 or @xmath213 . for @xmath245 , the greatest prime divisor of @xmath241 is equal to @xmath4 , and @xmath246 is not a multiple of @xmath197 , hence @xmath247 let @xmath248 . then @xmath249 , and the assertion follows from the inequalities : @xmath250 @xmath251 @xmath252 the lemma is proved . [ l : rpart ] let @xmath253 and @xmath254 be integers greater than @xmath255 , and @xmath14 . _ ( i ) _ if an odd prime @xmath49 divides @xmath256 , then @xmath257 . _ ( ii ) _ if an odd prime @xmath49 divides @xmath258 , then @xmath49 divides @xmath259 . _ ( iii ) _ if @xmath256 is a multiple of @xmath105 , then @xmath260 . see , for example , ( * ? ? ? * chapter ix , lemma 8.1 ) . let @xmath261 $ ] denote the integer part of a real number @xmath262 . [ l : eta ] let @xmath263 be positive integers , @xmath264 , and @xmath265 . if @xmath266 is even then @xmath267 $ ] . if @xmath266 is odd then @xmath268 $ ] . since the equation @xmath269 has two integer solutions for odd @xmath197 and one solution for even @xmath197 , the cardinality of @xmath270 is equal to the sum of @xmath55 and the quantity of odd numbers in the interval @xmath271 $ ] . if @xmath266 is even then this quantity is equal to @xmath272 $ ] , and if @xmath266 is odd then it equals @xmath273 $ ] . [ l : intervalwithprime ] if @xmath27 is a natural number and @xmath23 , then the interval @xmath274 contains a prime . if , in addition , @xmath275 , then the interval @xmath276 contains a prime . the first part is proved in @xcite . it is shown in @xcite that for every natural number @xmath277 , the interval @xmath278 $ ] contains a prime . since @xmath279 , the second part holds for all sufficiently large @xmath27 ( precisely , for @xmath280 ) . for smaller @xmath27 the assertion can be verified directly . the _ prime graph _ @xmath281 of a finite group @xmath1 is the nonoriented graph with the vertex set @xmath40 and two distinct vertices @xmath49 and @xmath282 are adjacent if and only if @xmath283 . the notion of prime graph was introduced by g.k . gruenberg and o. kegel ( for this reason it is also called the gruenberg kegel graph ) . they established that a finite group with disconnected prime graph is either frobenius or @xmath284-frobenius group , or has a unique nonabelian composition factor with disconnected prime graph . williams @xcite published this result and started the classification of finite simple groups with disconnected prime graph . the classification was completed by a.s . kondrat@xmath285ev @xcite . the full list of nonabelian simple groups with disconnected prime graph can be found , e.g. , in ( * ? ? ? * tables 1a-1c ) . recall that an independent set of vertices or a _ coclique _ of a graph @xmath286 is any subset of pairwise nonadjacent vertices of @xmath286 . we write @xmath287 to denote the independence number of @xmath286 , that is the greatest size of coclique in @xmath286 . for a group @xmath1 , put @xmath288 . by analogy , for each prime @xmath49 , define the @xmath49-independence number @xmath289 to be the greatest size of cocliques containing the vertex @xmath49 in @xmath281 . for convenience , we refer to a coclique containing @xmath49 as an @xmath290-coclique . in @xcite there was proved the following assertion which is , in some sense , a generalization of the gruenberg kegel theorem ( below we give the statement of this result from ( * ? ? ? * theorem 1 ) ) . [ l : structure ] let @xmath1 be a finite group with @xmath291 and @xmath292 . then the following hold : _ ( i ) _ there exists a nonabelian simple group @xmath6 such that @xmath293 , where @xmath294 is the maximal normal soluble subgroup of @xmath1 . _ ( ii ) _ for every coclique @xmath295 of @xmath281 of size at least @xmath29 , at most one prime of @xmath295 divides the product @xmath296 . in particular , @xmath297 . _ ( iii ) _ one of the following holds : _ ( a ) _ every prime @xmath298 nonadjacent to @xmath284 in @xmath281 does not divide the product @xmath299 in particular , @xmath300 _ ( b ) _ there exists a prime @xmath301 nonadjacent to @xmath284 in @xmath281 ; in which case @xmath302 , @xmath303 , and @xmath304 or @xmath305 for some odd @xmath253 . let @xmath3 and @xmath1 be isospectral finite groups . it follows by the definition of prime graph that @xmath306 . therefore , if @xmath3 satisfies the hypothesis of lemma [ l : structure ] , then so does @xmath1 . in @xcite , for every finite nonabelian simple group , an adjacency criterion of its prime graph is developed and all cocliques and @xmath307-cocliques of greatest size in this graph are found , as well as @xmath308-cocliques of greatest size for groups of lie type in characteristic @xmath4 . this information and williams s classification , imply the following assertion which is the first step toward a proof of main results of the present paper . [ p : uniques ] let @xmath3 be a finite simple group of lie type different from the groups @xmath31 , @xmath32 , and @xmath33 . if @xmath1 is a finite group isospectral to @xmath3 , then the following hold true for @xmath1 : _ ( i ) _ there exists a nonabelian simple group @xmath6 such that @xmath293 , where @xmath294 is the maximal normal soluble subgroup of @xmath1 . _ ( ii ) _ for every coclique @xmath295 of @xmath281 of size at least @xmath29 , at most one prime of @xmath295 divides the product @xmath296 . in particular , @xmath309 . _ ( iii ) _ every prime @xmath298 nonadjacent to @xmath284 in @xmath281 does not divide the product @xmath296 . in particular , @xmath310 if @xmath3 is a finite nonabelian simple group with connected prime graph and is different from an alternating group , then it satisfies the hypothesis of lemma [ l : structure ] by @xcite . if @xmath3 has a disconnected prime graph then the existence and uniqueness of a nonabelian composition factor @xmath6 follow from the gruenberg kegel theorem and @xcite . by ( * ? ? ? * propositions 2,3 ) , the inequality @xmath292 and the insolubility of @xmath1 imply that ( ii ) and ( iii ) of lemma [ l : structure ] hold . it remains to observe that the exceptional case ( b ) of assertion ( iii ) of lemma [ l : structure ] does not hold by ( * ? ? ? * theorem 2 ) . the proposition is proved . for a classical group @xmath3 , we put @xmath311 to denote its dimension if @xmath3 is a linear or unitary group , and its lie rank if @xmath3 is a symplectic or orthogonal group . observe that @xmath312 in theorems 13 in introduction . [ p : notexcept ] suppose that @xmath3 is a finite simple classical group , @xmath313 if @xmath3 is linear or unitary , and @xmath314 if @xmath3 is symplectic or orthogonal . suppose that @xmath1 is a finite group isospectral to @xmath3 , and @xmath6 is a unique nonabelian composition factor of @xmath1 . if @xmath6 is a group of lie type , then @xmath6 is a classical group , @xmath315 if @xmath6 is linear or unitary , and @xmath316 if @xmath6 is symplectic or orthogonal . applying ( * ? ? ? * tables 2,3 ) , it is easy to obtain that @xmath317 provided the hypothesis of the proposition . by proposition [ p : uniques](ii ) , we have @xmath318 . on the other hand , it follows from ( * ? ? ? * table 4 ) that @xmath319 for every simple exceptional group @xmath2 of lie type . thus , @xmath6 is a classical group . the required inequalities on @xmath320 hold by ( * ? ? ? * tables 2,3 ) . the proposition is proved . given a classical group @xmath3 over a field of order @xmath253 , put @xmath321 [ l : indepofcrit ] let @xmath3 be a simple classical group over a field of order @xmath253 and characteristic @xmath4 . suppose that @xmath49 and @xmath282 are distinct primes , @xmath322 , @xmath323 , and @xmath324 . _ ( i ) _ if @xmath325 , then @xmath326 for every distinct odd primes @xmath327 and @xmath328 . _ ( ii ) _ if @xmath329 , then @xmath330 for every odd prime @xmath331 . it follows from @xcite . thus , for two distinct primes @xmath332 , where @xmath333 , the answer to the question whether they are adjacent in @xmath334 depends only on @xmath335 , if @xmath336 , and @xmath335 , @xmath337 , if @xmath338 . in @xcite several functions of natural argument are used to formulate an adjacency criterion , one of them is @xmath339 defined in ( [ eq : eta ] ) , and two others we define here . @xmath340 for @xmath14 , put @xmath341 it is an easy observation that @xmath342 is a bijection and @xmath343 is an identity . for linear and unitary groups , we exploit also a reformulation of an adjacency criterion ( see ( * ? ? ? * lemmas 2.12.3 ) ) , if it is more convenient for our goals than an initial formulation from @xcite which used the function @xmath342 . this reformulation is based on the equality @xmath344 , which follows from lemma [ l : kipma ] and the definition of @xmath342 . now we introduce a new function in order to unify further arguments . namely , given a simple classical group @xmath3 over a field of order @xmath253 and a prime @xmath49 coprime to @xmath253 , we put @xmath345 it follows that @xmath346 observe that @xmath347 in the case of @xmath348 . [ l : fermat ] let @xmath3 be a simple classical group over a field of order @xmath253 and characteristic @xmath4 . if @xmath49 is an odd prime from @xmath349 then @xmath350 divides @xmath81 , and if @xmath3 is a symplectic or orthogonal group then @xmath351 divides @xmath81 . if @xmath3 is not unitary , then @xmath350 divides @xmath335 , which divides @xmath81 by fermat s little theorem . if @xmath3 is unitary , then @xmath350 does not divide @xmath335 only when @xmath335 is odd . but then @xmath352 and @xmath350 divides @xmath81 because @xmath49 is odd . let @xmath3 be a symplectic or orthogonal group . if @xmath350 is even then @xmath353 , and if not , then @xmath350 divides @xmath354 . [ l : auxvarphi ] let @xmath3 be a simple classical group over a field of order @xmath253 and characteristic @xmath4 , and let @xmath355 . _ ( i ) _ if @xmath356 , then @xmath357 . _ ( ii ) _ if @xmath49 and @xmath282 are distinct primes from @xmath349 with @xmath358 and @xmath359 , then @xmath49 and @xmath282 are adjacent in @xmath334 . _ _ if @xmath49 and @xmath282 are distinct primes from @xmath349 with @xmath360 and @xmath361 , then @xmath49 and @xmath282 are adjacent in @xmath334 if and only if @xmath362 . _ ( iv ) _ if @xmath49 and @xmath282 are distinct primes from @xmath349 and @xmath362 , then @xmath49 and @xmath282 are adjacent in @xmath334 . it follows from @xcite . let @xmath3 be a simple classical group over a field of order @xmath253 and characteristic @xmath4 . for @xmath363 , set @xmath364 . if @xmath365 then , by @xcite , every coclique @xmath295 of greatest size in @xmath334 does not contain @xmath4 , so the set @xmath366 is well - defined for @xmath295 . define @xmath367 as the union of sets @xmath366 , and @xmath368 as the intersection of these sets , where @xmath295 runs over all cocliques of greatest size in @xmath334 . the next lemma is a particular case of the main theorem of @xcite . [ l : tl ] let @xmath3 be a simple classical group over a field of order @xmath253 and characteristic @xmath4 , and let @xmath365 . let @xmath295 be a coclique of greatest size in @xmath334 . if @xmath369 then @xmath370 . if @xmath371 then @xmath372 for some @xmath373 . in particular , @xmath374 . the sets @xmath368 , @xmath375 and numbers @xmath376 are listed in table _ [ tab : tl]_. see ( * ? ? ? * tables 2,3 ) . .cocliques of greatest size [ cols="^,<,^,^,^",options="header " , ] let @xmath197 and @xmath44 be natural numbers and @xmath377 . observe that @xmath378 is an odd integer if and only if @xmath379 . [ l : kdividesj ] let @xmath380 . suppose that @xmath381 are different primes such that @xmath382 , @xmath383 , and @xmath384 , and suppose that @xmath385 and @xmath386 . if @xmath387 , then neither @xmath388 nor @xmath389 is divisible by @xmath390 . if @xmath6 is symplectic or orthogonal , then neither @xmath62 nor @xmath197 is congruent to @xmath44 modulo @xmath391 . in particular , if @xmath392 and @xmath393 , then @xmath388 is not divisible by @xmath390 when @xmath387 , and @xmath197 is not congruent to @xmath44 modulo @xmath391 when @xmath6 is symplectic or orthogonal . first , suppose that @xmath394 . to avoid unwieldy notation , we assume that @xmath6 is a linear group , that is @xmath342 is the identity function . the proof for unitary groups is analogous , we have only to use the function @xmath395 instead . recall that by lemma [ l : pdoesnotdivideum1 ] , the number @xmath4 does not divide @xmath396 , that is we have @xmath397 . it follows from the equality @xmath382 that @xmath6 has a semisimple element of order @xmath398 . now lemma [ l : criterion ] implies that @xmath6 has an element of order @xmath399/d$ ] , where @xmath400 is some divisor of @xmath396 . if @xmath44 divides either of @xmath197 and @xmath62 , then @xmath401 divides @xmath399/d$ ] , which contradicts the fact that @xmath402 . now let @xmath6 be a symplectic or orthogonal group . recall that @xmath403 . the equality @xmath382 implies that @xmath6 has a semisimple element of order @xmath398 . by lemma [ l : criterion ] , @xmath6 has an element of order @xmath404/d$ ] , where @xmath400 divides @xmath105 . assume that one of the numbers @xmath197 and @xmath62 , say @xmath197 , is congruent to @xmath44 modulo @xmath391 . then @xmath405 divides @xmath406 , and hence @xmath401 divides @xmath407/d$ ] , contrary to the hypothesis . let @xmath392 and let @xmath393 . by lemma [ l : rsnotprs ] , there is a number @xmath266 such that @xmath408 is large with respect to @xmath3 , @xmath409 lies in @xmath410 , and @xmath411 does not . writing @xmath412 , @xmath413 and observing that @xmath382 , we see that the assertion of the lemma holds for @xmath414 . [ l : lnotinx ] if @xmath415 , then the conclusion of theorem _ [ t : main ] _ holds . . then lemma [ l : sinyi ] implies that @xmath6 also lies in @xmath416 . suppose that @xmath417 . since @xmath415 , we have @xmath418 , and hence @xmath419 ( see table [ tab : kll ] ) . one of the numbers @xmath420 , @xmath421 , @xmath422 is a multiple of @xmath29 , and we denote this number by @xmath423 . by lemma [ l : tmaless2t3 ] , we have @xmath424 , and so @xmath425 . therefore , there is @xmath392 such that @xmath426 . then @xmath427 due to lemma [ l : transfer ] . by lemma [ l : tpll ] , the function @xmath428 is injective and @xmath429 , and hence @xmath430 . since @xmath431 , we see that @xmath432 is divisible by @xmath433 , which is a contradiction by lemma [ l : kdividesj ] . let @xmath6 be an orthogonal group lying in @xmath416 . then @xmath434 . by lemmas [ l : tplop ] and [ l : tplom ] , the set @xmath435 contains @xmath423 such that @xmath436 and @xmath437 . since @xmath438 , it follows that @xmath439 . thus there is @xmath392 such that @xmath426 . then @xmath440 , and hence @xmath441 , which is not the case by lemma [ l : kdividesj ] . [ l : sisnotx1 ] if @xmath442 , then the conclusion of theorem _ [ t : main ] _ holds . let @xmath380 . suppose that there is @xmath439 such that @xmath390 divides @xmath423 . as in the previous lemma , there is @xmath392 such that @xmath440 , and by lemma [ l : tpll ] , it follows that @xmath443 . hence @xmath432 is divisible by @xmath390 , which is a contradiction due to lemma [ l : kdividesj ] . thus it suffices to show that there is @xmath444 that is a multiple of @xmath390 . by the previous result , we may assume that @xmath445 . in this case , an additional difficulty is that the members of @xmath446 are not consecutive integers . nevertheless , for each group @xmath3 we will prove that @xmath446 contains a number with the desired property . let @xmath447 and @xmath27 be even . then @xmath448 . since @xmath449 , and therefore , the characteristic @xmath4 of @xmath3 does not divide @xmath450 by lemma [ l : pdoesnotdivideum1 ] , it follows that @xmath451 , and hence @xmath452 . if @xmath453 , then @xmath454 , and if @xmath455 , then @xmath456 . in each case , @xmath446 contains two consequent numbers , and one of them is a multiple of @xmath284 , as required . let @xmath24 and @xmath27 be even . then @xmath457 , whence @xmath458 . now the proof is somewhat similar to the proof in the previous paragraph if we take into account the following observation : @xmath446 contains the numbers @xmath459 when @xmath453 and @xmath459 when @xmath455 and @xmath460 , and the set of the residues of these numbers module 4 is all of @xmath461 ; furthermore , if @xmath462 , then @xmath463 is a multiple of @xmath105 and lies in @xmath376 , and if @xmath464 , then so does @xmath465 . the remaining cases are @xmath466 , @xmath467 with @xmath27 odd and @xmath22 with @xmath27 even , where we have @xmath418 , whence @xmath419 and @xmath433 . if @xmath466 , @xmath467 and @xmath468 , then @xmath469 . thus @xmath421 lies in @xmath446 and is a multiple of @xmath29 . if either @xmath466 , @xmath467 with @xmath470 or @xmath22 with @xmath453 , then @xmath471 , and @xmath422 lies in @xmath446 and is a multiple of @xmath29 . if @xmath22 and @xmath455 , then @xmath472 , and @xmath420 is the desired number . now we may assume that @xmath445 and @xmath473 , and in particular , both @xmath3 and @xmath6 are symplectic or orthogonal groups . moreover , by lemma [ l : sinyi ] , each @xmath474 satisfies the following condition : if @xmath475 , then @xmath476 . also observe that @xmath477 $ ] is an injective function , and hence @xmath478 is uniquely determined by @xmath479 . furthermore , if @xmath480 , then @xmath62 is uniquely determined by @xmath478 , and so by @xmath479 as well . [ l : tplneq3 ] if @xmath418 , then the conclusion of theorem _ [ t : main ] _ holds . let @xmath418 . then either @xmath466 , @xmath467 with @xmath27 odd or @xmath22 with @xmath27 even . furthermore , since @xmath6 is symplectic or orthogonal , we have @xmath434 . also note that if @xmath481 , @xmath482 , then @xmath483 ; if @xmath484 , then @xmath485 ; and if @xmath486 , then @xmath487 ( see table [ tab : kll ] ) . let @xmath466 , @xmath467 and @xmath468 . suppose that @xmath481 or @xmath488 . since @xmath489 , it follows that @xmath490 . now the congruence @xmath483 yields @xmath491 . but then @xmath492 , which is not the case because @xmath493 . if @xmath484 , then for every @xmath254 , we have that @xmath494 is not congruent to @xmath284 modulo @xmath29 . on the other hand , @xmath495 , and hence @xmath496 . similarly , @xmath497 , since otherwise @xmath487 and @xmath498 . let @xmath466 , @xmath467 and @xmath470 . suppose that @xmath481 or @xmath488 . the equality @xmath489 yields @xmath499 . both @xmath500 and @xmath501 are divisible by @xmath105 , therefore , one of them is congruent to @xmath105 modulo @xmath502 . denote this number by @xmath55 . since @xmath503 and @xmath504 , we have @xmath505 , whence @xmath506 . thus @xmath507 . as @xmath428 is invertible on the set of multiples of @xmath105 and @xmath6 has the same type as @xmath3 , it follows that @xmath508 is equal to @xmath55 , and so it is congruent to @xmath105 modulo @xmath502 . applying lemma [ l : kdividesj ] , we derive a contradiction . thus @xmath6 is an orthogonal group of even dimension . if @xmath499 , then @xmath509 . by lemmas [ l : tpls][l : tplom ] , both @xmath510 and @xmath511 are equal to @xmath421 , so @xmath512 . similarly , @xmath513 . repeating the previous argument we derive a contradiction by lemma [ l : kdividesj ] . if @xmath514 , then the equality @xmath489 implies that the only remaining possibility is @xmath515 . then @xmath492 , which is not the case because @xmath516 . let @xmath22 and @xmath455 . if @xmath517 , then @xmath489 yields @xmath518 , whence @xmath519 , which is a contradiction . if @xmath486 , then @xmath520 , and this is also a contradiction since @xmath254 must be congruent to @xmath29 modulo @xmath105 . if @xmath484 , then @xmath521 . since @xmath254 can not be congruent to @xmath255 modulo @xmath105 , it follows that @xmath522 . thus @xmath6 has the same type and the same lie rank @xmath27 as @xmath3 . both @xmath523 and @xmath524 are multiples of @xmath105 , so one of them is congruent to 4 modulo 8 , and we denote this number by @xmath55 . as @xmath525 , we have @xmath21 and @xmath526 . therefore @xmath392 , and applying lemma [ l : kdividesj ] , we derive a contradiction . let @xmath22 and @xmath453 . the equality @xmath489 leaves us with two possibilities : either @xmath527 or @xmath499 and @xmath528 . if @xmath6 is a symplectic group or an orthogonal group of odd dimension , then @xmath529 , while @xmath530 . suppose that @xmath528 , that is @xmath6 has the same type and the same lie rank @xmath27 as @xmath3 . since @xmath531 , @xmath532 are multiples of @xmath105 , we can choose a number @xmath55 of them with @xmath533 . furthermore , @xmath534 because @xmath525 and so @xmath21 . thus @xmath392 , and applying lemma [ l : kdividesj ] , we derive a contradiction . the remaining case for @xmath22 with @xmath453 , is @xmath535 , and this case requires the most effort . let @xmath536 and @xmath537 . similarly to the proof of lemma [ l : rsnotprs ] we conclude that @xmath538 and @xmath539 have the following properties : for @xmath540 , the number @xmath541 is large with respect to @xmath3 , @xmath542 , and @xmath543 . since @xmath453 , we conclude that @xmath544 has the same properties . now @xmath545 constitute a coclique in @xmath334 because for any @xmath546 we have @xmath547 and @xmath548 whenever @xmath68 . therefore these numbers are large with respect to @xmath6 and constitute a coclique in @xmath549 . since @xmath428 is injective and @xmath550 , the value of @xmath551 depends only on @xmath337 . if @xmath552 , then @xmath553 , and if @xmath554 , then @xmath555 . in both cases , @xmath556 is odd . let @xmath557 and write @xmath386 . as @xmath558 , we have @xmath559 . if @xmath560 , then since @xmath254 is odd , it follows that @xmath62 is even . the condition @xmath558 implies that @xmath6 has an element of order @xmath407/2 $ ] , which is not the case because @xmath62 and @xmath197 have opposite parity when @xmath484 , and the same parity when @xmath486 . thus @xmath561 . if @xmath562 , then @xmath6 has an element of order @xmath563 $ ] , which is divisible by @xmath401 , contrary to the fact that @xmath402 . let @xmath564 . then @xmath62 is even , and hence @xmath6 has an element of order @xmath565 $ ] , which is divisible by @xmath401 . finally , let @xmath566 . but then @xmath567 has only two different possible values , and this is a contradiction since the numbers @xmath568 with @xmath569 must be distinct . [ l : tplneq2 ] if @xmath448 , then the conclusion of theorem _ [ t : main ] _ holds . if @xmath448 , then @xmath570 and @xmath27 is even . since @xmath6 is symplectic or orthogonal , we have @xmath571 . . then @xmath530 . the equality @xmath489 implies that @xmath572 . let @xmath573 and @xmath574 . observe that @xmath575 is odd . as in the proof of lemma [ l : rsnotprs ] , we write @xmath576 . then @xmath577 and @xmath539 have the following properties : for @xmath540 , the number @xmath541 is large with respect to @xmath3 , @xmath542 , and @xmath543 . furthermore , @xmath578 and @xmath579 are not adjacent in @xmath334 . since @xmath580 and @xmath6 has the same type and the same lie rank as @xmath3 , it follows that @xmath581 . let @xmath582 and write @xmath386 . as @xmath558 , we have @xmath583 . if @xmath584 , then @xmath6 contains elements of orders @xmath585 $ ] and @xmath586 $ ] , one of which must be divisible by @xmath401 because @xmath4 divides either @xmath396 or @xmath587 . but then @xmath588 , a contradiction . thus @xmath589 . since @xmath55 is odd , there are two possible values for @xmath567 , and these are @xmath55 and @xmath590 . as @xmath591 , we have @xmath592 . if @xmath593 , then @xmath594 , and if @xmath595 , then @xmath596 . we denote by @xmath49 the number in the set @xmath597 for which @xmath598 . now @xmath282 and @xmath49 satisfy the hypothesis of lemma [ l : kdividesj ] , which is a contradiction because @xmath599 . now suppose that @xmath453 . the conditions @xmath489 and @xmath516 leave us with two further possibilities : either @xmath572 or @xmath600 with @xmath491 . in the first case , we use the argument of the previous paragraph with @xmath601 and @xmath602 . now assume that @xmath603 and write @xmath604 , @xmath605 , @xmath606 , @xmath577 , and @xmath539 . since @xmath607 is odd , it follows that @xmath578 and @xmath579 satisfy the conclusion of lemma [ l : rsnotprs ] , that is @xmath578 and @xmath579 are not adjacent in @xmath334 , they are large with respect to @xmath3 , and for @xmath540 the number @xmath608 belongs to @xmath410 , but @xmath609 does not . let @xmath582 , @xmath386 , and @xmath385 . then the equality @xmath610 yields @xmath611 when @xmath484 , and @xmath612 when @xmath486 . as @xmath558 , it follows that @xmath559 . if @xmath613 , then @xmath614 contains @xmath615 $ ] which is divisible by the odd part of @xmath616 , and so by @xmath401 , which is a contradiction . thus since @xmath617 , the number @xmath618 is even , and therefore there is only one possible value for @xmath567 , namely , @xmath619 . but then @xmath620 , a contradiction . [ l : tplneq4 ] if @xmath457 , then the conclusion of theorem _ [ t : main ] _ holds . note that the equality @xmath457 holds if and only if @xmath24 and @xmath27 is even . suppose that @xmath453 . if @xmath481 or @xmath488 , then @xmath489 yields @xmath499 . but then @xmath529 and @xmath493 , a contradiction . if @xmath486 , then @xmath490 . suppose that @xmath499 . then @xmath434 . since @xmath621 and @xmath622 lie in @xmath623 and are multiples of @xmath105 , it follows that one of the numbers @xmath624 and @xmath625 is congruent to @xmath105 modulo @xmath502 . but this contradicts lemma [ l : kdividesj ] . if @xmath484 , then @xmath626 . let @xmath627 . then @xmath628 , and , on the other hand , it follows from table [ tab : kll ] that @xmath629 . thus @xmath630 , and hence @xmath631 , @xmath632 for some integer @xmath423 . writing @xmath633 and noting that @xmath634 is even , we conclude that @xmath635=3c+3 $ ] by lemma [ l : eta ] . there is a coclique @xmath295 of size @xmath636 in @xmath334 such that @xmath637 for every @xmath638 . let @xmath638 . the prime @xmath49 is large with respect to @xmath3 , and so @xmath49 is large with respect to @xmath3 too , whence @xmath639 ( see table [ tab : tl ] ) . furthermore , there is @xmath640 such that @xmath641 . since @xmath642 and @xmath643 , we have @xmath644 . thus writing @xmath645 , we see that @xmath646 for every @xmath638 . by lemma [ l : eta ] , the size of @xmath647 is equal to @xmath648=3c+3=|x|$ ] , and therefore @xmath649 . it follows that @xmath295 contains @xmath49 and @xmath650 such that @xmath651 and @xmath652 . but then @xmath653 , a contradiction . it remains to consider the situation where @xmath600 , @xmath491 , and either @xmath434 or @xmath654 and @xmath655 . first , assume that @xmath654 and @xmath656 . observe that the congruence and the condition @xmath657 imply that @xmath658 . in particular , we have @xmath659 , and so @xmath660 with @xmath661 , @xmath662 , and @xmath663 . if @xmath484 , then it follows that the numbers @xmath664 , @xmath665 , and @xmath666 are congruent to @xmath284 modulo @xmath105 . moreover , they have different residues modulo @xmath29 , and hence one of them is congruent to @xmath667 modulo @xmath668 . we denote this number by @xmath197 and take @xmath55 to be such that @xmath441 . applying lemma [ l : kdividesj ] to @xmath392 , we conclude that @xmath669 , a contradiction . if @xmath486 , then the numbers @xmath670 , @xmath671 , and @xmath672 are odd and one of them is a multiple of @xmath29 . since @xmath673 , we again derive a contradiction by lemma [ l : kdividesj ] . thus we may assume that @xmath434 . let @xmath674 , @xmath675 , @xmath676 , @xmath677 , and @xmath678 . it follows that @xmath679 , the primes @xmath680 are large with respect to @xmath3 and for @xmath681 , we have @xmath682 , but @xmath683 . furthermore , @xmath684 . therefore , @xmath685 is equal to @xmath686 when @xmath484 and to @xmath687 when @xmath486 . let @xmath688 and @xmath386 . as for @xmath681 , we have @xmath689 , it follows that @xmath690 . if @xmath691 , then @xmath478 is even , and so @xmath692 . but then @xmath693 and either @xmath694 and @xmath484 or @xmath695 and @xmath486 , and by adjacency criterion , @xmath49 and @xmath696 are not adjacent in @xmath6 for @xmath681 , which is not the case . thus @xmath697 . if @xmath698 , then @xmath699 has an element of order @xmath700 $ ] and @xmath701 has an element of order @xmath702 $ ] . this is a contradiction because @xmath703 divides both @xmath55 and @xmath266 for @xmath681 . if @xmath704 , then it is easy to construct an element of required order . so @xmath705 and there are only two possible values for @xmath567 . this is impossible because @xmath706 is a coclique in @xmath334 . now suppose that @xmath455 . the group @xmath6 can not be isomorphic to @xmath699 since otherwise @xmath376 is congruent to @xmath284 modulo @xmath29 and @xmath494 does not . let @xmath707 . then @xmath489 yields @xmath708 . if @xmath499 , then @xmath492 , and if @xmath709 , then @xmath710 , which contradicts to the fact that @xmath516 . thus @xmath486 , and it follows from @xmath489 that @xmath499 . then @xmath628 , and so @xmath434 . assume that @xmath711 , that is @xmath712 . then @xmath713 , and hence @xmath714 with one of them being a multiple of @xmath105 but not of @xmath502 . now lemma [ l : kdividesj ] gives the desired contradiction . finally , assume that @xmath715 and @xmath716 . let @xmath717 , @xmath718 , @xmath719 , @xmath720 , and @xmath721 . then @xmath722 , @xmath706 is a coclique in @xmath334 and it consists of numbers large with respect to @xmath3 , and for @xmath681 , we have @xmath682 , but @xmath683 . writing @xmath723 , @xmath724 and using lemma [ l : transfer ] , we obtain @xmath725 . choose @xmath688 and take @xmath386 . since for @xmath681 , we have @xmath689 , it follows that @xmath726 . it is easy to see that if @xmath727 , then @xmath614 contains a number divisible by @xmath703 for each @xmath681 , which is impossible . so we assume that @xmath728 . if @xmath729 , then @xmath730 , and applying lemma [ l : kdividesj ] either to the pair @xmath731 or to the pair @xmath732 for some @xmath733 , we have the desired contradiction . let @xmath734 and @xmath735 . if @xmath736 , then since @xmath6 has no elements of order @xmath737/d$ ] with @xmath738 and does have element of order @xmath739 $ ] , it follows that @xmath740 . similarly , if @xmath741 then @xmath742 . if @xmath743 , then @xmath386 can take at most two different values , but there must be at least three of them as @xmath706 is a coclique in @xmath334 . and if @xmath744 , then @xmath745 , and this contradiction completes the proof . if @xmath17 with @xmath18 , @xmath447 with @xmath746 , @xmath24 with @xmath23 , or @xmath22 with @xmath21 , then @xmath525 . we eliminated all possibilities for the group @xmath6 under the assumption that @xmath525 , and so the proof of theorem [ t : main ] is complete . although the final part of the proof is technically complicated , its idea is transparent and based on an application of the well - known pigeonhole principle . here pigeons are from @xmath747 and holes are from @xmath748 . the number of pigeons and number of holes are almost the same and , what is sufficient , the difference between them is a small constant which does not depend on @xmath312 . however , there are prohibited holes , and the number of those holes increases when so does @xmath27 . it provides a contradiction that becomes evident with the growth of @xmath27 . as written in introduction , theorems [ t : l&u ] and [ t : s&o ] follow from theorem [ t : main ] and a series of previously obtained results . the next assertion summarizes the main results of @xcite . [ l : othercases ] let @xmath3 be a finite simple classical group over a field of characteristic @xmath4 , and @xmath749 . suppose that @xmath1 is a finite group with @xmath8 , and @xmath6 is a unique nonabelian composition factor of @xmath1 . then one of the following holds : _ ( 1 ) _ @xmath750 _ ( 2 ) _ @xmath751 , where @xmath752 , and @xmath753 _ ( 3 ) _ @xmath754 and @xmath755 _ ( 4 ) _ @xmath756 , @xmath20 and @xmath26 _ ( 5 ) _ @xmath757 is even , @xmath22 and @xmath758 _ ( 6 ) _ @xmath6 is a group of lie type over a field of characteristic distinct to @xmath4 . this assertion and theorem [ t : main ] yield theorem [ t : s&o ] immediately . a _ cover _ of a finite group @xmath1 is a finite group having @xmath1 as a factor group . a cover is proper if the corresponding factor group is proper . if @xmath759 for every proper cover @xmath2 of @xmath1 , then @xmath1 is said to be _ recognizable by spectrum among covers_. it follows from ( * ? ? ? * lemma 9 ) that a finite group @xmath1 is recognizable by spectrum among covers if and only if @xmath759 for every splitting extension @xmath760 , where @xmath761 is an absolutely irreducible finite - dimensional @xmath1-module over a finite field of characteristic @xmath49 . for every simple linear and unitary group @xmath17 with @xmath762 and every @xmath3-module @xmath761 , if the characteristic @xmath49 of @xmath761 coincides with the characteristic @xmath4 of @xmath3 , then it is proved that @xmath763 @xcite . if @xmath761 is defined over the field of characteristic distinct to @xmath4 and @xmath764 , then any extension @xmath765 contains an element whose order does not belong to @xmath410 ( * ? ? ? * lemma 11 ) . for other classical groups the following general result was recently obtained . [ l : grcov ] let @xmath3 be one of the simple groups @xmath766 , where @xmath767 , @xmath768 , where @xmath769 , @xmath467 , where @xmath770 , and @xmath771 , where @xmath767 . suppose that @xmath761 is a finite - dimensional @xmath3-module over a field of characteristic @xmath49 prime to @xmath253 . then either @xmath772 or @xmath773 and @xmath177 . if @xmath773 , then there is a @xmath774-dimensional @xmath3-module @xmath761 over a field of characteristic @xmath29 such that @xmath775 . if @xmath17 , @xmath18 , and @xmath1 is a finite group isospectral to @xmath3 , then theorem [ t : main ] and lemma [ l : othercases ] provide @xmath776 , where @xmath294 is the soluble radical of @xmath1 . an application of the aforementioned results on covers of linear and unitary groups completes the proof of theorem [ t : l&u ] . the author is grateful to m.a . grechkoseeva for her reading earlier drafts of this article . her valuable comments and helpful suggestions have enabled us to improve the text significantly . l. di martino , a. e. zalesskii , minimum polynomials and lower bounds for eigenvalues multiplicities of prime - power order elements in representations of quasi - simple groups , j. algebra 243 ( 2001 ) 228263 , corrigendum , j. algebra 296 ( 2006 ) , 249252 . e. i. khukhro , v. d. mazurov , automorphisms with centralizers of small rank , in : groups st . andrews 2005 , vol . 340 of london math . lecture note ser . , cambridge univ . press , cambridge , 2007 , pp .

we study the dynamics of a single frenkel exciton in a disordered molecular chain . the coherent - potential approximation ( cpa ) is applied to the situation when the single - molecule excitation energies as well as the transition dipole moments , both their absolute values and orientations , are random . such model is believed to be relevant for the description of the linear optical properties of one - dimensional aggregates . we calculate the exciton density of states , the linear absorption spectra and the exciton coherence length which reveals itself in the linear optics . a detailed analysis of the low - disorder limit of the theory is presented . in particular , we derive asymptotic formulas relating the absorption linewidth and the exciton coherence length to the strength of disorder . such expressions account simultaneously for all the above types of disorder and reduce to well - established form when no disorder in the transition dipoles is present . the theory is applied to the case of purely orientational disorder and is shown to agree well with exact numerical diagonalization .

cond-mat0302312

the optically active states in organic molecular aggregates @xcite the spatially - regular linear arrangements of dye molecules are one - dimensional frenkel excitons . @xcite the nature of the excitonic states to be extended over many molecules reveals itself in the narrowing of the absorption resonances and the shortening of the radiative lifetime upon aggregate formation ( effects known as exchange narrowing and superradiance " ) . @xcite the related coherence length coincides with the total number of molecules in the aggregate only if the size of the latter is sufficiently small . otherwise various exciton decoherence mechanisms , among which the static disorder is the most essential at low temperature , substantially reduce the coherence length that becomes independent from the actual size of the aggregate . considerable work has been done to analyze the aspects of disorder in the molecular excitation energies @xcite and positions . @xcite much less attention has been devoted to the disorder in orientations @xcite or , generally , in the transition dipole moments of individual molecules . however , from the limited available information on the actual structure of molecular arrangements one can not exclude that the transition dipoles are seriously influenced by disorder . moreover , the last type of disorder affects in a nontrivial way the separate components of the optical susceptibility tensor and , hence , would be observable in polarization - resolved absorption and luminescence experiments@xcite . the present work has been motivated by the need for a detailed theoretical analysis of orientational disorder , as well as clarification of certain aspects concerning the exciton coherence length in molecular aggregates . in this paper the linear optical properties of a disordered molecular chain are studied using the single - site coherent - potential approximation @xcite ( cpa ) . applied to various disorder problems , such as that of the electronic structure in a random alloy , this self - consistent approximation was shown to reproduce well the single - particle characteristics like the density of states ( dos ) . as for the systems in which elementary excitations are frenkel excitons , the cpa is in addition capable to provide the complete information on the linear optical response , because the latter is extracted only from the single - exciton green s function ( gf ) averaged over disorder realizations . the cpa has been successfully used to model the optical spectra of periodic molecular arrangements with random on - site energies @xcite and composition . @xcite however , if the transition dipoles are also affected by disorder , a proper modification of the theory is required . the key observation which allows us to do this is that the dipoles enter the off - diagonal part of the hamiltonian in a bilinear form . in this case a single - site approximation can be constructed according to a vector analog of the shiba ansatz . @xcite the resulting scheme is equivalent to the matrix extension of cpa derived by blackman , esterling and berk @xcite ( beb ) for compositionally - disordered alloys with random hopping energies . [ the analytical properties of beb cpa have been examined in refs . ] . persson and liebsch @xcite , and rozenbaum _ et al . _ @xcite have constructed a similar version of cpa to study , respectively , the susceptibility of polarizable particles and the vibrational modes of coupled oscillators randomly and isotropically oriented in two or three dimensions . however , their theory is still not applicable to the molecular aggregates in which one typically deals with an essentially anisotropic distribution of the transition dipoles . the paper is organized as follows . in the next section we define the model of a disordered molecular aggregate and set up the basic formalism for the forthcoming calculations . the exciton green s tensor is introduced to express the related single - particle quantities : the dos , the linear polarizability , and the coherence length . in sec . [ sec : cpa ] we present the general formulation of the beb - cpa scheme and in sec . [ sec : weakdis ] derive its analytic solution in the low - disorder limit . in sec . [ sec : results ] the theory is applied to a specific case of orientational disorder and compared to the results of the exact numerical diagonalization . an aggregate can be considered as a chain of @xmath1 identical two - level molecules . the linear optical properties in the resonance region are determined by the single - particle states of the frenkel hamiltonian @xcite @xmath2 the operator @xmath3 ( @xmath4 ) creates ( annihilates ) an electronic excitation of energy @xmath5 on the @xmath6th molecule ( whose excited state is supposed to be nondegenerate ) . the transfer terms @xmath7 result from the coulomb interaction between an excited molecule and one in the ground state . accounting only for the dipole - dipole contribution one has @xmath8 where @xmath9 are the vector components of the @xmath6th molecule transition dipole . the `` coupling kernel '' @xmath10 carries information on the dielectric function of the surrounding material and the location of the molecules , but not on their dipole moments . if the dynamics of the exciton subsystem is nondissipative ( and the molecules have no magnetic structure , so that @xmath11 can be chosen real ) @xmath10 has to be symmetric in the tensorial indices ( @xmath12 ) . we shall assume the molecules to form a one - dimensional regular lattice such that @xmath10 depends solely on the intermolecular distance @xmath13 ( measured in the units of lattice spacing ) . the theory will be presented in a general way without specifying a concrete form of this dependence , i.e. , for any screening and anisotropy of the coulomb interaction that can occur in a solvent or on a substrate interface . the disorder enters in our model both through the on - site energies @xmath5 and the dipole moments @xmath11 . we shall assume the families of random parameters @xmath14 corresponding to distinct sites to be mutually independent and to have identical probability distribution . at the same time we still allow for correlations between the transition energy and the dipole components of a given molecule . finally , we shall neglect everywhere the effects of a finite length of the aggregate considering the limit @xmath15 . this approximation is justified for sufficiently strong disorder when the exciton coherence length found for the infinite system does not exceed the actual length of the chain . for a given random realization the total information on the aggregate s optical dynamics is imbedded in the complete set of single - exciton eigenstates @xmath16 ; the vacuum @xmath17 is the direct product of the molecular ground states ; @xmath18 is the real - space exciton wave function with an eigenenergy @xmath19 . the local density of states ( ldos ) on the @xmath6th site and the total ( normalized ) exciton dos are found as @xmath20 @xmath21 where the real - space single - exciton gf is given by @xmath22 and the symbol @xmath23 indicates that summation is performed over all eigenstates . here and below the gf for a real - valued argument @xmath24 is found as the limit @xmath25 . assuming that around the single - exciton resonance the light wavelength is much larger than the spatial extent of a typical excitonic wave function the aggregate interacts with the optical field as a pointlike dipole . hence , the linear response to the external optical field is completely described by the polarizability tensor @xmath26 | 0 \rangle.\ ] ] in which the operator of total polarization represents the sum of local polarizations , @xmath27 , both positive- and negative - energy components : @xmath28 to set up a convenient formalism for the forthcoming discussion let us define the time - ordered two - point correlator of the local polarizations @xmath29 the newly introduced quantity will be referred through the paper as the green s tensor ( gt ) , as opposed to the `` scalar '' gf considered above . combining the previous formulas we arrive at the straightforward relation @xmath30,\ ] ] which indicates that ( up to the sign factor ) the gt coincides with the positive - energy counterpart of the local linear response function . as usual one is interested not in the solution of the entire dynamical problem for a given disorder realization , but in finding the configurational average of some basic observable parameters . in our case the quantities of interest are those defined in eqs . ( [ rhodef ] ) and ( [ absdef ] ) . the subsequent implementation of cpa suggests that the disorder - averaged dos @xmath31 and polarizability @xmath32 are to be expressed in terms of the conditionally - averaged on - site gt @xmath33 with fixed @xmath6th site variables @xmath34 , and the complete statistical average @xmath35 the only quantities accessible within a single - site effective theory . to proceed with the dos it is sufficient to notice that , because the @xmath6th - site dipoles @xmath9 remain fixed , the conditional average of the local gt and gf are still related to each other as established in eq . ( [ tenstoscal ] ) , @xmath36 . as a result the scalar gf is obtained via projection @xmath37 using this relationship the conditionally - averaged ldos can be expressed as @xmath38 while the total dos is found in terms of @xmath39 as a trivial single - site average @xmath40 remarkably , the statistical averaging in the last formula can be performed immediately if the probability distribution of the transition dipoles is of the `` purely orientational '' form . the trace over vector indices of the conditionally - averaged gt depends only on the deterministic , in this case , absolute value of the @xmath6th - site dipole moment @xmath41 . the averaged gt and gf are related to each other as @xmath42 and consequently , @xmath43 the straightforward application of the averaging procedure to eq . ( [ absdef ] ) gives the disorder - averaged linear polarizability in the form @xmath44.\ ] ] in order to reduce the double real - space summation we have used the translational invariance of @xmath45 that allows to replace in both terms of eq . ( [ dachi ] ) @xmath46 by @xmath47 . the last sum here is already converging and does not depend on its upper limit @xmath1 . equivalently , @xmath48,\ ] ] where the momentum - domain disorder - averaged gt is introduced according to @xmath49 @xmath50 being measured in units of the inverse lattice spacing . noticeably , the above polarizability scales linearly with the number @xmath1 of molecules constituting the aggregate . unlike the physics coming from the nontrivial dependence of @xmath45 on the energy variable @xmath51 and the intersite separation @xmath52 this elementary aspect is not related to the degree of exciton coherence . nevertheless it guarantees fulfillment of an important part of the general sum rule according to which the polarizability is an extensive quantity proportional to the total number of polarizable objects . in the literature dealing with the optics of molecular aggregates it is usually introduced the notion of the exciton coherence length @xcite to characterize the spatial extension of the exciton wave functions . the cited works , addressing the problem mainly with numerical methods , employ several definitions of this parameter to be extracted from the set of wave functions obtained with explicit diagonalization procedures . even though the quantitative estimates provided within all approaches can be in reasonable agreement with each other , the physical arguments used are somewhat different . from the viewpoint of the present work a natural definition of the exciton coherence length can be given on the basis of eq . ( [ dachi ] ) . in fact , entering it the disorder - averaged single - exciton gt contains all the information on the linear optical response of the excitonic system and depends nontrivially on the amplitude and phase coherence between the wave functions for different disorder realizations . these coherence properties partly reveal themselves in the dos and absorption spectra which can be essentially different from those of the disorder - free system , but it is also the dependence of @xmath53 on the intersite separation that gets strongly affected by the structural disorder within the chain . excluding some peculiar cases , @xcite each component of the disorder - averaged gt is characterized by an exponential behavior @xmath54 where @xmath55 is a complex - valued wave number . keeping in mind this asymptotics while performing the real - space summation in eq . ( [ dachi ] ) one concludes that the dominant contribution to the disorder - averaged polarizability @xmath56 comes from the pairs of sites with @xmath57 . the parameter @xmath58 indicating how many molecules contribute to a given component of the linear polarizability , provides a natural definition of the exciton coherence length . clearly , @xmath59 is a straightforward generalization of the conventional quasiparticle phase coherence length , defined in terms of the large - intersite - separation asymptotics of the gf , to incorporate the vectorial nature of the excitonic polarization . the quantity ( [ ncd1 ] ) can be also thought of as the nonlocality range of the linear optical response function . as already mentioned , in order to find the statistical averages ( [ cagamma ] ) and ( [ agamma ] ) we shall employ the single - site self - consistent approximation known as beb cpa.@xcite this theory is capable to address simultaneously both diagonal and off - diagonal disorder , provided the second enters the random hamiltonian in a generalized multiplicative form . the last condition is intrinsically fulfilled in the excitonic problem under consideration because the coupling ( [ transferterm ] ) is given by a bilinear combination of the random @xmath11 with @xmath10 being translationally - invariant deterministic matrix . the beb extension of the scalar cpa @xcite is based on the so - called beb transformation.@xcite the latter represents a vector generalization of the multiplicative ansatz originally implemented by shiba @xcite to construct a single - site self - consistent theory for bond - disordered alloys with mean - geometric relationship between hopping integrals . upon such transformation the problem of finding the resolvent of a hamiltonian with both diagonal and multiplicative off - diagonal disorder is equivalently reformulated for an operator which acts in a space with additional tensorial dimensions but contains only site - diagonal disorder . with application to frenkel excitons the beb transformation has been already realized in sec . [ sec : basic ] by considering instead of operators @xmath3 ( @xmath4 ) the polarizations @xmath60 ( @xmath61 ) equipped with vector indices . as a consequence the dynamical equation for the scalar gf , @xmath62 is replaced with that for the gt , @xmath63 [ here and below omitting the indices we assume the usual rules for multiplication and inversion of tensorial quantities . ] unlike the case of eq . ( [ greenmain ] ) , the disorder enters eq . ( [ gnjmatr ] ) only in the site - diagonal form , namely , through the tensor @xmath64 associated with the bare local gf @xmath65 at the same time , because the algebraic structure of equation ( [ gnjmatr ] ) remains similar to that of ( [ greenmain ] ) , the usual cpa self - consistency arguments can be employed to approximate the statistical averages defined in eqs . ( [ cagamma ] ) and ( [ agamma ] ) . the beb - cpa scheme@xcite is accomplished via straightforward generalization of the scalar - cpa equations within the tensorial formalism as outlined below . the disorder effect on the single - particle quantities is accounted in cpa by replacing the random local gts @xmath66 with a deterministic site - independent coherent - potential gt @xmath67 . the disorder - averaged gt in the momentum representation is found from eq . ( [ gnjmatr ] ) as @xmath68^{-1 } \gamma(z),\ ] ] where @xmath69 is the momentum - space coupling kernel . in turn , the conditionally - averaged on - site gt @xmath70 is approximated by assuming that the site , carrying its random parameters , is placed into the same coherent environment as the one of eq . ( [ cpa2matrix ] ) . thus , finding this conditional average reduces to solving a single - impurity problem realized by eq . ( [ gnjmatr ] ) in which the bare gts @xmath71 for @xmath72 are replaced by @xmath67 . a simple derivation @xcite leads to @xmath73^{-1 } \gamma_n(z),\ ] ] where the site - diagonal self - energy @xmath74 describes coupling of the selected molecule with the coherent - background molecules constituting the rest of the chain . clearly , being a functional of only the coherent - potential gt @xmath67 , this self - energy is the same as the one entering the complete disorder - average of the local gt @xmath75^{-1 } \gamma(z).\ ] ] provided both @xmath67 and @xmath76 are represented by nonsingular matrices one gets @xmath77 where the local gt is found in terms of @xmath67 from eq . ( [ cpa2matrix ] ) after the momentum - space integration : @xmath78^{-1 } \gamma(z).\ ] ] the unknown coherent - potential gt @xmath67 is to be determined in a self - consistent way . in spirit of the original cpa , one demands the single - site expectation of the conditionally - averaged gt to coincide with the completely - averaged local gt : @xmath79 the latter guarantees that the conditional and the complete average of the gt found above will provide identical estimates for the single - site quantities such as ldos or the local ( single - molecule ) polarizability . summarizing , the beb - cpa procedure amounts to solving equations which , including auxiliary definitions , are listed in formulas from ( [ g0matr ] ) to ( [ cpa1matr ] ) . for nontrivial disorder models and realistic forms of the coupling @xmath10 this can be done only with the use of numerical methods . an example of such numerical solution is given in sec . [ sec : results ] [ see also appendix [ bebexpl ] ] . the analytic treatment of the low - disorder limit of the theory is presented in sec . [ sec : weakdis ] . as clear from the general discussion of sec . [ sec : basic ] , the beb cpa allows to find configurational averages of the exciton dos and linear polarizability of a disordered molecular aggregate . it is natural to check the validity of such nonperturbative theory asking , in particular , how close are the estimates to the actual quantities and whether they meet certain fundamental physical requirements . this question can be partly answered by analyzing the structure of the beb - cpa equations as done in a number of publications . @xcite for completeness , let us outline the important facts concerning analyticity and accuracy of the beb cpa . gonis and garland @xcite have proved that @xmath53 found within beb cpa possesses the same properties as a function of the complex energy @xmath51 as would have the disorder - averaged gt calculated exactly . namely , it is analytic in the whole plane excluding branch cuts on the semi - axis @xmath80 , @xmath81 , while the tensor @xmath82,\ ] ] that at @xmath83 provides the exciton spectral density , is positively ( negatively ) defined at @xmath84 ( @xmath85 ) . as a consequence the disorder - averaged polarizability will preserve causality , while the dos and absorption will be nonnegative . the authors of ref . have studied the accuracy of the theory to reproduce the energy - domain moments of the spectral density , each given by a coefficient in the high - energy expansion of the gt : @xmath86 it was shown that the beb cpa provides correctly the first three moments of the spectral density , @xmath87 where the corresponding moments entering the high - energy expansion of the local coherent - potential gt @xmath67 are given by @xmath88 \\ { \displaystyle}\phantom { l^{\alpha\beta}_2 = } \times \int_{-\pi}^\pi \frac{dk}{2\pi } \sum_{\mu,\nu } \vartheta^{\gamma\mu}_k \left < p^\mu_n p^\nu_n \right > \vartheta^{\nu\delta}_k . \end{array } \end{array}\ ] ] the fact that the number of these moments is high enough guarantees fulfillment of the sum rules for the dos and polarizability , correctly accounting for redistribution of the total oscillator strength between components of the polarizability tensor in the presence of orientational disorder . it follows also that using beb cpa one gets an exact value of the absorption linewidth , _ provided _ the latter is defined from the weighted second centered moment of the spectral density [ @xmath89 . this means that , even though there exists no unique way to define the width of an inhomogeneously - broadened absorption line , the beb cpa still gives a reasonable estimate for this characteristics of the spectrum . let us also mention that the theory does not seem to violate the fundamental inequalities @xmath90 to be fulfilled for every eigenvalue @xmath19 . these , in particular , impose the lower , @xmath91 , and the upper , @xmath92 , bounds of the spectral region , where minimum and maximum are taken over all disorder realizations . even though no rigorous proof of this property is known for the case of beb cpa , it was shown to hold in the numerical solution of the beb - cpa equations for some random alloy models . @xcite as will be demonstrated in sec . [ sec : results ] , these spectral bounds are not violated in the case of orientational disorder either . schematic picture of the integration contour in the complex momentum plane . each couple of vertical lines with opposite orientations give zero contribution to the integral . the crosses mark poles of the momentum - space gt.,height=151 ] another important aspect of the theory is the behavior of the disorder - averaged gt in the complex momentum domain . considering @xmath93 for complex @xmath50 provides additional understanding about the exciton coherence length . since @xmath93 remains the same as @xmath94 is shifted by integer of @xmath95 , the integration contour in eq . ( [ gfoumatr ] ) can be modified as shown in fig . [ fig1 ] . provided the function is meromorphic ( i.e. , has only isolated poles ) in @xmath50 , the integration reduces to sum over poles in the upper ( lower ) half - plane for @xmath96 ( @xmath97 ) . [ due to the mirror symmetry of the problem each pole has a counterpart located symmetrically with respect to the origin @xmath98 . ] to establish connection with the coherence length defined above it is sufficient to notice that only a single pair of poles will contribute to the asymptotics ( [ ncohdef ] ) , namely , having minimal distance from the real axis @xmath99 . applying these arguments to the disorder - averaged gt in the beb - cpa form ( [ cpa2matrix ] ) one concludes that the complex wave numbers @xmath55 of eq . ( [ ncohdef ] ) are to be selected among the solutions @xmath50 of the characteristic equation @xmath100 = 0.\ ] ] here we assumed that the fourier - transformed kernel @xmath101 is analytic in @xmath50 to provide meromorphic @xmath93 . this assumption is valid , for instance , if the coupling extends only over a finite number of sites . however , the last equation may not lead to a correct coherence length for a more general nonanalytic coupling . for a disorder - free chain the gt obtained from the beb cpa coincides with the one known exactly . in order to establish the relation of our approach with the existing scaling theories of disordered one - dimensional excitons @xcite let us solve the beb - cpa equations in the limit of weak disorder . a similar calculation for the bond - disordered binary alloy has been already done in ref . . however , due to the rather involved structure of beb cpa the general treatment of the low - disorder regime is still missing . for the sake of simplicity we shall also relax generality imposing certain symmetries under which the theory is formulated in terms of diagonal tensors . to proceed one needs first to specify the low " disorder in terms of its probability distribution , entering the beb cpa through the self - consistency equation ( [ cpa1matr ] ) . we do this imposing smallness of the centered moments of the probability density which characterize deviations of the on - site energies and transition dipoles from their average values @xmath102 , @xmath103 . the conditionally - averaged local gt in eq . ( [ cpa1matr ] ) can be represented as a power series in @xmath104 and @xmath105 , so that its @xmath6th - site expectation can be written as @xmath106 a @xmath107th term in the last sum is proportional to the @xmath107th - order centered moments of the probability density . the first three terms are given explicitly by @xmath108 \right|_{\epsilon_n = \epsilon , p^\alpha_n = p^\alpha } , \end{array } \end{array}\ ] ] where in the last equation @xmath109 are the second - order centered moments . the terms not included in eq . ( [ mesexp ] ) will be neglected following our assumption about weakness of disorder . later it will be demonstrated that the information contained in the third- and higher - order centered moments is indeed not preserved in the low - disorder limit of beb cpa . we shall additionally impose some simplifying restrictions on the form of disorder and intermolecular coupling to be considered below . namely , let the system to be such that ( i ) in some ( orthogonal ) basis of the three - dimensional vector space the coupling @xmath10 is diagonal in the upper indices for all pairs of molecules , and ( ii ) the disorder probability density is symmetric under reflections of two basis vectors . we shall label tensorial components corresponding to either these two basis vectors by symbols `` @xmath110 '' , `` @xmath111 '' , and those of the remaining ( third ) vector by `` @xmath112 '' . the coupling energies @xmath7 and , hence , the scalar gf @xmath113 remain invariant , while every off - diagonal tensorial component of the gt @xmath114 changes its sign under one of the above reflections [ because from @xmath115 it follows that either @xmath116 or @xmath117 is different from @xmath112 ] . therefore , configurational averaging with a symmetric probability distribution will result to a diagonal @xmath119 . regarding the beb - cpa theory , from eqs . ( [ cpa2matrix ] ) and ( [ bargse ] ) it follows consecutively that both the coherent - potential gt @xmath67 and the self - energy @xmath74 reduce to the diagonal form . as concerns the low - disorder limit to be studied here , with our symmetry requirements the average dipole moment @xmath120 is directed strictly along the @xmath112-axis , the vectorial parameter @xmath121 defined in eq . ( [ abcdef ] ) has nonzero component only along the same axis , while the tensor @xmath122 is diagonal . using the explicit dependence of the conditionally - averaged gt on the @xmath6th site random parameters [ see eqs . ( [ g0matr ] ) , ( [ propg ] ) and ( [ gskobn ] ) ] the nontrivial ( diagonal ) components of the tensors ( [ mesexp ] ) can be found as @xmath123 @xmath124 ^ 2 } \\ { \displaystyle}\phantom{\left < \tilde \gamma^\parallel_{nn}(z ) \right>_2 = } + \frac{c_2^\parallel ( z-\epsilon)\bigl[z-\epsilon + 3 \sigma^\parallel(z)|p|^2\bigr]}{\bigl[z-\epsilon - \sigma^\parallel(z)|p|^2\bigr]^3 } \\ { \displaystyle}\phantom{\left < \tilde \gamma^\parallel_{nn}(z ) \right>_2 = } + \sum_{i=1,2 } \frac{c_2^{\perp_i } \sigma^{\perp_i}(z)|p|^2}{\bigl[z-\epsilon - \sigma^\parallel(z)|p|^2\bigr]^2 } , \end{array } \\ \\ { \displaystyle}\left < \tilde \gamma^{\perp_i}_{nn}(z ) \right>_2 = \frac{c_2^{\perp_i}}{z - \epsilon - \sigma^\parallel(z)|p|^2}. \end{array}\ ] ] combined with the beb - cpa self - consistency equation [ eq . ( [ cpa1matr ] ) ] these formulas provide the starting point for the analytical treatment of the low - disorder regime . as seen from eq . ( [ dp1 ] ) , the solution behaves quite differently in the two distinct spectral regions . namely , if @xmath51 is sufficiently far away from the singularities generated by denominator @xmath125 one can look for the unknown @xmath67 in an approximate form , replacing in ( [ dp1 ] ) the beb cpa self - energy @xmath74 with its expression @xmath126 for the disorder - free system . the solution found in this way will be valid only for energies not very close to the band edges of the disorder - free system . around these points one can no more neglect the disorder - induced corrections to the self - energy component @xmath127 in eq . ( [ dp1 ] ) . nevertheless , the last situation is still addressable analytically because near the band edges an explicit asymptotic relation between @xmath128 and @xmath67 can be used . away from the band edges of the disorder - free chain we can look for the coherent - potential gt in the iterative form @xmath129 the first term with components @xmath130 corresponds to the disorder - free system and @xmath131 is a small correction to be determined . from equation ( [ bargse ] ) linearized in the parameters ( [ abcdef ] ) one has @xmath132 ^ 2}{(z - \epsilon)^2 } \left < \tilde \gamma^{(0)\parallel}_{nn}(z ) \right>_2 , \\ \\ { \displaystyle}\gamma^{(1)\perp_i}(z ) = \left < \tilde \gamma^{(0)\perp_i}_{nn}(z ) \right>_2 , \end{array}\ ] ] where @xmath133 denotes the local gt on a single impurity site in the disorder - free chain , related to the disorder - free self - energy @xmath134 by eq . ( [ gskobn ] ) . the computation of the disorder - free self - energy @xmath134 is not straightforward due to the fact that components @xmath135 and @xmath136 vanish leading to an undeterminate expression in eq . ( [ bargse ] ) . to overcome this difficulty one can employ the following arguments . consider the disorder - free chain with a single impurity molecule placed on the @xmath6th site in such a way that its dipole moment @xmath11 has no component along the @xmath112-axis . since in this case the transfer energy @xmath137 vanishes , the @xmath6th site , decoupled from the rest chain , will be characterized by the local gt @xmath138 equal to the bare quantity @xmath139 . from eq . ( [ gskobn ] ) , for any of such @xmath11 , it follows that @xmath140 . therefore , being a diagonal tensor , the self - energy will have nonzero projection only on the @xmath112-axis , orthogonal to the subspace of the considered @xmath11 . using eq . ( [ sigmasimple ] ) for the nontrivial component of the self - energy we end up with a simple expression @xmath141 , \,\ , \sigma^{(0)\perp_i}(z ) = 0 , \!\!\ ] ] where @xmath142 denotes the on - site scalar gf of the disorder - free chain . the first of eqs . ( [ sigma0 ] ) represents a natural relationship between the scalar self - energy @xmath143 and the main component of the tensorial self - energy in the absence of disorder . having derived the disorder - free self - energy we get the desired correction ( [ gamma1 ] ) to the coherent - potential gt in the form @xmath144 , \end{array } \\ \\ { \displaystyle}\gamma^{(1)\perp_i}(z ) = c_2^{\perp_i } \bar g^{(0)}_{nn}(z ) . \end{array}\ ] ] the final expression for the gt @xmath93 , valid in the low - disorder regime away from the band edges , is obtained by inserting the total @xmath67 of eq . ( [ cpsepar ] ) into ( [ cpa2matrix ] ) . let us now analyze the beb - cpa equations near the lowest band edge of the disorder - free system . we shall use the standard ( for @xmath0 , but not @xmath158 aggregates ) assumption that this point corresponds to the center @xmath98 of the brilloiun zone , thus coinciding with the energy @xmath159 defined in ( [ zetadef ] ) . around @xmath160 it is convenient to represent the @xmath112-component of the coherent - potential gt as @xmath161 assuming the absolute value of the new unknown function @xmath162 to be small compared to the total width of the exciton band one can keep only a few significant terms in the expansion of the coupling kernel @xmath163 around @xmath98 while performing the momentum integration in eq . ( [ cpa2matr ] ) . provided the constant @xmath164 has a finite ( positive ) value , that is the case if the real - space coupling @xmath165 decays faster than @xmath166 at large intermolecular separation , we can use the effective mass approximation for the disorder - free exciton dispersion @xmath167 this leads to the asymptotic expression for the corresponding component of the local gt valid around the band edge : @xmath168 furthermore , in accordance with eq . ( [ sigmasimple ] ) , the respective self - energy can be represented as @xmath169 . \!\ ] ] concerning the two components @xmath170 , even though they share the singularity present in the above @xmath171 , one can demonstrate that it is still possible to use the simple relationship @xmath172 , or equivalently to set @xmath173 . writing the cpa self - consistency equation one can retain only the most singular terms of eq . ( [ dp1 ] ) near the band edge @xmath174 where for vanishing disorder also @xmath175 . the equation for @xmath176 reads simply @xmath177 as for the component @xmath178 , the only essential are the first and the third terms in the right - hand side of eq . ( [ dp1 ] ) . by straightforward algebra we arrive at the self - consistency condition @xmath179 ^ 2 } = \sigma(z),\ ] ] that represents an equation to be solved for @xmath162 . in this formula we have used the shorthand notation @xmath180 for a combination which can be thought of as an effective disorder strength parameter governing the low - disorder limit of beb cpa near the band edge . the algebraic structure of equation ( [ mainscaling ] ) is still too complicated to obtain its solution in a closed form . at the same time the scaling of @xmath162 with the disorder strength @xmath181 can be found easily . namely , in the interval @xmath182 the solution ( both real and imaginary part ) behaves as @xmath183 employing this scaling relation one can find the low - disorder asymptotic expressions for the absorption linewidth and the exciton coherence length in the resonance region . substituting ( [ sigmasmallintr ] ) into eq . ( [ cpa2matrix ] ) the main component of the disorder - averaged gt around the phase - space point @xmath184 , @xmath185 can be written as @xmath186 the corresponding polarizability @xmath187 obtained from the basic relation ( [ dachifour ] ) , is characterized by a resonant behavior near the band edge . the width of the resonance , estimated as @xmath188 , scales with the disorder strength @xmath181 as @xmath189 furthermore , one can make use of the effective mass approximation ( [ effmass ] ) to get a scaling of the exciton coherence length in the spectral region around the absorption resonance . according to the arguments of sec . [ sec : anal ] applied to expression ( [ gammascalreg ] ) , the complex wave number @xmath190 governing the asymptotic behavior of the disorder - averaged gt at large intermolecular separation is found from equation @xmath191 taking @xmath184 we end up with the following scaling of the coherence length in the resonance region : @xmath192 it should be also mentioned that combining eq . ( [ etascal ] ) with ( [ nscal ] ) leads to the scaling relation @xmath193 which does not involve the disorder strength @xmath181 . this dependence is universal in the sense that it follows only from the shape of the exciton dispersion in the disorder - free system in the vicinity of the band edge . indeed , the scaling ( [ netascal ] ) can be established without rigorous computation of the spectral density but by simply noticing that @xmath194 and @xmath195 can be considered , respectively , as uncertainties of the exciton energy and momentum brought about by the disorder scattering . the relation between these quantities imposed by the dispersion ( [ effmass ] ) is of the quadratic form ( [ netascal ] ) . it can be noticed that the terms of expansion ( [ ldmain ] ) proportional to the higher - order centered moments of the disorder distribution will bring stronger singularities in the vicinities of the band edges compared to ones of the second - order centered moments . nevertheless , the dominant contribution to the solution will come solely from the second - order moments . let us illustrate this for the simple situation when only diagonal disorder is present . a derivation analogous to that leading to eq . ( [ scalingsol ] ) shows that the centered moment @xmath196 of order @xmath197 alone would produce the scaling @xmath198 . if the probability to encounter a random energy @xmath5 falls off rapidly enough away from the mean value @xmath199 we have @xmath200 . thus , the contribution @xmath201 will be always suppressed by that of the second - order moment @xmath202 considered so far . when applied to a molecular chain with only diagonal disorder the asymptotic expressions ( [ etascal ] ) and ( [ nscal ] ) coincide with those obtained in ref . using different scaling arguments . furthermore , for the case of diagonal disorder the asymptotics ( [ etascal ] ) has been found in ref . by means of the scalar cpa . our theory thus provides an extension of these results treating both the on - site energy disorder and the disorder in transition dipoles on an equal footing through the combined parameter @xmath181 of eq . ( [ disordercombined ] ) . the fact that beb cpa is able to reproduce the known scaling relations is not surprising because this self - consistent approximation becomes exact in the disorder - free limit . in this section we demonstrate how the general theory discussed so far can be applied to a specific model of disordered aggregate : a chain with purely orientational disorder in the transition dipoles . in addition to having constant absolute values the random dipoles will be assumed to lie in parallel planes forming an angle @xmath203 with the chain s direction [ fig . [ fig2 ] ] . the random in - plane orientation angles @xmath204 will be modelled by a box probability with density @xmath205 in the interval @xmath206 and zero outside . we shall consider the molecules to interact as dipoles in an isotropic background , so that the tensorial kernel @xmath10 acquires the form @xmath207 for the long - range ( lr ) dipole - dipole coupling one would have @xmath208 . however , since in this case the momentum integral ( [ cpa2matr ] ) can not be reduced to a closed analytic form , we shall restrict ourselves to the more easily treatable nearest - neighbor ( nn ) [ @xmath209 or next - nearest - neighbor ( nnn ) [ @xmath210 couplings . for concreteness we shall always assume that @xmath211 , so that the disorder - free chain ( @xmath212 ) represents a @xmath0 aggregate . the singular case @xmath213 is considered separately in appendix [ app : sing ] . from now on the energy will be measured in the units of the nn interaction strength , and the dipoles will be chosen to have unit absolute values . we also shift the reference point of energy to the position of the bare molecular level . since the random dipoles are restricted in a single plane the theory can be conveniently reformulated in terms of tensors projected onto the corresponding two - dimensional subspace . we choose the new basis as @xmath214 i.e. , the first vector is parallel to dipoles of the disorder - free chain , the second is orthogonal to it and to the chain s axis . in terms of the projected components @xmath215 , @xmath216 the three - dimensional dipoles are given by @xmath217 the advantages of this representation come from the fact that our disorder distribution density is invariant under the change of sign of @xmath218 . hence , since upon projection the coupling ( [ ddkernel ] ) remains diagonal , @xmath219 the reduced beb cpa will involve only diagonal tensors . as a result we end up with the affordable equations , whose explicit form for the case of nn and nnn coupling is given in appendix [ bebexpl ] . moreover , as the additional symmetry requirements formulated in sec . [ sec : weakdis ] are fulfilled , the low - disorder solution derived there can be employed . since all components of a physically - measurable tensor can be straightforwardly reconstructed using eq . ( [ dmom ] ) from the two diagonal components of the corresponding projected tensor , the last two will be presented every time to illustrate the outcome of the theory . due to their explicit form the equations of appendix [ bebexpl ] can be efficiently solved using standard numerical methods . in order to test the accuracy of the theory we also performed exact diagonalization for an open chain of @xmath220 molecules with either nn or lr coupling . the statistical error was reasonably small after averaging over 5000 disorder realizations with the energy domain divided into 400 output intervals . the imaginary ( a , b ) , and real ( c , d ) parts of the beb - cpa self - energy for the case of nn coupling . several values of the disorder parameter are used : @xmath221 ( solid lines ) , @xmath222 ( dashed lines ) , @xmath223 ( dotted lines ) , @xmath224 ( dash - dotted lines ) . @xmath225 for all curves . the horizontal lines in panels ( c ) and ( d ) mark the diagonal elements of the nnn coupling matrix at momentum @xmath98 : @xmath226 , @xmath227.,width=302 ] the tensorial components of the coherent - potential gt are plotted against energy in fig . [ fig : self ] . this figure can be used to illustrate analyticity of the theory and to predict certain features of the physical quantities to be presented below . specifically , the functions @xmath228 and @xmath229 obey the correct sign that guarantees the positivity of dos and optical absorption . from eq . ( [ cpa2matrix ] ) it follows that the resonant denominators of the exciton spectral density components are given by @xmath230 ^ 2 + [ { \mathop{\rm im}}1/\gamma(\omega ) ] ^2 $ ] , where polarization labels @xmath112 or @xmath231 are assumed . hence , provided @xmath232 , the spectral density will be characterized by a sharp resonance located around the frequency @xmath24 to be found from @xmath233 . the solution of this equation for either polarization and momentum @xmath50 can be found graphically from fig . [ fig : self](c , d ) [ where , in particular , the positions of @xmath234 and @xmath235 at @xmath98 are shown with horizontal lines ] . actually , the assumption on smallness of the imaginary part of the inverse coherent - potential gt is valid only in the case of component @xmath236 and only for moderately weak disorder . as concerns @xmath237 , the above condition is never satisfied , so that one ends up with no well - defined quasiparticle peak in the corresponding component of the spectral density . the disorder - averaged exciton dos for several values of disorder parameter @xmath238 calculated with beb cpa ( dashed lines ) or exact diagonalization ( solid lines ) . ( a ) case of nn coupling ( both beb cpa and exact diagonalization ) . ( b ) case of nnn coupling ( beb cpa ) or lr coupling ( exact diagonalization ) . @xmath225 for all curves . the arrows mark the spectral bound estimates for the case of nn coupling and @xmath239 , namely , @xmath240.,title="fig:",width=302 ] the disorder - averaged exciton dos for several values of disorder parameter @xmath238 calculated with beb cpa ( dashed lines ) or exact diagonalization ( solid lines ) . ( a ) case of nn coupling ( both beb cpa and exact diagonalization ) . ( b ) case of nnn coupling ( beb cpa ) or lr coupling ( exact diagonalization ) . @xmath225 for all curves . the arrows mark the spectral bound estimates for the case of nn coupling and @xmath239 , namely , @xmath240.,title="fig:",width=302 ] the profiles of the disorder - averaged exciton dos are presented in fig . [ fig : rho ] . except for the well - known band - center dyson singularity@xcite in the case of nn coupling and a similar but nonsingular feature for the lr coupling@xcite the numerical dos is reproduced well within beb cpa . as a confirmation of the analyticity of the theory no unphysical behavior of the dos has been observed . the theory also captures properly the exact symmetry of the spectrum present in any tight - binding system without diagonal disorder and with only nn coupling . the asymmetric dos in the case of lr interactions is satisfactorily approximated already by that of the nnn coupling . furthermore , it can be observed that the upper and lower estimates of the spectrum region mentioned in sec . [ sec : anal ] are not violated in the presented solution . for instance , in the case of nn coupling the spectrum should not extend beyond the interval latexmath:[$|\omega| < 2 \max varying parameters @xmath204 within the interval @xmath206 is given differently in the distinct domains of the parametric space : if @xmath242 it equals to @xmath243 , while in the opposite case it becomes @xmath244 for @xmath245 , or @xmath246 for @xmath247 . the excitonic absorption spectra for nn coupling and several values of disorder parameter @xmath238 calculated with beb cpa ( dashed lines ) or exact diagonalization ( solid lines ) . @xmath225 for all curves . the arrows mark estimates of the spectral center for @xmath248 : @xmath249 , @xmath250 . for @xmath221 the absorption spectra resulting from the low - disorder limit of beb cpa are presented ( dotted lines).,title="fig:",width=302 ] the excitonic absorption spectra for nn coupling and several values of disorder parameter @xmath238 calculated with beb cpa ( dashed lines ) or exact diagonalization ( solid lines ) . @xmath225 for all curves . the arrows mark estimates of the spectral center for @xmath248 : @xmath249 , @xmath250 . for @xmath221 the absorption spectra resulting from the low - disorder limit of beb cpa are presented ( dotted lines).,title="fig:",width=302 ] the disorder - averaged absorption spectra ( normalized by the number of molecules in the chain ) along the two essential polarizations , @xmath251 and @xmath252 , are presented in fig . [ fig : chi ] . the component @xmath253 , corresponding to the direction of preferable orientation of the dipoles , demonstrates behavior typical for disordered @xmath0 aggregates : a sharp resonance acquiring inhomogeneous width upon increasing disorder is located near the bottom of the excitonic band . as for the polarization orthogonal to the preferred orientation of the dipoles , the absorption component @xmath254 is generated solely by orientational disorder . in contrast to the previous case no narrow resonance is observed , but rather a broad spectrum shifted to energies higher with respect to the bare molecular level . the positions of the absorption maxima can be roughly estimated from the zero and first moments of the spectral density as @xmath255 , @xmath256 . using eqs . ( [ moms ] ) for these parameters in the simplest case of nn coupling we end up with @xmath257 $ ] , @xmath258 . these expressions one more time confirm that the centra of the absorption spectra are located at essentially different energies for the two polarizations . @xmath259 @xmath260 @xmath261 @xmath262 to provide additional understanding of the features observed in the absorption spectra , the complete beb - cpa spectral density is plotted in fig . [ fig : gf](a , b ) . for the considered disorder parameter ( @xmath222 ) the component @xmath263 is characterized by a well - pronounced quasiparticle structure with most of the spectral wight located around the disorder - free exciton branch @xmath264 . as for the part @xmath265 , similarly to the already considered case of momentum @xmath98 ( the only directly accessible in the linear optics ) , the function has no resonant behavior in all the phase space . the spectral density transformed to the real - space representation is shown in fig . [ fig : gf](c , d ) . the component @xmath266 is essentially nonzero in the interval @xmath267 , and for fixed exciton energy has oscillatory behavior as a function of intermolecular distance @xmath268 . starting from zero near the bottom of the band the number of nodes monotonically increases as one moves towards the upper band edge . since absorption is obtained from the real - space spectral density after the summation ( [ dachi ] ) , the absence of oscillations explains the strong absorbance for the @xmath269-polarization near the band bottom . as concerns polarization orthogonal to the preferred orientation of the transition dipoles , the function @xmath270 is confined on a few ( @xmath271 ) sites . this implies the absence of the coherent optical response of the molecules resulting to a weak and very broad absorption for this polarization direction . energy dependence of the exciton coherence length for several values of the disorder parameter @xmath238 . the beb - cpa results are presented in the case of both @xmath272 and @xmath273 ( dashed lines ) and ones of the exact diagonalization ( solid lines ) only for @xmath272 . @xmath274 for all curves.,title="fig:",width=321 ] energy dependence of the exciton coherence length for several values of the disorder parameter @xmath238 . the beb - cpa results are presented in the case of both @xmath272 and @xmath273 ( dashed lines ) and ones of the exact diagonalization ( solid lines ) only for @xmath272 . @xmath274 for all curves.,title="fig:",width=321 ] figure [ fig : coh ] illustrates variation of the exciton coherence length across the spectrum . the beb - cpa coherence length has been found extracting the parameters @xmath55 of the large - intersite - separation asymptotics ( [ ncohdef ] ) from eq . ( [ chareq ] ) . owing to the symmetry of the problem the latter splits into the two independent equations @xmath275 , @xmath276 . consequently , each of the two components @xmath277 and @xmath278 is characterized by its own coherence length , to be denoted respectively as @xmath279 and @xmath280 . from the results presented in fig . [ fig : coh ] it follows that @xmath279 strongly depends on energy , having maximum around the band center and decreasing as @xmath24 approaches the edges , whereas @xmath280 is almost constant across the spectrum . beyond the spectral region the parameters @xmath281 are imaginary , and hence , the coherence length is finite , even in the uniform chain . remarkably , upon variation of the parameter @xmath238 the two lengths @xmath279 and @xmath280 evolve differently : the first decreases while the second grows with disorder . regarding the exact diagonalization approach , the coherence length can be extracted directly from envelope of the disorder - averaged real - space spectral density in accordance with eq . ( [ ncohdef ] ) . unlike the case of dos and absorption , much larger statistical error is present owing to the fact that the quantity to be averaged ( i.e. , a tensorial component of the spectral density ) depends on both energy and the real - space variables . here one should keep small the width of the energy integration interval [ @xmath282 of the total bandwidth in our simulations ] to ensure that the additional decoherence due to interference of waves separated by energies smaller than @xmath283 does not exceed the disorder - induced effect . if @xmath284 is the excitation group velocity the last condition can be expressed as @xmath285 . this inequality is not easy to fulfill around the band center [ @xmath279 is large ] and around the band edges [ @xmath284 is small ] . apart from mentioned the numerical procedure suffers from the finite - size quantization taking place as the coherence length approaches the total length of the chain . for the model under consideration we have also verified the accuracy of the low - disorder analytic solutions found in sec . [ sec : weakdis ] . since no diagonal disorder is present the centered moments @xmath286 and @xmath121 of the distribution vanish , and the only parameters which control the strength of disorder are @xmath287 and @xmath288 . to evaluate these constants we have used eq . ( [ abcdef ] ) and retained only the leading terms in the limit of small @xmath238 . in the magnitude of the averaged dipole moment one has to keep also the next - to - the - leading term , that results to the estimate @xmath289 . in fig . [ fig : chi ] ( panels with @xmath221 ) the analytic low - disorder absorption spectra are confronted with those calculated via direct solution of the beb - cpa equations and by exact diagonalization . for polarization perpendicular to the preferred orientation of dipoles the spectral weight of the band - edge singularities ( located at @xmath290 for nn coupling ) is negligible compared to the integrated spectral weight . to approximate such wide - bandwidth absorbance one can use the iterative solution derived in sec . [ sec : iter ] which results to expression @xmath291 . in contrast , the component @xmath253 , with most of the oscillator strength concentrated around the lower band edge , should be calculated using the approach of sec . [ sec : scaling ] . in particular , taking into account that the disorder parameter ( [ disordercombined ] ) is given by @xmath292 , the scalings ( [ etascal ] ) and ( [ nscal ] ) read @xmath293 , @xmath294 , respectively . the linewidth @xmath194 and the exciton coherence length @xmath272 at the absorption maximum extracted from the numerical solution of the beb - cpa equations are plotted against @xmath238 in fig . [ fig : scaling](a , b ) . the presented profiles confirm the fulfillment of the derived scaling laws . moreover , as illustrates fig . [ fig : scaling](c ) , the universal scaling ( [ netascal ] ) is observed in the whole range of @xmath194 and @xmath272 accessible by varying @xmath238 from @xmath295 to @xmath296 , even though the two above dependencies hold only for @xmath238 sufficiently close to zero . dependencies of the absorption linewidth ( a ) and the exciton coherence length at the resonance ( b ) on the disorder parameter @xmath238 . ( c ) the same data is plotted in the coordinates absorption linewidth coherence length " . the parameter @xmath203 is 0.26 ( solid line ) , 0.52 ( dashed line ) , 0.65 ( dotted line ) , 0.78 ( dash - dotted line).,title="fig:",width=198 ] dependencies of the absorption linewidth ( a ) and the exciton coherence length at the resonance ( b ) on the disorder parameter @xmath238 . ( c ) the same data is plotted in the coordinates absorption linewidth coherence length " . the parameter @xmath203 is 0.26 ( solid line ) , 0.52 ( dashed line ) , 0.65 ( dotted line ) , 0.78 ( dash - dotted line).,title="fig:",width=198 ] dependencies of the absorption linewidth ( a ) and the exciton coherence length at the resonance ( b ) on the disorder parameter @xmath238 . ( c ) the same data is plotted in the coordinates absorption linewidth coherence length " . the parameter @xmath203 is 0.26 ( solid line ) , 0.52 ( dashed line ) , 0.65 ( dotted line ) , 0.78 ( dash - dotted line).,title="fig:",width=198 ] in a typical experiment one deals not with a single molecular chain but rather with a macroscopic ensemble of such objects . the treatment reported here is applicable only if the coupling between molecules of distinct aggregates is weaker than the disorder - induced broadening in a single chain . the broadening mechanisms not related to the considered types of disorder can be accounted for via a phenomenological on - site dephasing rate @xmath297 , e.g. , due to exciton - phonon scattering . the results of the beb - cpa theory can be straightforwardly extended to include this case by adding imaginary contribution @xmath298 to the complex energy variable @xmath51 in all relevant functions . in most of the samples accessible in experiment , such as frozen solutions containing molecular aggregates , because of the randomness in the chain s orientations one is not able to measure the individual spatial components of the microscopic ( single - chain ) polarizability tensor . nevertheless , the available types of surface - deposited films@xcite with strictly in - plane alignment or recently synthesized bulk materials with smectic - like@xcite alignment of aggregates still allow for independent observation of certain spatial components of the single - chain polarizability . with a polarization - resolved spectroscopic study of such systems@xcite one can probe the orientational disorder within a single aggregate . in particular , a material obtained by stacking the chains described in fig . [ fig2 ] preserving perfectly parallel alignment of the axes and planes represents a biaxial medium . the optical axes are given by the directions @xmath251 , @xmath252 , with the corresponding polarizabilities proportional to @xmath299 and @xmath300 , and the one orthogonal to them , along which the optical response vanishes . in summary , using beb cpa we have investigated the linear optical properties of a disordered molecular aggregate with random single - molecule excitation energies and transition dipoles . the advantages of the used self - consistent approximation are its analyticity , exactness in the low - disorder limit and the capability to provide not only the dos and absorption spectra but also the coherence length . to the best of our knowledge , the last quantity has never been addressed in excitonic systems with the coherent - potential approximation . whereas the dos is on - site characteristics while the absorption spectrum is related to the spectral density at momentum @xmath98 , the calculation of coherence length requires knowledge of the spectral density at a generic @xmath50 . that is , the last quantity carries essentially new information on the single - particle properties of the system with respect to that contained in the first two . the interesting feature of the considered problem is the tensorial structure of equations brought about by orientational disorder in the transition dipoles . such type of disorder leads to redistribution of absorbance between different polarization components with quite distinct behavior of the corresponding coherence lengths . we thank professor f. bassani for helpful discussions and encouragement and dr d.m . basko for suggestions on improving the manuscript . financial support from u.v.o .- r.o.s.t.e . unesco , the european union rtn - hytec and miur ( prin-2001 `` advanced hybrid heterostructures '' ) is gratefully acknowledged . in the case of box probability distribution the configurational averaging in the self - consistency equation ( [ cpa1matr ] ) can be performed analytically leading to @xmath301 where function @xmath302 is to be found from @xmath303 from eq . ( [ sigmasimple ] ) the projected components of the self - energy can be expressed as @xmath304 performing momentum integration in eq . ( [ cpa2matr ] ) one relates the nontrivial components of the disorder - averaged on - site gt to the coherent - potential gt as @xmath305 the function @xmath306 entering these formulas is given by @xmath307 and @xmath308^{-1/2 } \right . \\ { \displaystyle}\left . + \left [ \left(\sqrt{8x } - \sqrt{4 + 9x } \right)^2 - 2x \right]^{-1/2 } \right\ } , \end{array}\ ] ] for the cases of nn and nnn coupling , respectively . for every disorder parameter @xmath238 and tilt @xmath203 the equation presented in this appendix can be solved numerically for @xmath178 and @xmath309 . the example of such solution is shown in fig . [ fig : self ] . in this appendix we study the single - particle properties of the system described in sec . [ sec : simplemodel ] when the angle @xmath203 formed by the plane of the transition dipoles with the chain s direction is such that @xmath310 ( i.e. , @xmath311 ) . the disorder - free situation for this value of @xmath203 can be considered as an intermediate case between that of @xmath0 and @xmath158 aggregate , because the transfer energy ( [ transferterm ] ) generated by the dipolar coupling ( [ ddkernel ] ) vanishes . neglecting higher multipole contributions to the transfer energy as well as all other broadening mechanisms which go beyond the present model , let us concentrate on the effect of orientational disorder to produce a finite excitonic bandwidth . [ considering instead the diagonal disorder alone results to a trivial physics : the off - diagonal part of the hamiltonian is always zero and the molecules remain uncoupled . ] our aim is to find using beb cpa the asymptotic form of the gt around @xmath312 , where the dos and certain components of polarizability tensor have a singularity . [ such singularity due to specific arrangement of coupling and disorder should not to be confused with the dyson singularity@xcite , not accessible within beb cpa ] . for simplicity we shall consider only the case of nn coupling . ( a ) exciton dos and ( b ) , ( c ) absorption spectra in the case @xmath310 . the beb - cpa results ( dashed lines ) are confronted with ones of the exact diagonalization ( solid lines ) and the derived asymptotic expressions ( dotted lines ) . the disorder parameter @xmath238 is equal to @xmath313 for all curves.,width=302 ] from eq . ( [ modelgnn ] ) with @xmath310 it follows that @xmath314 . according to eqs . ( [ models ] ) the corresponding component of the beb - cpa self - energy is zero , @xmath315 . on the other hand @xmath316 is nonzero at any finite @xmath51 . taking this into account it is easy to conclude that the auxiliary function defined by formula ( [ modelq ] ) behaves as @xmath317 upon @xmath318 . applying the same limit to eqs . ( [ modelsc22 ] ) we get @xmath319 @xmath320 combining formula ( [ sigmaas ] ) with eq . ( [ modelsc11 ] ) we obtain the asymptotics of the remaining unknown functions at @xmath321 : @xmath322 it turns out that , irrespectively on the strength @xmath238 of orientational disorder , the component @xmath171 of the disorder - averaged on - site gt is singular as @xmath318 , i.e. , at the energy of the bare molecular transition . strictly speaking , upon introducing some degree of diagonal disorder or a finite inelastic dephasing rate @xmath297 such singularity would be suppressed . nevertheless , one can still apply the derived asymptotic formulas if the absolute value of @xmath51 is much larger than @xmath297 but smaller than the total bandwidth generated by orientational disorder . the exciton dos , related to the on - site gt found above by formula ( [ orientdos ] ) , shows an analogous singularity around @xmath323 : @xmath324 as concerns the absorption spectra , for the two essential polarization directions these are characterized by completely distinct behavior : @xmath325 the dos and the absorption profiles calculated via numerical solution of the beb - cpa equations and ones of exact diagonalization are confronted with the derived asymptotic formulas in fig . [ fig : singul ] . all approaches agree well close to the band center . noticeably , unlike the usual case @xmath326 , the absorption component @xmath327 is symmetric with respect to the central energy @xmath328 ( still only the nn coupling is assumed ) . regarding the exciton coherence length , since @xmath329 it follows that @xmath330 for every @xmath51 . in other words , for polarization along the preferred orientation of the dipoles the disorder - averaged spectral density depends on the intermolecular separation as @xmath331 , i.e. , we encounter completely incoherent optical response of the molecules . for the orthogonal polarization in the vicinity of @xmath332 we have @xmath333 @xmath334 as a result the coherence length @xmath280 is finite within the band but diverges at its center . in the vicinity of @xmath328 the corresponding real - space spectral density behaves as @xmath335 , i.e. , coincides with the center - of - the - band spectral density of a finite - bandwidth uniform chain . f. domnguez - adame , phys . b * 51 * , 12801 ( 1995 ) ; f. domnguez - adame , b. mndez , a. snchez , e. maci , _ ibid . _ * 49 * , 3839 ( 1994 ) ; f. domnguez - adame , _ ibid . _ * 51 * , 12801 ( 1995 ) ; v.a . malyshev , a. rodrguez , f. domnguez - adame , _ ibid . _ * 60 * , 14140 ( 1999 ) d.a . papaconstantopoulos , a. gonis , p.m. laufer , phys . b * 40 * , 12196 ( 1989 ) ; k. koepernik , b. velick@xmath336 , r. hayn , h. eschrig , _ ibid . _ * 55 * , 5717 ( 1997 ) ; p.e.a . turchi , d. mayou , j.p . julien , _ ibid . _ * 56 * , 1726 ( 1997 ) ; j.p . julien , p.e.a . turchi , d. mayou , _ ibid . _ * 64 * , 195119 ( 2001 )

signs of hierarchy are prevalent in a wide range of systems in nature and society . one of the key problems is quantifying the importance of hierarchical organisation in the structure of the network representing the interactions or connections between the fundamental units of the studied system . although a number of notable methods are already available , their vast majority is treating all directed acyclic graphs as already maximally hierarchical . here we propose a hierarchy measure based on random walks on the network . the novelty of our approach is that directed trees corresponding to multi level pyramidal structures obtain higher hierarchy scores compared to directed chains and directed stars . furthermore , in the thermodynamic limit the hierarchy measure of regular trees is converging to a well defined limit depending only on the branching number . when applied to real networks , our method is computationally very effective , as the result can be evaluated with arbitrary precision by subsequent multiplications of the transition matrix describing the random walk process . in addition , the tests on real world networks provided very intuitive results , e.g. , the trophic levels obtained from our approach on a food web were highly consistent with former results from ecology .

1508.07732

hierarchical organisation is an ubiquitous feature of a large variety of systems studied in natural- and social sciences . examples of empirical studies on hierarchy are including the transcriptional regulatory network of escherichia coli @xcite , the dominant - subordinate hierarchy among crayfish @xcite , the leader - follower network of pigeon flocks @xcite , the rhesus macaque kingdoms @xcite , neural networks @xcite and technological networks @xcite , social interactions @xcite , urban planning @xcite , ecological systems @xcite , and evolution @xcite . naturally , hierarchy is a very relevant concept also in network theory @xcite . the network approach has become an ubiquitous tool for analysing complex systems ranging from the interactions within cells through transportation systems , the internet and other technological networks to economic networks , collaboration networks and the society @xcite . grasping the signs of hierarchy in networks is a non - trivial task with a number of possible different approaches . on the one hand , we may try the statistical inference of an underlying hierarchy based on the observed network structure , as suggested in ref.@xcite . on the other hand , the introduction of a hierarchy measure is also a natural idea @xcite . in general , a hierarchy measure , can be viewed as a function on the domain of graphs , @xmath0 , mapping a graph @xmath1 into a real number , @xmath2 . the value of the measure is actually @xmath3 $ ] or @xmath4 $ ] in most cases , with high values corresponding to hierarchical structures and low values indicating the absence of hierarchy in the examined network . one of the first methods was proposed by d. krackhardt , motivated by organisational hierarchy , defining the hierarchy measure simply as the number of ordered pairs divided by the number of connected pairs @xcite . in the approach introduced by a. trusina et al . , the position of the nodes in the hierarchy is assumed to be given by the degree , and the hierarchy measure is given by the fraction of directed shortest paths going strictly upwards in the hierarchy @xcite . another way for quantifying the possible asymmetry between nodes is to assume some sort of flow on the links , and examine whether the global map of flows in the system is revealing a kind of overall directionality or not . probably the simplest approach in this framework is to define the fraction of links not participating in any cycle as the measure of the hierarchy , as suggested by j. luo and c. l. magee @xcite . a further important property of a hierarchical system is that reaching the rest of the network should be relatively easy for the nodes high in the hierarchy , and more difficult for the nodes at the bottom of the hierarchy , as pointed out by e. mones et al . in ref.@xcite . the hierarchy measure based on this aspect is given by the global reaching centrality , characterising the inhomogeneity of the fraction of reachable nodes in at most @xmath5-steps in the network @xcite . a more elaborate quantification of hierarchy was proposed by b. corominas - murta et al . in ref.@xcite with the introduction of treeness , feedforwardness and orderability , projecting the studied network onto a point in a 3 dimensional space , where each dimension is aimed to capture a different aspect of hierarchy . treeness , @xmath6 , is measuring how ambiguous are the chain of commands in the network , while feedforwardness , @xmath7 is related to the size and position of the strongly connected components in the network . finally , the orderability , @xmath8 is simply the fraction of nodes not taking part in any directed cycles , i.e. , it is analogous to the hierarchy measure introduced by j. luo and c. l. magee . the 3d scatter plots of @xmath6 , @xmath7 and @xmath8 provided very interesting results , revealing different clusters of hierarchical networks @xcite . a more detailed description and comparison between the mentioned methods is given in the supplementary information s1 . although the methods listed above allow the examination of the hierarchical organisation from different perspectives , a noteworthy common aspect of these approaches that they all treat acyclic networks as already maximally hierarchical , independent of the further details of the graph structure . ( the global reaching centrality given in ref.@xcite is an exception , considering the star configuration as the most hierarchical ) . here we argue that different acyclic networks are not necessarily equally hierarchical . the general intuition of a hierarchy is usually corresponding to a multi level pyramidal structure , with levels becoming wider and wider as we descend from the root towards the bottom . on the one hand this way the top nodes in the hierarchy can reach most of the network in a very effective way , i.e. , via paths of average length scaling as @xmath9 , where @xmath10 denotes the number of nodes . on the other hand , in this structure all nodes can have a treatable number of direct subordinates . in contrast , if we consider a directed chain , all the levels are of size one , and this is leading to a large average distance scaling as @xmath10 . the other extreme limiting case of acyclic networks is given by the directed star configuration , where all the nodes have a single incoming link from a central hub , and no further out - links . in this case the hierarchy is consisting of only two levels , and the supposed leader in the network has to cope with a number of direct descendants scaling as @xmath10 . based on that , introducing a hierarchy measure preferring trees to chains and stars would be a substantial step towards achieving a more intuitive approach for evaluating the importance of hierarchy in a network structure . in this paper we tackle this problem with the help of random walks on the network . random walks provide a fundamental model for stochastic processes in a large variety of systems ranging from physics @xcite , chemistry @xcite and computer science @xcite through biology and ecology @xcite to economics @xcite and psychology @xcite . in the current problem of quantifying the extent of hierarchy in a network structure , random walkers can be used to evaluate the rank of the nodes in the hierarchy . the basic idea is assuming an information flow on the links from nodes high in the hierarchy towards the lower levels , in a similar fashion as in case of a company , where the management is likely to send information and instructions to the employees on a regular basis . given the network structure , the source of information in the system can be traced back by sending random walkers traversing the links in reverse direction from all nodes . in case the density of the random walkers is reaching a steady state , its value at a given node can be interpreted as the probability that the node was the source of information . consequently , high random walker density values indicate a high standing in the hierarchy , whereas low density values are corresponding to bottom nodes . the significance of hierarchical organisation in the network structure can be judged based on the inhomogeneity of the distribution : in a homogeneous distribution we can not pinpoint the source of information , thus , it is corresponding to a non - hierarchical network . in contrast , a very inhomogeneous distribution is indicating a strongly hierarchical structure . the details of the random walk process are the following . since the random walkers are traversing the links backwards , the transition probability for a walker from node @xmath11 to @xmath12 is proportional to the inverse of the in - degree of @xmath11 , i.e. , @xmath13 . another important factor to be taken into account is the limited capacity of the information sources for sending information : in general we can assume that the more out - neighbours a given node has , the less resource it can allocate for managing the communication over a given link . this effect can be taken into account by assuming that @xmath14 is also proportional to the out - degree of @xmath12 , i.e. , @xmath15.combining the above factors together is resulting in @xmath16 for the transition probability of the random walkers from node @xmath11 to @xmath12 . ( in case @xmath12 is not an in - neighbour of @xmath11 the transition probability @xmath14 is zero by definition ) . we note that due to the second factor on the right hand side of ( [ eq : trans_k ] ) , the probability for staying at the same node can be non - zero in general , given by @xmath17 . for weighted networks ( [ eq : trans_k ] ) can be naturally generalised to @xmath18 where @xmath19 denotes the weight of the link from @xmath12 to @xmath11 . in case of acyclic networks , all random walkers eventually converge into nodes with no incoming links , ( i.e. , the `` sources '' in the network ) . in order to avoid judging the importance of hierarchical organisation in the system solely based on these `` sources '' , we inject new random walkers into the network at every time step . the update rules are the following : * we insert @xmath20 random walkers into the system , increasing the random walker density on every node by @xmath21 , thus , the random walker density at node @xmath12 given by @xmath22 is changing as @xmath23 * we let all random walkers in the system proceed on step , governed by the transition probabilities given in ( [ eq : trans_k ] ) . by introducing a transition matrix @xmath24 with matrix elements @xmath25 , the density of random walkers on node @xmath12 after the transition can be expressed as @xmath26 * the total sum of random walkers has to be normalised , i.e. , we require @xmath27 . since the sum of new random walkers added to the system was @xmath20 , we have to simply divide the density of random walkers by @xmath28 in order to fulfil the normalisation condition , @xmath29 the above normalisation of the random walker density to unity after each iteration is equivalent to using `` decaying '' random walkers , having a weight decreasing by a factor of @xmath30 in each step . let us denote the characteristic distance under which the weight of a random walker is decreased to @xmath31 by @xmath32 , fulfilling @xmath33 based on that , @xmath20 can be also expressed as @xmath34 although @xmath32 , ( or equivalently , @xmath20 ) is a parameter of the method at the current stage , later on in the results we shall find a natural condition for fixing @xmath32 at an optimal value in general . our main object of interest is the stationary distribution of the random walkers in the network . by writing this distribution in a vector form of @xmath35 , where the @xmath12-th component of the vector , @xmath36 , is corresponding to the random walker density on node @xmath12 , we can derive a simple equation based on the update rules . adding @xmath21 new random walkers at each node is simply incrementing each vector component by @xmath21 , while the transition to the next site by the random walkers corresponds to multiplying by the transition matrix @xmath24 . finally , the normalisation of the random walker density simply multiplies each vector component by @xmath37 . based on the above the stationary distribution fulfils @xmath38=e^{-1/\lambda}\left[\mathbf{t}\left(\mathbf{p}^{\rm stat}+\frac{e^{1/\lambda}-1}{n}\mathbf{1}\right)\right ] , \label{eq : p_stac_alap}\ ] ] where @xmath39 is corresponding to a vector of size @xmath10 with all elements equal to 1 . by expressing @xmath35 we obtain @xmath40^{-1}\mathbf{t}\mathbf{1}=\frac{e^{1/\lambda}-1}{n}\left[e^{1/\lambda}\mathbf{i}-\mathbf{t}\right]^{-1}\mathbf{t}\mathbf{1 } , \label{eq : p_stac}\ ] ] where @xmath41 is denoting the identity matrix . since @xmath24 is a left stochastic matrix , the absolute value of its largest eigenvalue is 1 . consequently , the absolute value of the eigenvalues of @xmath42 are smaller than 1 , and therefore , ( [ eq : p_stac ] ) can also be written as @xmath43 this formula is very intuitive , showing explicitly that the stationary distribution of random walkers at a given node is given by the sum of the probabilities of all the paths ending on the node , where the contributions from longer paths are suppressed exponentially as a function of the path length . based on ( [ eq : p_stac]-[eq : q ] ) , @xmath35 can be computed very efficiently . if the size of the network is moderate , we can use ( [ eq : p_stac ] ) for obtaining exact results . however , if matrix inversion is becoming computationally expensive , a very good approximation of @xmath35 can be calculated according to ( [ eq : q ] ) . i.e. , by carrying out the summation up to a certain finite limit @xmath44 , the obtained result is converging to the exact @xmath35 exponentially fast . our hierarchy measure is based on the inhomogeneity of the stationary distribution of the random walkers . there are several different possibilities for quantifying the inhomogeneity of a probability distribution in general , here we choose the relative standard deviation , ( also called as the coefficient of variation ) . thus , the random walk hierarchy measure is defined as @xmath45 where @xmath46 and @xmath47 denote the mean and the standard deviation of @xmath36 respectively . since @xmath48 , the mean is given simply by @xmath49 , and our hierarchy measure can be expressed as @xmath50 for demonstrating the sensitivity of our measure to the topology also in case of acyclic networks , first we evaluate @xmath51 for chains , regular trees with a constant branching number @xmath52 , and stars . according to calculations detailed in methods , the corresponding hierarchy values can be expressed as @xmath53 where @xmath10 is the number of nodes in the networks , @xmath54 denotes the levels in case of the tree ( starting from @xmath55 at the root and ending with @xmath56 at the leafs ) , and @xmath57 is corresponding to the stationary distribution of the walkers on level @xmath54 in ( [ eq : rwh_tree_final ] ) , which can be obtained from a simple recursion written as @xmath58 in fig.[fig : chain_tree_star ] . we compare the hierarchy measures given in ( [ eq : rwh_chain_final]-[eq : rwh_star_final ] ) at @xmath59 ( fig.[fig : chain_tree_star]a ) and at @xmath60 ( fig.[fig : chain_tree_star]b ) . our construction algorithm for the trees with a branching number @xmath52 was to start by adding @xmath52 links to the root , then move to the second level and subsequently add @xmath52 links to every node on this level , and so on , move to the next level only when the given level was completed . whenever the number of nodes in the tree has reached @xmath10 , the algorithm terminates , and naturally , the resulting tree is not completely regular in most of the cases . nevertheless , the overall structure of the trees obtained in this way is getting closer and closer to regular trees as @xmath10 is increasing . as a function of @xmath10 for a chain ( black ) , a regular tree with branching number @xmath61 ( red ) , a regular tree with @xmath62 ( green ) , a regular tree with @xmath63 ( blue ) and a star ( orange ) at @xmath59 . b ) the same plot as in a ) when @xmath32 is set to @xmath60.,scaledwidth=63.0% ] according to fig.[fig : chain_tree_star ] . the @xmath51 for the chain and the star configurations has a peak at very small system sizes , and shows a decreasing tendency for growing @xmath10 . in contrast , for regular trees @xmath51 seems more or less converging to a finite value . thus , above a certain @xmath10 it is the structure of the tree , ( encoded in the branching number ) , what determines the hierarchy measure , not the size of the tree . this is indicating that @xmath51 is behaving similarly to intensive quantities in physics in some aspects . the `` intensive '' property of the hierarchy measure is analysed in more details in the supplementary information s2 , here we note that if we take a pair of graphs @xmath64 and @xmath65 which are not connected to each other , then @xmath51 for the union of the graphs is equal to the weighted quadratic mean of the @xmath51 values calculated for the graphs separately , @xmath66 ^ 2+n_2\left[h_{{\mathcal g}_2}\right]^2}{n_1+n_2}}.\ ] ] thus , in the special case of a pair of isomorphic graphs @xmath67 . we continue with the examination of the behaviour of @xmath51 in the thermodynamic limit . according to calculations detailed in methods , when the system size is diverging , @xmath68 , the hierarchies given in ( [ eq : rwh_chain_final]-[eq : rwh_star_final ] ) take the form of @xmath69 thus , the hierarchy measure is vanishing for a chain and a star in the thermodynamic limit . in contrast , @xmath70 is converging to a well defined finite limit value or @xmath71 when @xmath32 is smaller than a @xmath52 dependent critical value , and is diverging otherwise . in the methods we show that the critical @xmath32 value is given by @xmath72^{-1}. \label{eq : lambda_c}\ ] ] for infinitely large regular trees as a function of the branching number @xmath52 and the parameter @xmath32 . the formula for @xmath51 is given in ( [ eq : rwh_tree_inf ] ) . at @xmath73 we recover the infinitely large chain , while the infinitely large star is corresponding to the @xmath74 limit . the dashed line is showing the maximum place of @xmath51.,scaledwidth=70.0% ] the behaviour of the limiting @xmath70 given in ( [ eq : rwh_tree_inf ] ) is shown in a 2d plot in fig.[fig : rwh_inf_trees ] . as a function of @xmath52 and @xmath32 . at @xmath73 the tree becomes equivalent to an infinitely large chain , and according to ( [ eq : chain_inf ] ) the @xmath51 becomes zero . the 2d surface displayed in fig.[fig : rwh_inf_trees ] . is consistent with this result , as it starts from @xmath75 at @xmath73 for all @xmath32 values . similarly , the @xmath51 for an infinitely large star is also zero according to ( [ eq : rwh_star_inf ] ) . the surface shown in fig.[fig : rwh_inf_trees ] . is consistent with this result as well , as we can see a decreasing tendency in @xmath51 as a function of @xmath52 in the large @xmath52 regime . in the range of intermediate branching numbers we can observe an @xmath32 dependent maximum in @xmath51 . this behaviour is examined in more details in the supplementary information s3 . based on the behaviour of @xmath51 in the thermodynamic limit , we can also fix the @xmath32 parameter at an optimal value in general as follows . since @xmath32 is corresponding to the characteristic path length a random walker can traverse before `` decaying '' , on the one hand we would like to choose a @xmath32 as high as possible . i.e. , if @xmath32 is small , the random walkers can explore only within a very limited range from their origin , thus , the information we can retrieve via the random walkers is also very local . however , due to its self similar nature , hierarchical organisation can manifest on all length scales , therefore , we need random walkers travelling longer distances in order to be able to tell apart hierarchical and non - hierarchical networks . on the other hand , if @xmath32 is too large , we may run into diverging hierarchy values according to ( [ eq : rwh_tree_inf ] ) , which needs to be avoided in case of a well behaving hierarchy measure . therefore , we fix @xmath32 at a value as high as possible where a diverging @xmath51 is avoided for sure even in case of infinitely large regular trees . according to ( [ eq : lambda_c ] ) , the minimum of @xmath76 can be found at @xmath77 , where @xmath78 . since the path length traversed by a random walker is increasing by unity under every iteration , it is also natural to set @xmath32 to an integer value . based on the above , the optimal setting for @xmath32 is given by @xmath60 . in the rest of the paper we are assuming that @xmath32 is set to this optimal value , and thereby consider our approach a parameter free method for measuring the amount of hierarchical organisation in the structure of networks . since real world hierarchies are usually not as highly ordered as a regular tree with a constant branching number , we extended our comparison studies of acyclic graphs also to general directed trees . by applying a simple algorithm detailed in methods , we generated a large family of trees with branching numbers varying around a given average branching number @xmath79 according to a shifted poisson distribution . in fig.[fig : poisson_trees ] . we show the average of the random walk hierarchy measure , @xmath80 as a function of @xmath79 , calculated based on 100 realisations of trees consisting of @xmath81 nodes . according to the curve , the maximum of @xmath80 is at an intermediate average branching number , where the structure of the network is really tree like . i.e. , for low average branching numbers , ( where the structure is basically a chain ) , and also for very large branching numbers comparable to the system size , ( where the structure is close to a star ) , the obtained @xmath80 values are considerably lower . as a function of the average branching number @xmath79 for general trees of @xmath81 nodes , averaged over @xmath82 instances . when @xmath79 is close to one , the tree is basically a chain , whereas at very large branching number , its structure is close to a star.,scaledwidth=60.0% ] here we apply our method for analysing the hierarchy of the st . marks food web @xcite , representing a part of the ecosystem of goose creek bay , st . marks national wildlife refuge , florida , usa . the nodes of the network are corresponding to living compartments , ( group of species ) based on probable diet and life history characteristics . thus , compartments range from single species ( e.g. , pinfish ) through a couple of species ( e.g. , gulf flounder and needlefish ) to large groups of taxa , ( e.g. , bacterioplankton ) . the links between the nodes represent the feeding pathways , pointing from consumers to their food sources , where the link weights are corresponding to the fractions of the consumer s diet . ) , and the hierarchy levels are corresponding to groups of nodes for which the standard deviation of @xmath83 is at most @xmath84 , where @xmath47 denotes the standard deviation of @xmath36 over the whole network . the colour coding of the nodes reflects their effective trophic level published in @xcite . b ) listing of the abbreviations used in a).,scaledwidth=95.0% ] the static distribution of the random walkers on the network defined above can be calculated using ( [ eq : p_stac ] ) . however , @xmath83 is defining only a ranking between the nodes and does not provide the hierarchy levels in the first place . therefore , we sampled and aggregated nodes into levels so that in each level , the standard deviation of @xmath83 is lower than a pre - defined fraction of the standard deviation in the whole network . ( this type of procedure for obtaining the hierarchy levels was established in @xcite ) . in fig.[fig : st_marks ] . we show the resulting hierarchy between the compartments when the standard deviation of @xmath36 within the levels is at most @xmath85 . the hierarchy levels are consistent with the common sense about food webs as e.g. , benthic algae is on the lowest level , herbivorous ducks are somewhere in the middle , and raptors ( such as e.g. , the bald eagle ) are on the top of the hierarchy . the colour coding of the nodes is showing the effective trophic level of the compartments given in @xcite , ranging between 1.0 and 4.32 . apparently , the position of the nodes in the hierarchy and their colour are coherent in most of the cases , e.g. , the root has the highest effective trophic level , and the nodes with the lowest trophic level are at the bottom of the hierarchy . however , a small number of discrepancies can be also observed , ( e.g. , as in case of gulf flounder & needlefish ) , signing that the effective trophic levels and the random walk based hierarchy are catching slightly different aspects of the studied food web . finally , the spearman s rank correlation coefficient between the ranking of the compartments according to @xmath36 and the ranking according to the effective trophic levels is @xmath86 . in contrast , the spearman s rank correlation coefficient between the effective trophic levels and the hierarchy obtained after applying a degree preserving link randomisation to the network is only @xmath87 . based on the above , our hierarchy is highly consistent with former results from ecology . we also calculated the @xmath51 given in ( [ eq : rwh_final ] ) for numerous different systems ranging from metabolic and regulatory networks through citation , trust and language networks to the internet and the www . ( a detailed description of the networks is given in the supplementary information s4 ) . in fig.[fig : scatter ] . we show the obtained hierarchy values as a function of the network size , @xmath10 . .,scaledwidth=70.0% ] according to the figure food webs , electric circuits and regulatory networks provide the largest @xmath51 values , and in contrast , the informal networks of acquaintances in different organisations seem the least hierarchical . in the mean time , the www , the internet , the citation- , metabolic- , trust- and language networks appear to be moderately hierarchical . however , under certain circumstances we may obtain a moderate hierarchy measure even in a random graph . e.g. , the structure of the giant component in the erds rnyi graph @xcite is more or less tree - like if we are close to the percolation threshold , and tree - like structures are usually considered highly hierarchical . accordingly , in order to make a fair judgement on the importance of hierarchy in the topology of a real network , we need to compare the measured @xmath51 to the result we expect in a suitably chosen random network ensemble , modelling the structure of the given network under the assumption of random connections . in order to take into account of the degree distribution of the studied networks , we use the configuration model for evaluating the expected value of @xmath51 in the random network ensemble . a sample from this ensemble can be obtained by simply link randomising the given real network , keeping the degree of the nodes fixed under the random rewiring of the connections . the difference between @xmath51 obtained for the real networks and the expected value of @xmath51 in their random counterparts can be measured in terms of the @xmath88-score , defined as @xmath89 where @xmath80 and @xmath90 denote the expected value- and the standard deviation of @xmath51 in the random ensemble , respectively . thus , we basically scale the difference between the real @xmath51 and the average of @xmath51 over the random ensemble by the standard deviation of @xmath51 in the random ensemble . in fig.[fig : z_score ] . we show the @xmath88-scores corresponding to the @xmath51 values displayed in fig.[fig : scatter ] . according to the results , the citation networks and the network between the web pages of the nd.edu domain achieve outstandingly high @xmath88-scores . furthermore , all of the food webs , the internet networks and also the rest of the www networks obtain considerably large positive @xmath88-scores . this means that the structure of these networks is far more hierarchical compared to a random network with the same degree distribution . in contrast , all of the regulatory- and metabolic networks have negative @xmath88-scores , ( with rather large absolute values in the latter case ) . thus , these networks are less hierarchical compared to what we would expect on a random base . -score of @xmath51 for the networks shown in fig.[fig : scatter ] . the @xmath88-score , given by @xmath91 is plotted as a function of the system size @xmath10 . the random null - model for evaluating @xmath80 and @xmath90 is corresponding to the configuration model.,scaledwidth=67.0% ] finally , in case of the electric- , organisational- , and language networks we see a mixed picture , where both positive and negative @xmath88-scores occur . most of the organisational networks have positive @xmath88-scores , reaching to a quite high value in case consulting network , while in parallel we obtain a negative @xmath88-score for the enron network . the word adjacency network for the french , spanish and japanese languages have negative @xmath88-scores , opposed to a clearly positive @xmath88-score in case of the english language . a more detailed analysis of these results is provided in the supplementary information s4 . measuring the significance of hierarchical organisation in the structure of a complex network is a non - trivial problem with a number of different options available . here we have proposed a novel method based on random walks on the network . the basic idea behind our approach is that if nodes were sending instructions or information over the links to their subordinates , then the sources of the information could be traced using random walkers traversing the links backwards . the update rules of the dynamics are chosen in a way to make the density of the walkers on the nodes converge to a stationary distribution exponentially fast with the number of iterations . the position of the nodes in the hierarchy is determined by this distribution , with high random walker densities corresponding to top nodes , and low values of the distribution signalling bottom nodes . the overall measure of the hierarchy is given by the inhomogeneity of the stationary distribution . the calculation of the hierarchy measure can be carried out based on repeated multiplications of an @xmath10 by @xmath10 transition matrix , making the method computationally very efficient and opens up the possibility for gpu based parallelisation . the other main advantage of our approach is that it can differentiate between directed acyclic graphs of distinct nature , opposed to most other methods treating all directed acyclic networks as already maximally hierarchical . the random walk hierarchy measure provides higher scores for trees showing a multilevel pyramidal structure compared to chains and stars . this is consistent with a general intuitive picture about hierarchies : the multilevel pyramidal structure enables the leaders in the tip of the hierarchy to reach the rest of the system via relatively short paths , and also avoids the `` overloading '' of any nodes with a too large number of direct subordinates . in contrast , the distance between the top and the bottom becomes very large in a chain , while the number of direct subordinates for the central node is diverging with the system size in case of a star . a further interesting property of our measure is that it is behaving similarly to intensive quantities in physics . i.e. , for regular trees with a constant branching number the hierarchy measure is converging to a well defined value in the thermodynamic limit . thus , above a certain scale it is the structure , ( encoded in the branching number ) , what determines the hierarchy , not the size of the network . moreover , our tests on real world networks provided rather encouraging results . on the one hand , the detailed analysis of the st . marks food web resulted in hierarchy levels that are highly consistent with former results from ecology on the effective trophic levels in the system . on the other hand , the large scale analysis of numerous further real networks revealed that the value of the hierarchy measure on its own does not always provide a fair characterisation of the importance of hierarchy in the structure of the studied system . according to our results , in some cases a relatively low @xmath51 value can be accompanied by an outstandingly high @xmath88-score , when we compare the actual @xmath51 to the expected value of @xmath51 in a randomly rewired network with the same degree distribution . this leads to the conclusion that the basic network characteristics such as the link density , degree distribution , etc . can inflict some constrains on the possible range of @xmath51 and also on @xmath80 in the corresponding random network ensemble . however , the further analysis of these effects is out of the scope of the present paper and is providing interesting directions for further research . first we note that @xmath75 when the network is consisting of only a single directed cycle . since all the nodes are equivalent in this case , @xmath92 for all @xmath12 , thus , @xmath93 . now let us examine how does @xmath51 change if we move from a cycle to a chain by cutting a single link . since all the nodes have a unit in - degree except the first node , and all the nodes have a unit out - degree except the last node , the transition probability from level @xmath54 to level @xmath94 is unity . therefore , the stationary distribution on the last node is zero , @xmath95 , ( since the total amount of injected random walkers exit immediately , and there is no inflow of walkers from outside ) . based on ( [ eq : p_stac_alap ] ) , the stationary distribution of the walkers on the intermediate levels fulfils @xmath96 while in case of the first node the injected random walkers can not exit the node , resulting in @xmath97=e^{-1/\lambda}\left[p_2^{\rm stat}+p_1^{\rm stat}+2\frac{e^{1/\lambda}-1}{n}\right ] % \nonumber \\ % & & e^{-1/\lambda}\left[p_2^{\rm stat}+p_1^{\rm stat}+2\frac{e^{1/\lambda}-1}{n}\right ] \label{eq : chain_1}\ ] ] by solving ( [ eq : chain]-[eq : chain_1 ] ) we gain @xmath98=\frac{1-e^{-\frac{n - l}{\lambda}}}{n } \;\;\ ; ( 2\leq l\leq n ) , \\ p_1^{\rm stat}&=&\frac{p_2^{\rm stat}}{f}+\frac{2}{n}=\frac{p_2^{\rm stat}}{e^{1/\lambda}-1}+\frac{2}{n}.\end{aligned}\ ] ] by substituting into ( [ eq : rwh_final ] ) the result simplifies to ( [ eq : rwh_chain_final ] ) . the random walk hierarchy measure for general trees with varying branching number can not be given in a general formula , nevertheless it can be calculated exactly for any particular finite tree based on ( [ eq : p_stac ] ) and ( [ eq : rwh_final ] ) . however , in case of a regular tree with branching number @xmath52 , a simple recursion can be given for the stationary distribution of the random walkers , as the transition probability from any node to its `` leader '' in the level above is simply @xmath99 . the random walkers can not exit from the root , which we label as level @xmath55 , and there is an inflow of walkers from the second level , resulting in @xmath100= \nonumber \\ & = & e^{-1/\lambda}\left[p_2^{\rm stat}+p_1^{\rm stat}+2\frac{e^{1/\lambda}-1}{n}\right ] , \label{eq : p_tree_1}\end{aligned}\ ] ] which is exactly the same as in case of the first node in the chain , given in ( [ eq : chain_1 ] ) . for the intermediate levels , we have an inflow of walkers from the level below , and also a term corresponding to the probability of the walkers staying at the given level instead of moving to the level above , yielding altogether @xmath101= \nonumber \\ & & e^{-1/\lambda}\left[p_{l+1}^{\rm stat}+\frac{e^{1/\lambda}-1}{n}+\left(p_l^{\rm stat}+\frac{e^{1/\lambda}-1}{n}\right)\left(1-\frac{1}{b}\right)\right ] \label{eq : p_tree_l_meth}\end{aligned}\ ] ] finally , on the last level @xmath102 we have no inflow from other nodes , giving @xmath103 which provides an immediate solution for @xmath104 in the form given in ( [ eq : p_tree_lmax ] ) . based on ( [ eq : p_tree_lmax ] ) we can calculate the stationary distribution on the rest of the levels as well , i.e. , by rearranging ( [ eq : p_tree_1 ] ) and ( [ eq : p_tree_l_meth ] ) we gain ( [ eq : p_tree_l ] ) and ( [ eq : p_tree_first ] ) . the hierarchy measure for the star can be evaluated in a similar fashion to that of the chain . in this case @xmath105 peripheral nodes are connected to a central node , from which the random walkers can not exit . thus , the stationary distribution of the random walkers fulfil @xmath106=\nonumber \\ & & e^{-1/\lambda}\left[p_{\rm c}^{\rm stat}+p_{\rm p}^{\rm stat}+2\frac{e^{1/\lambda}-1}{n}\right ] , \label{eq : star_cent}\\ p_{\rm p}^{\rm stat}&=&\frac{1}{1+f}\left(p_{\rm p}^{\rm stat}+\frac{f}{n}\right)\left(1-\frac{1}{n-1}\right)=\nonumber \\ & & e^{-1/\lambda}\left(p_{\rm p}^{\rm stat}+\frac{e^{1/\lambda}-1}{n}\right)\left(1-\frac{1}{n-1}\right ) \label{eq : star_peri}\end{aligned}\ ] ] where @xmath107 denotes the density on the central node , and @xmath108 is equal to the density on the peripheral nodes . from ( [ eq : star_peri ] ) we can express @xmath108 directly as @xmath109 and by substituting ( [ eq : star_p_res ] ) into ( [ eq : star_cent ] ) we arrive to @xmath110 for the central node . according to ( [ eq : rwh_final ] ) , the random walk hierarchy measure can be given in this case as @xmath111 - 1}. \label{eq : rwh_star_alap}\ ] ] by substituting ( [ eq : star_p_res ] ) and ( [ eq : star_c_res ] ) into ( [ eq : rwh_star_alap ] ) , the resulting formula can be simplified to ( [ eq : rwh_star_final ] ) . taking the @xmath68 limit of @xmath112 given in ( [ eq : rwh_chain_final ] ) and of @xmath113 written in ( [ eq : rwh_star_final ] ) is trivial , the results are given in ( [ eq : chain_inf ] ) and in ( [ eq : rwh_star_inf ] ) respectively . in contrast , the evaluating the @xmath68 limit of @xmath70 given in ( [ eq : rwh_tree_final ] ) is more complicated and can be carried out as follows . first we separate the first term from the rest in the sum over the levels in ( [ eq : rwh_tree_final ] ) as @xmath114 in order to evaluate the remaining sum , we express @xmath57 given in ( [ eq : p_tree_l ] ) as @xmath115,\ ] ] where @xmath116 based on the above , @xmath117 ^ 2= \nonumber \\ & & \frac{1}{n}\sum_{l=2}^{l_{\rm max } } \left[a^2 b^{l-1 } + 2 ab b^{l-1}c^{l_{\rm max}-l+1 } + b^2 b^{l-1 } c^{2(l_{\rm max}-l+1)}\right]= \nonumber \\ & & \frac{1}{n}\left\{a^2\left(\frac{b^{l_{\rm max}}-1}{b-1}-1\right ) + 2 ab b^{l_{\rm max}}\left[\frac{\left(\frac{c}{b}\right)^{l_{\rm max}-l+1}}{\frac{c}{b}-1}-1\right]+ \right . \nonumber \\ & & \left . + b^2 b^{l_{\rm max}}\left[\frac{\left(\frac{c^2}{b}\right)^{l_{\rm max}-l+1}}{\frac{c^2}{b}-1}-1\right]\right\}. \label{eq : sum_temp}\end{aligned}\ ] ] by using that @xmath118 if @xmath119 , ( [ eq : sum_temp ] ) can be also written as @xmath120 + \nonumber \\ & & -\frac{2ab}{c / b-1}\left[n(b-1)+1\right]-\frac{b^2}{c^2/b-1}\left[n(b-1)+1\right]\bigg\}+ \nonumber \\ & & + \frac{b-1}{b^{l_{\rm max}}}\left\{\frac{2ab}{c / b-1}c^{l_{\rm max}}+\frac{b^2}{c^2/b-1}c^{2l_{\rm max}}\right\}=\nonumber \\ & & a^2-(2ab+b^2)(b-1)-\frac{2ab}{c / b-1}(b-1)-\frac{b^2}{c^2/b-1}(b-1)+\nonumber \\ & & + \mathcal{o}\left(\frac{1}{n}\right)+(b-1)\left[\frac{2ab}{c / b-1}\left(\frac{c}{b}\right)^{l_{\rm max}}+\frac{b^2}{c^2/b-1}\left(\frac{c^2}{b}\right)^{l_{\rm max}}\right ] \label{eq : sum_temp2}\end{aligned}\ ] ] according to ( [ eq : letter_def ] ) @xmath121 however , the similar inequality of @xmath122 holds if and only @xmath123 or in terms of @xmath32 if and and only @xmath124^{-1}. \label{eq : lam_cond}\ ] ] thus , when ( [ eq : b_cond ] ) , or equivalently ( [ eq : lam_cond ] ) are fulfilled , the last two terms in ( [ eq : sum_temp2 ] ) vanish if @xmath125 . by using ( [ eq : letter_def ] ) and neglecting the @xmath126 terms we obtain @xmath127 now let us examine the @xmath128 term in ( [ eq : sum_sep ] ) . according to ( [ eq : p_tree_1 ] ) we can write @xmath129 where @xmath130 since @xmath131 and @xmath132 , @xmath133 and we obtain that @xmath134 as a consequence , @xmath135 , which is also vanishing when @xmath136 . hence , by substituting ( [ eq : sum_fin ] ) into ( [ eq : sum_sep ] ) and neglecting the @xmath126 terms we arrive to @xmath137 equivalent to the formula given in ( [ eq : rwh_tree_inf ] ) . we used the following algorithm for generating a tree with varying branching number between @xmath10 nodes : 1 . initially the nodes are ordered , however , they are also completely isolated from each other 2 . we iterate over the nodes according to their order . for current node @xmath12 we draw a number @xmath138 from a poisson distribution with a fixed parameter @xmath139 , and assign the branching number @xmath140 to the node . ( this way it is guaranteed that the branching number of @xmath12 is @xmath141 ) . we scan the nodes coming after @xmath12 and stop at the first node @xmath11 with no incoming link . we attach directed links pointing from @xmath12 to the nodes starting from @xmath11 and ending at @xmath142 . 4 . we repeat steps ( ii)-(iii ) until all nodes become connected to the tree . the advantage of this algorithm is that it enables the study of the swift transition from a chain through a family of trees with increasing average branching number to a star . i.e. , if we set the parameter of the poisson distribution to @xmath143 , we obtain @xmath144 for all nodes , thus , the resulting graph is actually a chain . however , if @xmath139 is large enough compared to @xmath10 , the branching number drawn for the first node is already larger than @xmath10 , thus , we obtain a star . for intermediate parameter values the average branching number of the tree is of course @xmath145 . however , the branching numbers of the individual nodes in the tree will deviate from this average in a similar manner to real systems . the authors are grateful to tams vicsek and enys mones for the fruitful discussions and help in the data collection . the research was partially supported by the european union and the european social fund through project futurict.hu ( grant no.:tamop-4.2.2.c-11/1/konv-2012 - 0013 ) and by the hungarian national science fund ( otka k105447 ) . the funders had no role in study design , data collection and analysis , decision to publish , or preparation of the manuscript . gp developed the concept of the study . dc and gp worked out the definition of the hierarchy measure , dc derived the behaviour in the thermodynamic limit . dc collected the network data and carried out its analysis . dc and gp prepared the figures and contributed to the result interpretation . gp drafted the manuscript . supplementary information accompanies this paper . + the authors declare no competing financial interest . 10 url # 1`#1`urlprefix[2]#2 [ 2][]#2 , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . , , & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . , , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . _ _ , vol . of _ _ ( , , ) . , , , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . & . _ _ * * , ( ) . . _ _ * * , ( ) . & _ _ ( , , ) . & . _ _ * * , ( ) . & . _ _ * * , ( ) . _ _ ( , , ) . . _ _ * * , ( ) . _ _ * * , ( ) . , & . _ _ * * , ( ) . , , & . _ _ * * , ( ) . & . _ _ * * , ( ) . & _ _ ( , , ) . . in & ( eds . ) _ _ , ( , , ) . _ _ * * , ( ) . & . _ _ * * , ( ) . , & . _ _ * * , ( ) . , , & . _ _ * * , ( ) . _ _ ( , , ) . _ _ ( , , ) , edn . _ _ ( , , ) . & _ _ ( , , ) . , & . _ _ * * , ( ) . _ _ ( , , ) . . _ _ * * , ( ) . & . _ _ * * , ( ) . & . _ _ * * , ( ) . . , & . in _ _ , ( , ) . , & . _ _ * * , ( ) . . _ _ ( ) . & . _ _ * * , ( ) . , & . _ _ * * , ( ) . , , , & . _ _ * * , ( ) . , , , & _ _ ( , ) . ( ) . & _ _ ( , ) . , , , & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . , , , & _ _ * * , ( ) . , & . in _ _ , ( , ) . _ et al . _ . _ _ * * , ( ) . , & . in _ _ , ( , ) . & . _ _ * * , ( ) . , , , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . & . in _ _ , ( , ) . characterising the importance of hierarchy in a given network structure is a non - trivial problem with a large number of alternative approaches . without loss of generality we can formulate a few intuitive requirements a hierarchy measure should meet : first , we assume no a priori ordering between the nodes , the measure is evaluated purely based on the topology of the network . moreover , the hierarchy measure should should not be too sensitive to the local structure of the network , i.e. , the degree sequence alone should not provide enough information for complete evaluation of @xmath51 . in other words , we should be able to modify @xmath51 when rewiring the links in a network without changing the degree sequence . e.g. , let us suppose that the network is corresponding to a single directed cycle of @xmath10 nodes , ( a giant directed `` ring '' ) . since all the nodes are equivalent , the network lacks any hierarchy what so ever , thus , @xmath51 should be equal to 0 . however , if we take just a single link away , ( which is only a minor change if @xmath10 is large ) , the network becomes a directed chain , which is indeed hierarchical , thus , a significant jump should be observed in @xmath51 . one of the first hierarchy measure was proposed by d. krackhardt , motivated by organisational hierarchy @xcite . the main assumption here is that in a hierarchical company we can reach the lower levels of the hierarchy from the levels above via chains of commands , and in contrast , we can not reach the higher levels in a similar fashion from the levels below . thus , on the one hand , we count the total number of ordered pairs @xmath146 in the network for which there is directed path either from @xmath12 to @xmath11 or from @xmath11 to @xmath12 , but not both . on the other hand , we also evaluate the total number of connected pairs , for which at least one directed path exists between the two nodes , ( but we also allow paths in both directions ) . by denoting the number of ordered pairs by @xmath147 , and the number of connected pairs by @xmath148 , krackhardt s hierarchy measure @xcite is simply given by @xmath149 based on the definition above , @xmath150 for all acyclic networks , where all the node pairs are ordered . nevertheless , @xmath151 is a fine example for a hierarchy measure depending mostly on the global structure of the network : for a directed cycle of @xmath10 nodes @xmath152 , ( since we can reach any node from any other node , and hence , every pair is unordered ) . by deleting a single link we turn the network into a directed chain , ( where all the pairs are ordered ) , and suddenly the hierarchy measure jumps to @xmath150 . an approach motivated by similar intuitions compared to krackhard s hierarchy was introduced by e. mones et al . , assuming that reaching the rest of the network should be relatively easy for the nodes high in the hierarchy , and more difficult for the nodes at the bottom of the hierarchy @xcite . here the position of the node @xmath12 in the hierarchy is determined by its reaching centrality , @xmath153 , corresponding to the fraction of nodes that can be reached from @xmath12 , ( following directed paths ) . based on the @xmath154 associated to the individual nodes , the global reaching centrality of the whole network is defined as @xcite @xmath155}{n-1 } , \label{eq : grc}\ ] ] where the summation is running over the nodes , the size of the network is given by @xmath10 and the maximal reaching centrality is denoted by @xmath156 . the maximal possible value of this hierarchy measure is @xmath157 , which is obtained when the network is corresponding to a star with @xmath105 arms , with only the central node having a non zero reaching centrality . interestingly , for a chain of @xmath10 nodes , we obtain @xmath158 , which is still larger than @xmath159 . similarly to the previous measure , @xmath160 for a directed cycle . another way for quantifying the possible asymmetry between nodes is to assume some sort of flow on the links , and examine whether the global map of flows in the system is revealing a kind of overall directionality or not . probably the simplest approach in this framework is to define the fraction of links not participating in any cycle as the measure of the hierarchy . i.e. , the link flow hierarchy proposed by j. luo and c. l. magee can be formulated as @xcite @xmath161 where @xmath162 denotes the number of acyclic links , not part of any directed cycle , and @xmath163 is corresponding to the total number of links in the network . similarly to krackhardt s hierarchy , the link flow hierarchy is @xmath164 for all acyclic networks . furthermore , when the network is corresponding to a single directed cycle , @xmath165 . thus , when deleting a link from this cycle , we observe a jump from zero hierarchy to maximal hierarchy . a more elaborate quantification of hierarchy was proposed by b. corominas - murta et al . @xcite with the introduction of treeness , feedforwardness and orderability , projecting the studied network onto a point in a 3 dimensional space , where each dimension is aimed to capture a different aspect of hierarchy . treeness , @xmath6 , is measuring how ambiguous are the chain of commands in the network . i.e. , in a regular tree where links are pointing from higher levers to lower levels we obtain @xmath166 , whereas if revert the link directions , the obtained structure is considered anti - hierarchical , with @xmath167 . the calculation of @xmath6 is based on comparing forward and backward entropies @xcite . in the mean time feedforwardness , @xmath7 is related to the size and position of the strongly connected components in the network . since we can reach any node from any other node in a strongly connected component , we can not define an ordering amongst these nodes . furthermore , if the strongly connected component is found near the top of the overall hierarchy , its effect is more severe compared to a situation where it is occurring only at the deeper levels . finally , the orderability , @xmath8 is simply the fraction of nodes not taking part in any directed cycles , @xmath168 where @xmath169 is the number of nodes that are not members in any directed cycles , and @xmath10 is the total number of nodes . thus , orderability is analogous to the link - flow hierarchy @xmath170 given in ( [ eq : linkflow ] ) , the only difference is that here we measure the weight of the cycles in the network by the number of contained nodes instead of the number of contained links . similarly to @xmath170 , the oredrability is @xmath171 for all acyclic networks , while @xmath172 when the network is corresponding to a single directed cycle . we have seen on fig.1 . in the main paper that @xmath51 of finite regular trees is converging to the limit value for infinitely large trees already around @xmath81 when @xmath59 , while the convergence is somewhat slower when @xmath60 . nevertheless , above a certain @xmath10 it is the structure of the tree , ( encoded in the branching number ) , what determines the hierarchy measure , not the size of the tree . this indicates that @xmath51 is behaving similarly to intensive quantities in physics in some aspects . to investigate this property further , let us examine what happens to @xmath51 if we take a pair of disjoint graphs @xmath64 and @xmath173 , and unite them into a single graph @xmath174 with two isolated components . according to eq.(10 ) in the main paper , the hierarchy measure of the graphs when considered separately can be written as @xmath175 where @xmath36 and @xmath176 correspond to the stationary distribution of the random walkers on nodes @xmath177 and @xmath178 , while @xmath179 and @xmath180 denote the sizes of @xmath64 and @xmath173 respectively . when considering the union of the two graphs , the stationary distribution of the random walkers has to be normalised over the whole range of @xmath181 nodes . thus , the stationary distribution on node @xmath12 , originally in @xmath64 , now part of @xmath182 , can be given as @xmath183 . likewise , the stationary distribution on node @xmath11 , originally in @xmath173 , now part of @xmath182 , can be expressed as @xmath184 . therefore , the hierarchy measure of the union of the two graphs reads @xmath185 by rearranging ( [ eq : h_g_1]-[eq : h_g_2 ] ) we gain @xmath186 ^ 2 + 1}{n_1 } , \label{eq : p_1_sqr}\\ \sum_{i=1}^{n_1}(p_i^{\rm stat})^2&=&\frac{\left[h_{{\mathcal g}_1}\right]^2 + 1}{n_2 } , \label{eq : p_2_sqr}\end{aligned}\ ] ] and by substituting ( [ eq : p_1_sqr]-[eq : p_2_sqr ] ) into ( [ eq : h_g_u_mid ] ) we obtain @xmath66 ^ 2+n_2\left[h_{{\mathcal g}_1}\right]^2}{n_1+n_2}}. \label{eq : h_g_uni}\ ] ] thus , the hierarchy measure of the union of two isolated networks is simply the weighted quadratic mean of @xmath51 obtained for the individual graphs . a noteworthy consequence is that if @xmath187 , then this implies also that the hierarchy measure of the union is also the same , @xmath67 . furthermore , according to ( [ eq : h_g_uni ] ) in general , if we take the union of @xmath64 and @xmath173 which are isolated from each other and @xmath188 , then the hierarchy measure of the resulting network will be between the hierarchies measured for the two graphs separately , @xmath189 therefore , the rule for calculating @xmath51 for a system composed of isolated smaller parts based on @xmath51 obtained for these sub - systems is showing again a great deal of similarity to the behaviour of intensive physical quantities . according to our results in the main paper , when considering regular trees of infinitely large size , the hierarchy measure may either diverge or remain finite , depending on both the branching number @xmath52 and the value of the @xmath32 parameter . at a fixed @xmath52 , the critical @xmath190 separating the two regimes is given by eq.(23 ) in the main paper . here , in fig.[fig : h_phase_diag ] . we show @xmath191 as a function of @xmath52 on a semi - logarithmic plot , where the regime of diverging @xmath51 values is coloured grey . in case of regular trees in the thermodynamic limit . if @xmath32 and @xmath52 are falling into the region shown in grey , ( meaning that @xmath32 is larger than @xmath190 calculated from eq.(23 ) in the main paper ) , then @xmath51 is diverging . in contrast , for parameter settings falling in the white region , @xmath51 remains finite in the thermodynamic limit . the maximal @xmath51 in this region is obtained at a @xmath192 , given in ( [ eq : lam_star ] ) . the dashed horizontal line is showing the minimum of @xmath190.,scaledwidth=60.0% ] in the regime of finite @xmath51 , the limit value for the hierarchy measure is given by eq.(21 ) in the main paper . the maximum of this function can be located by solving @xmath193 based on that , at a fixed branching number @xmath52 , the @xmath32 parameter providing the maximal @xmath51 can be written as @xmath194^{-1 } , \label{eq : lam_star}\ ] ] while at a fixed @xmath32 parameter the tree with maximal hierarchy has a branching number of @xmath195 in fig.[fig : h_phase_diag ] . we also show @xmath196 given in ( [ eq : lam_star ] ) as a function of @xmath52 . in parallel , the @xmath197 expressed in ( [ eq : b_star ] ) is plotted as a function of @xmath32 in fig.[fig : max_helyek ] . for which @xmath51 is maximal , as a function of the parameter @xmath32 in case of infinitely large regular trees . the vertical line is corresponding to the minimum of @xmath190 given in eq.(23 ) in the main paper , thus , for @xmath32 values above that @xmath51 can become divergent in the thermodynamic limit.,scaledwidth=60.0% ] according to picture , the maximum of @xmath51 is at @xmath198 when @xmath32 is low . however , @xmath197 is steadily increasing as a function of @xmath32 , reaching to @xmath199 at @xmath60 , chosen to be the optimal @xmath32 value based on arguments given in the main paper . we have analysed the hierarchy in a large number of networks of different types and varying sizes , the scatter plot of the obtained @xmath51 values and the @xmath88-score of @xmath51 are given in figs.5 - 6 . in the main paper . in tables [ tab : results_real_prop]-[tab : results_real_rand ] . we provide the results in more details by listing also the size and the average degree beside the @xmath51 values and @xmath88-scores of the examined networks . .random walk hierarchy of real networks shown in fig.5 . in the main paper . the network type is given in the 1@xmath200 column , the meaning of the links in the @xmath201 column , the references to the data sources are listed in the @xmath202 column . the network size is given in the 4@xmath203 column , followed by the average degree in the 5@xmath203 column . the hierarchy measure @xmath51 ( calculated at the optimal @xmath60 parameter value ) is provided in the 6@xmath203 column . [ cols="^,^,^,^,^,^",options="header " , ] the peer to peer message networks over the internet and the metabolic networks , ( where the target of a link was corresponding to a product of the source ) showed moderate levels of hierarchy . for both types , the data points in fig.5 . in the main paper formed tight clusters . the intra organisational networks were the least hierarchical according to our measure , with the data points forming a cluster close to the origin . similarly to the food webs , the variance of the @xmath51 values obtained for the regulatory networks was relatively high . i.e. , the regulatory networks trn - yeast-2 and trn - yeast-1 are amongst the most hierarchical studied systems , whereas the trn - ec network shows only a modest hierarchy value . the trust networks obtained moderate hierarchy values , and the variance of their @xmath51 values is far smaller compared to the very large variance in their sizes . the language networks turned out to be moderately hierarchical as well . here the networks were originating from large text corpora , and a directed link from word @xmath147 to @xmath204 is signaled that @xmath204 was following @xmath147 in the text . according to the results , the english- french- and japanese language networks obtained @xmath51 values quite close to each other , while the hierarchy of the spanish language seemed considerably lower . finally , the networks between web pages showed again a moderate hierarchy , where @xmath51 for the google web and polblogs were quite close to each other , while network of nd.edu received a significantly higher hierarchy value . it is very interesting to see how the picture is changing when we switch from @xmath51 to the @xmath88-score shown in fig.6 in the main paper , corresponding to the difference between @xmath51 and @xmath80 in the configuration model , scaled by standard deviation of @xmath51 in the configuration model . the citation networks , the internet networks and the food webs are forming three rather tight clusters in fig.6 . in the main paper with significant positive @xmath88-score , ( which is outstandingly high in case of the citation networks ) . in contrast , the metabolic- and regulatory networks are forming clusters of significant negative @xmath88-score , indicating that their structure is far less hierarchical compared to what we would expect by assuming random connections between the nodes , ( at the same degree distribution as in the original network ) . the @xmath88-scores of www networks were all positive , ( where the nd.edu network showed an outstandingly high value ) . in case of the trust networks the @xmath88-scores are also almost always positive , with the exception of the prison network , where @xmath205 . the organisational- , electric- , and language networks showed a mixed picture , including both positive and negative @xmath88-scores .

in this paper , we extend the study of bremsstrahlung photon production in a quark - gluon plasma to the cases of soft static photons and hard real photons . the general framework of this study is the effective perturbative expansion based on the resummation of hard thermal loops . despite the fact that bremsstrahlung only comes at two loops , we find that in both cases it generates contributions of the same order of magnitude as those already calculated by several other groups at one loop . furthermore , a new process contained in the two - loop diagrams dominate the emission of a very hard real photon . in all cases , the rate of real or virtual photon production in the plasma is appreciably increased compared to the one - loop predictions . 1 . laboratoire de physique thorique lapth , + bp110 , f-74941 , annecy le vieux cedex , france 2 . physics department and winnipeg institute for theoretical physics , + university of winnipeg , winnipeg , manitoba r3b 2e9 , canada lapth678/98 , win98 - 05 , hep - ph/9804224

hep-ph9804224

we consider the production of a real photon or of a lepton pair in a quark - gluon plasma . the plasma is assumed to be in equilibrium at temperature @xmath1 . the theoretical framework used in the calculation is that of thermal field theory improved by the hard loop resummation @xcite of braaten and pisarski : in this approach one distinguishes hard momenta , of order @xmath1 , from soft momenta , of order @xmath2 , where @xmath3 is the quantum chromodynamics ( qcd ) coupling constant assumed to be small ( @xmath4 ) . after resummation of hard thermal loops , one is led to an effective field theory from which observables can be evaluated perturbatively . the production rates of real or virtual photons have already been evaluated , at the one loop level , in the effective theory @xcite . concerning soft virtual photons , it was found that the rate of production is considerably modified and enhanced compared to the result of the bare theory . besides the usual quark - antiquark annihilation process , there appear many production mechanisms , in particular processes where the photon is radiated off a ( hard ) quark in a scattering process where the quark is backward scattered in the plasma via soft quark exchange . this is to be contrasted to the result obtained in a semi - classical approximation @xcite where the photon is radiated off fast quarks in scattering processes mediated by a gluon exchange : we call such processes bremsstrahlung emission of a photon . in this study we reconcile the two approaches and show that the bremsstrahlung processes favored by the semi - classical approximation appear at the two - loop level in the effective theory and that , in fact , they contribute at the same order in @xmath2 as the processes in the one - loop effective theory . such a result should not be a surprise . consider the case of a soft virtual photon . the rate of production is related to the imaginary part of the vacuum polarization diagram @xcite . in the one - loop approximation of the effective field theory , it involves only effective fermion propagators and effective vertices @xcite . a dominant contribution to the rate arises when the internal fermion momentum is soft and , therefore , all effective propagators and effective vertices have the same order of magnitude as their bare counterparts . a close examination of the final result shows , however , that it has a logarithmic sensitivity to scales of order @xmath1 ( see section [ bpycomp ] ) : this means that such a diagram also receives a dominant contribution from hard fermion momenta . when the momentum becomes large , the hard thermal loop ( htl ) corrections to propagators and vertices are suppressed by , at least , a factor @xmath3 with respect to their bare counterparts . this suppression factor @xmath3 can easily be compensated by the larger phase space available to a hard fermion ( @xmath5 ) compared to soft fermion ( @xmath6 ) , thereby leading to a contribution of the same order of magnitude from the soft region and the hard region of phase - space . now , when an observable is sensitive to the thermal corrections of hard vertices and propagators , it is obvious that all such corrections should be taken into account for the calculation to be complete . some of these thermal corrections are naturally included in the lowest order of the effective theory via the resummation of hard thermal loops . but , even if the htl approximation is correct for soft external particles , it does not account for all thermal corrections to hard vertices and hard propagators . for instance , neglecting the external momenta as one does in the calculation of the hard thermal loops is no longer justified when these momenta are not soft . besides , equally important may be the contribution arising from soft gluons in the loop giving the htl when the external momenta are hard , due to the bose enhancement of the soft gluon term . within the effective theory , both types of additional thermal corrections to a hard propagator or vertex are taken into account by considering a one - loop correction to this propagator or vertex . in the calculation of the virtual photon production rate in the effective theory , soft gluon exchange appears in two - loop diagrams . it will be seen that the bremsstrahlung production mechanism is precisely given by these diagrams when the exchanged gluon is space - like . the evaluation of these diagrams is discussed below . these contributions are clearly not included in the effective one - loop diagram . this is obvious when looking at the corresponding physical processes and it manifests itself in the result by the calculated rate being proportional to the square of the thermal gluon mass @xmath7 , in contrast to the one - loop result where only the thermal quark mass appears @xcite . another important contribution of the two - loop diagrams is associated with time - like gluon exchanges : physically this represents qcd compton scattering and quark - antiquark annihilation to produce a gluon and a photon . to evaluate this properly requires care since the two - loop diagrams with a hard time - like gluon exchanged are already part of the one - loop diagram with effective propagators and vertices . taking into account the contribution of the appropriate counterterms in the effective lagrangian will prevent double - counting and allow the correct evaluation of the soft , time - like , gluon contribution . the case of soft real photon production follows essentially the same pattern , except for the crucial fact that , the external line in the vacuum polarization diagram being massless , collinear divergences appear when evaluating the two - loop diagrams : the quasi - overlap of two such divergences , associated with the fermion propagators , leads to an enhancement factor of @xmath8 @xcite . the paradoxical result then follows that the one - loop contribution is relatively suppressed by a factor @xmath9 compared to the two - loop one ! the latter is entirely dominated by the bremsstrahlung process since the kinematical constraints require the gluon to be space like for the enhancement factor to occur . the compton and annihilation mechanisms are sub - dominant . the case of a hard real photon , of momentum of @xmath10 , shares features with both cases above . the one - loop approximation has a logarithmic sensitivity to the hard fermion momentum in the loop @xcite . the two - loop bremsstrahlung has a @xmath11 collinear enhancement , as in the real soft photon case , which however is compensated by a factor @xmath12 , where @xmath13 is the photon momentum , leaving the bremsstrahlung contribution at the same order in @xmath2 as the one - loop contribution . in the following we are concerned mainly with the bremsstrahlung part of the two - loop diagrams , and leave the discussion of the compton and annihilation processes and their interplay with the counterterms of the effective theory to future work . we do not discuss the production of soft real photons since this has already been studied in detail in @xcite . in the next section we derive the general expression for the ( real or virtual ) photon production rate at the two loop level . then we consider the case of soft virtual photons produced at rest in the plasma and derive the leading behavior analytically . we compare to the one - loop results and show that the bremsstrahlung contribution is numerically dominant although both contributions are technically of the same order in @xmath3 . the semi - classical approach is then discussed and it is shown that even though the approximations inherent in the semi - classical approach are not really justified in the case of soft photon production in a quark - gluon plasma , it leads to a result quite comparable to that obtained in thermal field theory . turning to the case of hard real photons , it is shown that the bremsstrahlung mechanism is of the same order as the already calculated one - loop result . carrying out a more detailed comparison with the latter approach , it is found that the bremsstrahlung process dominates over the one - loop result for photon momentum of @xmath10 but is relatively suppressed by a logarithmic factor for hard enough photons . for very hard energies , we find that the photon production is in fact dominated by a new process consisting of a @xmath14 annihilation where the quark or antiquark undergoes a scattering in the medium . we summarize all the thermal field theory results concerning real and virtual photon production in a concluding section . the role of counterterms in the application of the effective theory up to two - loops is discussed in a first appendix where the problem is also illustrated in a simple example . in a second appendix , the importance of phase space factors in thermal calculations is emphasized and the difference with the zero temperature phase space is made clear . let us first recall the relationship between the photon production rate and the imaginary part of the retarded polarization tensor of the photon , as given by thermal field theory ( we follow the notations of @xcite ) . for real photons , this relation gives the number of photon emitted per unit time and per unit volume of the plasma as @xcite : @xmath15 whereas for the production of a photon of invariant mass @xmath16 decaying into a lepton pair we have : @xmath17 basically , the above two formulae differ only by the allowed phase space for the photon , by an extra qed coupling constant when the photon decays into a lepton pair , and by the propagator of a heavy photon . it is worth recalling that these relations are valid only at first order in the qed coupling @xmath18 since they do not take into account the possible re - interactions of the photon on its way out of the plasma nor the simultaneous emission of more than one photon . nevertheless , they are true to all orders in the strong coupling constant @xmath19 . this should not be a serious limitation from a practical point of view since @xmath20 . let us now examine in which topologies the bremsstrahlung can appear . it is worth recalling at this point that the retarded imaginary part in eqs . ( [ realphot ] ) and ( [ virtualphot ] ) can be expressed as a sum over possible cuts through the diagram @xcite . therefore , we need to look for diagrams that will give bremsstrahlung processes once cut . a simple inspection of the processes involved in one - loop contributions ( see @xcite for instance ) shows that bremsstrahlung does not appear at this order . to see bremsstrahlung processes , one should consider the two - loop contributions of fig . [ fig:2loop ] . the diagrams have been obtained via a strict application of the feynman rules of the effective theory @xcite , giving _ a priori _ effective vertices and propagators and diagrams with counterterms in order to avoid any double counting of thermal corrections already included at the one - loop level via the resummation of hard thermal loops , as outlined in appendix [ app : counterterms ] . to make the connection with previous works @xcite easier , we mention that looking at two loop diagrams in the effective theory is just a more rigorous way of doing what we might call `` calculating one - loop diagrams beyond the htl approximation '' . our present formulation is indeed more rigorous since it takes care of the counterterms , and also more positive since it does not assume _ a priori _ that one needs to go beyond the effective theory . among all the possible cuts through the diagrams , those that correspond to bremsstrahlung necessarily cut the gluon propagator . moreover , if @xmath21 is the gluon 4-momentum , only the landau damping part @xmath22 gives bremsstrahlung , the @xmath23 part rather giving compton effect or @xmath14-like annihilations @xcite . there is another reason why the region @xmath23 deserves a separate treatment : in the @xmath24 kinematical domain , it is obvious that we can not have contributions coming from the htl counterterms since these counterterms involve only bare gluon propagators that do nt have any imaginary part in the space - like region . on the contrary , in the @xmath23 region , one should pay special attention to the counterterm diagrams in order to avoid any double - counting . indeed , when the gluon becomes hard , we have a hard loop that may reproduce what is already included in the one - loop diagram via the effective vertices and propagators . from now on , we limit ourselves to the region where @xmath24 and only to the true two - loop diagrams , leaving the @xmath23 region and the discussion of counterterms to future work . moreover , since our main focus is on bremsstrahlung , we must retain from the cut quark propagators only the pole part and reject the landau damping part , which would correspond to a different physical process . for the same reason , the cut should avoid going through an effective vertex . in order to obtain a bremsstrahlung contribution of the same order of magnitude as the already calculated one - loop contributions , we need a hard phase space for the quark circulating in the loop , as explained in appendix [ app : phasespace ] . we will see later that we have such contributions in the diagrams depicted in fig . [ fig:2loop ] . therefore , at leading order , we can use `` bare '' . ] vertices and propagators everywhere except for the gluon propagator since the gluon can be soft . the diagrams we have to consider are therefore the simplified versions of the previous ones represented on fig . [ fig:2loopsimple ] . in the same figure , we have depicted the relevant cuts as well as the arrangement of circlings that enables one to calculate the corresponding contribution in the framework of the thermal cutting rules . we have checked that the two cuts represented form a gauge independent set of terms , to which one should add the symmetric cut for the vertex diagram and a third diagram with the self - energy correction on the upper quark line . since these two other terms give the same contribution as the previous two , we simply take them into account by multiplying the final result by an overall factor of 2 . a straightforward application of the cutting rules valid for the `` r / a '' formalism , with the notations of @xcite , gives for the vertex correction : & & ^^ra _ ^(q_o , q)_|_vertex=- im^^ar _ ^(q_o , q)_|_vertex= nc__f2 + & & e^^(q , p+l ,- r - l ) g^^(-p - l , p , l ) + & & g^^aar(r+l ,- l ,- r ) e^^aar(r ,- p ,- q ) + & & d^^a_(l ) + & & = -nc__f2e^2g^2 + & & + & & ( n__f(r_o)-n__f(p_o ) ) ( n__b(l_o)+n__f(r_o+l_o ) ) + & & s(r)s(p+l)p_^^t , l(l)^_|_vertex , [ cutvertex ] where , following @xcite , we denote the fermion propagator : @xmath25 and the effective gluon propagator in a linear covariant gauge : @xmath26\\ & & \pi_{_{l}}(l)\equiv3m_{\rm g}^2(1-x^2)\left [ 1-{{x}\over{2}}\ln\left ( { { x+1}\over{x-1}}\right)\right]\ ; , \label{notations : gluon}\end{aligned}\ ] ] with @xmath27 the usual transverse and longitudinal projectors in linear covariant gauges @xcite , @xmath28 @xcite the asymptotic thermal mass of the quark , and @xmath29/9 $ ] the soft gluon thermal mass . in this formula , @xmath30 is the electric charge of the quark and therefore depends on its flavor . likewise , we obtain for the second diagram : & & ^^ra _ ^(q_o , q)_|_self=- im^^ar _ ^(q_o , q)_|_self= nc__f2 + & & e^^(q , p ,- r ) g^^(r , l ,- r - l ) + & & g^^aar(r+l ,- l ,- r ) e^^aar(r ,- p ,- q ) + & & d^^a_(l ) + & & = -nc__f2e^2g^2 + & & p_^^t , l(l)^_|_self + & & ( s(r))^2 ( n__f(r_o)-n__f(p_o ) ) ( n__b(l_o)+n__f(r_o+l_o ) ) . [ cutself ] in the previous formulae , a factor @xmath31 without any @xmath32 or @xmath33 superscript simply denotes the principal part of the propagator . in other words , for these factors , the @xmath32 or @xmath33 prescription is irrelevant because the corresponding delta function is incompatible with the other delta functions present and therefore vanishes . we may notice the similarity between eqs . ( [ cutvertex ] ) and ( [ cutself ] ) . in particular , the same combination of statistical weights appear in both formulae , while the expressions in the square brackets simply express the cuts on internal lines . moreover , when plugged into eq . ( [ realphot ] ) in order to obtain the production rate , the sum of eqs . ( [ cutvertex ] ) and ( [ cutself ] ) gives the more intuitive eq . ( [ eq : lastform ] ) in appendix [ app : phasespace ] ( to which one should add similar terms to take into account all the processes included in our diagrams ) . therefore , the `` r / a '' formalism appears just as an efficient method to reorder the factors of the integrand in order to make it more compact and more convenient for the subsequent integrations . the drawback of this formalism is that it generates less intuitive expressions . the calculation of the dirac s traces is of course common to both cases . we obtain for the self - energy insertion : & & trace^_|_self -4 , and for the vertex correction : & & trace^_|_vertex -4 . it is worth recalling that these expressions are obtained by anticipating the use of the relation @xmath34 in order to drop any @xmath35 or @xmath36 in the expression of the dirac s traces . since this identity is not true for the gauge dependent part of the gluon propagator , one should not use these expressions of the traces to check the independence of the rate with respect to the gauge parameter @xmath37 . moreover , we discarded terms that will be killed later by the delta functions such as the one contained in @xmath38 . since a 4-vector like @xmath39 is not a linear function of the momentum @xmath40 , we used some approximations to simplify the calculations , the effect of which is to neglect only terms that are always subdominant ] ] . we also notice that the statistical weights and delta functions present in eqs . ( [ cutvertex ] ) and ( [ cutself ] ) are invariant under the change of variables @xmath41 , @xmath42 . therefore , in the remaining factors of the integrand , we are allowed to drop the parts which are antisymmetric under this transformation . collecting contributions from the two topologies , this symmetrization gives : & & _ sym-4 , [ eq : tracetotal ] which will be our starting point in the following sections . when the 3-momentum @xmath43 of the emitted photon is zero , a lot of simplifications occur . first of all , since there is one 3-vector less in the problem , we need only one angular variable which simplifies considerably the angular integrations . moreover , as shown in @xcite , the non vanishing invariant mass of the emitted photon regulates all the potential collinear divergences when @xmath44 . therefore , one can simply forget about the quark asymptotic thermal mass since the purpose of such a mass is precisely to regularize collinear singularities . this means that we can everywhere identify @xmath40 and @xmath45 at this level of approximation , since furthermore @xmath40 is hard . from the identity @xmath46 , we extract the values @xmath47 and @xmath48 . for the second cut quark propagator , we have the identity @xmath49 from which we extract the cosine of the angle @xmath50 between @xmath51 and @xmath52 : @xmath53 of course , we must require that this value be in @xmath54 $ ] , which will reduce the available phase space . this requirement leads to the following two inequalities : @xmath55 which lead to a phase space reduction that can be seen in fig . [ fig : kine ] , where the region excluded by the requirement @xmath56 $ ] has been shaded in dark gray . other regions are excluded also by our choice of looking only at bremsstrahlung , _ i.e. _ excluding areas where @xmath23 @xcite . finally , the only regions we have to consider are the unshaded ones . having taken these constraints into account , the independent variables we are left with are for instance ( the choice is not unique ) @xmath57 , @xmath58 and @xmath59 , everything else being a function of these three . in particular , the denominators appearing in the calculations are @xmath60 the @xmath61 sign corresponding to @xmath62 and the @xmath63 sign to @xmath64 . it is worth examining more closely to which physical processes the regions i , ii and iii correspond . this is done just by looking at the signs of @xmath65 and @xmath66 . examples are shown in fig . [ fig : processes ] . besides the bremsstrahlung present in regions i and iii , we see a new process in region ii . this process corresponds to an annihilation of a quark - antiquark pair , one of the particles having previously undergone a scattering in the medium . since the scale for the quark momentum is given by the temperature @xmath67 , and since we are looking here at soft photons , it is obvious from fig . [ fig : kine ] that processes i and iii have support of order @xmath68 in the @xmath69 plane while process ii has only support of order @xmath70 , the integrand being the same . therefore , we expect and we have checked that process ii is suppressed by a factor of order @xmath71 compared to bremsstrahlung . as a consequence , bremsstrahlung appears to be the dominant contribution as far as the @xmath24 domain is concerned . in the remaining part of this section , we limit our study to the regions i and iii . we can obtain a further reduction of the phase space by noticing that regions i and iii give the same contribution since they are equivalent by a change of variables ( indeed , after the symmetrization in eq . ( [ eq : tracetotal ] ) , the integrands are invariant under the change of variables @xmath41 and @xmath42 ) . physically , this means that photons are produced equally by quarks and by antiquarks . therefore , we just consider region iii ( _ i.e. _ @xmath72 and @xmath73 ) and multiply the result by an extra factor @xmath74 . hence , the contribution of bremsstrahlung is given by & & im^^ar_^(q_o,0 ) e^2g^2dpldl dl_o + & & + & & _ a = t , l_a(l ) p^a_(l ) _ sym . let us now concentrate on the matrix element to be integrated over the phase space depicted above . in order to perform the contraction of the dirac s trace obtained in eq . ( [ eq : tracetotal ] ) with the longitudinal projectors , we use the following expressions @xcite @xmath75 where @xmath76 is the 4-velocity of the plasma in its rest frame . an important simplification is obtained in the case of static photons since we have @xmath77 taking into account this simplification , we obtain after some algebra : & & _ a = t , l_a(l ) p^a_(l ) _ sym + & & -4 . [ eq : static1 ] it is worth noticing that all the potential poles in @xmath78 ( see eq . ( [ eq : projectors]-b ) ) have disappeared in this formula . this is nothing but a consequence of the gauge invariance of the set of diagrams we are looking at . now , in order to extract the order of magnitude of each term in eq . ( [ eq : static1 ] ) , we can use the following very rough rules : @xmath79 @xmath80 which is correct for our purpose even if @xmath59 is hard since @xmath59 is bounded by a quantity proportional to @xmath81 ( see the previous paragraph on kinematics ) . @xmath79 @xmath82 which always gives the correct order of magnitude , even when @xmath58 is hard . @xmath79 @xmath83 . @xmath79 @xmath84 . @xmath79 @xmath85 . @xmath79 the details of the @xmath86 dependence are irrelevant to obtain the correct order of magnitude . @xmath79 @xmath87 behaves like @xmath88 if one neglects its @xmath86 dependence . moreover , using the variable @xmath89 , we can see @xmath90 is a sum of terms like @xmath91 where @xmath92 is a dimensionless function and where @xmath93 in order to give the correct overall dimension . a close inspection of eq . ( [ eq : static1 ] ) shows that @xmath18 ranges from @xmath94 to @xmath95 , taking all the integer values between these bounds . ( [ eq : static1 ] ) shows also that @xmath96 . from the previous structure and since we can factorize a factor @xmath97 out of the spectral functions @xmath98 , it is obvious that we can write the result as @xmath99 where @xmath100 is a polynomial of two variables , of total degree @xmath101 . under the assumption that @xmath102 , we are allowed to truncate this polynomial and keep only terms of degree @xmath103 in the variable @xmath104 , leaving a polynomial of @xmath105 only . moreover , the above rules show that the term of eq . ( [ eq : static1 ] ) proportional to @xmath106 contributes to the constant term of this polynomial , and gives an integral that behaves like @xmath107 for hard @xmath59 . this means that a logarithm of order @xmath108 shows up in this coefficient . the argument of this logarithm can be written as @xmath109 where @xmath110 is a function of dimension two . this function depends on both @xmath111 and @xmath7 since there can be a competition between @xmath111 which appears as a kinematical infrared cut - off in the integral over @xmath112 and @xmath7 which appears in @xmath98 and can also play the role of an infrared cut - off for the same integral . using the same tools , it is quite easy to check that all the other coefficients of this polynomial are of order @xmath113 ( _ i.e. _ do not contain any large logarithm ) . therefore , under the assumption that @xmath114 , we can formally put the result into the compact form : @xmath115\ ; , \label{generic}\ ] ] where @xmath116 is a numerical constant and @xmath117 is a polynomial . under some more restrictive assumptions , we can go further analytically . more precisely , it is possible to extract analytically the constant @xmath116 in front of the logarithm of eq . ( [ generic ] ) , as well as the function @xmath118 in the limit @xmath119 . the assumption @xmath120 ( _ i.e. _ @xmath121 and @xmath102 ) ensures that the argument of the logarithm is large , so that the logarithmic term should be a fairly good approximation of the whole expression is essential to have a large logarithm , whereas the extra inequality @xmath119 is necessary just to be able to calculate analytically the function @xmath122 . ] . as mentioned before , the logarithm we are looking at comes from the term in @xmath106 in eq . ( [ eq : static1 ] ) . using eq . ( [ eq : cosine ] ) , we obtain the following expression for the imaginary part of the photon polarization tensor : & & im^^ar_^(q_o,0 ) -4nc__fe^2g^2 t _ 0^+dp n^__f(p ) _ -1^+1dx ( 1-x^2)^2 + & & _ q_o^2p dll^3 ( _ _ t(l , x)-__l(l , x ) ) , where @xmath123 is the spectral function in the space - like region . this is the place where the condition @xmath119 enters the picture . indeed , if we do nt make this assumption , the infrared regulator of the integral over @xmath112 will be a complicated combination of @xmath111 contained in the lower bound and @xmath7 contained in the spectral functions . on the contrary , when @xmath119 , only the largest regulator ( _ i.e . _ @xmath7 ) plays a role , and the argument of the logarithm is quite simple . the integral over @xmath112 is elementary and yields an @xmath124 function and a logarithm . keeping only the latter term we discarded is convergent when performing the subsequent integration over @xmath125 since the @xmath124 is bounded by @xmath126 . ] , we obtain : & & im^^ar_^(q_o,0 ) _ 0^+dp n^__f(p ) _ -1^+1dx + & & ( im__l(x)- im__t(x))(p^4 ) , the terms neglected in that procedure show up only in the polynomial that would accompany the @xmath108 in a more complete calculation , and are not tractable analytically . moreover , for the same reason and because of the statistical weight in the @xmath127 integral that will cut off everything above @xmath128 , we can replace the remaining logarithm integral without this further approximation . doing so leads to a result in which the logarithm of eq . ( [ eq : staticfinal ] ) is replaced by @xmath129 where @xmath130 is the euler constant . however , the additional constant is not complete since we have already neglected contributions to it in earlier approximations . ] by @xmath131 . now , the @xmath125 and @xmath127 integrals are trivial and give : @xmath132 the production rate is then given by ( see eq . ( [ virtualphot ] ) ) : @xmath133 where the sum runs over the flavor of the quarks in the loop ( @xmath134 is the electric charge of the quark of flavor @xmath135 , in units of the electron electric charge ) . a comment is relevant concerning the sensitivity of the exchanged gluon to the hard scale . indeed , the discontinuity of the effective propagator is used here in the space - like region , and the htl approximation used to obtain this propagator may inaccurately reflect the phenomenon of landau damping for a hard gluon . the consequence of this remark is that a loop correction on the gluon propagator may lead to an important three - loop correction to the photon emission - rate . before comparing this analytical result with numerical estimates of the unapproximated expression , let us recall the domain in which this expression is expected to be valid . firstly , we need the logarithm to be large in order to be dominant , which requires @xmath120 , _ i.e. _ @xmath121 and @xmath102 . the additional _ purely technical _ condition is that @xmath119 , in order to keep simple the argument of the logarithm . on the plots of the figure [ fig : plotstatic ] , we show the ratio numerical / theoretical " , where theoretical " is the formula given in eq . ( [ eq : staticfinal ] ) while numerical " denotes a numerical evaluation of the contribution to bremsstrahlung of the complete matrix element as given in eq . ( [ eq : static1 ] ) . the left plot shows that @xmath136 must be smaller than @xmath137 in order to have an agreement between our approximations and the complete expression with an accuracy better than 5% . if @xmath136 is not small enough , then the polynomial that comes with the @xmath108 can not be neglected anymore . on the second plot , we see that the approximations we performed inside the logarithm by assuming the smallness of @xmath138 are in fact valid far outside their expected domain of validity , since we still have a reasonable accuracy with @xmath139 . in a previous paper , we gave asymptotic formulae for the same quantity in the case where the photon invariant mass satisfies @xmath140 ( see eqs . ( 89),(90 ) and ( 93 ) of @xcite ) . by extrapolating these estimates outside of their apparent range of validity towards the case of static photons for which @xmath141 , we obtain exactly the formula of eq . ( [ eq : staticfinal ] ) . such an agreement means that , for a given energy @xmath111 , the formulae established for the production of low invariant mass photons are very robust since they remain valid when extrapolated to the case of heavy photons at rest . from a technical point of view , this is made possible by the fact that the term that contains the collinear singularity in which we were interested for low mass photons and the term that develops the logarithm we extracted analytically for heavy photons are the same . at the one loop level in the effective perturbative expansion , the production rate of soft static photons has been evaluated by braaten , pisarski and yuan in @xcite . the purpose of this paragraph is to present an analytic comparison of their final result ( eq . ( 11 ) of @xcite ) with eq . ( [ eq : staticfinal ] ) , in the domain @xmath142 for which our expression has been justified . in this domain , we can retain from the result of @xcite only the terms having the most singular behavior in @xmath111 . such terms are found only in the `` cut - cut '' part . moreover , some of these terms develop a large logarithm @xmath108 which is simple to extract analytically , in a way very similar to the method leading to eq . ( [ eq : staticfinal ] ) . applying these approximations to bpy s result leads to the following estimate for the one - loop production rate : @xmath143 where @xmath144 is the soft quark thermal mass . therefore , comparing with eq . ( [ eq : staticfinal ] ) , we obtain the ratio : @xmath145 which for 2 light flavors and 3 colors becomes @xmath146 this ratio is rather large , which means that bremsstrahlung is definitely an essential contribution to the soft static photon production rate by a hot plasma . the bremsstrahlung production of a soft virtual photon has been considered in the context of the semi - classical approximation by cleymans _ et al._. in their approach@xcite , they took into account the effect of the multiple scattering of the quark in the plasma ( landau - pomeranchuck - migdal effect ) . in order to compare with our thermal field theory result we need to `` undo '' the effect of rescattering and consider only one collision of the ( photon emitting ) quark in the plasma . _ use several simplifying hypotheses : the energy of quarks or gluons is much larger than the temperature so that boltzmann distributions are used for particles entering the interaction region and a factor 1 is assigned to those leaving it . the scattering of quark in the plasma is treated as in the vacuum , the only modification being the introduction of a phenomenological debye mass @xmath147 to screen the forward singularity of the quark scattering amplitude . neglecting furthermore the virtual photon momentum compared to the momenta of the constituents in the plasma the production rate can be factorized into a quark scattering term and a photon emission term so that the lepton pair rate can be written ( see @xcite for a similar expression in the case of a real photon ) & & _ i=1,2 e^-p_i / t dp_i 2 p_i _ i=1,2 dp_i 2 p_i + & & ( 2)^4 ( p_1+p_2-p_1-p_2 ) + & & |m|^2(p_1,p_2;p_1;p_2 ) e^2 _ ( p_1- p_1 ) ^2 , [ factorization ] where @xmath148 is the square of the matrix element of the quark scattering process . we have folded in the above expression the appropriate factor describing the decay of the virtual photon of mass @xmath149 into the lepton pair . with the above mentioned approximations and keeping the most singular term in the @xmath150channel @xcite we have @xmath151 with @xmath152 for quark - quark scattering and @xmath153 for quark - gluon scattering . we then find for the rate of production of the pair at rest the following expression : @xmath154 where @xmath155 and @xmath156 are the degeneneracy factors introduced in @xcite . comparing with eq . ( [ eq : staticfinal ] ) we find for two light flavors @xmath157 _ i.e. _ the semi - classical result agrees with the thermal field theory result in its functional dependence but over - estimates the rate of production by about @xmath158 . this difference appears to be due to the very approximate treatment of thermal effects of the dynamics of the plasma : for example , the ratio of quark - gluon scattering to quark - quark scattering is estimated to be @xmath159 in thermal field theory ( this is the ratio of the gluon contribution to the quark contribution in a hard thermal loop ) compared to @xmath160 ( taking account of the multiplicity factors associated to the quarks and the gluons ) in the semi - classical approach . let us now concentrate on the case of hard real photons ( @xmath161 ) . the kinematics for this situation is much more complicated than for static soft photons for two reasons . first , since @xmath149 is hard , we are no longer allowed to neglect @xmath149 in front of @xmath40 or @xmath32 . moreover , since we are looking this time at real photons , we may encounter collinear divergences ( like in the case of soft real photons @xcite ) , and we must carefully keep the quark asymptotic thermal mass @xmath162 in the expressions . now , from the identity @xmath163 , we extract the value @xmath164 , where we denote @xmath165 and @xmath166 . the second delta function constraint provides us with the angle @xmath50 between the 3-vectors @xmath167 and @xmath52 , via the relation : @xmath168 again , we must enforce the requirement @xmath169 , which implies the following set of inequalities : @xmath170 and leads to a reduction of the allowed domain in the @xmath69 plane . the above two inequalities may be rewritten as @xmath171 the result of these inequalities is shown in fig . [ fig : kinehard ] , where the regions excluded by the requirement @xmath56 $ ] are shaded in dark gray . in order to make the comparison easier with the case of static photons , we have also reproduced the boundaries of the allowed region for that case ( _ i.e . _ the frontiers that one obtains in the limit @xmath172 ) . one can see that the old boundaries are asymptotes for the news ones . again , we have three regions allowed by the above two inequalities for @xmath173 . again , regions i and iii give the same contribution , as can be seen by the change of variables @xmath41 performed at an earlier stage of the calculation . from now on , we will drop region i and multiply by a factor @xmath74 the contribution of region iii . we start the discussion with the study of region iii ( bremsstrahlung processes ) and then turn to region ii ( @xmath14 annihilation with scattering in the plasma , see fig . [ fig : processes ] ) . as will be seen , this region can no longer be neglected , contrary to the case of soft photon production . since the delta function @xmath174 makes more convenient the choice of @xmath175 as an independent variable ( instead of @xmath176 ) , the quantities @xmath177 that appear on the previous figure are to be understood as functions of @xmath81 and the angle @xmath178 between @xmath51 and @xmath43 . therefore , this restriction of the allowed phase space is in fact a constraint that relates the independent variables @xmath81,@xmath178,@xmath59 and @xmath58 . in the following , doing the integration over @xmath178 first will prove to be convenient since , due to collinear divergences , the result of this angular integral controls the order of magnitude of the result . if @xmath179 is the variable over which we integrate first , then the bounds on @xmath180 depend on the other variables @xmath181 and @xmath81 because of the identities of eqs . ( [ eq : hardconstraints2]-a ) and ( [ eq : hardconstraints2]-b ) . in particular , the inequality eq . ( [ eq : hardconstraints2]-a ) can be written in terms of @xmath179 : @xmath182 which gives if one assumes that @xmath58 and @xmath59 are soft @xmath183 where we denote @xmath89 . as a consequence , since the relevant values of the momentum @xmath59 are controlled by thermal masses of order @xmath184 , this upper bound is of order @xmath3 . it is now worth giving expressions for the denominators that enter in the rate , since they are potentially dangerous when the photon is emitted collinearly to the quark . to that purpose , we need also another angular variable , which is left unconstrained by the previous considerations . this variable can be the azimuthal angle @xmath185 between @xmath43 and @xmath52 when projected on a plane orthogonal to @xmath167 . therefore , if we denote @xmath186 the angle between @xmath43 and @xmath52 , we can calculate this angle by @xmath187 where @xmath188 is the angle between @xmath167 and @xmath43 . this last angle is not independent of @xmath178 , to which it can be related by @xmath189 so that eq . ( [ eq : theta2 ] ) can be rewritten as @xmath190 using only the variables @xmath191 , @xmath81 , @xmath59 and @xmath192 , we can write@xcite @xmath193\ ; , \label{eq : r2}\ ] ] and @xmath194^{1/2}}}\ ; , \label{eq : pl2}\ ] ] where we performed the integration over @xmath185 at this stage since this denominator is the only place where @xmath185 appears at the dominant order . as one can see , the first formula remains exactly the same as in the case of soft real photons @xcite , while the second one is only slightly modified by an extra @xmath195 factor . as a consequence , the discussion made in @xcite concerning the enhancement that one gets when performing the @xmath180 integral is still valid . in particular , the terms where both denominators are present are enhanced by a factor of order @xmath196 , while those where only one of them appear will get only a logarithm of this quantity . again , our starting point is eq . ( [ eq : tracetotal ] ) . the order of magnitude of each term is evaluated by taking into account the fact that the momentum @xmath149 is now hard , as well as the quark momentum . moreover , we must take into account the enhancement by a factor of order @xmath11 for terms having two denominators . since the emitted photon is assumed to be real , we have @xmath197 . therefore , one can check that only one term dominates in this matrix element : @xmath198 which is again the same as in the case of soft real photons . since the relevant values of @xmath180 are of order @xmath199 , we have @xmath200{\rm\ \ for\ } { p_o=+\omega_p}\ ; . \label{eq : diff1}\ ] ] therefore , one can see in fig . [ fig : kinehard ] that the extra requirement @xmath173 does not restrict significantly region iii due to eq . ( [ eq : diff1 ] ) ( this is a result already obtained in @xcite according to which the kinematics confines the phase space to the @xmath173 region in the collinear limit if @xmath21 is soft ) . since the term to consider is the same , and since the denominators have very similar expressions , it is completely straightforward to reproduce the calculation for the variable @xmath180 . since we are considering enhanced angular integrals for which the relevant values of @xmath180 are of order @xmath199 , this upper bound does not appear in the result of the @xmath180 integral . ] that has been performed in @xcite . the only minor difference lies in the contraction of @xmath201 with the projectors , which gives a factor @xmath202 approximated by @xmath203 in the collinear limit . introducing the dimensionless quantities : @xmath204 we find & & im^^ar_^(q)(j__t - j__l ) t + & & _ 0^+dp ( p^2+(p+q_o)^2 ) , [ eq : hardresult ] where we denote @xmath205 as one can see , the integral over @xmath86 and @xmath206 can be factorized out of the expression and is the same as in the case of soft real photons @xcite . in this previous work , we noticed that the integral over the gluon momentum could potentially be singular when the gluon is transverse , but is regularized by the _ thermal mass instead of the gluon thermal mass as one might expect . the interpretation of this result is in fact rather simple . indeed , a close look at kinematics shows that the delta functions @xmath174 and @xmath207 become incompatible in the limit @xmath208 if @xmath209 . therefore , the quark thermal mass prevents an infrared divergence by reducing to zero the region of phase space where this divergence can occur . the only major difference with @xcite concerns the integral over @xmath81 , because of the fact that @xmath111 is now hard . nevertheless , we can still give a rather compact expression for this integral in terms of poly - logarithms : & & _ 0^dp ( p^2+(p+q_o)^2 ) [ n_f(p ) - n_f(p+q_0 ) ] + & & = t^3 , [ eq : polylogs ] where the poly - logarithm functions are defined via @xmath210 eq . ( [ eq : polylogs ] ) simplifies in the limit of extremely hard photons @xmath211 : @xmath212 \build{\approx}\over{q_o\gg t}q_o^2 t \ln(2)\ ; .\ ] ] therefore , in this asymptotic regime , we find for the production rate of hard real photons : @xmath213 where @xmath134 is the electric charge of the quark of flavor @xmath135 expressed in units of the electron charge . besides this simple asymptotic result , it is worth adding that eqs . ( [ eq : hardresult ] ) and ( [ eq : polylogs ] ) provide a generalization of the analogous formula of @xcite . indeed , the result we provide in the present paper for the bremsstrahlung contribution to real photon production is valid over the whole range of photon energies , and in particular reduces to eq . ( 55 ) of @xcite in the limit of soft @xmath111 . the discussion of the contribution of region ii can be carried out in a similar way . we recall that @xmath214 so that eq . ( [ eq : r2 ] ) becomes @xmath215\ ; , \ ] ] with the notation @xmath216 . for hard enough @xmath111 , the statistical weight @xmath217 is equal to @xmath113 for @xmath218 and equals @xmath103 everywhere else and @xmath113 takes place in a range of width @xmath1 in the variable @xmath65 . these side - effects are negligeable if the condition @xmath211 is satisfied since they provide corrections of relative order @xmath219 . ] . therefore , we can restrict to the range @xmath220 and we check that eq . ( [ eq : pl2 ] ) remains valid except for the changes @xmath221 and @xmath222 . the enhancement mechanism in the terms that contain the two denominators goes through as before due to their behavior near @xmath223 . such terms are dominated by the region where @xmath223 , which enables us to obtain @xmath224 and @xmath225 . as a consequence , the boundaries in the @xmath69 plane are @xmath226 and @xmath227 , where @xmath89 . therefore , the integral over @xmath59 and @xmath86 gives the same factor @xmath228 as before . the integral over @xmath81 in eq . ( [ eq : hardresult ] ) is now to be replaced by @xmath229 which leads to the following asymptotic contribution for the region ii : @xmath230 therefore , it appears that region ii dominates over bremsstrahlung in the asymptotic regime . the production rate of hard real photons has already been calculated at the one - loop order in @xcite as an application of the effective theory based on the resummation of hard thermal loops . we now compare the contribution of bremsstrahlung obtained above with their result . for hard real photons , the predictions of @xcite are @xmath231 with the constant @xmath232 . it is worth recalling here that the quantities @xmath233 that appear in eqs . ( [ eq : hardfinal ] ) and ( [ eq : hardiifinal ] ) are functions of the ratio @xmath234 when @xmath197 , _ i.e. _ depend only on @xmath235 and @xmath236 . for @xmath237 colors and @xmath238 light flavors , we can evaluate numerically @xmath239 and @xmath240 . the following plot shows a comparison of bremsstrahlung , @xmath14 annihilation with scattering , and one - loop contributions . in this figure , the bremsstrahlung contribution is taken from eq . ( [ eq : hardresult ] ) ; for the region ii , we use the asymptotic result obtained for large @xmath111 , . nevertheless , we expect it to decrease like @xmath111 since its support in the @xmath69 plane decreases like @xmath111 , so that region ii is certainly subdominant for @xmath241 . therefore , the asymptotic result is sufficient in this comparison . ] and the one - loop result is a numerical evaluation of the diagrams considered in @xcite . the value used here for the coupling constant is @xmath242 . we can see that the bremsstrahlung is the dominant contribution for the smaller values @xmath111 , whereas the region ii becomes dominant for hard enough @xmath111 . for intermediate photon energies ( around @xmath1 ) , the three contributions have equivalent orders of magnitude . for higher values of the coupling constant @xmath3 , the relative importance of the one - loop contribution tends to decrease . we can also see that the sum of the three contributions is significantly above the one - loop contribution considered alone . we have discussed the production of a real photon or of a lepton pair in a hot quark - gluon plasma in thermal equilibrium . we assumed that the plasma is described by thermal quantum chromodynamics and we worked in the framework of the effective theory obtained after resummation of hard thermal loops . we have shown that , at leading order in @xmath2 , it is necessary to include the one - loop as well as the two - loop diagrams . the one - loop diagrams of the effective theory correctly account for the contribution of soft fermion momenta in the loop , whereas in the two - loop diagrams hard momenta play a dominant role . many physical processes are contained in the effective theory calculated up to two - loop order . to simplify , it can be said that , at one - loop , scattering processes mediated by ( soft ) fermion exchanges are of paramount importance for the emission of the photon . the corresponding photon production rate consequently strongly depends on the soft fermion thermal mass @xmath243 . on the other hand , the two - loop topologies account , among other possibilities , for bremsstrahlung processes where the photon is emitted by a hard quark scattering in the plasma via a ( space - like ) gluon exchange : in that case the rate is proportional to the square of the gluon mass @xmath7 . we have calculated the contribution of such processes . for the production a soft virtual photon ( @xmath244 , @xmath245 ) the bremsstrahlung processes largely dominate over the one - loop result in the range @xmath246 . for the case of real photon production an interesting enhancement phenomenon occurs in the bremsstrahlung processes . because of the vanishing photon mass , the fermion propagators become infinite when a quasi forward - scattered quark emits a collinear photon : the singularities are regularized by an interplay between the thermal masses of the fermion and the gluon . it was seen before that this leads to an enhancement by a factor @xmath11 of the bremsstrahlung contributions to the real soft photon rate so that the two - loop diagrams entirely dominate over the one - loop contribution . on the other hand , the bremsstrahlung production of a hard real photon occurs at the same order in @xmath3 as the one - loop result . a rather simple analytic expression , valid for soft , hard and ultra - hard photon has been derived . for ultra - hard photons , another process becomes dominant , consisting of @xmath14 annihilation where one of the fermions undergoes a scattering in the medium . in all cases , the calculated two - loop contributions considerably increase the rate of photon or lepton pair emission . our results can easily be numerically extended to cover , on the one hand , the case of a soft lepton pair at non - vanishing momentum , and on the other hand , the case of lepton pairs produced at large momentum with a small invariant mass . all the features of the present results should survive in these more general situations . a word of caution should be given concerning the bremsstrahlung contribution to the soft virtual photon rate . the result shows a sensitivity to a hard space - like gluon and therefore one may suspect that the extrapolation of the effective gluon propagator in the hard region is not complete : to be consistent may require taking into account three - loop diagrams in very much the same way two - loop diagrams were needed besides the one - loop diagram with a hard space - like fermion propagator . on the contrary , for the case of a real photon emission , the bremsstrahlung rate is sensitive only to soft gluon momenta , and the calculation is therefore expected to be complete . our study does not cover all the physical processes included in the two - loop diagrams . in particular , the contribution with a time - like gluon should be added . however such processes when the gluon is hard are already contained in the one - loop diagram : counterterms should therefore be included . to do this will be a very interesting practical exercise in the use of the effective theory up to two - loop . we thank r. baier for useful discussions . the work of rk was supported by the natural sciences and engineering research council of canada . we also acknowledge support by nato under grant crg . when using effective theories based on the summation of hard thermal loops at higher orders , there is potentially a possibility to have multiple counting of thermal corrections that should be there only once . as a first illustration of this problem in a trivial context , let us use the example of a massless real scalar field with a @xmath247 interaction in @xmath101 dimensions . the lagrangian of such a model is @xmath248 after the calculation of the one - loop tadpole , one realizes that this diagram a hard thermal loop in the terminology of @xcite generates a thermal mass @xmath249 that can be important for the phenomenology of soft modes . therefore , the idea of the htl resummation is to include this thermal mass in an effective lagrangian @xmath250 which is the lagrangian of a real scalar field of mass @xmath251 . let us assume now that one uses this effective theory instead of the bare one to calculate the same tadpole diagram . the result would be @xmath252 and since we start now from a propagator with squared mass @xmath253 , the resummation of the self - energy @xmath254 would lead us to a propagator with a squared mass @xmath255 , approximately twice larger than the correct thermal mass . obviously , the above result arises due to multiple countings of the same thermal correction . stated differently , this is a consequence of the fact that this effective theory is more than a mere reordering of the perturbative expansion of the bare theory since its lagrangian is different . to solve this problem , one must write @xmath256 and treat the counterterm @xmath257 as an interaction term , just like @xmath258 . the effect of this counterterm is of course to subtract at higher order the thermal corrections that had already been included at the tree level via the effective lagrangian , in order to avoid multiple countings . for instance , in the above example , the correct answer for the tadpole @xmath254 when one takes care of the counterterm is @xmath259 which is a perturbative correction to the thermal mass found at the previous step , as it should be . when this counterterm is taken properly into account , the effective theory is nothing more than a reordering of the perturbative expansion , since the overall lagrangian remains unmodified . this problem also arises in effective gauge theories , where the situation is a bit more complicated since we need there an infinite series one for each hard thermal loop of non - local counterterms . these counterterms are defined to be the opposite of the htl contribution to the corresponding function . then , to a given diagram obtained in the effective theory , one should add the diagrams obtained by collapsing each loop in turn and replacing it by the corresponding counterterm . in order to be more definite , let us consider the example of the two - loop diagram of figure [ fig:2loop]-[b ] . in the figure [ fig : counterterms ] , we have represented next to the one - loop contribution to the polarization tensor of the photon some of its two - loop corrections . it should be clear from the figure that when the boldface loop of the second diagram carries a hard momentum , then this loop reproduces the htl part of the effective vertex in the dotted box already included at one - loop , plus new sub - dominant perturbative corrections . the purpose of the third diagram is precisely to subtract a quantity equal to the htl contribution to this vertex , so that what remains constitutes only new contributions . the net effect of this procedure is thus to reorder the terms of the perturbative expansion . let us now explain why , despite their conceptual importance , the counterterms do not contribute in the case of bremsstrahlung . as said in section [ sec : bremss ] , the contributions of bremsstrahlung to photon production come from two - loop diagrams in which the gluon propagator is cut , and where one retains only the landau damping part ( @xmath24 ) of the cut gluon propagator . technically , the gluon included in the 3-point htl that appears on the third diagram of figure [ fig : counterterms ] is a bare one , and therefore its discontinuity has support totally included in the time - like region . as a consequence , the diagram with counterterms contributes only in the region of phase - space where @xmath260 , and can not contribute to bremsstrahlung . even if they are not a worry in the case of bremsstrahlung , it was important to discuss the potential effect of counterterms since , having shown that the second diagram gives an important contribution , it is not enough to conclude that the two - loop order is important , as it may be canceled by counterterms . the purpose of this appendix is to emphasize the importance of being in a thermal bath in order to have a hard phase space for the quark loop . this is indeed crucial in the calculations performed in the previous sections since this feature of the diagrams considered in this paper enables them to be of the same order of magnitude as one - loop diagrams . in order to make the following discussion more intuitive , let us first transform the bremsstrahlung photon production rate in a way that separates more clearly the phase space from the amplitude of the process producing the photon . the tools to do that have already been presented in @xcite ( see the section devoted to the comparison with semi - classical methods ) , so that we only give the result here : 1=to 5 cm where @xmath261 . this formula tells us that in order to calculate the contribution of some process to the photon production rate , we just have to integrate the amplitude squared of this process over the phase space of unobserved particles ( here , the incoming and outgoing quarks ) . when doing this integration , the external particles are put on their mass shells , and accompanied by the appropriate statistical weight . the main advantage of this formula is that it exhibits a clear separation in two factors : the amplitude squared of the process producing the photon , and the phase space of the quarks . therefore , this formula can be used to separate the estimation of the order of magnitude of the diagram in two steps : the order of magnitude of the amplitude , and the size of the phase space over which it must be integrated . looking at this formula , it is clear that the effect of the thermal bath appears only in the phase space . indeed , the amplitude that appears in eq . ( [ eq : lastform ] ) is nothing but a zero temperature one ( it does not contain any statistical weight ) . stated differently , if , instead of looking at photon production by a thermal bath , we were looking at bremsstrahlung photon production in p - p collisions ( _ i.e. _ the two scattering quarks come from protons in colliding beams ) , the statistical factors of incoming quarks would have to be replaced by structure functions of a quark inside a proton , and the statistical weights of outgoing quarks would be replaced by 1 . this difference is precisely the point which makes the thermal bath dramatically different from the p - p collision . indeed , in the case of the thermal bath , the statistical functions have a support which is 3-dimensional since the plasma is an isotropic medium . on the other hand , the structure function of the quark inside protons of the beam is vanishing if the quark has a direction different from that of the beam . this will make a difference when one performs the integral over the momentum @xmath81 . indeed , if one has something like @xmath262 in the case of the thermal bath , the integral would be @xmath263 for the case of p - p collisions . as a consequence , the integral over the quark momentum is more likely to be sensitive to hard momenta in the case of the thermal bath . therefore , even if it is not a rigorous proof , these considerations show why some higher order processes which are not dominant at zero temperature may become dominant in a plasma , due to a bigger size for their phase space . moreover , eq . ( [ eq : lastform ] ) may help to understand why two - loop contributions may be as important as one - loop ones . indeed , it shows that the final order of magnitude of a contribution results from a competition between two effects . the first one is the order of magnitude of the amplitude , which usually becomes smaller when the number of loops increases since the number of coupling constants increases also . the second aspect of the problem is the size of the phase space , since it may happen that due to kinematical constraints , the one - loop phase space is much smaller than the two - loop one . both effects can compensate so that two - loop diagrams contribute also at the dominant level . this is precisely what happens in the case of photon production by a plasma : the one - loop phase space is soft due to kinematical constraints , while the two - loop thermal phase space can be hard ( as explained above , this is possible because we are in an isotropic medium ) . therefore , the smallness of the two - loop amplitude is compensated by the size of the two - loop phase space .

we examine the computation of the nucleation barrier used in the expression for false vacuum decay rates in finite temperature field theory . by a detailed analysis of the determinantal prefactor , we show that the correct bounce solution used in the computation of the nucleation barrier should not include loop corrections coming from the scalar field undergoing decay . temperature corrections to the bounce appear from loop contributions from other fields coupled to the scalar field . we compute the nucleation barrier for a model of scalar fields coupled to fermions , and compare our results to the expression commonly used in the literature . we find that , for large enough self - couplings , the inclusion of scalar loops in the expression of the nucleation barrier leads to an underestimate of the decay rate in the neighborhood of the critical temperature . pacs number(s ) : 98.80.cq , 64.60.qb . + e - mail : gleiser@peterpan.dartmouth.edu ; marques@phys.tamu.edu ; rudnei@northstar.dartmouth.edu .

hep-ph9304234

for the past decade or so , the study of first - order phase transitions in cosmology has been the focus of much interest due to their possible relevance to the physics of the early universe . some well - known examples are inflationary models @xcite , the quark - hadron transition @xcite and , more recently , the generation of the cosmological baryon asymmetry in the electroweak phase transition @xcite . in a first - order phase transition , the initial metastable phase decays to the stable phase by the nucleation of bubbles larger than a critical size . this decay may be triggered by either quantum or thermal fluctuations , depending on how the ambient temperature compares to the nucleation barrier @xcite . within a cosmological context , the cooling is provided by the expansion of the universe ; the long - wavelength modes of the order parameter responsible for the symmetry breaking transition are coupled to the `` environment '' , which is assumed to be in local thermal equilibrium at some temperature @xmath0 . ( here , we are mainly concerned with `` late '' transitions , for which the typical fluctuation time - scales are much shorter than the expansion rate . ) of great relevance to the understanding of the evolution of the phase transition is the determination of the bubble nucleation rate per unit volume . this is a well - known problem in classical statistical mechanics , with an long history @xcite . phenomenological field - theoretic treatments were developed by cahn and hilliard @xcite , and by langer @xcite in the context of a time - dependent coarse - grained ginzburg - landau model . classical homogeneous nucleation theory within a field theoretic context has been recently shown by numerical experiments to successfully predict the nucleation barrier @xcite . in the case of zero - temperature quantum field theory , the study of metastable vacuum decay was initiated with the work of voloshin , kobzarev , and okun @xcite , and was put onto firm theoretical ground by coleman and callan in the late seventies @xcite . finite temperature corrections to the vacuum decay rate were first considered by linde @xcite , who argued that temperature corrections to the nucleation rate are obtained recalling that finite temperature field theory ( at sufficiently high temperatures ) in @xmath1-dimensions is equivalent to @xmath2-dimensional euclidean quantum field theory with @xmath3 substituted by @xmath0 . thus , in @xmath1 dimensions , the nucleation rate is proportional to @xmath4 $ ] , where @xmath5 is the @xmath2-dimensional euclidean action evaluated at its extremum ( specifically , a saddle point ) , the critical bubble , or bounce , @xmath6 . the usual expression for the nucleation rate per unit volume used in the literature is @xcite @xmath7 } { \det ' [ - \nabla^{2 } + v_{\rm { eff}}''(\varphi_b(r , t),t ) ] } \right\}^ { \frac{1}{2 } } { \rm exp}\left [ - { { s_e^3 ( \varphi_b(r , t),t)}\over t}\right ] \ : , \label{badrate}\ ] ] where @xmath8 is the value of the field @xmath9 at the metastable minimum , and the prime in the determinant in the denominator is a reminder that one should omit the zero and negative eigenvalues , associated with the translation symmetry of the bubble and with its instability ( being a saddle point configuration ) , respectively . @xmath10 is the one - loop approximation to the finite temperature effective potential , and @xmath11 . there are three important points here . the first is that in order to estimate the determinantal prefactor ( the `` equilibrium '' part of the prefactor ; there is a dynamic factor which can not be obtained by using equilibrium arguments ) one usually proceeds by invoking dimensional arguments to approximate it by a term of order @xmath12 ( @xmath13 is the critical temperature ) . how good is this approximation ? clearly , in most cases it is impossible to evaluate the determinants exactly . but can one obtain a better approximation than the simple use of dimensional arguments ? the second , and most important , point is that the critical bubble configuration @xmath14 used to evaluate the nucleation barrier , denoted above by @xmath15 , was obtained from an effective potential which includes corrections coming from scalar loops . hence the temperature dependence in @xmath16 . we will show here that this procedure is not in general justified and is only a good approximation if the corrections from scalar loops are negligible . finally , in the expression for the temperature corrected barrier in eq . ( [ badrate ] ) , @xmath17 , one uses the _ temperature corrected _ effective potential , @xmath18 as opposed to the tree level potential . thus , it is claimed that @xmath17 is equivalent to the free energy of the temperature dependent bounce , given by @xmath19 \ : . \label{badact}\ ] ] as far as we know , apart from the work of affleck in the context of finite temperature quantum mechanics @xcite , this point has never been properly addressed in the literature . how did the temperature corrected potential appear in the exponent ? is the exponentiation of the massless modes sufficient ? in fact , most of the work done on cosmological phase transitions in which temperature effects are important ( including some by the present authors ) simply invokes linde s results . given the many applications of finite temperature vacuum decay in cosmology , we feel that this important question should not be left unscrutinized . this concern has also been expressed in recent works by csernai and kapusta @xcite , and by buchm@xmath20ller , helbig , and walliser @xcite . both works attempted to improve on linde s results , by generalizations of langer s work . csernai and kapusta obtained an expression for the dynamical prefactor by using a relativistic hydrodynamic approach , while buchm@xmath20ller _ et al . _ obtained an approximate expression for the decay rate in scalar electrodynamics ( and more recently , with z. fodor , in the standard electroweak model ) by integrating out the electromagnetic degrees of freedom from the partition function . however , a more detailed analysis of the nucleation barrier and how it compares to the usual result is still lacking . in this paper we address the three points raised above . we will be mostly interested in the regime in which thermal fluctuations are much larger than quantum fluctuations . this way we avoid the question of how to match continuously the two regimes , although we believe this to be a very important question @xcite . ( see also refs . @xcite and @xcite . ) the hope is that in most situations of interest the transition will be dominated by one or the other regime . by a saddle - point evaluation of the partition function in the case of a self - interacting scalar field , it is possible to show that the temperature corrections to the nucleation barrier can be interpreted as entropic contributions due to stable vibrational modes on the _ tree - level _ bounce configuration , @xmath6 . in other words , the first corrections to the energy of the critical bubble configuration come from temperature induced stable fluctuations on the bubble , which will modify its volume and surface energies . we will show that these corrections are given by the temperature corrected effective action evaluated at the tree - level bounce @xmath6 . in expression ( [ badrate ] ) , the bounce is obtained from the effective potential which includes scalar loops . the difference between the two nucleation barriers will be important whenever scalar loops are not negligible . we will show that they become particularly important within the so - called thin - wall limit , that is , in the vicinity of the critical temperature for the transition . this is perfectly consistent with the fact that large entropic corrections are expected near the critical temperature . we will obtain this result by a perturbative evaluation of the determinantal prefactor . in principle , the determinantal prefactor can be evaluated in two ways . clearly , the computation can be done directly if we know the eigenvalues related to a given bubble configuration . this method is not very useful in practice , since we in general do not know the eigenvalues . ( unless , of course , we obtain them numerically . ) writing down explicitly the eingenvalue equations , and using a thin - wall approximation to the bubble configuration , we show how the temperature corrections to the nucleation barrier originate from fluctuations about the critical bubble configuration . even though the thin - wall approximation is not very useful in realistic situations , there is no reason to believe that thicker wall bubbles will behave any differently . ( unless the transition becomes too weak , in which case nucleation of critical bubbles may not be the relevant mechanism for the transition @xcite . ) the second approach we use to evaluate the prefactor relies on a perturbative expansion of the determinants . within first - order , it is again possible to show how the prefactor accounts for the temperature corrections to the nucleation barrier . the paper is organized as follows . in section 2 we briefly review langer s formalism for obtaining nucleation rates , adapted to field theory at finite temperatures . that is , we obtain the partition function for the metastable phase plus a nucleating fluctuation by a saddle - point evaluation of the functional integral . in section 3 we show how the determinantal prefactor can account for the finite temperature corrections to the nucleation barrier . for simplicity , the calculation is performed in the context of the thin - wall approximation for the bubble profile , although in principle one could obtain results for any configuration . in sections 2 and 3 , for the sake of clarity , the discussion is somewhat oversimplified . we _ assume _ that the system is initially in a metastable state and study only the scalar degrees of freedom in the problem . this situation is not unrealistic , as it can be reproduced in numerical simulations of vacuum decay @xcite . in section 4 we study a model of a scalar field coupled to fermions . this example is particularly interesting as it illustrates how a thermal state evolves into a metastable state due to radiative corrections , very much like in the standard electroweak model . ( recall that in a cosmological context the cooling is provided by the expansion of the universe . ) a related problem has been recently studied by e. weinberg , in the context of massless scalar models for which symmetry breaking occurs due to radiative corrections . in the regime dominated by quantum fluctuations , weinberg showed how radiative corrections induce a metastable state and how it is possible to evaluate its decay rate@xcite . we show how to obtain an effective partition function for the scalar field by integrating out the fermionic modes , and proceed to obtain the decay rate . we then compare our results for the nucleation barrier to the results obtained using eq . ( [ badrate ] ) . conclusions are presented in section 5 and two appendices are included to clarify a few technical points . consider a scalar field model with four - dimensional euclidean action @xmath21 where @xmath22 is the euclidean lagrangian density given by @xmath23 and the potential in ( [ lphih ] ) has a metastable minimum at @xmath24 and a stable minimum at @xmath25 , as shown in fig . 1 . [ we will only consider potentials with two minima here . ] assume _ that the system is prepared initially in the metastable phase , without worrying for the moment about how this is done . ( see section 4 . ) the metastable phase will decay into the stable phase by the nucleation of bubbles larger than a critical size . ( for a review see , e.g. , @xcite . ) as is well - known , in order to study the decay of the false vacuum at finite temperature we impose the periodic boundary condition ( anti - periodic for fermions ) @xmath26 , so that the euclidean action becomes @xcite @xmath27=\int_0^{\beta}d\tau\int d^3x\left [ { 1\over 2 } \left ( { { \partial \phi}\over { \partial \tau}}\right ) ^2+{1\over 2 } \left ( \nabla\phi \right ) ^2 + v(\phi ) \right ] \ : . \label{seuclt}\ ] ] the partition function of the system is given by a functional integral over all possible field configurations weighted by their euclidean action , @xmath28 following langer @xcite , we describe the nucleation of bubbles of the stable phase inside the metastable phase under the assumption that a dilute gas approximation for these droplets is valid . unless the transition is weakly first - order , this should be a very good approximation to describe the early stages of the transition , when bubble collisions and other complicated kinetic effects can be neglected . the critical configuration is an extremum of the euclidean action , @xmath29 being thus a solution of the equation of motion , @xmath30 with boundary conditions , @xmath31 , and @xmath32 . coleman , glaser , and martin @xcite , have shown that the configuration with minimum energy , _ i.e. _ , the minimum of all the maxima , will have @xmath33-symmetry . as argued by linde @xcite , for sufficiently high temperatures the problem becomes effectively three - dimensional , and the saddle point will be given by the @xmath34-symmetric , or static , solution of @xmath35 with boundary conditions , @xmath36 and @xmath37 . note that the potential in ( [ eulersph ] ) is the _ tree - level _ potential . for future reference we note that when the false vacuum energy density [ @xmath38 is much smaller than the barrier height [ @xmath39 , see fig . 1 ] , the bubble radius @xmath40 is much larger than the wall thickness @xmath41 , where @xmath42 is a typical mass scale in the problem . in this case , the solution to eq . ( [ eulersph ] ) can be estimated by the so - called thin - wall approximation and is given by @xmath43 which describes a bubble of radius @xmath40 of the stable phase @xmath44 embedded in the metastable phase @xmath45 . @xmath46 describes the bubble wall separating the two phases . the main advantage of langer s approach is that in a dilute gas of bubbles one can infer the thermodynamics of the system from the knowledge of the partition function of a single bubble . we can write the partition function for the system with a bubble of the stable vacuum inside the metastable vacuum as @xmath47 where @xmath48 and @xmath49 are the partition functions of the system for the vacuum field configuration @xmath45 and for the bubble field configuration @xmath50 , respectively @xcite . the generalization of ( [ z1bubb ] ) for several bubbles is given by @xmath51 + z ( \varphi_{f } ) \frac{1}{2 ! } \left [ \frac{z(\varphi_{b})}{z(\varphi_{f } ) } \right]^{2 } + \ldots \nonumber \\ & \simeq & z(\varphi_{f } ) \exp \left [ \frac{z(\varphi_{b})}{z(\varphi_{f } ) } \right ] \ : . \label{zmany}\end{aligned}\ ] ] the proof of ( [ zmany ] ) can be found in the work of arnold and mclerran @xcite , who studied the properties of a dilute gas of sphalerons . they expressed the multiple - sphaleron configurations as the superposition of many single sphalerons , with partition function approximated as above . the partition functions in ( [ zmany ] ) can be evaluated by the saddle - point method , expanding the lagrangian field in ( [ zphih ] ) as @xmath52 for @xmath49 and @xmath53 for @xmath48 . @xmath54 and @xmath55 are small perturbations around the classical field configurations @xmath56 and @xmath57 , respectively . these perturbations around each configuration bring the temperature corrections to the nucleation barrier into the problem , as we shall see . up to 1loop order one keeps the quadratic terms in the fluctuations @xmath54 and @xmath55 in the expansion of the scalar field in the partition function ( [ zphih ] ) . this way one can write the following expressions for @xmath49 and @xmath48 , respectively , @xmath58 \eta \right\ } \label{zphib}\ ] ] and @xmath59 \zeta \right\ } \ : , \label{zphif}\ ] ] where @xmath60 and @xmath61 . performing the functional gaussian integrals in ( [ zphib ] ) and ( [ zphif ] ) one gets the following expression for the ratio between the partition functions , @xmath62 , appearing in ( [ zmany ] ) : @xmath63^{-\frac{1}{2 } } e^{-\delta s } \ : , \label{ratiodet}\ ] ] where @xmath64^{- \frac{1}{2 } } \equiv \int d \eta \exp \left\ { - \int_{0}^{\beta } d \tau \int d^{3 } x \frac{1}{2 } \eta [ m ] \eta \right\}$ ] and @xmath65 is the difference between the euclidean actions for the field configurations @xmath50 and @xmath45 . note that @xmath66 , and hence @xmath67 , _ does not include any temperature corrections . _ from ( [ zmany ] ) and ( [ ratiodet ] ) , the free energy of the system , @xmath68 can be written as @xmath69^{-\frac{1}{2 } } e^{-\delta s } \ : . \label{energy}\ ] ] as is well - known , the determinantal prefactor evaluated for the bounce configuration has one negative eigenvalue , signalling the presence of a metastable state , and also three zero eigenvalues related with the translational invariance of the bubble in three - dimensional space . because of the negative eigenvalue , the free energy @xmath70 is imaginary . as shown by langer @xcite ( see also ref . @xcite ) , the decay rate is proportional to the imaginary part of @xmath70 @xmath71 where @xmath72 is the single negative eigenvalue . in general it depends on non - equilibrium aspects of the dynamics , such as the coupling strength to the thermal bath . in this section we compute the ratio of the determinants appearing in the decay rate , and show how it provides a finite temperature correction to the nucleation barrier . first recall that for static field configurations @xmath67 is given by @xmath73 = \frac{\delta e}{t } \ : , \label{deltas}\ ] ] where @xmath74 is simply the nucleation _ energy _ barrier , that is , the energy of a critical nucleation within the metastable phase . for example , in the thin - wall approximation of eq . ( [ bubble ] ) , @xmath74 is @xmath75 where @xmath76 is the tree - level surface tension of the bubble wall ( _ i.e. _ with no corrections due to fluctuations around the bubble wall field configuration @xmath77 ) @xmath78 \ : . \label{sigma0}\ ] ] using ( [ deltas ] ) in eqs . ( [ energy ] ) and ( [ gamma ] ) we can write the decay rate as @xmath79^{-\frac{1}{2 } } { \rm exp } \left ( -{{\delta e } \over t}\right ) \ : . \label{gamma2}\ ] ] the determinantal prefactor in ( [ gamma2 ] ) will provide the temperature corrections to the nucleation barrier . this should come as no surprise , given that the determinant is obtained from integrating over thermally induced fluctuations about the tree - level bubble configuration . ( recall that we are only considering the regime in which thermal fluctuations are much larger than quantum fluctuations . ) once the negative and zero eigenvalues are taken care of , the positive eigenvalues are easily associated with entropic contributions to the activation energy due to stable deformations of the bubble s shape , as in classical nucleation theory . we now proceed to show how to incorporate temperature corrections to the nucleation barrier . this can be done without an explicit evaluation of the eigenvalues of the operators in the determinants . as we show next , all that we need is to separate consistently the positive eigenvalues from the negative and zero eigenvalues , and then show how the former can be exponentiated . in principle , the computation of the determinantal prefactor in ( [ energy ] ) can be performed by two different methods . the first involves obtaining directly ( analytically , or more realistically , numerically ) the eigenvalues for the determinants in eq . ( [ energy ] ) . the second method consists in developing a consistent perturbative expansion for the ratio of the determinants . we now examine both these possibilities . consider the eigenvalue equations for the differential operators that appear in the determinantal prefactor , @xmath80 \phi_{f}(i ) = \varepsilon_{f}^{2}(i ) \phi_{f}(i ) \label{valuephif}\ ] ] and @xmath81 \phi_{b}(i ) = \varepsilon_{b}^{2}(i ) \phi_{b}(i ) \ : . \label{valuephib}\ ] ] in momentum space one writes , @xmath82 , where @xmath83 , for bosons ( for fermion fields @xmath84 ) . from ( [ valuephif ] ) and ( [ valuephib ] ) one can write the determinant ratio in ( [ energy ] ) as @xmath85^{\frac{1}{2 } } & = & \exp \left\ { \frac{1}{2 } \ln \left [ \frac { \det ( -\box_{e } + v''(\varphi_{f}))_{\beta } } { \det ( -\box_{e } + v''(\varphi_{b}))_{\beta } } \right ] \right\ } = \nonumber \\ & = & \exp \left\ { \frac{1}{2 } \ln \left [ \frac { \prod_{n=- \infty}^{+ \infty } \prod_{i } \left ( \omega_{n}^{2 } + e_{f}^{2}(i ) \right ) } { \prod_{n= - \infty}^{+ \infty } \prod_{j } \left ( \omega_{n}^{2 } + e_{b}^{2}(j ) \right ) } \right ] \right\ } \ : . \label{ratio}\end{aligned}\ ] ] using the identity , @xmath86 and taking into account that we have in the denominator of ( [ ratio ] ) one negative and three zero eigenvalues , one obtains ( for details see appendix a ) @xmath85^{\frac{1}{2 } } & = & { \cal v}\frac{t^{4}}{i |e_{-}| } \frac{\beta \frac{|e_{-}|}{2 } } { \sin \left ( \beta \frac{|e_{-}|}{2 } \right ) } \left [ \frac{\delta e}{2 \pi t } \right]^{\frac{3}{2 } } \nonumber \\ & \times & \exp \left\ { \sum_{i } \left [ \frac{\beta}{2 } e_{f}(i ) + \ln \left ( 1 - e^{- \beta e_{f}(i ) } \right ) \right ] \right . \nonumber \\ & - & \left . \sum_{j } \ ; ' \left [ \frac{\beta}{2 } e_{b}(j ) + \ln \left ( 1 - e^{- \beta e_{b}(j ) } \right ) \right ] \right\ } \ : . \label{eigenratio}\end{aligned}\ ] ] the factor @xmath87^{\frac{3}{2 } } $ ] in the right hand side of eq . ( [ eigenratio ] ) comes from the contribution of the zero eigenvalues , which can be handled as in ref . @xcite , through the use of collective coordinates corresponding to the position of the bubble . @xmath88 is the volume of three space , while the prime in @xmath89 is a reminder that we have excluded the negative and the three zero eigenvalues from the sum . thus , the argument of the exponential incorporates only the contributions from the stable vibrational modes on the bubble , very much like in langer s result @xcite . substituting eq . ( [ eigenratio ] ) in ( [ gamma2 ] ) one obtains the following expression for the nucleation rate per unit volume , @xmath90 , as defined in eq . ( [ gamma ] ) : @xmath91 \ : , \label{endgamma}\ ] ] where we have denoted by @xmath92 the dimensionless factor @xmath93 note that the zero - mode contribution in the prefactor depends on the energy barrier of the critical nucleation , and not on its free energy as in eq . ( [ badrate ] ) . in eq . ( [ endgamma ] ) we have incorporated the exponential contribution from the prefactor into the definition of the temperature corrected nucleation barrier , which we call @xmath94 . within the thin - wall approximation we can write @xmath95 where @xmath96 - \nonumber \\ & - & t \int \frac{d^{3 } k}{(2 \pi)^{3 } } \ln \left [ 1 - e^{- \beta \sqrt { \vec{k}^ { 2 } + m^{2}(\varphi_{t } ) } } \right ] \label{deltau}\end{aligned}\ ] ] and @xmath97 - \right . \nonumber \\ & - & \left . \sum_{i } \left [ \frac { \beta}{2 } e_{f}(i ) + \ln \left ( 1 - e^{- \beta e_{f}(i ) } \right ) \right ] \right\ } \ : . \label{sigmat}\end{aligned}\ ] ] in eq . ( [ deltau ] ) we have substituted the discrete sums by integrals over momenta . [ for field configurations @xmath45 and @xmath44 , we have the continuum eigenvalues , @xmath98 and @xmath99 , respectively , with @xmath100 and @xmath101 , from eq . ( [ deltau ] ) since they can always be handled by the introduction of the usual counterterms @xcite . in eq . ( [ sigmat ] ) @xmath102 are the eigenvalues related with the bubble wall field configuration @xmath77 . thus , within the thin - wall approximation , the problem is reduced to the computation of these eigenvalues for a field configuration describing the bubble wall , a non - trivial task . it is possible , however , that the temperature corrections to the surface density are negligible . for example , in the context of the qcd transition , the surface density is obtained from lattice calculations , and is shown not to be very sensitive to temperature @xcite . it is easy to see from eq . ( [ deltau ] ) that @xmath103 is the usual 1-loop approximation to the finite temperature false vacuum energy density @xcite . the second term in the right hand side of ( [ sigmat ] ) comes from the finite temperature 1-loop contribution to the surface tension @xmath76 , due to thermal fluctuations on the bubble wall . thus , by exponentiating the contribution from the stable modes , the activation energy for the critical bubble becomes indeed an activation free energy . note however , that contrary to eq . ( [ badrate ] ) , the free - energy functional is evaluated for the tree - level bounce . in section 4 we will compare the results obtained with the two approaches . a second approach to the computation of the determinantal prefactor in ( [ gamma2 ] ) consists in developing a perturbative expansion for it . first we write the determinantal ratio as @xmath104^{\frac{1}{2 } } & = & \exp \left\ { \frac{1}{2 } { \rm tr } \ ; \ln \left[-\box_{e } + v''(\varphi_{f } ) \right]_{\beta } - \right . \nonumber \\ & - & \left . \frac{1}{2 } { \rm tr ' } \ ; \ln \left[-\box_{e } + v''(\varphi_{b } ) \right]_{\beta } \right\ } \ : , \label{fieldratio}\end{aligned}\ ] ] where we have used in ( [ fieldratio ] ) the identity @xmath105 and the prime in both sides denote that the negative and the zero modes have been omitted . ( they are treated as in previous section . ) we rewrite ( [ fieldratio ] ) as @xmath106^{\frac{1}{2 } } = \exp \left\ { - \frac{1}{2 } { \rm tr } \ ; \ln \bigl [ 1 + g_{\beta}(\varphi_{f } ) \left [ v''(\varphi_{b } ) - v''(\varphi_{f } ) \right ] \bigr ] \right\ } \ : , \label{expratio}\ ] ] where @xmath107 is just the propagator for the scalar field @xmath9 , with @xmath108 . expanding the logarithm in ( [ expratio ] ) in powers of @xmath109 $ ] , we obtain the graphic representation , @xmath110 \right \ } = \ols \ ! \ ! \ ! \ ! \ ! \ ! \ ! \ ! \ ! \ ! \ ! \ ! \ ! + \olr + \olk + \ : \ : \ldots \ : , \label{graphic}\ ] ] where the dashed lines correspond to the background `` field '' @xmath111 $ ] and the internal lines denote the propagator @xmath112 . the expression ( [ graphic ] ) can be written as @xmath113 \right\ } & = & \sum_{m=1}^{+ \infty } \frac { ( -1)^{m+1 } } { m } \int d^{3 } x \left[v''(\varphi_{b } ) - v''(\varphi_{f } ) \right]^{m } \times \nonumber \\ & \times & \sum_{n= -\infty}^{+ \infty } \int \frac { d^{3 } k } { ( 2 \pi)^{3 } } \frac{1 } { \left [ \omega_{n}^{2 } + \vec{k}^{2 } + m^{2}(\varphi_{f } ) \right]^{m } } \ : . \label{traceln}\end{aligned}\ ] ] the sum in @xmath42 can be performed and one obtains @xmath114 \right\ } = \int d^{3 } x \sum_{n= -\infty}^{+ \infty } \int \frac { d^{3 } k } { ( 2 \pi)^{3 } } \ln \left [ 1 + \frac { v''(\varphi_{b } ) - v''(\varphi_{f } ) } { \omega_{n}^{2 } + \vec{k}^{2 } + m^{2}(\varphi_{f } ) } \right ] \ : . \label{endtrln}\ ] ] this expression can be further simplified by means of the identity , ( @xmath115 , and @xmath116 ) @xmath117 & = & \beta e_{b}(\vec{k } ) + 2 \ln \left ( 1 - e^{- \beta e_{b}(\vec{k } ) } \right ) - \nonumber \\ & - & \beta e_{f}(\vec{k } ) - 2 \ln \left ( 1 - e^{- \beta e_{f}(\vec{k } ) } \right ) \ : . \label{sumln}\end{aligned}\ ] ] the terms proportional to @xmath118 can be renormalized . we are left with the familiar temperature corrected effective potential ( we neglect terms coming from zero temperature quantum corrections ) @xcite , @xmath119 within the thin - wall limit , using ( [ sumln ] ) into ( [ endtrln ] ) , and substituting into eq . ( [ expratio ] ) , we obtain , from eq . ( [ gamma2 ] ) the temperature corrected barrier , @xmath120 where @xmath121 \label{deltaveff}\ ] ] and @xmath122 \ : . \label{fieldsigmat}\ ] ] expression ( [ deltaveff ] ) for the temperature corrected false vacuum energy density can be easily identified with @xmath103 , as defined in eq . ( [ deltau ] ) . also , expression ( [ fieldsigmat ] ) for @xmath123 can be identified with the temperature corrected surface tension of the bubble wall , by writing it as @xmath124 \ : , \label{defsigmat}\ ] ] with @xmath125 where @xmath126 is the free energy for the field configuration @xmath127 . @xmath123 , as given by ( [ defsigmat ] ) , is then the free energy difference ( per unit area of the bubble ) between the two configurations ( @xmath77 and @xmath45 ) , which defines the surface tension . as shown in ref . @xcite , the expressions for the free energies in ( [ defsigmat ] ) , at 1-loop order give ( [ fieldsigmat ] ) . in the previous two sections we have obtained the temperature effects on the nucleation barrier by _ assuming _ that the system becomes trapped in a metastable phase as the temperature drops below the critical temperature @xmath13 . this assumption is not very realistic , and was addopted so that we could stress the main points of the calculation without having to worry about the effects of interactions of the order parameter with other fields . however , in models of interest within a cosmological context , such as the standard electroweak model and some of its extensions , the universe becomes trapped in a metastable phase due to the interactions of the higgs field ( or effective scalar order parameter ) with other massless fields , such as gauge bosons or fermions . in fact , it is possible to formulate this statement in terms of conditions that a general effective potential must satisfy , if it is to develop a metastable phase below @xmath13 @xcite . basically , the condition states that the mass gap between the symmetric and broken - symmetric phases must be large enough so that massive fluctuations away from the symmetric minimum are suppressed . since the mass gap is given in terms of the ( temperature dependent ) vacuum expectation value of the higgs field and of its coupling to other fields , the condition states that the transition can not be too weakly first order . the question then is what effective potential should be considered when calculating the decay rate . as recently shown by e. weinberg for the zero - temperature coleman - weinberg models ( which exhibit symmetry breaking only due to radiative corrections ) , the effective potential relevant in the calculation of the bounce solution is obtained by integrating over all fields _ but _ the scalar field ; the radiative corrections coming form these fields modify the vacuum structure of the model , making metastability possible . in this section , we argue that the same procedure must be followed when calculating the nucleation rate at finite temperature . we first review , in general terms , how to take into account the interactions of the scalar field with other fields in the calculation of the decay rate . ( for details see , ref . @xcite . ) then we apply our method to an example involving fermions , explicitly comparing our results to those obtained from eq . ( [ badrate ] ) . let us consider a system described by a scalar field @xmath9 and a set of fields @xmath128 ( bosonic or fermionic fields ) which are coupled to @xmath9 . the partition function of the system is given by @xmath129 where @xmath130 is the euclidean action of the system and the functional integration is carried over field configurations subject to periodic boundary conditions , @xmath131 , for bosons or antiperiodic , @xmath132 , for fermion fields . if one integrates out the @xmath128 fields in ( [ zphixi ] ) , the partition function can be written as @xmath133 where @xmath134 @xmath135 can be viewed as an effective action for the scalar field @xmath9 , where only @xmath128-loop terms are included . note that these @xmath128-loop terms introduce finite temperature corrections in ( [ wfunc ] ) . @xmath136 in ( [ zphixi ] ) and ( [ wfunc ] ) includes renormalization counter - terms and , if one or more of the @xmath128-fields is a gauge field , @xmath137 also includes the gauge fixings and the corresponding ghost terms . the procedure is now , in principle , straightforward . we evaluate the partition function in eq . ( [ zwphi ] ) semiclassically by expanding the effective action @xmath135 around its extremum configuration , which will be the bounce . note that the bounce will include the radiative corrections coming from the fields that couple to @xmath9 , _ but not from @xmath9 itself . _ the determinantal prefactor can be evaluated as before , by considering the negative and zero eigenvalues separately from the positive eigenvalues . however , as pointed out by weinberg @xcite , @xmath135 can not always be obtained in closed form , being in general a nonlocal functional . he proposes to resolve this difficulty by considering a local action @xmath138 which is close enough to the original one . from a derivative expansion of @xmath135 , @xmath139 is found to be @xmath140 \ : , \label{wphi0}\ ] ] where @xmath141 includes the tree level potential @xmath142 and the 1-loop contributions coming _ only _ from the @xmath128-field integration . the bounce solution can then be obtained from ( [ wphi0 ] ) . @xmath139 should be a good approximation to @xmath135 as long as the typical interaction length scale set by the field(s ) @xmath128 is shorter than the scale for the nonlocality of @xmath135 . in what follows we will apply the above formalism to a specific example of a scalar field coupled to fermions . for large enough yukawa couplings , this model has been shown to satisfy the conditions for metastability specified in ref . [ gleiser ] . consider a model of a real scalar field @xmath9 coupled to fermion fields with a lagrangian density @xmath143 with @xmath142 given by @xmath144 where @xmath145 and @xmath146 are positive , dimensionless constants and @xmath147 is a ( mass)@xmath148 parameter . a large enough coupling to fermions guarantees that a metastable phase is possible as the system is cooled below @xmath13 ; the high temperature minimum of the 1-loop effective potential , @xmath149 , lies to the left of the maximum of the potential at @xmath150 , and thermal fluctuations away from the symmetric minimum are suppressed . numerical values for the couplings satisfying these conditions can be found in ref . [ gleiser ] . the partition function for this model is given by @xmath151 as the fermion fields appear quadratically in ( [ zphipsi ] ) , they can be integrated out , giving @xmath152 where @xmath153 - { \rm tr } \ln ( - \not{\ ! \partial } - i g \phi)_{\beta } \ : , \label{wbetaphi}\ ] ] and @xmath154 thus , @xmath139 can be written as @xmath155 \ : , \label{wbeta}\ ] ] where , for sufficiently smooth fields ( see appendix b ) , the effective potential obtained after integrating over the fermions and renormalizing is ( we will drop all zero temperature quantum corrections ) @xmath156 the temperature dependent term accounts for finite temperature corrections coming from fermion loops . @xmath157 is the potential we should use to compute the bounce . note that , neglecting the @xmath158 quantum corrections , the high - temperature limit of @xmath157 is approximately @xmath159 so that for @xmath160 , a condition which is easily satisfied for reasonable values of the couplings , the high - temperature minimum is @xmath161 . once we have the action @xmath139 , the bounce is obtained as a solution of @xmath162 thus , for a static , spherically symmetric configuration , the bounce configuration @xmath6 will be a solution of @xmath163 with boundary conditions , @xmath164 and @xmath165 . ( from now on @xmath45 and @xmath44 should be understood as being the minima of ( [ vbetaphi ] ) and not the minima of the tree - level potential @xmath166 . ) the procedure is now identical to that of sections 2 and 3 . having a bounce solution we can evaluate the partition function written in eq . ( [ zwscalar ] ) semiclassically , exactly as was done in eqs . ( [ zphib ] ) and ( [ zphif ] ) , by expanding around @xmath167 and @xmath168 . we then obtain , from eqs . ( [ energy ] ) and ( [ gamma ] ) , the nucleation rate , @xmath169^{-\frac{1}{2 } } e^{- \delta w_0 } \ : , \label{gammaw}\ ] ] where @xmath170 , with @xmath171 obtained above . @xmath172 is given by @xmath173 where @xmath174 was defined in eq . ( [ wbeta ] ) . in order to proceed , we must rewrite the determinantal prefactor explicitly isolating the negative and zero modes from the positive modes . this is done following the same steps of section 3 , although now we must handle the fermionic contribution to the determinants . the details of the perturbative expansion for the fermionic determinantal prefactor are given in appendix b. we can then write the nucleation rate per unit volume as @xmath175^{\frac{3}{2 } } \exp \left [ - \frac { \delta f(t ) } { t } \right ] \ : , \label{endgammaw}\ ] ] where @xmath94 , the bubble activation free energy in the 1-loop approximation , is given by @xmath176 \right\ } \ : . \label{dft}\ ] ] as usual , the sum over @xmath177 can be performed and we get , @xmath178 \ : , \label{veff}\ ] ] where the effective potential @xmath179 , is given by ( neglecting zero temperature quantum corrections ) @xmath180 where @xmath166 is the tree level potential ( [ vphi ] ) and the mass term appearing in the scalar loop contribution is @xmath181 in leading order in the fermion loops . it is instructive to contrast this result with that obtained for self - coupled scalars , eq . ( [ veffren ] ) . the coupling to fermions modifies the scalar mass propagating in the loops . this effect naturally improves the infrared behavior of the theory , and can be of importance in weak first - order transitions . we will say more about this later . note also a crucial difference between this expression for the free energy barrier and the expression for the free energy barrier in eq . ( [ badrate ] ) : here , the bounce is obtained with the effective potential that does not include the corrections coming from scalar loops . the corrections from scalar loops which appear in the last term of eq . ( [ hatveff ] ) are thermally induced fluctuations about the bounce solution computed with @xmath182 . in the usual expression for the nucleation barrier the bounce is obtained from the full effective potential including the scalar loops . the two expressions are definetely not equivalent , even though for small scalar self - couplings the differences are negligible . in order to illustrate the differences let us look at a specific example . in fig . 2 we contrast the two approaches by comparing the nucleation barriers as a function of the temperature for a fixed set of coupling constants . the barriers in the figure were obtained by a numerical integration of the bounce equation including the relevant loops according to each approach . for clarity let us call the nucleation barrier obtained in the usual approach , _ i.e. _ , by including the scalar loops in the bounce calculation , the scalar barrier . the nucleation barrier obtained without including the scalar loops we call the fermionic barrier . in fig . 2 we take @xmath183 , and @xmath184 . since @xmath145 controls the strength of the scalar corrections , we expect the differences between the two barriers to be noticeable . we find that this is indeed the case , noting that as we approach the critical temperature ( that is , as we move closer to the thin - wall limit ) the differences between the two barriers increase , with the scalar barrier _ always larger _ than the fermionic barrier . this is precisely what one expects if the scalar corrections are entropic corrections to the nucleation barrier . thus , the nucleation barrier used in expression eq . ( [ badrate ] ) is overestimated for large enough scalar corrections . finally , we point out two additional differences between the results . first note that the contribution from the zero modes to the prefactor depends on @xmath185 , as opposed to @xmath94 . this could be important for weak transitions in which the prefactor may play a relevant rle . most importantly , the expression for @xmath186 , eq . ( [ hatveff ] ) , differs from the usual 1-loop finite temperature effective potential by the mass term for the scalar field loops , @xmath187 . since we have used the stationary points of @xmath139 , eq . ( [ wbeta ] ) , as opposed to the stationary points of @xmath188 , as the effective `` background '' fields in the saddle - point evaluation of the partition function , the scalar field propagator carries the finite temperature mass @xmath189 . the propagator is dressed by the quantum corrections due to fermion loops . in the usual 1-loop finite temperature effective potential , the stationary points are obtained from the tree level action , with mass term for scalar loops , @xmath190 . this results in the usual negative mass terms related to the change in convexity of the effective potential between the inflection points , and , in the case of very shallow potentials , in bad infrared behavior near @xmath8 . the incorporation of the fermionic corrections to the scalar propagator , which is demanded by our method of calculation atenuates these problems . in the example above , the scalar mass gets dressed by fermionic loops , being given by @xmath191 , where @xmath166 is the tree - level potential ( [ vphi ] ) . the temperature term in @xmath192 works as the infrared regulator for small values of @xmath190 . this result is independent of the particular model studied . similar conclusions have been obtained in ref . @xcite for scalar electrodynamics . in this paper we examined in some detail the computation of false vacuum decay rates at finite temperatures in the regime in which quantum fluctuations are negligibly small compared to thermal fluctuations . we have shown that temperature corrections to the nucleation barrier can be obtained from a saddle - point evaluation of the partition function in a dilute gas approximation . in fact , the temperature corrections are simply due to the positive eigenvalues from stable fluctuations around the critical bubble . that is , they are the entropic contributions due to thermally induced deformations on the bubble . even though this result has been known in classical statistical mechanics for more than two decades @xcite , we believe that a consistent treatment within field theory is still lacking . although we left many questions unanswered , we hope to have clarified some of the issues involved in the calculation of finite - temperature decay rates . of particular importance is the fact that the bounce is _ not _ obtained from the full 1-loop corrected effective potential , but from the potential excluding the scalar loops . thus , for a self - interacting scalar , the bounce is obtained from the tree - level potential . the full finite temperature potential appears in the exponent only after properly accounting for the positive eigenvalues of the determinantal prefactor . that is , the scalar contributions account for entropic corrections to the nucleation barrier . we obtained a temperature corrected nucleation barrier which can differ from the usual result . we showed this to be particularly true for sufficiently large scalar self - couplings in the vicinity of the critical temperature for the transition . also , we found that the interaction with other fields gives rise to a potential which is better behaved in the infrared . ( see also ref . this result is the finite - temperature equivalent to what e. weinberg found for the zero - temperature case , once the integration over the other fields is performed @xcite . the reader may be wondering if our results will have any consequences to current work on the electroweak phase transition . the answer depends on the higgs mass . for a sufficiently light higgs it is consistent to neglect the contribution from scalar loops to the effective potential . in this case , the usual estimate for the nucleation barrier is a valid approximation . however , the situation may change for a heavier higgs . given that the experimental lower bound on the higgs mass is now above 60 gev , we believe it worthwhile to study this question in more detail , keeping in mind that the transition becomes weaker as the higgs mass increases . we would like to thank a. linde for many important discussions on these and related issues . we would like to thank the institute for theoretical physics in santa barbara where , during the program on cosmological phase transitions , this work begun . at itp this work was supported in part by a national science foundation grant no . phy89 - 04035 at itp . ( mg ) was supported in part by a national science foundation grant no . ( gcm ) aknowledges financial support from fapesp ( so paulo , brazil ) and ( ror ) from conselho nacional de desenvolvimento cientfico e tecnolgico - cnpq ( brazil ) . * figure 1 : * a typical asymmetric double - well potential . + * figure 2 : * a comparison of the nucleation barrier as a function of temperature , in units of mass parameter @xmath193 , obtained by including ( stars ) and excluding ( dots ) scalar loops in the computation of the bounce . the parameters in the tree - level potential are @xmath194 . + * figure 3 : * a comparison of the terms @xmath195 and @xmath196 appearing in appendix b. the parameters in the tree - level potential are @xmath197 . + where the prime in @xmath199 means that the negative eigenvalue , @xmath200 , and the three zero eigenvalues , @xmath201 , are now excluded from the product . the term for @xmath202 in @xmath203 , can be handled as in ref . @xcite , resulting in the factor @xmath204^{\frac{3}{2}}$ ] in eq . ( [ eigenratio ] ) . separating the @xmath202 modes both in the numerator and the denominator of ( [ aratio ] ) , and using the identity ( [ idenpi ] ) , we get , @xmath205^{\frac{3}{2 } } \exp\left\{\left(-4 + \sum_{i } - \sum_{j}\ ; ' \right ) { \rm ln } \prod_{n=1}^{+\infty } \omega_{n}^{2 } - { \rm ln } \left ( e_{-}^{2 } \right)^{1/2 } - { \rm ln } \left [ \frac{\sin(\frac{\beta}{2 } |e_{-}|)}{\frac{\beta}{2 } |e_{-}| } \right ] + \right . & + & \left . \left(\sum_{j}\ ; ' - \sum_{i}\right ) { \rm ln } \beta + \sum_{i } \left [ \frac{\beta}{2 } e_{f}(i ) + { \rm ln } \left ( 1 - e^{- \beta e_{f}(i ) } \right ) \right ] + \right . \nonumber \\ & - & \left . \sum_{j } \ ; ' \left [ \frac{\beta}{2 } e_{b}(j ) + { \rm ln } \left ( 1 - e^{- \beta e_{b}(j ) } \right ) \right ] \right\ } \ : . \label{aexplicity}\end{aligned}\ ] ] & + & \left . \sum_{i } \left [ \frac{\beta}{2 } e_{f}(i ) + { \rm ln } \left ( 1 - e^{- \beta e_{f}(i ) } \right ) \right ] - \sum_{j } \ ; ' \left [ \frac{\beta}{2 } e_{b}(j ) + { \rm ln } \left ( 1 - e^{- \beta e_{b}(j ) } \right ) \right ] \right\ } \label{aendexplicitly}\end{aligned}\ ] ] @xmath213^{2 } & = & \det ( - \not{\ ! \partial } -i g \varphi).\det ( - \not{\ ! \partial } + i g \varphi ) = \nonumber \\ & = & \det \left [ ( -\box_{e } + g^{2 } \varphi^{2 } ) 1_{4 \times 4 } - i g \gamma_{e}^{\mu } \partial_{\mu } \varphi \right ] \ : , \label{bident}\end{aligned}\ ] ] @xmath227_{\beta } + { \rm tr } \ln \left [ - \box_{e } + g^{2 } \varphi_{b}^{2 } - g \frac { \partial \varphi_{b}}{\partial r } \right]_{\beta } - \right . \nonumber \\ & - & \left . 2 { \rm tr } \ln \left [ - \box_{e } + g^{2 } \varphi_{f}^{2 } \right]_{\beta } \right\ } \ : . \label{bdetbf}\end{aligned}\ ] ] @xmath228 \right ] + \right . \nonumber \\ & + & \left . { \rm tr } \ln \left [ 1 + s_{\beta}(\varphi_{f } ) \left [ g^{2 } ( \varphi_{b}^{2 } - \varphi_{f}^{2 } ) - g \frac { \partial \varphi_{b}}{\partial r } \right ] \right ] \right\ } \ : , \label{bexpratio}\end{aligned}\ ] ] @xmath230 \right ] & = & \sum_{m= 1}^{+ \infty } \frac{(-1)^{m+1}}{m } \int d^{3}x \left [ g^{2 } ( \varphi_{b}^{2 } - \varphi_{f}^{2 } ) \pm g \frac { \partial \varphi_{b}}{\partial r } \right]^{m } \times \nonumber \\ & \times & \sum_{n= -\infty}^{+ \infty } \int \frac{d^{3 } k } { ( 2 \pi)^{3 } } \frac{1 } { \left [ \bar{\omega}_{n}^{2 } + \vec{k}^{2 } + g^{2 } \varphi_{f}^{2 } \right]^{m } } \ : , \label{btraceln}\end{aligned}\ ] ] where @xmath231 . as before , ( [ btraceln ] ) can be expressed as a graphic expansion similar to ( [ graphic ] ) , with the propagators @xmath112 replaced now by @xmath232 and the external lines given by @xmath233 or @xmath234 . the determinant factor in ( [ gammaw ] ) , coming from the functional integration of the scalar field , can be evaluated by the same methods of sec . 3 . in ( [ gammaw ] ) , the determinant term @xmath235_{\beta}$ ] , with @xmath236 , has a negative eigenvalue , @xmath200 , associated with the instability of the critical bubble , and the three zero eigenvalues , associated with the translational invariance of the bubble . these eigenvalues can be handled as usual , giving the preexponential term in ( [ endgammaw ] ) . the part of the determinant involving the positive eigenvalues can be written as an expansion exactly as in ( [ traceln ] ) , @xmath237^{- \frac{1}{2 } } = \exp \left\ { - \frac{1}{2 } { \rm tr } \ ; \ln \bigl [ 1 + \hat{g}_{\beta}(\varphi_{f } ) \left [ \hat{v}_{\psi}''(\varphi_{b } ) - \hat{v}_{\psi}''(\varphi_{f } ) \right ] \bigr ] \right\ } \ : , \label{bdetbos}\ ] ] @xmath239 \right\ } & = & \sum_{m=1}^{+ \infty } \frac { ( -1)^{m+1 } } { m } \int d^{3 } x \left[\hat{v}_{\psi}''(\varphi_{b } ) - \hat{v}_{\psi}''(\varphi_{f } ) \right]^{m } \times \nonumber \\ & \times & \sum_{n= -\infty}^{+ \infty } \int \frac { d^{3 } k } { ( 2 \pi)^{3 } } \frac{1 } { \left [ \omega_{n}^{2 } + \vec{k}^{2 } + m_{\beta}^{2 } ( \varphi_{f } ) \right]^{m } } \ : . \label{btrlnw}\end{aligned}\ ] ] the sum in @xmath42 in both ( [ btraceln ] ) and ( [ btrlnw ] ) can be performed as in eq . ( [ traceln ] ) ) . therefore , from eqs . ( [ bexpw ] ) , ( [ btraceln ] ) and ( [ btrlnw ] ) , we can write the relevant part of eq . ( [ gammaw ] ) as @xmath240^{- \frac{1}{2 } } e^{- \delta w_0 } = \left [ \frac { \det ( -\box_{e } + \hat{v}_{\psi}''(\varphi_{b}))_{\beta } } { \det ( -\box_{e } + \hat{v}_{\psi}''(\varphi_{f}))_{\beta } } \right]^{- \frac{1}{2 } } \ : \ : \frac { \det ( - \not{\ ! \partial } - i g \varphi_{b } ) _ { \beta } } { \det ( - \not{\ ! \partial } - i g \varphi_{f } ) _ { \beta } } \ : \ : e^{-\delta s } = } \nonumber \\ & & = { \cal v}\frac{t^{4}}{i |e_{-}| } \frac{\beta \frac{|e_{-}|}{2 } } { \sin \left ( \beta \frac{|e_{-}|}{2 } \right ) } \left [ \frac{\delta w_0 } { 2 \pi } \right]^{\frac{3}{2 } } \exp \left\ { - \delta s + \int d^{3 } x \sum_{n= -\infty}^{+ \infty } \int \frac{d^3 k}{(2 \pi)^3 } \left [ \ln \left ( 1 + \frac{g^2 ( \vp_{b}^{2 } - \vp_{f}^{2 } ) + g \frac{\partial \vp_{b}}{\partial r}}{\bar{\omega}_{n}^{2 } + \vec{k}^2 + g^2 \vp_{f}^{2 } } \right ) + \right . \nonumber \\ & & + \left . \left . \ln \left ( 1 + \frac{g^2 ( \vp_{b}^{2 } - \vp_{f}^{2 } ) - g \frac{\partial \vp_{b}}{\partial r}}{\bar{\omega}_{n}^{2 } + \vec{k}^2 + g^2 \vp_{f}^{2 } } \right ) - \frac{1}{2 } \ln \left ( 1 + \frac{m_{\beta}^2 ( \vp_{b } ) - m_{\beta}^{2 } ( \vp_{f}^{2 } ) } { \omega_{n}^{2 } + \vec{k}^2 + m_{\beta}^{2 } ( \vp_{f}^{2 } ) } \right ) \right ] \right\ } \ : , \label{bdeterminantal}\end{aligned}\ ] ] apart from the derivative terms @xmath243 , the momentum integral reproduces the finite temperature corrections to the the tree - level potential appearing in @xmath67 . when we wrote the expression for @xmath244 in eq . ( [ dft ] ) , these terms were not included in the effective potential @xmath245 . there are two reasons for negleting this term . first , due to the graphic expansion we used for the determinants , it is easy to see that at least at the tadpole level , their contribution cancels . since the tadpole gives the dominant temperature contribution to the potential , terms that depend on @xmath243 will be sub - dominant . second , it is possible to explicitly compare the terms @xmath246 and @xmath247 , by obtaining @xmath6 numerically . we have performed this comparison for the same set of parameters used in figs . 2 and 3 , and convinced ourselves that the derivative term will indeed be sub - dominant . a typical example is shown in fig . neglecting the term @xmath243 , we can use eqs . ( [ bdeterminantal ] ) and ( [ finaldeltaw ] ) to obtain the expression for @xmath94 in ( [ dft ] ) . m. gleiser and e. w. kolb , _ phys . lett . _ * 69 * , 1304 ( 1992 ) ; fermilab preprint no . pub-92/222-a ; _ int . phys . _ * c3 * , 773 ( 1992 ) ; m. gleiser , e. w. kolb , and r. watkins , _ nucl . phys . _ * b364 * , 411 ( 1991 ) . c. a. carvalho , g. c. marques , a. j. silva and i. ventura , _ nucl . * b265 * , 45 ( 1986 ) ; c. a. carvalho , d. bazeia , o. j. p. eboli , g. c. marques , a. j. silva and i. ventura _ phys . rev . _ * d31 * , 1411 ( 1985 ) ;

several tensor eigenpair definitions have been put forth in the past decade , but these can all be unified under generalized tensor eigenpair framework , introduced by chang , pearson , and zhang ( 2009 ) . giventh - order ,-dimensional real - valued symmetric tensors and , the goal is to find and such that . different choices for yield different versions of the tensor eigenvalue problem . we present our generalized eigenproblem adaptive power ( geap ) method for solving the problem , which is an extension of the shifted symmetric higher - order power method ( ss - hopm ) for finding z - eigenpairs . a major drawback of ss - hopm was that its performance depended in choosing an appropriate shift , but our geap method also includes an adaptive method for choosing the shift automatically . tensor eigenvalues , e - eigenpairs , z - eigenpairs ,-eigenpairs , generalized tensor eigenpairs , shifted symmetric higher - order power method ( ss - hopm ) , generalized eigenproblem adaptive power ( geap ) method 15a18 , 15a69 adaptive shifted power method for computing generalized tensor eigenpairs

1401.1183

suppose @xmath2 is a real - valued , @xmath0th - order , @xmath1-dimensional tensors and @xmath8 is a real - valued @xmath1-vector . we let @xmath9 denote the @xmath1-vector defined by @xmath10 we let @xmath11 denote the scalar defined by @xmath12 . we say the tensor @xmath2 is _ symmetric _ if its entries are invariant under permutation . we say the tensor @xmath2 is _ positive definite _ if @xmath13 for all @xmath14 . the notion of generalized eigenpairs has been defined for tensors by chang , pearson , and zhang @xcite as follows . let @xmath2 and @xmath3 be real - valued , @xmath0th - order , @xmath1-dimensional symmetric tensors . assume further that @xmath0 is even and @xmath3 is positive definite . we say @xmath15 is a _ generalized eigenpair _ ( also known as a _ @xmath3-eigenpair _ ) if @xmath16 taking the dot product with @xmath8 , it is clear that any solution satisfies @xmath17 the advantage of the generalized eigenpair framework is that it nicely encapsulates multiple definitions of tensor eigenvalues , as follows . a _ z - eigenpair _ @xcite is defined as a pair @xmath18 such that @xmath19 this is equivalent to a generalized tensor eigenpair with @xmath20 , the identity tensor such that @xmath21 for all @xmath22 @xcite . note that , unlike ordinary tensor z - eigenpairs , generalized tensor eigenpairs allow arbitrary rescaling of the eigenvector @xmath8 with no effect on the eigenvalue @xmath23 . in this way , the generalized tensor eigenvalue problem preserves the homogeneity of the corresponding matrix eigenproblem . an _ h - eigenpair _ is defined as a pair @xmath15 such that @xmath24}.\ ] ] here @xmath25}$ ] denotes elementwise power , i.e. , @xmath26})_i \equiv { { { \bm { \mathbf{\makelowercase{x}}}}}}_i^{m-1}$ ] , for @xmath27 . this is equivalent to a generalized tensor eigenpair with @xmath28 @xcite . let @xmath29 be a symmetric @xmath30 matrix and assume @xmath31 . we say @xmath32 is a _ d - eigenpair _ @xcite if @xmath33 this is equivalent to a @xmath3-eigenpair where @xmath3 is the symmetrized tensor outer product of @xmath29 with itself @xcite . in this paper , we describe a method for computing generalized eigenpairs . our method is a generalization of the shifted symmetric higher - order power method ( ss - hopm ) that we previously introduced for computing z - eigenvalues @xcite . in addition to generalizing the method , we have also significantly improved it by adding an adaptive method for choosing the shift . to derive the method , we reformulate the generalized eigenproblem , , as a nonlinear program such that any generalized eigenpair is equivalent to a kkt point in . we develop an adaptive , monotonically convergent , shifted power method for solving the optimization problem in . we call our method the generalized eigenproblem adaptive power ( geap ) method . in , we show that the geap method is much faster than the ss - hopm method for finding z - eigenpairs due to its adaptive shift selection . furthermore , the geap method is shown to find other types of generalized eigenpairs , by illustrating it on examples from related literature as well as a randomly generated example . this is the only known method for finding generalized eigenpairs besides direct numerical solution ; we survey related work in . a symmetric tensor has entries that are invariant under any permutation of its indices . more formally , a real - valued , @xmath0th - order , @xmath1-dimensional tensor @xmath2 is _ symmetric _ if @xmath34 where @xmath35 denotes the space of all @xmath0-permutations . we let @xmath36}}$ ] denote the space of all symmetric , real - valued , @xmath0th - order , @xmath1-dimensional tensors . let @xmath37}}$ ] , then we can define the following tensor - vector products . @xmath38 observe that the derivatives of the tensor - vector product w.r.t . @xmath8 are given by @xmath39 we say a tensor @xmath37}}$ ] is _ positive definite _ if @xmath40 we let @xmath36}}_+$ ] denote the space of positive definite tensors in @xmath36}}$ ] . we use the symbol @xmath41 to mean symmetrized outer product , i.e. , @xmath42 let @xmath43 denote the unit sphere , i.e. , @xmath44 let @xmath37}}$ ] and @xmath45}}_+$ ] . then we may define the nonlinear program @xmath46 the constraint makes the term @xmath47 in @xmath48 superfluous ; nevertheless , we retain this form since choosing @xmath20 yields @xmath49 , as in @xcite . the details of computing the derivatives are provided in . here we simply state the results as a theorem . let @xmath48 be as defined in . for @xmath50 , the gradient is @xmath51.\ ] ] for @xmath50 , the hessian is @xmath52 \\ - \frac{m}{({{{\boldsymbol { \mathscr{\makeuppercase{b}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^m})^2 } \biggl [ ( m-1 ) { { { \boldsymbol { \mathscr{\makeuppercase{a}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^m}{{\boldsymbol { \mathscr{\makeuppercase{b}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^{m-2 } + m \bigl ( { { { { \boldsymbol { \mathscr{\makeuppercase{a}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^{m-1}}\circledcirc { { { \boldsymbol { \mathscr{\makeuppercase{b}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^{m-1 } } } \bigr ) \\ + m { { { \boldsymbol { \mathscr{\makeuppercase{a}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^m}\bigl ( { { { { \bm { \mathbf{\makelowercase{x}}}}}}\circledcirc { { { \boldsymbol { \mathscr{\makeuppercase{b}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^{m-1 } } } \bigr ) \biggr].\end{gathered}\ ] ] these complicated derivatives reduce for @xmath20 . in that case , we have @xmath53 and @xmath54 , so these equations become . @xmath55 these match the derivatives of @xmath49 , as proved in @xcite . note that we have used the fact that @xmath56 for all @xmath50 . we are considering the nonlinear program in because there is a correspondence between it and the generalized tensor eigenvalue problem in . note that the @xmath8 in can be arbitrarily rescaled . any pair @xmath57 is a solution to iff the scaled version with @xmath58 is a kkt point of with @xmath23 as the lagrange multiplier . first , assume @xmath57 is a solution to . let constraint @xmath50 be expressed as @xmath59 . then the lagrangian is @xmath60 hence , using the derivatives in , we have @xmath61 - m \lambda { { { \bm { \mathbf{\makelowercase{x}}}}}}= 0.\ ] ] so , @xmath8 is a kkt point of with lagrange multiplier @xmath23 as defined in . to prove the reverse , assume @xmath8 is a kkt point of with lagrange multiplier @xmath23 . then , must hold . if we multiply each term in by @xmath8 , then the third and fourth terms cancel out , and we conclude that @xmath23 satisfies . substituting that back into , we see that is satisfied . hence , the claim . from the previous theorem , there is an equivalence between generalized tensor eigenpairs and kkt points of . hence , solving yields eigenpairs . an eigenpair may correspond to a local maximum , a local minimum , or a saddle point . for a given eigenpair @xmath57 normalized so that @xmath50 , we can categorize it by considering the projected hessian of the lagrangian , i.e. , @xmath62 where @xmath63 is an orthonormal basis for @xmath64 . we can then say the following : @xmath65{(\ref*{eq : opt } ) } } } , \\ { { \bm { \mathbf{\makeuppercase{c}}}}}(\lambda,{{{\bm { \mathbf{\makelowercase{x } } } } } } ) & \text { negative definite } & \rightarrow & \text { local maximum of { \hyperref[eq : opt]{(\ref*{eq : opt } ) } } } , \\ { { \bm { \mathbf{\makeuppercase{c}}}}}(\lambda,{{{\bm { \mathbf{\makelowercase{x } } } } } } ) & \text { indefinite } & \rightarrow & \text { saddle point of { \hyperref[eq : opt]{(\ref*{eq : opt})}}}. \end{aligned}\ ] ] the argument is very similar to that presented in @xcite and so is omitted . optimization approaches can not easily find saddle points , but they can find local minima and maxima . we describe such an optimization approach in the next section . we propose to use a property of convex functions of the sphere to develop a monotonically convergent method . we consider an idea originally from @xcite ; see also @xcite for a proof . we have modified the theorem here to focus on its _ local _ applicability by considering just an open neighborhood of @xmath66 rather than all of @xmath67 . [ thm : cvx ] let @xmath68 be a given function , and let @xmath69 such that @xmath70 . let @xmath71 be an open neighborhood of @xmath66 , and assume @xmath72 is convex and continuously differentiable on @xmath71 . define @xmath73 . if @xmath74 and @xmath75 , then @xmath76 . [ cor : cnv ] let @xmath77 be a given function , and let @xmath69 such that @xmath70 . let @xmath71 be an open neighborhood of @xmath66 , and assume @xmath72 is concave and continuously differentiable on @xmath71 . define @xmath78 . if @xmath74 and @xmath75 , then @xmath79 . hence , if @xmath72 is locally convex , then a simple algorithm , i.e. , @xmath80 will yield ascent . conversely , if @xmath72 is locally concave , we can expect descent from @xmath81 . unfortunately , the function @xmath48 in may not be convex or concave . to fix this , we work with a shifted function , @xmath82 from @xcite , we have that for @xmath50 , @xmath83 if we choose @xmath84 appropriately , then we can ensure that @xmath85 is positive or negative definite , ensuring that @xmath86 is locally convex or concave . in @xcite for the special case of @xmath87 , we proposed choosing a single value for @xmath84 in ss - hopm that ensured convexity on the entire sphere . but it is difficult to choose a reasonable value in advance , and poor choices lead to either very slow convergence or a complete lack of convergence . in this work , we propose to choose @xmath84 adaptively . for an arbitrary matrix @xmath30 symmetric matrix @xmath88 , the following notation denotes its eigenvalues : @xmath89 . assume @xmath50 . let @xmath90 and @xmath85 be defined as in and , respectively . for @xmath91 , the eigenvalues of @xmath92 are bounded as @xmath93 for @xmath94 . likewise , for @xmath95 , the eigenvalues of @xmath92 are bounded as @xmath96 for @xmath94 . the proof follows immediately from weyl s inequality . in the convex case , our goal is to choose @xmath84 so that @xmath92 is positive semi - definite in a local neighborhood of the current iterate , @xmath8 . by the smoothness of @xmath86 when @xmath8 is away from zero , we may argue that for every @xmath97 , there exists and @xmath98 such that @xmath92 is positive semi - definite for all @xmath99 whenever @xmath100 . in other words , @xmath101 is the threshold for positive definiteness . assume @xmath50 . let @xmath97 . if @xmath102 then @xmath103 . in the concave case , our goal is to choose @xmath84 so that @xmath92 is negative semi - definite in a local neighborhood of the current iterate . assume @xmath50 . let @xmath97 . if @xmath104 then @xmath105 . from , if @xmath84 is selected to make @xmath86 locally convex , we have @xmath106 so long as @xmath107 , the convex neighborhood of @xmath8 . even though we are adaptively changing @xmath84 , we see increase in the original function at each step . a similar argument applies in the concave case , with the function decreasing at each step . the potential problem with this approach is that it may be the case that @xmath108 . if that happens , we may observe that the function values ( i.e. , @xmath109 ) are not increasing ( or decreasing ) monotonically as expected . to fix this , we make a more conservative choice for @xmath101 ( at least temporarily ) , which will in turn enforce a more conservative choice of @xmath84 . if @xmath101 is large enough , then we will satisfy the lower bound on @xmath84 that guarantees convergence for the shifted algorithm ( this is proven for z - eigenvalue in @xcite ; the proof for the general problem is similar and so omitted ) . thus far in our experiments , such contingencies have not been necessary , so we have not included the details in the algorithm . the full algorithm is presented in . given tensors @xmath37}}$ ] and @xmath45}}_+$ ] and an initial guess @xmath110 . let @xmath111 if we want to find local maxima ( and the function is convex ) ; otherwise , let @xmath112 , indicating that we are seeking local minima ( and the function is concave ) . let @xmath101 be the tolerance on being positive / negative definite . @xmath113 precompute @xmath114 , @xmath115 , @xmath116 , @xmath117 , @xmath118 , @xmath119 @xmath120 @xmath121 @xmath122 @xmath123 @xmath124 the cost per iteration of is as follows . assuming @xmath2 and @xmath3 are dense , the dominant cost is computing products with these tensors . computing the hessian requires six products : @xmath125 . recall that @xmath126 is given by @xmath127 the cost is @xmath128 multiplies and @xmath129 additions per each of @xmath130 entries ; therefore , the total cost is @xmath131 operations . exploiting symmetry yields reduces the cost to @xmath132 @xcite . we can compute @xmath133 for an additional cost of @xmath134 and @xmath135 operations , respectively . ( these can also be computed directly , but at a cost of @xmath131 operations each . ) these values have to be computed for both @xmath2 and @xmath3 at every iteration for a total cost ( ignoring symmetry ) of @xmath136 . once these six products are computed , the cost for computing @xmath137 is a series of matrix operations , for a total cost of @xmath138 . the cost of computing the eigenvalues of the symmetric matrix @xmath137 is @xmath139 , which is less than the cost of the products . updating @xmath8 requires a 5 vector operations at a cost of @xmath1 operations each . hence , cost of the method is dominated by the computation of @xmath126 and @xmath140 , at an expense of @xmath132 . in , we show the specialization of the method to the z - eigenvalue problem . this is the same as ss - hopm , except for the adaptive shift . note that unlike , this algorithm can be used even when @xmath0 is odd . the cost per iteration of is the same order as for , but it does not need to do any computations with @xmath141 . given tensor @xmath37}}$ ] and an initial guess @xmath110 . let @xmath111 if we want to find local maxima ( and the function is convex ) ; otherwise , let @xmath112 , indicating that we are seeking local minima ( and the function is concave ) . let @xmath101 be the tolerance on being positive / negative definite . @xmath113 precompute @xmath114 , @xmath116 , @xmath118 @xmath142 @xmath143 @xmath122 @xmath144 @xmath124 all numerical tests were done using matlab version r2012b and the tensor toolbox version 2.5 @xcite . the experiments were performed a laptop computer with an intel dual - core i7 - 3667ucpu ( 2ghz ) and 8 gb of ram . in all numerical experiments , we used the following settings . we set @xmath145 , where @xmath101 is the tolerance on being positive or negative definite . we consider the iterates to be converged once @xmath146 . the maximum iterations is 500 . the following example is originally from @xcite and was used in evaluating the ss - hopm algorithm in @xcite . our goal is to compute the z - eigenpairs using the z - eigenpair adaptive power method in and show that it is faster than ss - hopm @xcite using a fixed value for the shift . kofidis and regalia @xcite [ ex : kore02ex1 ] our objective is to compute the z - eigenpairs . let @xmath147}}$ ] be the symmetric tensor given by kofidis and regalia ( * ? ? ? * example 1 ) whose entries are specified in ( ) . since we are computing z - eigenpairs , we have @xmath20 . a complete list of the 11 z - eigenpairs is provided in ( ) ; there are three maxima and three minima . a comparison of the fixed and adaptive shift results are provided in . there are six different experiments looking at maxima ( @xmath111 ) and minima ( @xmath112 ) and different shifts ( @xmath148 ) in . note that using a fixed shift means that is equivalent to ss - hopm and no adaptive update of the shift is performed in step 5 . we used 100 random starting guesses , each entry selected uniformly at random from the interval @xmath149 $ ] ; the same set of random starts was used for each set of experiments . for each eigenpair , the table lists the number of occurrences in the 100 experiments , the median number of iterations until convergence , the number of runs that violated monotonicity , the average error and standard deviation in the final result , and the average run time and standard deviation . the error is computed as @xmath150 . the two monotinicity violations were both extremely small , i.e. , o(@xmath151 ) . these violations indicate that a step went outside the region of local convexity . @xmath84 adaptive , @xmath1113 53 & 0.8893 & 0.6672 & 0.2471 & -0.7027 & 30 & & 9e-09 & 3e-09 & 0.05 & 0.02 + 29 & 0.8169 & 0.8412 & -0.2635 & 0.4722 & 34 & & 1e-08 & 3e-09 & 0.04 & 0.01 + 18 & 0.3633 & 0.2676 & 0.6447 & 0.7160 & 26 & & 7e-09 & 2e-09 & 0.03 & 0.00 + @xmath152 , @xmath1113 53 & 0.8893 & 0.6672 & 0.2471 & -0.7027 & 49 & 1 & 1e-15 & 2e-08 & 3e-09 & 0.07 & 0.03 + 29 & 0.8169 & 0.8412 & -0.2635 & 0.4722 & 45 & & 2e-08 & 3e-09 & 0.04 & 0.00 + 18 & 0.3633 & 0.2676 & 0.6447 & 0.7160 & 57 & & 2e-08 & 2e-09 & 0.06 & 0.00 + @xmath153 , @xmath1113 48 & 0.8893 & 0.6672 & 0.2471 & -0.7027 & 192 & & 5e-08 & 6e-09 & 0.24 & 0.12 + 29 & 0.8169 & 0.8412 & -0.2635 & 0.4722 & 185 & & 5e-08 & 5e-09 & 0.17 & 0.02 + 18 & 0.3633 & 0.2676 & 0.6447 & 0.7160 & 261 & & 5e-08 & 2e-09 & 0.24 & 0.02 + 5 & & & & 0.43 & 0.01 + @xmath154 , @xmath1123 22 & -0.0451 & 0.7797 & 0.6135 & 0.1250 & 18 & & 4e-09 & 2e-09 & 0.02 & 0.00 + 37 & -0.5629 & 0.1762 & -0.1796 & 0.9678 & 17 & & 6e-09 & 2e-09 & 0.02 & 0.00 + 41 & -1.0954 & 0.5915 & -0.7467 & -0.3043 & 17 & & 6e-09 & 3e-09 & 0.02 & 0.01 + @xmath155 , @xmath1123 22 & -0.0451 & 0.7797 & 0.6135 & 0.1250 & 34 & & 1e-08 & 2e-09 & 0.03 & 0.00 + 37 & -0.5629 & 0.1762 & -0.1796 & 0.9678 & 20 & & 7e-09 & 2e-09 & 0.02 & 0.00 + 41 & -1.0954 & 0.5915 & -0.7467 & -0.3043 & 21 & 1 & 1e-15 & 8e-09 & 4e-09 & 0.02 & 0.00 + @xmath156 , @xmath1123 22 & -0.0451 & 0.7797 & 0.6135 & 0.1250 & 186 & & 5e-08 & 2e-09 & 0.16 & 0.01 + 37 & -0.5629 & 0.1762 & -0.1796 & 0.9678 & 103 & & 4e-08 & 4e-09 & 0.09 & 0.01 + 41 & -1.0954 & 0.5915 & -0.7467 & -0.3043 & 94 & & 4e-08 & 6e-09 & 0.09 & 0.01 + the first three experiments use @xmath111 to look for local maxima . the first experiment varies @xmath84 , the second uses @xmath152 ( as in @xcite ) , and the third uses @xmath153 . all three variations find all three local maxima . the results for @xmath152 and the adaptive method are nearly identical they find the same local maxima with the same frequency . the difference is that @xmath152 uses more iterations than the adaptive shift . choosing @xmath153 is similar , except now five of the runs do not converge within the allotted 500 iterations . there was no breakdown in monotonicity , and these runs would converge eventually . if the shift is too small ( e.g. , @xmath157 ) , then some or all of the runs may fail to converge @xcite . the last three experiments use @xmath112 to find local minima . again , we vary @xmath84 using an adaptive choice along with @xmath155 and @xmath156 . the adaptive method requires the fewest number of iterations . each experiments finds all three local minima with the exact same frequencies . to compare the convergence in terms of the number of iterations , shows sample results for one run for computing z - eigenpairs of @xmath2 from . the left hand plot shows the selected shift values at each iteration . the right hand plot shows the convergence of the eigenvalue . the adaptive shift is the fastest to converge . here we demonstrate that the geap method in calculates h - eigenpairs with an appropriate choice for @xmath141 . [ ex : heig ] we generate a random symmetric tensor @xmath158}}$ ] as follows : we select random entries from @xmath149 $ ] , symmetrize the result , and round to four decimal places . the tensor entries are specified in ( ) . since we are computing h - eigenpairs , we specify @xmath141 as @xmath28 . a complete list of the h - eigenpairs is provided in ( ) ; there are five maxima and five minima . a summary of the results are provided in . there are two different experiments looking at maxima ( @xmath111 ) and minima ( @xmath112 ) . we used 1000 random starting guesses , each entry selected uniformly at random from the interval @xmath149 $ ] ; the same set of random starts was used for each experiment . the columns are the same as for . the error is computed as @xmath159 } \|_2 $ ] . @xmath1114 211 & 14.6941 & 0.5426 & -0.4853 & 0.4760 & 0.4936 & 28 & 124 & 2e-15 & 2e-09 & 2e-09 & 0.08 & 0.02 + 144 & 9.6386 & 0.5342 & -0.5601 & 0.5466 & -0.3197 & 110 & 53 & 4e-15 & 9e-09 & 3e-09 & 0.27 & 0.04 + 338 & 8.7371 & 0.4837 & 0.5502 & 0.6671 & -0.1354 & 100 & 143 & 4e-15 & 1e-08 & 5e-09 & 0.26 & 0.06 + 169 & 5.8493 & 0.6528 & 0.5607 & -0.0627 & -0.5055 & 54 & 19 & 2e-15 & 8e-09 & 3e-09 & 0.13 & 0.02 + 138 & 4.8422 & 0.5895 & -0.2640 & -0.4728 & 0.5994 & 66 & 13 & 8e-01 & 6e-09 & 1e-09 & 0.16 & 0.01 + @xmath1124 130 & -2.9314 & 0.3161 & 0.5173 & 0.4528 & -0.6537 & 76 & 3 & 2e-15 & 7e-09 & 1e-09 & 0.18 & 0.02 + 149 & -3.7179 & 0.6843 & 0.5519 & 0.3136 & 0.3589 & 59 & 10 & 1e-15 & 7e-09 & 2e-09 & 0.15 & 0.02 + 152 & -4.1781 & 0.4397 & 0.5139 & -0.5444 & 0.4962 & 99 & 7 & 2e-15 & 5e-09 & 1e-09 & 0.23 & 0.03 + 224 & -8.3200 & 0.5970 & -0.5816 & -0.4740 & -0.2842 & 65 & 73 & 2e-15 & 8e-09 & 3e-09 & 0.16 & 0.02 + 345 & -10.7440 & 0.4664 & 0.4153 & -0.5880 & -0.5140 & 47 & 181 & 2e-15 & 4e-09 & 3e-09 & 0.12 & 0.03 + the first experiment uses @xmath111 to look for local maxima . we find all five local maxima . there are several monotonicity violations , including at least one for @xmath160 that is relatively large . these violations indicate that a step went outside the region of local convexity . nevertheless , in all cases the algorithm is able to recover and converge , as can be seen from the small error . in general , these monotonicity violations do not cause the algorithm to fail . however , such violations can be avoided by increasing @xmath101 , the tolerance on the definiteness of the hessian matrix . once @xmath101 is large enough , the shift will be so great that the function will be convex over the entire unit sphere . the downside of choosing a large value for @xmath101 ( and the shift @xmath161 ) is that convergence will be slow . the second experiment uses @xmath162 to look for local minima . we find all five local minima . there are several monotinicity violations , but they are all small . next we consider a different type of tensor eigenapair that also conforms to the generalized tensor eigenpair framework . d - eigenpairs @xcite [ ex : deig ] qi , wang , and wu @xcite propose d - eigenpairs for diffusion kurtosis imaging ( dki ) . the tensors @xmath163}}$ ] are specified in ( and , respecitively ) . we consider this example here since it can be expressed as a generalized tensor eigenproblem . and @xmath29 are provided in @xcite . we were unable to validate the solutions provided in the original paper . it is not clear if this is to a lack of precision or a typo in paper . here , the problem is rescaled as well : @xmath29 is multiplied by @xmath164 , @xmath23 is divided by @xmath165 , and @xmath8 is divided by @xmath166 . ] there are a total of 13 distinct real - valued d - eigenpairs , computed by solving the polynomial equations using mathematica and listed in ( ) ; there are four maxima and three minima . shows the eigenpairs calculated by . the error is computed as @xmath167 . with 100 random starts , we find the four local maxima with @xmath111 . likewise , with 100 random starts , we find the three local minima with @xmath112 . there are no violations to monotonicity . @xmath1113 31 & 0.5356 & 0.9227 & -0.1560 & -0.3526 & 39 & & 4e-08 & 5e-09 & 0.08 & 0.01 + 19 & 0.4359 & 0.5388 & 0.8342 & -0.1179 & 48 & & 3e-08 & 4e-09 & 0.09 & 0.02 + 25 & 0.2514 & 0.3564 & -0.8002 & 0.4823 & 67 & & 4e-08 & 3e-09 & 0.13 & 0.01 + 25 & 0.2219 & 0.2184 & 0.3463 & 0.9124 & 34 & & 6e-08 & 8e-09 & 0.07 & 0.01 + @xmath1123 39 & -0.0074 & 0.3669 & 0.5346 & -0.7613 & 13 & & 1e-08 & 4e-09 & 0.03 & 0.01 + 37 & -0.1242 & 0.9439 & 0.1022 & 0.3141 & 51 & & 5e-08 & 5e-09 & 0.10 & 0.01 + 24 & -0.3313 & 0.2810 & -0.9420 & -0.1837 & 27 & & 2e-08 & 4e-09 & 0.06 & 0.01 + here we consider a randomly generated problem . we use the randomly generated @xmath158}}$ ] described in . however , we need a method to generate a positive definite @xmath3 . we use the notation @xmath168}}$ ] to denote tensor - matrix multiplication in which the tensor @xmath169 is multiplied by a matrix @xmath170 in every mode , i.e. , @xmath171 [ thm : posdef ] let @xmath172 be symmetric . for @xmath0 even , define @xmath45}}$ ] as @xmath173 . if @xmath174 is a real - valued eigenpair of @xmath175 and @xmath58 , then @xmath57 is a z - eigenpair of @xmath141 with @xmath176 . furthermore , @xmath177 for any @xmath178 with @xmath179 . let @xmath174 be an eigenpair of @xmath175 such that @xmath180 . noting that @xmath181 , we have @xmath182 to prove the lower bound , let @xmath178 with @xmath179 . we can write @xmath183 as a linear combination of the eigenvectors of @xmath175 , i.e. , @xmath184 and @xmath185 . then @xmath186 hence , the claim . random[ex : random ] we use the same randomly - generated symmetric tensor @xmath187}}$ ] as for and listed in ( ) . to generate a random positive definite symmetric tensor @xmath188}}_+ = { \mathbb{s}^{[6,4]}}_+$ ] , we use . ( note that this approach samples a convenient subset of @xmath36}}_+$ ] and does not draw from the entire space . ) we compute a matrix @xmath189 , where @xmath190 is a random orthonormal matrix and @xmath191 is a diagonal matrix with entries selected uniformly at random from @xmath192 \cup [ \gamma,1]$ ] with @xmath193{0.1 } = \sqrt[6]{0.1}$ ] . we let @xmath194 , so that @xmath3 has all its z - eigenvalues in @xmath195 $ ] and is positive definite . in this case , the randomly selected diagonal for @xmath29 is @xmath196 . the random @xmath141 is then rounded to four decimal places , and the entries are given in ( ) . its minimum z - eigenvalue ( computed by geap ) is @xmath197 , as expected . there are a total of 26 real - valued @xmath3-eigenpairs of @xmath2 , listed in ( ) . there are three maxima and four minima . shows the generalized eigenpairs calculated by . the error is computed as @xmath167 . with 1000 random starts , we find the three local maxima with @xmath111 . likewise , with 1000 random starts , we find the four local minima with @xmath112 . there are only small violations to monotonicity ; the maximum of any violation was o(@xmath198 ) . @xmath1114 683 & 11.3476 & 0.4064 & 0.2313 & 0.8810 & 0.0716 & 59 & 420 & 4e-15 & 5e-09 & 4e-09 & 0.14 & 0.03 + 128 & 3.7394 & 0.2185 & -0.9142 & 0.2197 & -0.2613 & 140 & 11 & 2e-15 & 1e-08 & 3e-09 & 0.30 & 0.04 + 189 & 2.9979 & 0.8224 & 0.4083 & -0.0174 & -0.3958 & 23 & 9 & 1e-15 & 3e-09 & 1e-09 & 0.06 & 0.01 + @xmath1124 151 & -1.1507 & 0.1935 & 0.5444 & 0.2991 & -0.7594 & 88 & & 8e-09 & 8e-10 & 0.19 & 0.02 + 226 & -3.2777 & 0.6888 & -0.6272 & -0.2914 & -0.2174 & 33 & 14 & 1e-15 & 6e-09 & 2e-09 & 0.08 & 0.01 + 140 & -3.5998 & 0.7899 & 0.4554 & 0.2814 & 0.2991 & 22 & 21 & 1e-15 & 2e-09 & 1e-09 & 0.05 & 0.01 + 483 & -6.3985 & 0.0733 & 0.1345 & 0.3877 & 0.9090 & 82 & 73 & 2e-15 & 9e-09 & 3e-09 & 0.17 & 0.03 + like its predecessor ss - hopm @xcite , the geap method has the desirable qualities of guaranteed convergence and simple implementation . additionally , the adaptive choice of @xmath84 in geap ( as opposed to ss - hopm ) means that there are no parameters for the user to specify . also like ss - hopm , the geap method can only converge to local maxima and minima of and so will miss any saddle point solutions . nevertheless , the largest and smallest magnitude eigenvalues can always be discovered by geap since they will not be saddle points . an alternative to geap is to solve or using a numerical nonlinear , homotopy , or optimization approach . the advantage of geap is that is guarantees decrease at each iteration _ without _ any globalization techniques ( like line search or trust region ) and is generally as cheap or cheaper per iteration than any competing numerical method . the disadvantage is that the rate of convergence of geap is only linear , as opposed to quadratic for , say , newton s method . han @xcite proposed an unconstrained variations principle for finding generalized eigenpairs . in the general case , the function to be optimized is @xmath199 the @xmath200 has the same meaning as for geap : choosing @xmath111 finds local maxima and @xmath112 finds local minima . for comparison , the final solution is rescaled as @xmath201{{{{\boldsymbol { \mathscr{\makeuppercase{b}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^m}}$ ] , and then we calculate @xmath202 ( since @xmath53 ) . the computational experiment settings are the same as specified in . han used the matlab optimization toolbox , and we use version 2.6.1 . folliwng han , we use the ` fminunc ` function and the default settings from calling ` optimset(fminunc ) ` except that we explicitly specify * ` gradobj : on ` * ` largescale : off ` * ` tolx:1e-10 ` * ` tolfun:1e-8 ` * ` maxiter:10000 ` * ` display : off ` this means that the toolbox uses a quasi - newton method with a line search which should have superlinear convergence . the results of han s method for are shown in . for each eigenpair , the table lists the number of occurrences in the 1000 experiments , the median number of function evaluations ( fevals ) until convergence , the average error and standard deviation in the final result , and the average run time and standard deviation . the error is computed as @xmath167 ; both methods achieved comparable errors . @xmath1114 718 & 11.3476 & 0.5544 & 0.3155 & 1.2018 & 0.0977 & 45 & 1e-08 & 2e-08 & 0.17 & 0.06 + 134 & 3.7394 & 0.2642 & -1.1056 & 0.2657 & -0.3160 & 31 & 4e-09 & 7e-09 & 0.12 & 0.05 + 144 & 2.9979 & 1.0008 & 0.4969 & -0.0212 & -0.4817 & 31 & 4e-09 & 5e-09 & 0.12 & 0.05 + 4 & & 0.21 & 0.10 + @xmath1124 72 & -1.1507 & 0.2291 & 0.6444 & 0.3540 & -0.8990 & 34 & 9e-09 & 3e-08 & 0.14 & 0.06 + 150 & -3.2777 & 0.8349 & -0.7603 & -0.3532 & -0.2635 & 33 & 5e-09 & 7e-09 & 0.14 & 0.07 + 148 & -3.5998 & 1.0486 & 0.6046 & 0.3736 & 0.3971 & 41 & 6e-09 & 8e-09 & 0.16 & 0.08 + 624 & -6.3985 & 0.1003 & 0.1840 & 0.5305 & 1.2438 & 48 & 7e-09 & 1e-08 & 0.19 & 0.08 + 4 & & 0.10 & 0.11 + 2 & & 0.23 & 0.02 + for @xmath111 , han s method finds all three local maxima , though it fails to converge within 10,000 iterations for four starting points . there are no consistent results with respect to time . han s method is faster than geap for @xmath203 but slower for the other two eigenpairs . this is consistent if we compare the number of function evaluations and the number of iterations for geap , which are measuring comparable amounts of work . for @xmath112 , han s method finds all four local minima , but it fails to converge for two starting points and converges to wrong solutions for four starting points . in those four cases , it terminated because the gradient was small ( flag = 1 ) for three cases and the fourth it stopped because it could no improve the function value ( flag = 5 ) . in this case , geap was faster on average for all eigenpairs . in general , han s method represents an alternative approach to solving the generalized tensor eigenpair problem . in @xcite , han s method was compared to ss - hopm with a fixed shift ( for z - eigenpairs only ) and was superior . however , geap is usually as fast as han s method and perhaps a little more robust in terms of its convergence behavior . the speed being similar is thanks to the adaptive shift in geap . it may be that han s method could avoid problems of converging to incorrect solutions with tighter tolerances , but then the speed would be slower . since is a polynomial system of equations , we can also consider a polynomial solver approach . this does not scale to larger problems and may be slow even for small problems . nevertheless , it finds _ all _ solutions ( even saddle points ) . we have used the grbner basis polynomial solver ` nsolve ` in mathematica to compute the full set of solutions for the problems discussed in this paper . in terms of methods specifically geared to tensor eigenvalues , most work has focused on computing the largest h - eigenvalue for a _ nonnegative _ tensor : @xcite . the method of liu , zhou , and ibrahim @xcite is guaranteed to always find the largest eigenvalue and also uses a `` shift '' approach . the paper has proposed two improvements to the ss - hopm method @xcite . first , we have adapted the method to the _ generalized _ tensor eigenproblem . second , we have proposed a method for adaptively and automatically selecting the shift , overcoming a major problem with the ss - hopm method because choosing the shift too large dramatically slows convergence whereas choosing it too small can cause the method to fail completely . we have tested our method numerically on several problems from the literature , including computing of z- , h- , and d - eigenpairs . we have also proposed a novel method for generating random symmetric positive definite tensors . as this paper was in review , a new method has been proposed to compute _ all _ real general eigenvalues using jacobian semidefinite programming relaxations @xcite . comparing to this method will be a topic of future study . first , we consider the gradient and hessian of the general function @xmath204 let @xmath205 denote @xmath206 . from matrix calculus , the gradient of @xmath48 is @xmath207 here we have dropped the argument , @xmath8 , to simply the notation . let @xmath208 denote @xmath209 . the hessian of @xmath48 is @xmath210\\ -\frac{1}{{f_3{}}^2 } \bigl [ { f_1{}}{f_2{}}{{{\bm { \mathbf{\makeuppercase{h}}}}}_3{}}+ { f_2{}}({{{{\bm { \mathbf{\makelowercase{g}}}}}_1{}}\circledcirc { { { \bm { \mathbf{\makelowercase{g}}}}}_3 { } } } ) + { f_1{}}({{{{\bm { \mathbf{\makelowercase{g}}}}}_2{}}\circledcirc { { { \bm { \mathbf{\makelowercase{g}}}}}_3 { } } } ) \bigr].\end{gathered}\ ] ] now we specialize @xmath48 to : let @xmath211 , @xmath212 , and @xmath213 . the following derivatives are proved in @xcite : @xmath214 we need only consider the case for @xmath50 , so we may assume @xmath215 putting everything together , we have for @xmath50 , @xmath216.\ ] ] for the hessian , assuming @xmath50 , we have @xmath217 \\ - \frac{m}{({{{\boldsymbol { \mathscr{\makeuppercase{b}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^m})^2 } \biggl [ ( m-1 ) { { { \boldsymbol { \mathscr{\makeuppercase{a}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^m}{{\boldsymbol { \mathscr{\makeuppercase{b}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^{m-2 } + m \bigl ( { { { { \boldsymbol { \mathscr{\makeuppercase{a}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^{m-1}}\circledcirc { { { \boldsymbol { \mathscr{\makeuppercase{b}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^{m-1 } } } \bigr ) + m { { { \boldsymbol { \mathscr{\makeuppercase{a}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^m}\bigl ( { { { { \bm { \mathbf{\makelowercase{x}}}}}}\circledcirc { { { \boldsymbol { \mathscr{\makeuppercase{b}}}}}{{{\bm { \mathbf{\makelowercase{x}}}}}}^{m-1 } } } \bigr ) \biggr].\end{gathered}\ ] ] the tensor for comes from example 1 in kofidis and regalia @xcite and is specified in . @xmath218 the tensor @xmath219 used for is randomly generated as described in ; its entries are specified in . @xmath220 the tensor used in . the dki tensor @xmath147}}$ ] ( called @xmath221 in the original paper ) is the symmetric tensor defined by the unique elements shown in . the tensor @xmath141 is the symmetrized outer product of the matrix @xmath29 with itself where @xmath222 so @xmath3 is the tensor whose unique elements are given in @xmath223 @xmath224 the tensor @xmath219 used in is the same as is used in and specified in . the tensor @xmath141 is a random positive definite tensor from ; it entries are specified in . a polynomial system solver ( ` nsolve ` ) using a grbner basis method is available in mathematica and has been employed to generate a complete list of eigenpairs for the examples in this paper in tables [ tab : kore02ex1_ep][tab : random_eigs ] . -1.0954 & 0.5915 & -0.7467 & -0.3043 & 1.86 & 2.75 & minima + -0.5629 & 0.1762 & -0.1796 & 0.9678 & 1.63 & 2.38 & minima + -0.0451 & 0.7797 & 0.6135 & 0.1250 & 0.82 & 1.25 & minima + 0.1735 & 0.3357 & 0.9073 & 0.2531 & -1.10 & 0.86 & saddle + 0.2433 & 0.9895 & 0.0947 & -0.1088 & -1.19 & 1.46 & saddle + 0.2628 & 0.1318 & -0.4425 & -0.8870 & 0.62 & -2.17 & saddle + 0.2682 & 0.6099 & 0.4362 & 0.6616 & -1.18 & 0.79 & saddle + 0.3633 & 0.2676 & 0.6447 & 0.7160 & -1.18 & -0.57 & maxima + 0.5105 & 0.3598 & -0.7780 & 0.5150 & 0.59 & -2.34 & saddle + 0.8169 & 0.8412 & -0.2635 & 0.4722 & -2.26 & -0.90 & maxima + 0.8893 & 0.6672 & 0.2471 & -0.7027 & -1.85 & -0.89 & maxima + @xmath226 & @xmath227 & @xmath228 & @xmath229 & @xmath230 & @xmath231 & @xmath232 & @xmath233 & minima + @xmath234 & @xmath235 & @xmath236 & @xmath237 & @xmath238 & @xmath239 & @xmath240 & @xmath241 & minima + @xmath242 & @xmath243 & @xmath244 & @xmath245 & @xmath246 & @xmath247 & @xmath248 & @xmath249 & minima + @xmath250 & @xmath251 & @xmath252 & @xmath253 & @xmath254 & @xmath255 & @xmath256 & @xmath257 & minima + @xmath258 & @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath263 & @xmath264 & @xmath265 & saddle + @xmath266 & @xmath267 & @xmath268 & @xmath269 & @xmath270 & @xmath271 & @xmath272 & @xmath273 & saddle + @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 & @xmath279 & @xmath280 & @xmath281 & minima + @xmath282 & @xmath283 & @xmath284 & @xmath285 & @xmath286 & @xmath287 & @xmath288 & @xmath289 & saddle + @xmath290 & @xmath291 & @xmath292 & @xmath293 & @xmath294 & @xmath295 & @xmath296 & @xmath297 & saddle + @xmath298 & @xmath299 & @xmath300 & @xmath301 & @xmath302 & @xmath303 & @xmath304 & @xmath305 & saddle + @xmath306 & @xmath307 & @xmath308 & @xmath309 & @xmath310 & @xmath311 & @xmath312 & @xmath313 & saddle + @xmath314 & @xmath315 & @xmath316 & @xmath317 & @xmath318 & @xmath319 & @xmath320 & @xmath321 & saddle + @xmath322 & @xmath323 & @xmath324 & @xmath325 & @xmath326 & @xmath327 & @xmath328 & @xmath329 & saddle + @xmath330 & @xmath331 & @xmath332 & @xmath333 & @xmath334 & @xmath335 & @xmath336 & @xmath337 & saddle + @xmath338 & @xmath339 & @xmath340 & @xmath341 & @xmath342 & @xmath343 & @xmath344 & @xmath345 & saddle + @xmath346 & @xmath347 & @xmath348 & @xmath349 & @xmath350 & @xmath351 & @xmath352 & @xmath353 & saddle + @xmath354 & @xmath355 & @xmath356 & @xmath357 & @xmath358 & @xmath359 & @xmath360 & @xmath361 & saddle + @xmath362 & @xmath363 & @xmath364 & @xmath365 & @xmath366 & @xmath367 & @xmath368 & @xmath369 & saddle + @xmath370 & @xmath371 & @xmath372 & @xmath373 & @xmath374 & @xmath375 & @xmath376 & @xmath377 & saddle + @xmath378 & @xmath379 & @xmath380 & @xmath381 & @xmath382 & @xmath383 & @xmath384 & @xmath385 & saddle + @xmath386 & @xmath387 & @xmath388 & @xmath389 & @xmath390 & @xmath391 & @xmath392 & @xmath393 & saddle + @xmath394 & @xmath395 & @xmath396 & @xmath397 & @xmath398 & @xmath399 & @xmath400 & @xmath401 & saddle + @xmath402 & @xmath403 & @xmath404 & @xmath405 & @xmath406 & @xmath407 & @xmath393 & @xmath408 & saddle + @xmath409 & @xmath410 & @xmath411 & @xmath412 & @xmath413 & @xmath414 & @xmath415 & @xmath416 & saddle + @xmath417 & @xmath418 & @xmath419 & @xmath420 & @xmath421 & @xmath422 & @xmath423 & @xmath424 & saddle + @xmath425 & @xmath426 & @xmath427 & @xmath428 & @xmath429 & @xmath430 & @xmath431 & @xmath432 & saddle + @xmath433 & @xmath434 & @xmath435 & @xmath436 & @xmath437 & @xmath438 & @xmath439 & @xmath440 & saddle + @xmath441 & @xmath442 & @xmath443 & @xmath444 & @xmath445 & @xmath446 & @xmath447 & @xmath448 & maxima + @xmath449 & @xmath450 & @xmath451 & @xmath452 & @xmath453 & @xmath454 & @xmath455 & @xmath456 & saddle + @xmath457 & @xmath458 & @xmath459 & @xmath460 & @xmath461 & @xmath462 & @xmath463 & @xmath464 & maxima + @xmath465 & @xmath466 & @xmath467 & @xmath468 & @xmath469 & @xmath470 & @xmath471 & @xmath472 & maxima + @xmath473 & @xmath474 & @xmath475 & @xmath476 & @xmath477 & @xmath478 & @xmath479 & @xmath480 & saddle + @xmath481 & @xmath482 & @xmath483 & @xmath484 & @xmath485 & @xmath486 & @xmath487 & @xmath488 & maxima + @xmath489 & @xmath490 & @xmath491 & @xmath492 & @xmath493 & @xmath494 & @xmath495 & @xmath496 & maxima + -0.3313 & 0.2309 & -0.7741 & -0.1509 & 1.02 & 2.11 & minima + -0.1242 & 0.6577 & 0.0712 & 0.2189 & 0.35 & 1.25 & minima + -0.0074 & 0.2161 & 0.3149 & -0.4485 & 0.36 & 0.46 & minima + 0.0611 & 0.6113 & -0.4573 & 0.1181 & -0.63 & 1.14 & saddle + 0.1039 & 0.3314 & 0.5239 & 0.3084 & -0.46 & 0.63 & saddle + 0.2009 & 0.2440 & -0.1250 & 0.4601 & -0.32 & 0.07 & saddle + 0.2056 & 0.1211 & -0.2367 & -0.4766 & -0.29 & 0.13 & saddle + 0.2219 & 0.1143 & 0.1812 & 0.4773 & -0.08 & -0.20 & maxima + 0.2431 & 0.0943 & -0.6840 & 0.2905 & 0.18 & -1.11 & saddle + 0.2514 & 0.2485 & -0.5579 & 0.3363 & -0.14 & -0.71 & maxima + 0.3827 & 0.6236 & 0.3954 & -0.1678 & -1.58 & 0.32 & saddle + 0.4359 & 0.4336 & 0.6714 & -0.0949 & -0.43 & -1.64 & maxima + 0.5356 & 0.6638 & -0.1123 & -0.2537 & -0.48 & -1.43 & maxima + -6.3985 & 0.0733 & 0.1345 & 0.3877 & 0.9090 & 20.43 & 4.93 & 11.20 & minima + -3.5998 & 0.7899 & 0.4554 & 0.2814 & 0.2991 & 8.05 & 10.39 & 12.41 & minima + -3.2777 & 0.6888 & -0.6272 & -0.2914 & -0.2174 & 8.27 & 3.65 & 5.95 & minima + -1.7537 & 0.6329 & -0.2966 & -0.6812 & -0.2180 & -4.25 & 3.00 & 5.56 & saddle + -1.1507 & 0.1935 & 0.5444 & 0.2991 & -0.7594 & 0.73 & 3.54 & 4.20 & minima + -1.0696 & 0.1372 & 0.5068 & 0.0665 & -0.8485 & -1.54 & 3.30 & 3.64 & saddle + -1.0456 & 0.2365 & 0.4798 & -0.7212 & 0.4402 & -1.16 & 1.54 & 2.57 & saddle + -0.7842 & 0.5409 & 0.3388 & 0.4698 & 0.6099 & 16.02 & 8.79 & -12.47 & saddle + -0.7457 & 0.6348 & 0.5354 & -0.4388 & 0.3434 & 2.49 & 0.94 & -1.59 & saddle + -0.2542 & 0.3900 & -0.1333 & 0.4946 & -0.7652 & -2.51 & 2.99 & 0.93 & saddle + -0.2359 & 0.6956 & -0.1369 & 0.3550 & -0.6094 & 6.38 & 2.23 & -1.27 & saddle + 0.0132 & 0.3064 & 0.0541 & 0.3111 & -0.8980 & -5.33 & -2.36 & 2.21 & saddle + 0.1633 & 0.4278 & -0.6578 & -0.2545 & 0.5652 & -2.42 & 3.86 & 2.36 & saddle + 0.3250 & 0.5265 & 0.4653 & 0.0927 & 0.7055 & 7.50 & -12.05 & -3.41 & saddle + 0.5206 & 0.3738 & -0.4806 & -0.6066 & 0.5111 & 3.19 & -2.27 & -1.47 & saddle + 0.5463 & 0.5157 & -0.3055 & -0.3313 & -0.7287 & -9.91 & -3.67 & 1.37 & saddle + 0.5945 & 0.4015 & 0.8447 & 0.1782 & -0.3058 & -3.70 & 4.95 & 1.87 & saddle + 0.6730 & 0.9634 & -0.0009 & 0.2396 & -0.1204 & -5.84 & 7.88 & 1.78 & saddle + 0.8862 & 0.3559 & 0.8571 & -0.1675 & -0.3326 & 3.55 & -2.24 & -2.63 & saddle + 1.2962 & 0.9849 & 0.0018 & -0.1681 & 0.0419 & 2.20 & -5.97 & -3.18 & saddle + 1.4646 & 0.7396 & 0.4441 & 0.4009 & -0.3083 & 8.41 & -2.08 & -7.72 & saddle + 2.9979 & 0.8224 & 0.4083 & -0.0174 & -0.3958 & -4.00 & -5.46 & -6.56 & maxima + 3.5181 & 0.4494 & -0.7574 & 0.4502 & -0.1469 & -9.40 & 1.89 & -2.83 & saddle + 3.6087 & 0.0340 & -0.8989 & -0.0373 & -0.4353 & 0.87 & -8.03 & -5.77 & saddle + 3.7394 & 0.2185 & -0.9142 & 0.2197 & -0.2613 & -8.72 & -0.90 & -3.34 & maxima + 11.3476 & 0.4064 & 0.2313 & 0.8810 & 0.0716 & -7.20 & -18.98 & -21.53 & maxima + , _ an unconstrained optimization approach for finding real eigenvalues of even order symmetric tensors _ , numerical algebra , control and optimization ( naco ) , 3 ( 2012 ) , pp . 583599 , http://dx.doi.org/10.3934/naco.2013.3.583[doi : ] . , _ on the best rank-1 approximation of higher - order supersymmetric tensors _ , siam journal on matrix analysis and applications , 23 ( 2002 ) , pp . 863884 , http://dx.doi.org/10.1137/s0895479801387413[doi : ] . , _ singular values and eigenvalues of tensors : a variational approach _ , in camsap05 : proceeding of the ieee international workshop on computational advances in multi - sensor adaptive processing , 2005 , pp . 129132 , http://dx.doi.org/10.1109/camap.2005.1574201[doi : ] . , _ an always convergent algorithm for the largest eigenvalue of an irreducible nonnegative tensor _ , journal of computational and applied mathematics , 235 ( 2010 ) , pp . 286292 , http://dx.doi.org/10.1016/j.cam.2010.06.002[doi : ] .

we report the results of an extensive imaging and spectroscopic survey in the goods - north field completed using deimos on the keck ii telescope . observations of 2018 targets in a magnitude - limited sample of 2911 objects to yield secure redshifts for a sample of 1440 galaxies and agn plus 96 stars . in addition to redshifts and associated quality assessments , our catalog also includes photometric and astrometric measurements for all targets detected in our-band imaging survey of the goods - north region . we investigate various sources of incompleteness and find the redshift catalog to be 53% complete at its limiting magnitude . the median redshift of is lower than in similar deep surveys because we did not select against low - redshift targets . comparison with other redshift surveys in the same field , including a complementary hawaii - led deimos survey , establishes that our velocity uncertainties are as low as km s for red galaxies and that our redshift confidence assessments are accurate . the distributions of rest - frame magnitudes and colors among the sample agree well with model predictions out to and beyond . we will release all survey data , including extracted 1-d and sky - subtracted 2-d spectra , thus providing a sizable and homogeneous database for the goods - north field which will enable studies of large scale structure , spectral indices , internal galaxy kinematics , and the predictive capabilities of photometric redshifts .

astro-ph0401353

in 1995 , the hubble deep field ( hdf ) became the most well - studied extragalactic area in all of observational astronomy . extremely deep optical exposures taken with the _ hubble space telescope ( hst ) _ were eventually joined by additional imaging and spectroscopic observations from many observatories covering all wavelengths ( cf . @xcite and references therein ) . this combined dataset which was ultimately expanded to include a complementary field in the southern sky painted an extraordinarily detailed picture of a `` blank '' region of sky which turned out to be teeming with galaxies , providing astronomers with an exquisite window into the high - redshift universe . the result was a series of important advances in the properties of distant galaxies and large - scale structure in the universe . many of these key discoveries depended critically upon knowledge of galaxy redshifts , the vast majority of which were obtained by a variety of observers using the lris spectrograph @xcite on the twin 10 m telescopes of the w. m. keck observatory @xcite . nearly ten years later , attention has returned to the northern hdf field ( hdf - n ) via an ambitious plan to use a new generation of ground- and space - based instruments for probing the distant universe in even greater detail . the great observatories origins deep survey ( goods ; @xcite ) project involves a coordinated effort by three of nasa s great observatories to obtain ultra - deep images of two selected fields in the optical ( via the _ hst _ advanced camera for surveys ) , infrared ( with the _ spitzer space telescope _ ) , and x - ray ( using _ chandra _ ) wavelength regimes . these two fields include a northern target area coincident with ( but significantly larger than ) the hdf - n plus a similarly - sized southern field coincident with the chandra deep field - south . again , intensive ground - based followup observations in all wavelength regimes are underway to exploit this high - resolution space - based imaging . high - quality spectra of the many galaxies in the goods fields will , as before , be essential to understand both large - scale structure and galaxy evolution within these regions . keck remains the most powerful northern - hemisphere observatory for completing deep surveys of faint galaxies . its efficient new deimos spectrograph @xcite on the keck ii telescope was designed specifically for acquiring multislit spectroscopy of over 100 faint galaxies at once , and with its @xmath6 field of view is ideally suited to survey the @xmath7 goods - north ( hereafter goods - n ) field . presented with this unique opportunity to make an important contribution to the community , the wmko director and scientific staff elected to complete a deep deimos survey of this region and committed to releasing the resulting data to the community as rapidly as possible . we report herein the results of this `` team keck '' treasury redshift survey ( hereafter , tkrs ) in the goods - n field . our survey is complemented by the hawaii - led keck / deimos survey in goods - n described in the accompanying report @xcite , and also by a gemini - north survey completed with gmos @xcite . in [ sxn - photo ] of this report we review the construction of the photometric catalog of the goods - n field on which we based our slitmask designs . section [ sxn - spec ] discusses the acquisition and reduction of the spectra and the determination of redshifts . section [ sxn - catalog ] concerns the redshift catalog , including the effects of target characteristics on the sample s completeness , the comparison of our survey with others in the same area , the accuracy of our redshifts and quality assessments , and the accuracy of photometric redshifts in this field . section [ sxn - analysis ] presents the redshift and rest - frame color distribution for our targets , and [ sxn - summary ] concludes the paper . to ensure timely access to a catalog of astrometric positions and approximate magnitudes for completing the spectroscopy , we generated our own independent catalog of sources in the goods - n field . we acquired a series of @xmath8 s @xmath1-band exposures using deimos in imaging mode on keck ii during the night of 2002 dec 30 ( ut ) , and complemented these with additional images obtained on 2003 feb 02 ( ut ) to complete coverage of the full acs imaging region . the exposures were offset and dithered to compensate for the gaps in the deimos ccd mosaic . all images were bias subtracted and gain corrected , then flatfielded with sky flats generated from the set of 14 images acquired in this field . we employed an automated procedure to detect cosmic rays on each image and ignored the affected pixels in all subsequent processing . the mean sky level on each image was measured via an iterative sigma clipping algorithm and subtracted to eliminate sky brightness variations from the data set . next , we used deimos images of an astrometric field to derive a distortion correction map which we applied to the images of the goods - n field . this allowed us to shift and co - add the science images , yielding a mosaic image that covered the entire @xmath7 first - epoch acs imaging field . for this survey , we only considered the portion of this mosaic image coincident with the first epoch _ hst _ goods - n observations . we obtained a first - order plate solution for this mosaic using several stars measured in the usno astrometric catalog @xcite . to generate our target catalog we used the sextractor package @xcite to detect positive fluctuations at least 20 px in size with a threshold of 1.0 times the rms sky noise . because of vignetting near the edges of the deimos imaging field of view , the final mosaic image contains a few regions with spurious detections . these bogus objects are generally very faint , and most were excluded from the final catalogue because they lie below the faint magnitude cutoff of the sample . although the usno catalog provided an effective first - order astrometric solution for the goods - n mosaic image , the small number of sources available in this field limited the accuracy of the solution . the resultant rms positional uncertainty of approximately @xmath9 is a significant fraction of the @xmath10 slit width for spectroscopy , and hence a more accurate solution was needed to produce acceptable slitmask designs . in order to improve the object positions , we used a catalogue of objects in this region kindly provided by the sdss team @xcite which produced a list of 130 stars common to both catalogs . this permitted us to refine the astrometric solution and obtain a final rms residual of @xmath11 , more than adequate for our needs . comparing our derived astrometric coordinates to those obtained independently by @xcite , we find no measureable offset in ( @xmath12 ) and a dispersion of @xmath13 , indicating excellent agreement . our coordinates agree less well in absolute zero point with those @xcite obtained from _ hst _ imaging , against which the dispersion is @xmath14 but the offsets are @xmath15 and @xmath16 ; this suggests a slight error in absolute astrometric positions for the acs catalog . for several saturated stars which we wanted to use for rough slitmask alignment purposes , we derived estimated celestial coordinates positions visually by estimating the ( @xmath12 ) coordinates for the location at the convergence points of the diffraction spikes . the unsaturated stars also served as photometric calibrators . the magnitudes we used were ` mag_best ` calculated by sextractor , which are designed to estimate the total light . since only @xmath17 magnitudes were available from the sdss database , we were unable to include any color terms in our photometric calibration . for the 130 stars shared between the tkrs and sdss catalogs , the rms difference is @xmath18 mag . the final magnitudes are in the ab system @xcite , roughly equivalent to the sdss @xmath17 . figure [ fig - filters ] compares the sdss @xmath17 and deimos @xmath1 instrumental response functions . system ( telescope , filter , and detector ) response curves of @xmath1 band filter used in the deimos imaging ( _ solid line _ ) and the sdss @xmath17 filter ( _ dotted line _ ) . the @xmath17 filter is part of the system devised to calibrate the sdss survey @xcite , and is shown here to illustrate the differences in bandpass of the filters . note that in the actual calibration of the tkrs catalogue , the @xmath19-band data from the sdss dr1 release were used.,scaledwidth=47.5% ] sextractor also provides for each object an estimate of its orientation as represented by the major axis position angle . this information can be used to orient spectroscopic slits to the major axis of the galaxy and thus obtain both improved @xmath20 spectra and more accurate rotation curves . we compared the position angles derived from the deimos mosaic image with independent measurements obtained from _ images by the ucsc deep1 team in the hubble deep field and flanking fields using two - dimensional two - parameter fits to the images @xcite . a histogram comparing the difference in position angles using @xmath21 bins is shown in fig . [ fig - posang ] . the solid histogram shows the distribution measured for galaxies with @xmath22 , while the dotted line represents the fainter galaxies . both distributions have been normalized by the total number of galaxies in each subsample . although both distributions are peaked at values close to @xmath23 , indicating that the two datasets generally agree , the peak is far more pronounced for the brighter galaxy sample . we conclude that the position angles measured by sextractor are adequate for the brighter galaxies . distribution of the difference in position angle measured from ground - based deimos @xmath1-band images vs. _ hst _ wfpc2 @xmath24-band images in the hdf and flanking fields @xcite , binned in @xmath21 intervals . the solid histogram represents brighter galaxies with @xmath25 , while the solid line represents the position angle differences for galaxies fainter than 23.5 . both bright and faint galaxies show a peak at position angle differences at @xmath23 , but whereas this peak is well defined for the brighter galaxies , for fainter ones the smoother distribution indicates a poor correlation . , scaledwidth=47.5% ] the selection of objects for spectroscopic observations followed roughly the same guidelines as used by the deep2 redshift survey @xcite , with one major difference : no attempt was made to use brightness , color , or other information to exclude stars and lower - redshift galaxies from the sample . all objects brighter than the limiting magnitude were potential spectroscopic targets , regardless of whether their appearance was stellar or galactic . at the magnitude limit of this survey the number of stars relative to the number of galaxies is small , so that the adopted strategy avoids excluding agn - dominated galaxies . deimos uses slitmasks to record over 100 broadband spectra at once . the limiting magnitude of the spectroscopic sample , the number of available nights and uniformity of wavelength coverage were all considered in designing slitmasks for our survey . the magnitude cutoff was set at @xmath26 in order to provide sufficient source density per slitmask , yielding a sample of 2911 candidate targets for inclusion on masks . to improve consistency of wavelength coverage , objects were preferentially placed near the center of the @xmath27 available in the spatial direction on each deimos mask . automated slitmask design software developed for the deep2 survey then selected appropriate targets from the catalog to produce a set of 18 slitmasks , each displaced from the next by @xmath28 , to cover the goods - n field . the resulting distribution in equatorial coordinates of the 2018 objects in the spectroscopic sample is shown in fig . [ fig - radec - all ] . all slits were tilted by a minimum amount with respect to the ccd rows in order to improve the wavelength sampling of the spectra ; the slight offset between adjacent columns in the spectra serves to dither the night sky lines and thus permit a more accurate reconstruction of the background spectrum . for objects exhibiting significant elongation in the deimos images , slits were typically aligned with the major axis to improve both @xmath20 and rotational velocity measurements . for closely spaced objects , slit position angles were chosen such that the spectra of both objects were recorded . however , slits were never tilted by more than @xmath29 from the nominal position angle of the slitmask . spectroscopic observations of the goods - n field were completed using deimos on the keck ii telescope during early - to - mid 2003 as detailed in table [ tab - obslog ] . each observation used the 600 l mm@xmath4 grating blazed at 7500 , with the gg455 order - blocking filter eliminating all flux below 4550 . the central wavelength was set at 7200 , providing nominal spectral coverage of 46009800 at a fwhm resolution of @xmath30 . eleven masks were observed during time allocated to the wmko director , and the remaining seven were generously completed by the deep2 redshift survey team . each slitmask was observed for a total on - source integration time of 3600 s , broken up into @xmath31 s integrations to allow rejection of cosmic rays . no dithering was performed between exposures because of the use of tilted slits and because the minor fringing pattern present in deimos images is sufficiently corrected by the use of flat field images . ccccrrrr[tbp ] 1 & 2003 may 02 & 12 36 30.81 & 62 17 01.9 & 44&117 & 72 & 62 + 2 & 2003 may 02 & 12 36 33.44 & 62 16 33.9 & 44&118 & 91 & 77 + 3 & 2003 may 02 & 12 36 36.99 & 62 16 20.2 & 44&118 & 94 & 80 + 4 & 2003 may 03 & 12 36 39.63 & 62 15 52.2 & 44&111 & 97 & 87 + 5 & 2003 may 03 & 12 36 43.18 & 62 15 38.5 & 44&112 & 99 & 88 + 6 & 2003 may 03 & 12 36 45.81 & 62 15 10.5 & 44&103 & 94 & 91 + 7 & 2003 may 04 & 12 36 49.37 & 62 14 56.9 & 44&103 & 69 & 67 + 8 & 2003 may 05 & 12 36 52.00 & 62 14 28.8 & 44&105 & 80 & 76 + 9 & 2003 may 06 & 12 36 55.56 & 62 14 15.2 & 44&110 & 89 & 81 + 10 & 2003 mar 26 & 12 36 58.19 & 62 13 47.1 & 44&105 & 62 & 59 + 11 & 2003 mar 26 & 12 37 01.75 & 62 13 33.5 & 44&114 & 88 & 77 + 12 & 2003 mar 26 & 12 37 04.38 & 62 13 05.5 & 44&112 & 93 & 83 + 13 & 2003 mar 26 & 12 37 07.93 & 62 12 51.8 & 44&102 & 73 & 72 + 14 & 2003 mar 26 & 12 37 10.57 & 62 12 23.8 & 44&118 & 75 & 64 + 15 & 2003 may 28 & 12 37 14.12 & 62 12 10.1 & 44&120 & 104 & 87 + 16 & 2003 may 28 & 12 37 16.75 & 62 11 42.1 & 44&125 & 98 & 78 + 17 & 2003 may 29 & 12 37 20.31 & 62 11 28.5 & 44&116 & 86 & 74 + 18 & 2003 may 29 & 12 37 19.86 & 62 11 21.3 & @xmath32&109 & 72 & 66 + to enable wavelength calibration and the removal of instrumental signatures , calibration data consisting of three internal quartz flats and a single arc lamp spectrum ( kr , ar , ne , xe ) were obtained for each mask in the daytime ; the deimos flexure compensation system ensured that these calibration images matched the science frames to better than @xmath33 px . the transparency and seeing conditions on mauna kea varied from fair to excellent during the observations , affecting the resultant data quality . all spectra were extracted using the fully - automated pipeline developed by the deep2 redshift survey team @xcite and generously made available to us . each spectral exposure was divided by its corresponding flatfield image to remove small amounts of ccd fringing . the three exposures for each mask were then combined to yield an image cleaned of cosmic rays . for each slitlet , a 2-d sky - subtracted spectrum is obtained by modeling the night - sky with a b - spline and subtracting it from the combined 2-d spectrum . wavelength calibration was achieved by fitting to the arc lamp emission lines . the sky - subtracted spectrum of the target was produced by summing over the rows containing the flux from the object as predicted by the tabulated position relative to the slit endpoints . the first phase of measuring redshifts involved fully automated processing of each spectrum using the same algorithm as the sdss @xcite , which calculates for each lag position ( _ i.e. _ , redshift ) the best representation of the source spectrum through the use of a set of template eigenspectra @xcite . these templates include several types of stars , a galaxy absorption spectrum , a galaxy emission - line spectrum , and an agn spectrum . the best fit is determined from the minimum @xmath34 in the redshift - lag and eigenspectra parameter space . the ten best fits are stored in the 1-d image header . in the second phase , all spectra were visually inspected by at least two team members , who would select both a redshift and a corresponding quality code ; objects for which no unambiguous redshift could be determined were also noted . due to low @xmath20 spectra or to poor matches between the spectra of the templates and some program objects , the automated redshift determination algorithm sometimes selected an incorrect redshift ( newman , private communication ) . however , all redshifts reported herein were agreed upon by two reviewers , and can thus be considered secure . the adopted redshift quality codes ( hereafter , @xmath35 ; see table [ tab - zqual2 ] ) are the same as used by the deep2 team @xcite . we consider the accuracy of these quality codes in [ sxn - zquality ] . rp0.5rrrrr[btp ] 4 & very secure redshift ( @xmath36 ) ; at least two spectral features identified & 1180 & 58.5 & 154 & 145 & 94.2 + 3 & secure redshift ( @xmath37 ) ; one strong line and another weak feature identified _ or _ single wide line , typically [ ] @xmath38 blend & 260 & 12.9 & 83 & 76 & 91.6 + 2 & uncertain redshift ; signal is present but no unambiguous spectral line identified & 268 & 13.3 & 69 & 26 & 37.7 + 1 & no redshift ; @xmath20 too poor & 183 & 9.1 & 28 & 1 & 3.6 + @xmath39 & star & 96 & 4.8 & 50 & 49 & 98.0 + @xmath40 & no redshift measured because of instrumental artifacts in spectrum & 31 & 1.5 & 6 & 0 & 0.0 + the results from the visual inspections were then collated and discrepant cases noted and reassessed by one individual . here , `` discrepant '' means that one observer assigned a @xmath35 value of 3 or 4 ( corresponding to a `` secure '' redshift ) whereas another would select @xmath41 or less , or that two observers would assign secure @xmath35 values but discrepant redshifts . the latter cases resulted when spectra of two different galaxies overlapped in the same slit . the distribution of redshift quality codes among the 2018 observed targets is shown in table [ tab - zqual2 ] . the number of secure redshift measurements ( @xmath42 ) is 1536 . the @xmath43 objects are technically not counted as failures ; hence , the overall efficiency is 77% ( 1536/1987 ) , which is comparable to the success rate of deep2 and similar redshift surveys ( _ e.g. _ , @xcite ) . the measurements resulting from our survey are presented in table [ tab - tkrs ] , a portion of which is reproduced here . this table appears in its entirety in the electronic version of the journal and also on the website devoted to the survey . photometric and spectroscopic data for all targets detected in our imaging survey appear here , in addition to alternate redshifts from other surveys when available . redshifts for targets in our survey are only given when they are classified as secure ( @xmath42 ) . also available on the survey website is the master table of measurements which combines the results of this survey with those of the parallel effort by @xcite . users of these data should note that the tabulated redshifts for our survey currently lack a heliocentric correction ; an adjustment will be applied in future releases , thus changing the zero point for the redshifts by no more than a few tens of km s@xmath4 . in the present section we examine the systematics of the present sample as a function of location in the deimos field of view , apparent magnitude , color , and surface brightness . the number of objects placed on each mask appears in table [ tab - obslog ] , which also shows both the number of targets for which secure redshifts were obtained ( @xmath44 ) and the success fraction per mask . on average , each mask contained about 112 objects , of which 76 ( 68% ) produced secure redshifts . however , these numbers do not take into account the @xmath43 objects representing failures due to instrumental effects , so these numbers understate the survey s redshift measuring efficiency . [ [ spatial - effects . ] ] spatial effects . + + + + + + + + + + + + + + + + figures [ fig - zqual - x ] and [ fig - zqual - y ] depict the distribution of galaxies as a function of the deimos mosaic @xmath45 and @xmath46 coordinates , respectively . the figures compare the distribution of all targets ( _ dark grey histograms _ ) to those which yielded secure redshifts ( _ light grey histogram _ ) . in fig . [ fig - zqual - x ] , the periodic depressions result in part from a lack of slits placed in the inter - ccd gap regions ( a product of the slitmask design ) and in part due to the decreased success rate for spectra taken acquired too close to the gaps and ruined . the decrease in number of objects at the edges of the deimos mosaic @xmath46 ( shorter ) axis is clearly seen in fig . [ fig - zqual - y ] , and this is probably an effect caused by the slitmask design algorithm . for galaxies and stars located at the edges there is only one chance of being placed on masks , in contrast to objects further from the edge . number of targets placed on masks ( _ dark grey histogram _ ) , and number of successful redshift measures ( _ light grey _ ) as a function of the deimos catalog @xmath47 coordinate , which is parallel to the long axis of the goods - n field . as explained in the text , the depressions at @xmath48 , 4000 and 6000 correspond to the gaps between ccds.,scaledwidth=47.5% ] coordinate , which is parallel to the goods - n shorter axis . the fall - off at both edges is caused by the fact that galaxies and stars in these positions have a smaller probability of being placed on masks , since they are only considered as potential mask candidates a smaller number of times . [ fig - zqual - y ] , scaledwidth=47.5% ] we show the spatial distribution of objects for which redshifts were not successfully measured in fig . [ fig - zfailure ] . objects which produced no measurable spectrum ( @xmath43 ) because of instrumental effects are shown as circles . since these failures are primarily due to spectra falling on the gaps between ccds in the deimos detector mosaic or to objects placed near the mask edges , they cluster in several lines across the goods - n field . targets whose redshifts are either completely unknown ( @xmath49 ) or highly uncertain ( @xmath41 ) are distributed randomly across the field , as expected . [ [ magnitude - effects . ] ] magnitude effects . + + + + + + + + + + + + + + + + + + our survey s sampling and success rates as a function of apparent magnitude are shown in fig . [ fig - completeness2 ] and table [ tab - completeness ] . for each 0.5 mag bin , the shaded histogram represents the differential quantity and the solid line ( when present ) indicates the integrated completeness down to that magnitude . figure [ fig - completeness2](a ) shows the number of available targets in our photometric catalog of the goods - n field , with the number in the rightmost bin reduced by the effect of our @xmath0 cutoff . panel ( b ) shows the fraction of targets we observed spectroscopically . our slitmasks targeted 7080% of the available objects over the range @xmath50 that contains the majority of the galaxies , falling significantly below this level at the bright end ( where sources were scarce ) and in the faintest bin ( containing faint targets against which selection was weighted during the slitmask design process ) . panel ( c ) displays the fraction of these spectra which yielded secure redshifts ( cf . fig . 4 in @xcite ) . our success rate is high ( @xmath51 ) for galaxies brighter than @xmath52 and quickly decreases to below 60% at the faint limit . panel ( d ) gives the fraction of all targets in the sample for which we obtained secure redshifts , and can be compared to the corresponding figures in @xcite and @xcite . overall , our survey obtained secure redshifts for 53% of all targets brighter than @xmath53 in the goods - n field . completeness indicators as a function of @xmath54 magnitude in the tkrs sample . ( a ) the number of available targets @xmath55 per 0.5 mag bin in our goods - n field target catalog , to the limiting magnitude of @xmath26 . ( b ) differential fraction of available targets ( relative to the total sample available within each 0.5 mag bin ) for which spectra were acquired ( _ shaded histogram _ ) , or the integrated fraction to that magnitude bin ( _ connected diamonds _ ) . ( c ) differential and integrated fraction of _ spectra _ which yielded secure redshifts . ( d ) differential and integrated fraction of _ available targets _ which yielded secure redshifts , including objects not observed.,scaledwidth=47.5% ] crrrrrrr[tbp ] 17.017.5 & 5 & 0 & 0 & 0 & 0 & 0 & 0 + 17.518.0 & 17 & 0 & 0 & 0 & 0 & 0 & 0 + 18.018.5 & 7 & 0 & 0 & 0 & 0 & 0 & 0 + 18.519.0 & 10 & 2 & 2 & 20 & 100 & 20 & 5 + 19.019.5 & 21 & 12 & 12 & 57 & 100 & 57 & 23 + 19.520.0 & 33 & 14 & 14 & 42 & 100 & 42 & 30 + 20.020.5 & 29 & 16 & 16 & 55 & 100 & 55 & 36 + 20.521.0 & 57 & 44 & 43 & 77 & 98 & 75 & 49 + 21.021.5 & 111 & 80 & 74 & 72 & 93 & 67 & 56 + 21.522.0 & 152 & 119 & 112 & 78 & 94 & 74 & 62 + 22.022.5 & 231 & 169 & 156 & 73 & 92 & 68 & 64 + 22.523.0 & 330 & 248 & 222 & 75 & 90 & 67 & 65 + 23.023.5 & 466 & 345 & 273 & 74 & 79 & 59 & 63 + 23.524.0 & 671 & 500 & 353 & 75 & 71 & 53 & 60 + 24.024.4 & 775 & 438 & 259 & 57 & 59 & 33 & 53 + [ [ color - effects . ] ] color effects . + + + + + + + + + + + + + + the recent release of multicolor photometry for the goods - n field @xcite enables tests of color dependence in our sample . in the following analyses , we have matched our target catalog derived from the deimos photometry with the giavalisco et al . multicolor photometry derived from the v1.0 goods - n acs images . objects in both catalogs number 2816 , with a detection rate of 2816/2911=90% . most non - matching galaxies in the deimos catalog are found in regions just outside that covered by the v1.0 goods - n catalog . figure [ fig - zqual - vi ] displays the distribution of the @xmath56 colors measured within a circular aperture of radius @xmath57 for all galaxies placed on mask , galaxies with secure redshifts ( @xmath58 ) and galaxies with unmeasurable or insecure redshifts ( @xmath35=1,2 ) . the objects with failed redshift measurements are preferentially bluer than the successful targets . figure [ fig - zqual - vi - panels ] compares the colors of targets with successful and failed redshift measurements in different apparent magnitude ranges , and shows that the distribution of failed measurements changes from relatively flat to become bluer than the overall distribution near the faint limit . number of galaxies placed on masks ( _ solid black line _ ) , and number of redshifts measured ( _ grey histogram _ ) , as a function of the @xmath56 color . also shown is the distribution of galaxies with redshift qualities @xmath49 and 2 ( _ dashed line _ ) . for the bluest galaxies the success rate ( ratio of grey histogram to solid line ) is under 50% , and , as discussed in the text and fig . [ fig - apparent - cm ] , is likely due to unidentified high - redshift galaxies . , scaledwidth=47.5% ] histogram showing the number of galaxies placed on masks ( _ solid black line _ ) and number of redshifts measured ( _ grey histogram _ ) , as a function of the @xmath59 color . also shown is the distribution of failed redshift measurements ( _ dashed line _ ) . panels ( a ) through ( c ) show the color distributions for different magnitude ranges , and panel ( d ) shows the distribution for the entire sample . while the color distributions of galaxies placed on masks and of those successfully observed do not change significantly with magnitude , the color distribution of redshift failures shows a clear trend of becoming bluer at fainter apparent magnitude.,scaledwidth=47.5% ] the apparent color - magnitude distribution of the sample in common with the goods - n v1.0 catalogue is shown in fig . [ fig - apparent - cm ] , where panel ( a ) shows the entire sample ( stars and galaxies ) . panel ( b ) shows the subsample of objects that were placed on slits , while panel ( c ) discriminates targets in various redshift quality classes . failed reshifts ( @xmath49 ) are predominantly found at the faintest and bluest limits of the sample . a similar analysis made for the deep2 survey @xcite , shows that most of these failed redshifts are likely to be of galaxies at high redshift , for which the [ ] lines fall outside the wavelength range of the deimos setup used in the tkrs . the sampling rate as a function of apparent @xmath56 color and magnitude is shown in fig . [ fig - sampling ] . the sampling rate was measured in bins of 0.25 mag both in magnitude and color , and the shading in the figure represents different sampling rates as described in the key . panel ( a ) shows the ratio between number of objects placed on masks , relative to the number of objects within the same color - magnitude limits in the sample . the sampling rate of the tkrs is higher than 65% for the majority of the color - magnitude bins , though this rate falls for fainter and bluer galaxies . panel ( b ) shows the success rate ; _ i.e. _ , the number of spectra that produced redshifts compared to the total placed on masks . the completeness level is higher than 80% to about @xmath60 , beyond which it falls due to the lower @xmath20 of absorption line spectra , and the loss of key spectral features outside the wavelength range of the spectra . [ [ surface - brightness - effects . ] ] surface brightness effects . + + + + + + + + + + + + + + + + + + + + + + + + + + + figure [ fig - zmag - area ] shows isophotal area plotted against @xmath61 magnitude for all of the spectroscopic targets in our survey . both photometric parameters are from the _ hst _ acs imaging catalog of @xcite , with the magnitudes measured in the outermost aperture ( @xmath57 ) . in this plot we again distinguish objects in the various redshift quality classes of our survey . the stellar and galaxy sequences are well differentiated to the limiting @xmath61 magnitude . objects with uncertain redshifts include faint stars as well as galaxies of varying brightness . many of the brighter objects with uncertain redshifts ( @xmath41 ) failed because of poorly defined extraction windows when creating the 1-d spectra . this problem will be revisited in a later re - analysis of the 2-d spectra and a corrected version of the catalog will be issued . failed redshifts ( @xmath49 ) , as opposed to uncertain ones , start to occur at @xmath62 . however , the diagram provides no strong evidence that the @xmath49 galaxies have lower surface brightnesses than the subsample with secure redshifts . apparent object size ( acs isophotal image area ) vs. apparent @xmath61 magnitude measured in the outermost aperture ( @xmath57 ) on acs images @xcite for the 1981 sources observed in our redshift survey . different symbols distinguish objects with secure redshifts ( @xmath63 ) , uncertain redshifts ( @xmath41 ) , no measured redshifts ( @xmath49 ) , and stars ( @xmath64).,scaledwidth=47.5% ] areas within the goods - n field have already been the subject of intensive study by deep surveys of the hdf - n region that lies within it . as mentioned in [ sxn - intro ] , several research teams have obtained redshifts for galaxies in this part of the sky , among them @xcite . recent work by other observing teams using keck ii with deimos to survey the goods - n field include the companion survey of @xcite and the higher - resolution survey of red galaxies by @xcite . we can compare our measurements with the findings of these independent surveys to gauge the accuracy of our redshifts and our quality assessments . the results of the redshift comparisons are depicted graphically in fig . [ fig - zhist ] and key results are summarized in table [ tab - zcompare ] . for each survey , the table provides the number of objects in common with the tkrs sample in addition to robust measures of both the difference in measured redshift and the dispersion in that difference . for purposes of this discussion we defined two redshift measurements to `` agree '' if they differ by less than @xmath65 ( @xmath66 km s@xmath4 ) ; this excludes galaxies with severely discrepant redshifts and does not significantly bias the resulting measures of redshift offset and dispersion . to quantify the redshift differences with our work we computed the offsets via the biweight , while the median absolute deviation from the median ( mad , with multiplicative scaling factor 9 ) served as a robust estimator of the dispersion in the offsets @xcite . we now discuss the results of the comparison for individual surveys . histogram showing the redshift differences between galaxies with secure measurements in the tkrs vs. four other extensive samples in the goods - n region . each panel shows the bi - weight location ( _ dotted line _ ) and its value @xcite . ( a ) redshift differences for galaxies with secure measurements from lris in @xcite and @xcite . ( b ) redshift differences between tkrs and secure measurements of @xcite and @xcite , also using lris . ( c ) redshift differences with the @xcite sample of galaxies , which used similar instrumental setup and integration times as the tkrs . ( d ) redshift differences with the @xcite sample of galaxies observed with deimos but using higher resolution and longer integration times than this work.,scaledwidth=47.5% ] lcrrr[btp ] cohen et al . hdf - n & 1,2 & 254 & 52.9 & 174.3 + deep1 hdf - n & 3,4 & 60 & 27.2 & 99.8 + cowie et al . goods - n & 5 & 378 & @xmath67 & 140.5 + treu et al . goods - n & 6 & 131 & @xmath68 & 42.2 + in an lris survey covering a region centered on the hdf and flanking fields , cohen et al . obtained spectra for most galaxies with @xmath22 in the flanking fields and @xmath69 in the hdf - n @xcite . the redshift catalogue of these works also contains measurements obtained from the literature ; however , in the present comparison we concentrate on the uniform sample of 413 galaxies measured by j. cohen using keck i with lris . for 254 galaxies with redshift in rough agreement with ours ( fig . [ fig - zhist](a ) ) the biweight difference is @xmath70 km s@xmath4 with a dispersion of 174 km s@xmath4 . @xcite obtained secure redshifts for 52 targets measured poorly in the tkrs sample ( @xmath49 or 2 ) . another lris sample in the hdf - n and flanking fields is that observed by the deep1 survey team , who targeted primarily @xmath71- and @xmath72-band dropouts @xcite and compact galaxies @xcite . of 114 galaxies in common , 60 have redshifts within @xmath73 of the tkrs measurement , and the distribution of redshift differences is shown in fig . [ fig - zhist](b ) . the offset in redshift is @xmath74 km s@xmath4 with a dispersion of 100 km s@xmath4 . the deep1 sample provides 7 reliable measurements of galaxy redshifts for which the tkrs redshift has a quality of 1 or 2 . the @xcite sample , which used the same instrument , configuration , integration time , and survey region as the tkrs sample , features 830 galaxies in common with tkrs , of which 452 have secure redshifts in both works ; see fig . [ fig - zhist](c ) . among the 378 redshifts in agreement , the offset is @xmath75 km s@xmath4 with a dispersion of 141 km s@xmath4 . a total of 56 galaxies for which tkrs was unable to obtain a redshift ( _ i.e. _ , @xmath76 ) were successfully measured by cowie et al . conversely , tkrs was able to measure redshifts for 50 galaxies that were observed by cowie et al . but for which they were unable to get a reliable redshift . finally , in fig . [ fig - zhist](d ) we compare our redshifts with measurements obtained by @xcite , who targeted preferentially red galaxies in the goods - n field using keck ii with deimos . their higher - resolution survey used the 1200 l mm@xmath4 grating to acquire exposures ranging from 510 h , as compared to 1 h for tkrs . of 199 galaxies in common with our catalog , 178 have secure redshifts in the treu et al . sample and 131 have secure redshifts in both samples ; only one redshift measurement is significantly discrepant ( @xmath77 km s@xmath4 with a dispersion of 42 km s@xmath4 , so that this sample represents the lowest dispersion with respect to our redshifts . these measurements from treu et al . also provide an estimate on how the increase of exposure time affects the redshift success rate . of the 18 objects in common whose redshifts in the tkrs were uncertain ( @xmath41 ) or could not be measured , 7 ( 39% ) agree with the measurements with longer exposures . this number corresponds to @xmath78 ( 7/179 ) of galaxies within the sample in common . an additional 6% ( 11/179 ) of objects yielded successful redshift measurements with the longer integration times . the independent redshift measurements available for a significant number of galaxies in our catalog allow us to estimate the uncertainty in our redshifts . of the four comparable catalogs described in [ sxn - other - surveys ] , the most useful ones for comparison are the deimos surveys of treu et al . and cowie et al . , since they use the same instrument as our survey . of these two , the treu et al . sample is expected to be more accurate by virtue of its higher dispersion ( twice that of tkrs ) and longer exposure time ( typically 5 times longer ) . the dispersion in redshift measurements between these two surveys should thus be dominated by errors in the tkrs sample . as shown in the last column of table [ tab - zcompare ] , the robust estimate of dispersion in the redshift differences of these surveys is 42 km s@xmath4 . assuming that the tkrs velocity uncertainties are at least a factor of two larger than those of treu et al . due to the difference in dispersion , we estimate an average uncertainty of about 38 km s@xmath4 for tkrs , with the caveat that the sample of galaxies used in the comparison consists almost exclusively of early - type galaxies and is thus not representative of the goods - n region as a whole . the most comparable of the existing surveys to tkrs is the cowie et al . sample made at the same dispersion and wavelength interval with deimos . the dispersion in redshift measurements for the sample of 363 common targets between these two surveys is 141 km s@xmath4 . assuming the two surveys have comparable uncertainties , this implies an error of 99 km s@xmath4 for each . this value is over a factor of two higher than that measured compared to the treu et al . sample . this may be due to cowie et al . surveying fainter galaxies over a wider range of spectral types than treu et al . , as well as different slit orientations used by cowie et al . vs. tkrs . being conservative , we conclude that the characteristic uncertainty in our redshift measurements is @xmath79 km s@xmath4 or better . the existence of independent redshift measurements for many objects in our catalog allows us to assess the accuracy of our redshift quality categories . in other words , are the quality code 3 redshifts ( which by definition should be correct at the 95% confidence level ) truly correct 95% of the time ? in table [ tab - zqual2 ] we match the tkrs catalogue with a list of redshifts measured by other authors . this list collated measurements from @xcite and @xcite . the collated catalog was then correlated against the deimos catalog , providing a list of matching objects . in assessing the number of correct redshift identifications , we assume that all discrepant redshifts result from errors in the present sample , thus obtaining a firm _ lower limit _ on the quality of our measurements . the final column in this table shows that the @xmath80 redshifts agree with previous measurements in 94% of cases , establishing this as the lower limit to the accuracy of our most secure redshifts . for @xmath81 , the corresponding lower limit on the success probability is 91.6% . if we make the more reasonable assumption that the mis - identifications are equally split between tkrs and other samples , the success rate increases to 96% ( @xmath81 ) and 97% ( @xmath80 ) . we conclude that the respective intended confidence levels of 95% and 99% for the category 3 and 4 redshift measurements are reasonably accurate . we have compared the spectroscopic redshifts with the photometric redshift catalog of @xcite . these redshift estimates are based on the ultradeep imaging in eight colors from the u to the near ir which is described in @xcite and are computed using the bayesian code of @xcite . this assigns odds that the redshift lies close to the estimated value and we have only used those cases where these odds lie above 90% . the imaging data also saturate at bright magnitudes so we have restricted the comparison to sources with @xmath82 . this gives a sample of 1078 sources with both spectroscopic and photometric redshifts which we compare in fig . [ fig - photo - z ] . of the sources , 8 are broad line agn where the photometric estimates may be poorer since the templates do not properly model such objects . we have distinguished these objects with larger symbols . nearly all ( 97% ) of the photometric redshifts agree to within 30% with the spectroscopic redshift , consistent with the expectation from the probabilities assigned to the photometric redshifts . this agreement is independent of magnitude over the @xmath83 range . photometric redshift versus spectroscopic redshift for the 1078 sources with @xmath84 where both have been measured ( _ solid squares _ ) . objects with broad - line agn spectra are distinguished with a larger symbol . only 38 of the photometric redshifts differ by more than 30% from the spectroscopic redshift . five of these cases are broad - line agn.,scaledwidth=47.5% ] the distribution of galaxies in our survey as a function of @xmath54 and redshift appears in fig . [ fig - zmedian ] . also shown in this plot and in table [ tab - zmedian ] is the median redshift as a function of magnitude . the plot compares these values to the median redshifts in the @xcite survey ; both samples are uncorrected for incompleteness . for bins brighter than @xmath85 , our median redshifts are significantly lower than those of cohen et al .. we note that the present survey has a cumulative completeness of just over 60% at @xmath85 , compared to @xmath51 for cohen et al .. in the faintest magnitude bin of the cohen et al . hdf sample ( 23.524.0 ) the differential completeness drops to the same 50% value as our survey , and as a result the median redshifts in this magnitude range ( @xmath86 for cohen et al . vs. 0.80 in our sample ) are consistent . redshift vs. @xmath1 magnitude for targets with secure redshifts . small pluses indicate individual objects in the tkrs survey . boxes indicate the interquartile range of redshifts within respective 0.5 mag bins ; each has a horizontal line and @xmath87 which indicate the median redshift within that bin . the median redshifts of @xcite are shown for comparison . the vertical dotted line at right represents the apparent magnitude limit of the present survey at @xmath0.,scaledwidth=47.5% ] crccc[tbp ] 18.519.0 & 2 & & 0.09 & + 19.019.5 & 11 & 0.11 & 0.14 & 0.28 + 19.520.0 & 13 & 0.20 & 0.27 & 0.32 + 20.020.5 & 15 & 0.20 & 0.28 & 0.36 + 20.521.0 & 33 & 0.30 & 0.44 & 0.51 + 21.021.5 & 57 & 0.32 & 0.46 & 0.51 + 21.522.0 & 93 & 0.43 & 0.51 & 0.60 + 22.022.5 & 133 & 0.47 & 0.52 & 0.68 + 22.523.0 & 193 & 0.48 & 0.63 & 0.84 + 23.023.5 & 254 & 0.51 & 0.77 & 0.95 + 23.524.0 & 335 & 0.53 & 0.80 & 0.96 + 24.024.4 & 248 & 0.60 & 0.85 & 1.02 + figure [ fig - gdwzhist ] shows the redshift distribution of galaxies with secure measurements . this plot shows the number of galaxies contained within a velocity window of width @xmath88 km s@xmath4 at any given redshift and thus isolates concentrations of galaxies in redshift space . the window width of 1500 km s@xmath4 is well - matched to the numerous structures with velocity dispersions of @xmath89 500700 km s@xmath4 reported by @xcite in the hdf . numerous concentrations of galaxies are seen in the sample , most notably those at @xmath90 , 0.56 , 0.64 , 0.85 , and 1.02 . these structures are equally prominent in the pie diagrams shown in fig . [ fig - pie](a ) and ( b ) which plot the projected linear offset from the field center ( along the @xmath91 and @xmath92 axes , respectively ) vs. redshift . these figures illustrate that some of the structures span the goods - n field , while others ( particularly those at @xmath93 and 1.02 ) are spatially clumped . + + in fig . [ fig - absolute - cmd ] we show the rest - frame color - magnitude diagram based on the _ hst _ acs photometry for all galaxies with secure tkrs redshifts appearing in the @xcite catalog . the rest - frame parameters were calculated by convolving the ccd and filter throughputs of acs with spectral energy distributions of galaxies in the @xcite atlas , a more detailed description of which will be presented elsewhere @xcite . as seen in fig . [ fig - absolute - cmd](b ) , the rest - frame galaxy colors are bi - modal about a minimum near @xmath94 ( corresponding to about @xmath95 in the vega system ) . this effect has also been noted in the deep1 survey @xcite in the groth survey strip ( which also used _ hst _ photometry ) , as well as in the sloan digital sky survey @xcite and combo-17 survey @xcite . figure [ fig - absolute - cmd](a ) illustrates that the the color bi - modality is clearly present to @xmath96 , which is beyond the limits both of @xcite and @xcite . the lack of low - luminosity red galaxies , also noted by @xcite , is most likely to be a real effect . rest - frame color - magnitude diagram and color distribution . ( a ) rest - frame brightness vs. color for galaxies with tkrs redshifts in the @xcite catalog . the three dashed lines show the limiting absolute magnitude as a function of restframe color at three different redshifts : @xmath97 ( _ lower curve _ ) , @xmath98 ( _ middle curve _ ) and @xmath99 ( _ top curve _ ) . the symbols are also coded by redshift interval as shown in the figure . ( b ) distribution of @xmath100 colors in 0.05 mag bins . the trough at @xmath94 is consistent with that seen in other samples of galaxies . , scaledwidth=47.5% ] we have described our deep redshift survey of the goods - n region using deimos , the result of which is a catalog of photometry , astrometry , redshifts and redshift quality assessments for 1440 confirmed galaxies / agn and stars in the @xmath7 goods - n region . this sample is 53% complete to a limiting magnitude of @xmath0 and has a median redshift of @xmath101 . in conjunction with an accompanying deimos survey by @xcite , it provides a substantial and homogeneous sample of new redshifts and spectra in this important area of the sky . the tkrs catalog will continue to evolve as deimos pipeline processing algorithms are refined , and we expect that a significant number of galaxies may be promoted from redshift quality class 2 ( uncertain ) to 3 ( @xmath51 secure ) . to ensure that researchers obtain access to the revised data , we have established a website to disseminate all data products resulting from this survey . in addition to our most recent catalog of deimos photometric and spectroscopic measurements of all targets in the goods - n field , astronomers can currently obtain the wavelength - calibrated 1-d spectra and will soon be able to download the sky - subtracted 2-d images from which the 1-d spectra were extracted . these resources will allow others to measure velocity widths and spectral indices for the sample , promoting the study of fundamental galaxy properties at moderate redshift . this project was the brainchild of david koo , who conceived of the survey , formulated a detailed plan for executing it , and ( with support from sandra faber ) served as its principal advocate . the survey would not have been completed without the generosity of the deep2 redshift survey team , which not only developed the sophisticated data reduction pipeline for deimos , but also donated a significant amount of their deimos observing time to complete the survey . we are indebted to garth illingworth and pieter van dokkum for acquiring several of the deimos images required for the photometric survey . additionally , we thank drew phillips for helping process the slitmask designs and for use of his cosmic ray rejection algorithm , ben weiner for useful discussions and for running his redshift measuring code independently on this sample , plus tommaso treu and richard ellis for sharing redshifts from their goods - n survey prior to publication . the national science foundation supported our survey through grant as-0331730 . 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 also relied upon data from the sloan digital sky survey ( sdss ) . funding for the creation and distribution of the sdss archive has been provided by the alfred p. sloan foundation , the participating institutions , the national aeronautics and space administration , the national science foundation , the u.s . department of energy , the japanese monbukagakusho , and the max planck society . the sdss is managed by the astrophysical research consortium ( arc ) for the participating institutions . the participating institutions are the university of chicago , fermilab , the institute for advanced study , the japan participation group , the johns hopkins university , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , university of pittsburgh , princeton university , the united states naval observatory , and the university of washington . our special thanks to andy connolly , who helped arrange our access to the sdss data prior to the dr1 release . we thank the anonymous reviewer for helpful suggestions which improved the quality of the manuscript . c. n. a. w. thanks the w. m. keck observatory for its hospitality throughout the visit during which much of this work was completed . 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 . bell , e. f. , wolf , c. , meisenheimer , k. , rix , h .- w . , borch , a. , dye , s. , kleinheinrich , m. , & mcintosh , d. h. 2004 , , submitted , astro - ph/0303394 beers , t. c. , flynn , k. , & gebhardt , k. 1990 , , 100 , 32 benitez , n. 2000 , , 536 , 571 bertin , e. , & arnouts , s. 1996 , , 117 , 393 bunker , a. j. , stern , d. , spinrad , h. , dey , a. , steidel , c. c. 1998 , , 30 , 927 capak , p. , et al . 2003 , , in press , astro - ph/0312635 capak , p. , et al . 2004 , in preparation cohen , j.g . , cowie , l. l. , hogg , d. w. , songaila , a. , blandford , r. , hu , e. m. , & shopbell , p. 1996 , , 471 , l5 cohen , j.g . , hogg , d. w. , blandford , r. , cowie , l. l. , hu , e. m. , songaila , a. , shopbell , p. & richberg , k. 2000 , , 538 , 29 cohen , j. g. 2001 , , 121 , 2895 cohen , j. g. 2002 , , 567 , 672 cowie , l. l. , songaila , a. , hu , e. m. & cohen , j. g. 1996 , , 112 , 839 cowie , l. l. , barger , a. j. , hu , e. m. & capak , p. 2004 , , submitted coleman , g. d. , wu , c .- c . , weedman , d. w. 1980 , , 43 , 393 davis , m. , et al . ( the deep2 team ) 2003 , spie , 4834 , 161 dawson , s. , stern , d. , bunker , a. j. , spinrad , h. , & dey , a. 2001 , , 122 , 598 dickinson , m. 1998 in the hubble deep field , ed . m. livio , s. m. fall , & p. madau ( new york : cambridge univ . press ) , 219 dickinson , m. , et al . 2001 , , 33 , 820 dickinson , m. , et al . 2004 , in preparation faber , s. m. , et al . 2003 , spie , 4841 , 1657 ferguson , h. c .. dickinson , m. , & williams , r. 2000 , , 38 , 667 giavalisco , m. , et al . 2004 , in preparation glazebrook , k. , offer , a. r. , deeley , k. 1998 , , 492 , 98 hogg , d. w. , et al . ( the sdss collaboration ) 2002 , , 124 , 646 i m , m. , simard , l. , faber , s. m. , koo , d. c. , gebhardt , k. , willmer , c. n. a. , phillips , a. c. , illingworth , g. d. , vogt , n. p. , & sarajedini , v. l. 2002 , , 571 , 136 kinney , a. l. , calzetti , d. , bohlin , r. c. , mcquade , k. , storchi - bergmann , t. , & schmitt , h. r. 1996 , , 467 , 38 le fvre , o. , crampton , d. , lilly , s. j. , hammer , f. , & tresse , l. 1995 , , 455 , 60 lowenthal , j. d. , koo , d. c. , guzmn , r. , gallego , j. , phillips , a. c. , vogt , n. p. , faber , s. m. , illingworth , g. d. , & gronwall , c. 1997 , , 481 , 673 marleau , f. r. , & simard , l. 1998 , , 507 , 585 monet , d. g. , et al . 2003 , , 125 , 984 moustakas , l. a. , zepf , s. e. , & davis , m. 1997 , in the hubble space telescope and the high redshift universe , ed . n. tanvir , a. aragon - salamanca , & j. wall ( singapore : world scientific ) , 273 newman , j. a. , et al . ( the deep2 team ) 2004 , in preparation oke , j. b. 1974 , , 27 , 21 oke , j. b. , cohen , j. g. , carr , m. , cromer , j. , dingizian , a. , harris , f. h. , labrecque , s. , lucinio , r. , schaal , w. , epps , h. , & miller , j. 1995 , , 107 , 135 phillips , a. c. , guzmn , r. , koo , d. c. , lowenthal , j. d. , vogt , n. p. , faber , s. m. , & illingworth , g. d. 1997 , , 489 , 543 simard , l. , et al . 2002 , , 142 , 1 smith , j. a. , et al . ( the sdss collaboration ) 2002 , , 123 , 2121 steidel , c. c. , giavalisco , m. , dickinson , m. , & adelberger , k. l. 1996 , , 112 , 352 steidel , c. c. , adelberger , k. l. , shapley , a. e. , pettini , m. , dickinson , m. , & giavalisco , m. 2003 , , 592 , 728 springel , v. , & white , s. d. m. 1988 , , 298 , 143 strateva , i. , et al . ( the sdss collaboration ) 2001 , aj , 122 , 186 strauss , m. a. , et al . ( the sdss collaboration ) 2002 , , 124 , 1810 treu , t. et al . 2004 , in preparation weiner , b. , et al . ( the deep2 team ) 2004 , in preparation willmer , c. n. a. , et al . ( the deep2 team ) 2004 , in preparation york , d. g. , et al . ( the sdss collaboration ) 2000 , , 120 , 1579 zepf , s. e. , moustakas , l. a. , & davis , m. 1997 , apj , 474 , l1 rrrrrrrrrrrrrrrrrrr 1468 & 12 36 18.450 & 62 16 01.53 & 23.18 & 01 & 046 & 0.79803 & 4 & 0.7970 & 20 & 3291.7 & 466.3 & 43 & 4574.6 & 7529.9 & 0.521 & 0.008 & 63.9 & j123618.46 + 621601.9 + 1473 & 12 36 01.805 & 62 14 05.19 & 24.20 & 01 & 025 & 0.43592 & 4 & & & 1897.3 & 462.3 & 53 & 272.6 & 3663.6 & 0.674 & 0.225 & 133.4 & j123601.80 + 621405.6 + 1475 & 12 36 05.932 & 62 14 35.81 & 22.16 & 01 & 029 & 0.40836 & 4 & & & 2253.8 & 452.9 & 43 & 7490.8 & 4678.5 & 0.636 & 0.028 & 19.2 & j123605.97 + 621436.1 + 1479 & 12 36 25.632 & 62 16 47.94 & 24.61 & & & & & & & 3870.3 & 489.7 & 44 & 2897.1 & 886.6 & 0.372 & 0.003 & -34.8 & j123625.65 + 621648.3 + 1482 & 12 37 00.626 & 62 20 54.44 & 23.79 & 02 & 096 & 0.79390 & 4 & & & 6809.1 & 473.9 & 35 & 2969.6 & 904.3 & 0.572 & 0.009 & 117.4 & j123700.59 + 622054.9 +

we study the concept of the fourth skein module of-manifolds , that is a skein module based on the skein relation and a framing relation ( , , invertible ) . we give necessary conditions for trivial links to be linearly independent in the module . we investigate the behavior of elements of the skein module under the-move and compute the values for-torus links and twist knots as elements of the skein module . using the idea of mutants and rotors , we show that there are different links representing the same element in the skein module . we also show that algebraic links ( in the sense of conway ) and closed-braids are linear combinations of trivial links . we introduce the concept of-algebraic tangles ( and links ) and analyze the skein module for-algebraic links . as a byproduct we prove the montesinos - nakanishi-moves conjecture for-algebraic links ( including-bridge links ) . in the case of classical links ( i.e. links in ) our skein module suggests three polynomial invariants of unoriented framed ( or unframed ) links . one of them generalizes the kauffman polynomial of links and another one can be used to analyze amphicheirality of links ( and may work better than the kauffman polynomial ) . in the end , we speculate about the meaning and importance of our new knot invariants . jzef h. przytycki and tatsuya tsukamoto in memory of rodica simion ( january 18 , 1955 - january 7 , 2000 )

math0010282

to understand the structure of @xmath0-manifolds and links inside them , the first author introduced the concept of the skein module @xcite motivated by the conway idea of linear skein " . skein modules are quotients of a free module of formal linear combinations of links in a @xmath0-manifold by properly chosen ( local ) skein relations . in search of appropriate skein relations , we analyze deformations of moves on links previously studied in knot theory . the simplest of such moves are @xmath6 ( twist ) moves ( figure @xmath9 ) . the simplest skein modules are deformations of @xmath9-moves , @xcite ( a linear skein relation ) . the conway type skein modules are deformations of @xmath10-moves ( a quadratic skein relation ) and they are studied extensively ( e.g. @xcite ) . in this paper we start a systematic study of skein modules based on a deformation of @xmath0-moves ( a cubic skein relation ) . since there are four terms involved in the relation , we call such a skein module the fourth skein module and denote it by @xmath11 , following the notation of @xcite . in section [ secmnc ] , we discuss the montesinos - nakanishi conjecture on @xmath0-moves . this conjecture , and its partial solutions , make the study of fourth skein modules feasible . in section [ secfsm ] , we define the fourth skein module and we make the general conjecture about generators of the fourth skein module in @xmath8 and @xmath12 with @xmath6 boundary points ( @xmath6-tangles ) . in section [ seclitl ] , we discuss the necessary conditions for coefficient ring so that trivial links are linearly independent in the fourth skein module . we prove the independence in two , rather degenerated cases . in section [ secgfsm ] , we consider the case in which trivial links differ only by a multiplicative constant . we show that after proper substitution the skein module reduces to the kauffman skein module , at least for links generated by trivial links . generally we have always epimorphism into the kauffman skein module . in section [ sectorusandtwists ] , we give an example of computation , finding eigenvectors of dehn twists on @xmath10-tangles and applying this to describe the effect of @xmath6-moves on elements of the skein module . as a corollary we compute the values for @xmath7-torus links and twist knots . in section [ secmrfsm ] , we use the idea of mutants and rotors to construct different links representing the same element in the fourth skein module . in section [ secnalgtl ] , we introduce a notion of an @xmath6-algebraic tangle and link , generalizing the notion of an algebraic link in the sense of conway . we show that @xmath6-bridge links and closed @xmath6-braids are @xmath6-algebraic . we prove conjectures [ generalconjmn ] and [ conjgenerator ] for @xmath10- and @xmath0-algebraic tangles . in the last section , we speculate on existence of fourth " polynomial link invariants of classical links ( in @xmath8 ) and show connection of our work with investigation of coxeter , assion and wajnryb on finite quotients of braid groups . let @xmath13 be a link ( possibly a relative link ) in a @xmath0-manifold @xmath14 ( considered up to ambient isotopies ) . an @xmath6-move is a local change of a link which adds @xmath6 positive or negative half - twists to @xmath13 ( @xmath15 ) . in an oriented @xmath0-manifold we may distinguish a move @xmath16 from a move @xmath17 . we occasionally use the notation @xmath18-move and @xmath19-move in these cases ( see figure @xmath9 ) . the first part of conjecture [ generalconjmn ] ( montesinos - nakanishi conjecture ) motivated , in part , this work . 1 . @xmath20 @xmath21 @xcite@xmath22 $ ] every link is @xmath0-equivalent to a trivial link @xmath23i.e . it can be converted to a trivial link by @xmath0-moves@xmath24 . 2 . @xmath23@xcite@xmath24 every @xmath10-tangle is @xmath0-equivalent to a tangle with no more than one crossing ( equivalently to one of the four basic " @xmath10-tangles , with possible trivial components ) ; see figure @xmath10 . 3 . every @xmath0-tangle is @xmath0-equivalent to one of the @xmath25 basic " @xmath0-tangles , with possible trivial components ; see figure @xmath0 . note that every basic @xmath0-tangle has no more than four crossings . 4 . there is a finite number , @xmath26 , of basic " @xmath6-tangles , such that every @xmath6-tangle is @xmath0-equivalent to one of the @xmath26 basic " @xmath6-tangles , with possible trivial components . @xmath26 @xmath27 @xmath28 @xmath23e.g . in this section we start a systematic study of the fourth skein module based on a linear relation among links @xmath30 and @xmath31 of fig.4 ( the module was only shortly mentioned in @xcite ) . [ deffsm ] let @xmath14 be an oriented @xmath0-manifold , @xmath32 a commutative ring with identity , and @xmath33 the set of ambient isotopy classes of unoriented framed links in @xmath14 . let @xmath34 be the free @xmath32 module generated by @xmath33 and @xmath35 elements in @xmath32 such that @xmath36 and @xmath5 are invertible . we define the fourth skein module , @xmath37 , or shortly @xmath11 as the quotient of @xmath33 by a submodule generated by the framing relations @xmath38 and the fourth skein expression : @xmath39 . similarly , we can define a relative fourth skein module starting from an oriented manifold @xmath14 with @xmath40 points @xmath41 , @xmath42 , @xmath43 , @xmath44 chosen on @xmath45 and the set of ambient isotopy classes of unoriented relative framed links in @xmath46 such that @xmath47 @xmath27 @xmath48 @xmath27 @xmath49 for each relative link @xmath13 . note that the following conjecture is a generalization of conjecture [ generalconjmn ] . in fact , conjecture [ conjgenerator ] reduces to conjecture [ generalconjmn ] for @xmath50 and @xmath51 . 1 . @xmath52 is generated by trivial links . the fourth skein module of @xmath10-tangles in a disk is generated by tangles with no more than one crossing . the fourth skein module of @xmath0-tangles in a disk is generated by the @xmath25 basic @xmath0-tangles described in figure @xmath0 with possibly trivial components . there is a function @xmath53 such that the fourth skein module of @xmath6-tangles in a disk is generated by tangles with no more than @xmath53 crossings . in this paper , we prove the conjecture for @xmath0-algebraic tangles and @xmath0-algebraic links ( including @xmath0-bridge links and closures of balanced @xmath0-tangles ) . the notion of an @xmath6-algebraic tangle ( and link ) is a new concept defined for the first time in this paper , section 8 . we end this section by listing several useful properties of the fourth skein module , including the _ universal coefficient property _ , which our skein module shares with other skein modules @xcite . [ thmucp ] 1 . an orientation preserving embedding of @xmath0-manifolds @xmath54 yields a homomorphism of skein modules @xmath55 . the above correspondence leads to a functor from the category of @xmath0-manifolds and orientation preserving embeddings @xmath23up to ambient isotopy@xmath24 to the category of @xmath32-modules @xmath23with specified elements @xmath56 , @xmath57 , @xmath58 invertible@xmath24 . 2 . @xmath23universal coefficient property@xmath24 + let @xmath59 be a homomorphism of rings @xmath23commutative with @xmath60 . we can think of @xmath61 as an @xmath32 module . then the identity map on @xmath33 induces the isomorphism of @xmath61 @xmath23and @xmath62 modules : @xmath63 in particular @xmath64,1,x ,- x,-1 ) \otimes_{z[x]}z$ ] , where @xmath65 . 3 . let @xmath66 where @xmath67 is an oriented surface . then @xmath11 is an algebra , where @xmath68 is obtained by placing @xmath69 above @xmath70 with respect to the product structure . the empty link @xmath71 is the neutral element of the multiplication . every embedding @xmath72 yields an algebra homomorphism @xmath73 . @xmath74 we analyze the general question : for which @xmath3 and @xmath75 are trivial links in @xmath8 linearly independent in @xmath52 ? we give necessary conditions and conjecture that they are also sufficient . from the relation @xmath76 ( figure @xmath77 ) , we get the relation @xmath78 for any framed link @xmath79 ( including the trivial knot ) . thus to have the linear independence of trivial links we should assume : [ condtntwistrelation ] @xmath80 now consider two ambient isotopic diagrams @xmath81 and @xmath82 ( figure @xmath83 ) . let @xmath84 be the trivial @xmath6 component link . for @xmath81 we compute : @xmath85 and for @xmath82 we have : @xmath86 . since @xmath87 and @xmath88 can be any link , therefore in order to have linear independence we require : [ condtnchainrelation ] 1 . @xmath89 , and 2 . @xmath90 condition [ condtntwistrelation ] can be rewritten as ( using condition [ condtnchainrelation ] @xmath91 to eliminate @xmath92 ) : @xmath93 and condition [ condtnchainrelation ] @xmath94 can be written as ( again using condition [ condtnchainrelation ] @xmath91 to eliminate @xmath92 ) : @xmath95 if we eliminate @xmath96 , then we get : [ condtnzerodivisor ] @xmath97 it is convenient to work with a ring without complicated zero divisors so we consider two cases of condition [ condtnzerodivisor ] separately . 1 . assume @xmath98 . our conditions reduce now to @xmath99 which again leads to two cases : 1 . let @xmath100 . then @xmath96 is a free variable , and @xmath101 . as @xmath4 is invertible and relations are homogeneous , we can put @xmath102 . let also write @xmath103 ; then @xmath104 , @xmath105 . we work in this case with a ring @xmath106[a]/(a^4 - 1)$ ] . in particular we measure framing only mod 4 . our skein relation has a form : @xmath107 . 2 . let @xmath108 . then @xmath109 and after substituting @xmath110 one obtains the skein relation : @xmath111 . 2 . assume @xmath112 . this leads to the following solution @xmath23for simplicity we put @xmath102 and @xmath113@xmath24 : @xmath114 , @xmath115 , and @xmath116 . the skein relation has now the form @xmath117 , and the framing relation @xmath118 . @xmath119 $ ] . we have considered conditions which are necessary for trivial links to be linearly independent in the fourth skein module , @xmath11 . we conjecture that these conditions are also sufficient . in the case of condition [ condtnzerodivisor ] @xmath231@xmath24@xmath23ii@xmath24 we are able to prove it , and the skein module of @xmath8 seems to be rather trivial . [ thmlinindep ] 1 . if @xmath98 , @xmath120 , @xmath121 then trivial links , @xmath122 , are linearly independent in @xmath52 . 2 . consider a homomorphism @xmath123 @xmath124/(a^4 - 1)$ ] given by @xmath125 @xmath126 where @xmath127 is the number of components of a link @xmath13 and @xmath128 is for oriented framed link a difference between framing of @xmath13 and @xmath129-framing in @xmath8 . for unoriented links @xmath128 is well defined @xmath130 . then @xmath131 yields a homomorphism @xmath132 from the fourth skein module described by condition @xmath133 @xmath134@xmath135 @xmath23i.e . @xmath136 to @xmath124/(a^4 - 1)$ ] . if we allow the empty link , @xmath137 , in @xmath33 and conjecture @xmath138 @xmath23i@xmath24 holds then @xmath132 is an algebra isomorphism , where product of two links is defined to be their disjoint sum . to show @xmath23ii@xmath24 it suffices to show that @xmath131 sends the skein relation @xmath139 to @xmath129 we have @xmath140 because @xmath141 and @xmath142 . the last statement follows from the fact that if @xmath13 is an oriented framed link and @xmath143 its diagram such that the framing of @xmath13 is the flat framing of @xmath143 then @xmath144 where the sum is taken over all crossings of @xmath143 . @xmath131 and @xmath132 are clearly algebra epimorphisms and @xmath145 and @xmath146 , where @xmath147 are left inverses of @xmath131 and @xmath132 respectively . thus @xmath132 is an isomorphism on the subspace of @xmath148 generated by trivial links . in fact our proof works for any rational homology sphere , and assuming @xmath149 , for any @xmath0-manifold . @xmath23i@xmath24 follows immediately from @xmath23ii@xmath24 , because @xmath150 are linearly independent in the ring of polynomials . [ conjlinindep ] the trivial links , @xmath122 , are linearly independent in @xmath52 in the following cases : 1 . @xmath98 , @xmath151 and @xmath152 2 . @xmath153 , @xmath154 , @xmath155 notice that in all cases of conjecture [ conjlinindep ] and in theorem [ thmlinindep ] , we assumed that @xmath89 ( compare section [ secgfsm ] ) . conjecture [ conjlinindep ] @xmath231@xmath24@xmath23i@xmath24 leads to the polynomial invariant of unframed links ( @xmath156 ) in @xmath8 ( for links generated by trivial links ) , @xmath157 . if @xmath158 in our skein module , then @xmath159 . conjecture [ conjlinindep ] @xmath23ii@xmath24 leads to the polynomial invariant of framed links in @xmath8 ( for links generated by trivial links ) , @xmath160 . if @xmath161 in our skein module , then @xmath162 . for @xmath6-tangles the analogue of conjecture [ conjlinindep ] is more involved but very interesting . we will consider below the case @xmath23i@xmath24 . [ conjrelsm ] 1 . consider the skein module @xmath163 with @xmath156 and the skein relation @xmath164 where @xmath165 is a disk with @xmath40 boundary points . let @xmath166 be a set of @xmath6-tangles , one from each @xmath0-move equivalence class . then @xmath166 is a base for the fourth skein module @xmath163 , for @xmath167 . 2 . consider the @xmath168-linear space of all @xmath0-colorings of a tangle , @xmath169 , denoted by @xmath170 and the homomorphism @xmath171 where @xmath172 is the space of all colorings of boundary points @xcite . then different elements of @xmath166 can be distinguished by their @xmath0-colorings . precisely , if @xmath173 then either @xmath174 or ranks are the same but @xmath175 . by definition , @xmath134 holds for @xmath176 , and so the theorem says here that the deformation ( @xmath177 ) does not change the module . we exclude the case @xmath178 in the conjecture because we know that our skein module of @xmath8 ( or of @xmath179 ) behave differently for @xmath176 and @xmath178 ; in the first case we identify @xmath0-move equivalence classes and in the second case ( as far as we know , compare theorem [ thmlinindep ] ) links with the same number of components . it is known that each @xmath0-move is preserving a @xmath0-coloring of a tangle and we conjecture the inverse holds , that is , @xmath6-tangles with the same @xmath0-coloring structure are related by @xmath0-moves . conjecture 4.6 suggests the importance of analyzing @xmath170 and @xmath180 . we have proved recently that @xmath180 are lagrangians in the space of colorings of the boundary of the tangle @xmath169 with respect to the properly chosen symplectic structure @xcite . in the previous section we showed that in all cases in which @xmath122 are ( conjecturally ) linearly independent , one has @xmath182 . we show that if @xmath183 is invertible then one has : [ formulalinkwithtrivial ] @xmath184 in particular : [ formulatriviallinks ] @xmath185 namely the relation given by reversing the hopf link summand gives : @xmath186 . thus @xmath187 @xmath188 . this gives the formulas [ formulalinkwithtrivial ] and [ formulatriviallinks ] . we should also remember that we have found the restriction ( condition [ condtntwistrelation ] ) : @xmath189 @xmath27 @xmath129 . for @xmath190 and the substitution @xmath191 our skein module ( polynomial ) reduces to the kauffman polynomial for links generated by trivial links ( we get in particular @xmath192 ) . more precisely , our substitutions give the skein relations : [ formula4srrelksm ] @xmath193 , @xmath194 . on the other hand the kauffman relations are : [ formulaksrelation ] @xmath195 , @xmath196 we work then with the coefficients in a commutative ring with identity , @xmath32 , where @xmath197 and @xmath3 are chosen invertible elements in @xmath32 . the kauffman skein module @xmath198 is defined to be @xmath199(_kauffman relations _ ) . as a consequence of kauffman relations we have @xmath200 in @xmath198 . by applying [ formulaksrelation ] twice we get : @xmath201 @xmath202 from this we get the relation : @xmath203 which is equivalent to [ formula4srrelksm ] . thus we have an epimorphism from the fourth skein module onto the kauffman skein module ( proposition [ propepitoksm ] ) . [ propepitoksm ] let @xmath14 be an oriented @xmath0-manifold , @xmath32 a commutative ring with identity , and @xmath204 invertible elements in @xmath32 . then 1 . we have an epimorphism @xmath205 2 . for @xmath206 invertible in @xmath32 and @xmath207 , the epimorphism @xmath208 restricted to the subspace generated by trivial links is a monomorphism . if conjecture @xmath138 @xmath23i@xmath24 holds then @xmath208 is an isomorphism . part @xmath91 follows from the fact that relations [ formula4srrelksm ] follow from the kauffman relations , [ formulaksrelation ] . if the trivial link @xmath209 is linearly independent in @xmath198 as is the case for @xmath207 or more generally for @xmath210 ( i.e. the product of a surface and the interval ) , and as is conjectured for any @xmath0-dimensional manifold , then we have a monomorphism @xmath211 from @xmath212 , the submodule generated by @xmath209 to @xmath213 [ formula4srrelksm]@xmath24 , which is left inverse to @xmath208 ( @xmath214 ) . as observed before , for @xmath206 invertible @xmath215 in the fourth skein module , thus @xmath94 of the proposition follows . we analyze in this section the behavior of the value of a link as an element of the fourth skein module @xmath52 under @xmath6-moves and as a corollary compute the values of @xmath7-torus links and of twist knots . we show an example in the case of the skein relation @xmath216 ( we discussed the importance of this relation , for @xmath98 , in section [ seclitl ] ) . let @xmath217 denote the positive half twist on a tangle @xmath79 , taking framing into account we can write @xmath218 , or generally @xmath219 . we will find the expression for @xmath220 in the basis @xmath221 . of course there is nothing sophisticated in our computation but it is useful to be able to evaluate our invariants for several classes of links . because @xmath222 , @xmath223 , and @xmath224 , therefore the matrix , @xmath225 of our linear map has the form : @xmath226 the characteristic polynomial of @xmath217 is therefore @xmath227 . as @xmath228 , it is convenient to put @xmath229 . thus 1 is an eigenvalue of @xmath217 and two other eigenvalues are @xmath230 , which gives @xmath231 and @xmath232 . from this we get the eigenvectors of @xmath217 to be : @xmath233 @xmath234 @xmath235 from these we obtain @xmath236 , and in general , we have the formula for @xmath220 . [ formulatwisting ] @xmath237 @xmath238 @xmath239 @xmath240 @xmath241 @xmath242 [ cordehntwist ] consider the dehn twist @xmath243 on the tangle . it has eigenvalues @xmath9 , @xmath244 , and @xmath245 and the same eigenvectors as @xmath217 . one can use this observation to construct a torsion in a manifold with an incompressible @xmath246 ( e.g. @xmath247 ) . in the spirit of @xcite we can check the condition at which an @xmath6-move changes our invariant only by a constant . [ cortwisting ] if @xmath248 and @xmath249 for @xmath6 even and @xmath250 for @xmath6 odd then @xmath251 . in particular , as we know , if @xmath252 , @xmath253 . to find the expression for torus links , @xmath254 ( figure @xmath255 ( @xmath3 ) ) , in the fourth skein module , we observe that @xmath256 and @xmath257 . [ cortorus ] the invariant of the torus link of type @xmath7 is given by : @xmath258 @xmath259 notice that for unoriented links @xmath260 which gives @xmath261 . we have already discussed the condition @xmath98 which is necessary if we want trivial links to be linearly independent . the condition @xmath262 leads to @xmath263 ( this agrees with [ formulatriviallinks ] ) . a similar calculation for twist knots , @xmath264 $ ] ( figure @xmath255 ( @xmath169 ) ) uses the formula [ formulatwisting ] and the initial conditions @xmath265=a^2t_1 $ ] , @xmath266 = a^{-1}t_1 $ ] , and @xmath267= t_{2,3}= a^{-3}\frac{(s^3-s^{-3})(a^{-2}(s^{2}-s^{-2 } ) + a^2(s^{4 } - s^{-4}))t_1 -(s+s^{-1})(s^{2}-s^{-2})(s^{4 } - s^{-4})t_2}{(s+s^{-1})(s - s^{-1})^2}.$ ] more generally we can write a straightforward formula for the pretzel link @xmath268 $ ] , figure @xmath269 by first using formula [ formulatwisting ] for each column to get a linear combination generated by torus links and then using the formula again for these torus links to get a linear combination generated by trivial links to write a formula for a @xmath10-bridge link ( and tangle ) or for an algebraic tangle , we should first decompose an algebraic tangle into @xmath6-twist tangles and then , from the table of multiplication of the four basic @xmath10-tangles , we can read off the result . using the table and this formula , we can also figure out the automorphism of the rotation " . a similar formula can be built for @xmath0-algebraic tangles ( with knowledge of the @xmath270 multiplication table , and the rotation " formula ; see section [ secnalgtl ] ) . using the idea of mutants and rotors , we show that there are different links representing the same element in the fourth skein module . if we have a @xmath10-tangle , @xmath88 , we consider three involutions @xmath271 , @xmath272 , and @xmath273 on it which denote the rotation of the tangle along the axis denoted in the subscript , figure @xmath274 . if we have a link diagram with a @xmath10-tangle in it , a mutation of the diagram along the tangle is obtained by performing one of the described involutions on the tangle ( see figure @xmath274 ) . [ propmutants ] consider a @xmath10-tangle part of a link diagram . if it is generated by tangles of figure @xmath10 @xmath23with possible trivial components@xmath24 , then any mutation of the link preserves the element of the fourth skein module . any tangle of figure @xmath10 is symmetric with respect to @xmath275 , @xmath276 , and @xmath197 axes . consider a tangle @xmath32 in a regular @xmath6-gon @xmath277 in @xmath246 ( @xmath278 ) . we call @xmath32 an ( @xmath6-)rotor if @xmath32 is invariant under the rigid rotation @xmath279 by angle @xmath280 around the center of @xmath277 . take a link @xmath13 and let @xmath13 have a tangle decomposition @xmath281 , i.e. , @xmath13 has a projection which intersects @xmath282 transversely in @xmath10 points on each edge of @xmath277 . if @xmath32 is an @xmath6-rotor , then we call @xmath283 a stator . consider a line which passes through the center of @xmath277 and either a corner or middle point of an edge . let @xmath284 be a @xmath285rotation through the third dimension with the line as an axis . then we call @xmath286 a rotant of @xmath13 . here note that we do not assume that @xmath32 is invariant under the operation @xmath287 . if we consider a similar rotation @xmath288 for a disk @xmath289 , then we have the same result @xmath290 . see @xcite for more details . regard each tangle of figure @xmath0 as a stator of a @xmath0-rotor . then @xmath291 of the @xmath25 basic @xmath0-tangles are invariant under an operation @xmath288 for an axis of a symmetry of the polygon . exceptions are @xmath292 , @xmath293 , and their mirror images as they have only plane symmetries . thus we have the following proposition . [ proprotants ] assume that a link @xmath13 has a tangle decomposition @xmath281 , where @xmath32 is a @xmath0-rotor and @xmath283 is a @xmath0-stator . if @xmath283 is generated by @xmath291 tangles in figure @xmath0 as described above ( with possible trivial components ) , then its rotant and @xmath13 are the same in the fourth skein module . @xmath74 in this section , we study the fourth skein module of the @xmath0-sphere and the relative skein module of the 3-dimensional disk . to understand the structure better , we also study fourth skein algebra for @xmath6-braid groups . here note that each skein module in this section has its tangle elements ( including links as @xmath129-tangles ) in @xmath8 or in @xmath294 . therefore we can analyze their diagrams on @xmath295 ( @xmath296 @xmath246 ) or @xmath297 and count the number of crossings of these diagrams . we use @xmath298 for a crossing of a braid , as usual and use @xmath299 in the case we do not specify whether the crossing is positive or negative . we use three kinds of equality symbols to denote various relations between a pair of n - tangle diagrams ( including link diagrams ) @xmath3 and @xmath169 . we say that @xmath3 and @xmath169 are _ skein module equivalent _ , denoted by @xmath3 @xmath27 @xmath169 , if they are equal in the skein module . we say that @xmath3 and @xmath169 are _ strongly equivalent _ , denoted by @xmath3 @xmath300 @xmath169 , if they are ambient isotopic . finally , we say that @xmath3 and @xmath169 are _ dot equivalent _ , denoted by @xmath3 @xmath301 @xmath169 , if the difference of @xmath3 and @xmath169 is presented in the skein module as a linear combination of elements with fewer crossings than @xmath3 and @xmath169 have . if every element of the linear combination has no more than @xmath6 crossings , then we denote the linear combination by @xmath302 . let @xmath303 denote the category of unoriented tangles and @xmath304 a semigroup of @xmath6-tangles . we read " morphisms from left to right and compose tangles in the same manner ( it corresponds to the notation @xmath305 denoting composition of tangles @xmath81 and @xmath82 ) . @xmath304 allows the natural action by the @xmath306 group , where @xmath307 is the action of the dihedral group ( preserving orientation ) of @xmath308 elements , generated by the rotation @xmath309 along the @xmath197-axis by the angle @xmath310 , and the rotation @xmath311 along the @xmath276-axis denoted by @xmath272 and often called a mutation . the factor @xmath312 corresponds to the mirror image map with respect to the @xmath275 , @xmath276 plane . the semigroup of @xmath6-tangles has two important subsemigroups : the braid group @xmath313 , generated by crossings ( tangles ) @xmath314 , and a semigroup of balanced @xmath6-tangles , @xmath315 , generated by @xmath314 and tangles @xmath316 creating maxima and minima ( for a visual notation , @xmath317 might be good . see figure @xmath0 for an example ) . in fact @xmath318 would suffice , as @xmath319 . representations of @xmath315 ( their finite dimensional quotients ) were studied by b.westbury . we are introducing also another semigroup , generalizing conway s algebraic tangles and allowing successful induction . [ defnalgtangles ] 1 . we define an n - algebraic tangle inductively as follows : 1 . an @xmath6-tangle with no more than one crossing is @xmath6-algebraic . inductive step : if @xmath81 and @xmath82 are two @xmath6-algebraic tangles then their twisted sum , @xmath320 is also @xmath6-algebraic . 2 . in a more restrictive way , we define @xmath321-algebraic tangles if in step @xmath23ii@xmath24 @xmath82 has no more than @xmath322 crossings . 3 . an @xmath6-algebraic link @xmath23resp . @xmath321-algebraic link@xmath24 is a link with a diagram obtained by closing an @xmath6-algebraic @xmath23resp . @xmath321-algebraic@xmath24 tangle by pairwise disjoint arcs . we denote by @xmath323 the monoid generated by @xmath6-algebraic tangles . [ lemn1algebraic ] 1 . \(a ) tangles in @xmath313 , @xmath324 and @xmath6-bridge tangles are @xmath325-algebraic . \(b ) @xmath6-bridge links and closed n - braids are @xmath325-algebraic links . the fact that elements of @xmath313 and @xmath324 are @xmath325-algebraic follows from the definition . for an @xmath6-bridge tangle , we start from the @xmath6-tangle which is equal to @xmath326 for @xmath327 or @xmath328 , and then we add crossings one by one ( compare figure @xmath329 , the figure is drawn to stress the fact that we deal with a 3-bridge tangle . in the algebra of 3-tangles we would describe it as @xmath330 ) . our main result in this section is the following . [ thmthreelinks ] any @xmath0-algebraic link can be generated as a linear combination of trivial links in @xmath331 . our proof works for any coefficients thus as a corollary we have : [ thmm - nconjecuture ] the montesinos - nakanishi conjecture is true for @xmath0-algebraic links . @xmath74 * remark . * what concerns the montesinos - nakanishi conjecture , we have two candidates for a possible counterexample . one is the double of borromean rings proposed by y. nakanishi @xcite ; it has 24 crossings . the other is proposed by q. chen in @xcite and it has @xmath332 crossings . chen s link is a reduction of the closure of the 5-braid @xmath333 by @xmath0-moves . in @xcite , it is shown that the double of borromean rings and chen s link are @xmath334-algebraic ( in fact , @xmath335-algebraic ) up to @xmath0-moves , which tell us that it is hard to prove theorems [ thmthreelinks ] and [ thmm - nconjecuture ] for @xmath334-algebraic links ( it is shown in @xcite that these theorems hold for @xmath336-algebraic links ) . each of the propositions and theorems presented in this section holds also for @xmath337 ( we take @xmath338 , @xmath339 and @xmath340 instead of @xmath341 and @xmath342 , respectively ) . since proofs are much simpler in that case , we omit them . first of all , let us define the fourth skein algebras for @xmath323 and @xmath313 . [ def4saforbg ] let @xmath32 be commutative ring with identity . let @xmath343 be the @xmath32-module with the basis @xmath313 . furthermore , let @xmath344 be the ideal of @xmath343 generated by the skein relation @xmath345 @xmath346 @xmath347 @xmath346 @xmath348 @xmath346 @xmath349 , where @xmath4 and @xmath5 are invertible in @xmath32 . we define the fourth skein module @xmath350 for @xmath313 as the quotient @xmath351 . the product for elements of @xmath313 induces a bilinear map @xmath350 @xmath352 @xmath350 @xmath353 @xmath350 so that , with respect to this product , @xmath350 becomes an algebra . thus we call @xmath350 the fourth skein algebra for @xmath313 . we define the forth skein algebra @xmath354 for @xmath323 as a subalgebra of the algebra of @xmath6-tangles modulo the fourth skein relations and the framing relations generated by @xmath6-algebraic tangles . note that in @xmath355 , any braid which contains @xmath356 or any non - alternating expression of @xmath357 or @xmath358 can be generated as a linear combination of elements with fewer crossings than those we start from . from the following proposition , we also see that @xmath359 and @xmath360 can be generated as a linear combination of elements with fewer crossings than those we start from . we call a braid in @xmath355 _ reducible _ if the braid contains one of those configurations . otherwise , we call the braid _ irreducible_. [ prop3braidoflength4 ] four configurations @xmath361 , @xmath362 , @xmath363 , and @xmath364 are dot equivalent to each other in @xmath355 . it is sufficient to show that @xmath362 @xmath301 @xmath361 , @xmath364 @xmath301 @xmath361 , and @xmath363 @xmath301 @xmath364 using the fourth skein relations . @xmath362 @xmath300 @xmath365 @xmath27 @xmath366@xmath367 @xmath368 @xmath346 @xmath366@xmath369 @xmath370 @xmath346 @xmath371@xmath367 @xmath372 @xmath346 @xmath371@xmath369 @xmath373 @xmath346 @xmath374 @xmath27 @xmath366@xmath367 @xmath368 @xmath346 @xmath374 . @xmath364 @xmath300 @xmath375 @xmath27 @xmath366@xmath367 @xmath368 @xmath346 @xmath366@xmath369 @xmath376 @xmath346 @xmath371@xmath367 @xmath377 @xmath346 @xmath371@xmath369 @xmath378 @xmath346 @xmath374 @xmath27 @xmath366@xmath367 @xmath368 @xmath346 @xmath374 . @xmath363 @xmath300 @xmath379 @xmath27 @xmath366@xmath367 @xmath380 @xmath346 @xmath366@xmath369 @xmath381 @xmath346 @xmath371@xmath367 @xmath382 @xmath346 @xmath371@xmath369 @xmath383 @xmath346 @xmath374 @xmath27 @xmath366@xmath367 @xmath380 @xmath346 @xmath374 . define @xmath341 as the set of @xmath384 invertible ( braid type ) basic tangles in figure @xmath0 . then we have the following proposition . [ propirreducible3braids ] for any irreducible element @xmath169 of @xmath385 , there exists an element @xmath386 of @xmath341 such that @xmath386 is dot equivalent to @xmath169 . let @xmath6 be the number of crossings of @xmath169 . if @xmath6 @xmath387 @xmath0 , then there exists an element @xmath386 of @xmath341 such that @xmath386 is strongly equivalent to @xmath169 . if @xmath6 @xmath27 @xmath334 , then @xmath169 is an alternating expression of @xmath357 or @xmath358 . then , @xmath169 is dot equivalent to @xmath361 from proposition [ prop3braidoflength4 ] . there is no irreducible element of @xmath385 with more than four crossings . using this proposition , we obtain the following theorem . [ thmthreebraids ] any element of @xmath355 can be generated as a linear combination of elements of @xmath341 . we prove the theorem by an induction on the number of crossings of the element . if @xmath388 , then clearly the statement holds . assume that the statement holds in the case @xmath6 fewer than @xmath322 . consider the case @xmath389 . if the element is reducible , then it is generated by elements with fewer crossings than @xmath322 , which can be generated by elements of @xmath341 from the assumption of the induction . if the element is irreducible , then it is generated by an element of @xmath341 and elements with fewer crossings than @xmath322 from proposition [ propirreducible3braids ] . this induces that the irreducible element is also generated by elements of @xmath341 . define @xmath342 as the set of @xmath390 non - invertible basic tangles in figure @xmath0 . then we have the following . [ propproduct ] a product of any pair of elements of @xmath341 and @xmath342 can be generated as a linear combination of elements of @xmath391 ( with possible trivial components ) . if both elements of the pair are @xmath0-braids , then the statement follows from theorem [ thmthreebraids ] . the other cases are easy to be checked . for simplicity , we ignore trivial components in the consideration . [ proprotation ] any rotation of elements of @xmath341 and @xmath342 can be generated as a linear combination of elements of @xmath391 . we need to show only for @xmath392 and @xmath393 . from figure @xmath394 , we have @xmath395 @xmath396 @xmath397 @xmath398 since any @xmath0-tangle with no more than one crossing can be expressed as @xmath399 , where @xmath400 , @xmath401 is an element of @xmath402 , and @xmath3 reflects a possible framing change yielded by a kink " , we obtain the following theorem from propositions [ propproduct ] and [ proprotation ] . [ thmthreetangles ] any element of @xmath403 can be generated as a linear combination of elements of @xmath391 ( with possible trivial components ) . @xmath74 _ proof of theorem [ thmthreelinks ] . _ from theorem [ thmthreetangles ] , we know that every @xmath0-algebraic link can be generated as a linear combination of @xmath0-algebraic links obtained from elements of @xmath391 . for each of these links , it is easy to see that it can be generated as a linear combination of trivial links . @xmath74 in this section we present results of calculations that suggest that even for @xmath8 the fourth skein module is more powerful ( can distinguish more links ) than the third ( jones - conway ) and kauffman skein modules . in particular , if the fourth skein polynomial exists ( trivial links are linearly independent ) , then the polynomial is ( at least sometimes ) more powerful than the jones - conway ( homflypt ) and kauffman polynomials . our calculation shows that under these assumptions we can distinguish the @xmath404-knot from its mirror image @xmath405 ( @xcite ) and some 2-bridge links that share the same jones - conway and kauffman polynomials ( @xcite ) . in the second part of this section we present general ideas that can lead to a solution of montesinos - nakanishi conjecture and its generalizations . in particular we speculate about the geometrical meaning of the finite quotients of the braid group described by j.assion @xcite ( symplectic and unitary cases ) . [ ] 1 . there is a polynomial invariant of unoriented links , @xmath406 $ ] that satisfies : * initial conditions : @xmath407 , where @xmath84 is the trivial link of @xmath6 components . * skein relation , @xmath408 , where @xmath409 is a standard , unoriented skein quadruple @xmath410 is obtained from @xmath411 by a right - handed half twist on two arcs involved in @xmath412 . there is a polynomial invariant of unoriented framed links , @xmath413 $ ] which satisfies : 1 . initial conditions : @xmath414 , 2 . framing relation : @xmath415 , where @xmath416 is obtained from a framed link @xmath13 by a positive half twist on its framing . skein relation : @xmath417 . 3 . in each case the polynomial is uniquely defined . [ 10.1 ] consider the @xmath404-knot , in notation of @xcite . then the polynomial @xmath418 distinguishes @xmath404 from its mirror image @xmath405 . in fact , we obtain the following for @xmath404 . we use the fact that we have the formula : @xmath419 . @xmath420 @xmath401 @xmath421 @xmath422 @xmath423 @xmath424 @xmath346 @xmath425 @xmath426 . [ 10.2 ] consider two @xmath10-bridge knots @xmath427 and @xmath428 as shown in the following figure . according to kanenobu @xcite , they share the jones - conway @xmath23homflypt@xmath24 and kauffman polynomials . we can , however , distinguish them by @xmath429 polynomial . in fact , we obtain the following for @xmath430 ; @xmath431 and @xmath432 . we speculate that conjectures [ generalconjmn ] and [ conjgenerator ] can be approached by using results of coxeter and of assion and by interpreting them using some elementary moves on oriented links . coxeter showed that @xmath433 is finite iff @xmath434 , @xcite . assion found two basic cases in which @xmath435 is finite : the symplectic " and unitary " cases @xcite . let @xmath436 be a generator of the center of the braid group , @xmath437 . it is easy to check that assion s ideals are generated by @xmath438 and @xmath439 , respectively ( see also @xcite appendix iii exercise 1.3 ) . 1 . symplectic case " . @xmath440 is a finite group . 2 . unitary case " . @xmath441 is a finite group . we should remark that @xmath442 in @xmath443 , and @xmath444 is a simple group @xmath445 ( projective symplectic group ) . one could try to incorporate deformation of relations @xmath446 or @xmath447 into skein module relations , but we would speculate that another approach may give appropriate skein modules . we will work with oriented links and the following useful definition and conjecture @xcite . [ 10.3 ] 1 . the @xmath448-move is a local change of an oriented link which adds @xmath6 positive half - twists to @xmath13 @xmath449 as in figure @xmath390 @xmath23i@xmath24 . 2 . the @xmath450-move is a local change of an oriented link which adds @xmath322 right handed half - twists to @xmath13 @xmath451 as in figure @xmath390 @xmath23ii@xmath24 , where @xmath322 is an even integer @xmath23notice anti - parallel orientation of the strings involved in the move@xmath24 . two links are @xmath452 equivalent if one can start from the first one and reach the second one by using @xmath448-moves , @xmath450-moves , and their inverses . [ 10.4 ] 1 . every oriented link in @xmath8 is @xmath453 equivalent to a trivial link . every oriented link in @xmath8 is @xmath454 equivalent to a trivial link . we speculate that there is the following connection between our moves and assion ideals . [ 10.5 ] 1 . the 5-tangle associated with the 5-braid @xmath438 is @xmath0-move equivalent to the trivial 5-braid tangle . the 5-tangle associated with the 5-braid @xmath439 is @xmath454 equivalent to the trivial 5-braid tangle . note that from ( i ) ( and @xcite ) it follows immediately that closed 5-braids are @xmath0-equivalent to trivial links . b.wajnryb proved ( see @xcite , p.694 ) that every link can be reduced to a trivial link by @xmath455 moves and @xmath438-moves ( wajnryb s move @xmath456 is @xmath455-equivalent to @xmath438-move ) . thus the positive answer to ( i ) yields the positive answer to the montesinos - nakanishi conjecture . generally the positive solution to problem [ 10.5 ] would allow partial solution to conjectures [ generalconjmn ] and [ conjgenerator ] . one should comment , at least shortly , on the background of problem [ 10.5 ] . we follow approach and notation of @xcite ( compare @xcite , @xcite ) . let @xmath277 be an oriented link diagram . we associate @xmath277 with the alexander - burau module over the ring @xmath32 ( with invertible element @xmath401 ) , as follows . to every arc of the diagram we associate a generator ( variable ) @xmath457 and every crossing , @xmath458 , of the diagram yields the relation @xmath459 where @xmath460 is the sign of the crossing @xmath458 , see figure @xmath461 . if we think of a neighborhood of a crossing as a @xmath10-tangle ( read from left to right ) , our relation yields a @xmath462 matrix of a linear map ( we can say that we consider a contra - variant functor from the category of 2-tangles to @xmath32-modules category ) . for tangles of figure @xmath461 matrices are : @xmath463 , \ \ a_2 = a_1^{-1 } = \left [ \begin{array}{cc } 0 & 1 \\ t & 1-t \end{array } \right]\ ] ] + in particular the matrix corresponding to the move of figure @xmath464 is : @xmath465 ^ 3 = \left [ \begin{array}{cc } 1-t^{-3}(t^2-t+1 ) & t^{-3}(t^2-t+1 ) \\ t^{-2}(t^2-t+1 ) & 1- t^{-2}(t^2-t+1 ) \end{array } \right ] $ ] @xmath466\ mod ( t^2-t+1 ) $ ] generally we get that the @xmath455-move preserves the alexander - burau module iff @xmath467 in the ring @xmath32 . we can repeat our analysis for @xmath468-moves . in particular the figure @xmath469 illustrates an example of @xmath470-moves , with the matrix @xmath471 \left [ \begin{array}{cc } 1-t^{-1 } & t^{-1 } \\ 1 & 0 \end{array } \right ] = \left [ \begin{array}{cc } 2-t^{-1 } & t^{-1}-1 \\ 1-t^{-1 } & t^{-1 } \end{array } \right]\ ] ] we can now analyze the equivalences @xmath472 and @xmath473 noting that @xmath474\ ] ] our results are summarized in the following lemma ( compare @xcite ) . [ ] 1 . the alexander - burau module @xmath23of an oriented link or a tangle@xmath24 is unchanged by a @xmath455-move if and only if @xmath475 in the ring @xmath32 . the alexander - burau module is unchanged by @xmath455 and @xmath476-moves iff @xmath475 and @xmath477 in @xmath32 . notice that @xmath478 follows from @xmath477 . the alexander - burau module is unchanged by @xmath455 and @xmath479-moves iff @xmath475 and @xmath480 in @xmath32 . [ 10.8 ] let @xmath481 denote the @xmath322-fold cyclic branched cover of @xmath8 @xmath23resp . @xmath482 with the branching set the link @xmath23resp . tangle@xmath24 @xmath13 . then : 1 . @xmath483 is unchanged by a @xmath455 and @xmath476-moves @xmath23more generally by 3-moves@xmath24 . @xmath484 is unchanged by a @xmath455 and @xmath479-moves . for our heuristic argument for a positive answer to conjecture [ 10.4 ] it remains to show that @xmath438-move preserves alexander - burau matrix modulo @xmath485 and @xmath439-move preserves alexander - burau matrix modulo @xmath486 . it is a tedious but easy task . in particular figure @xmath332 illustrates the fact that the matrix @xmath487 for @xmath488 is @xmath489\ ] ] we get modulo @xmath490 : @xmath491\ ] ] which is the identity matrix iff @xmath492 and @xmath493 . in fact we have more generally : @xmath494 . one should stress here that the @xmath495 move is not preserving the alexander - burau module modulo @xmath496 therefore this move is not a combination of @xmath455 and @xmath476 moves ( we conjecture only that it is a combination of 3-moves ) . we have also ( still modulo @xmath490 ) : @xmath497\ ] ] which is the identity matrix iff @xmath498 . it follows from coxeter work that @xmath499 . we have more generally : @xmath500 . we would like to thank bruce westbury for very helpful discussion . anstee , j.h . przytycki , d. rolfsen , _ knot polynomials and generalized mutation _ , topology appl . * 32 * ( 1989 ) , 237249 . m. aschbacher , m. hall , _ groups generated by a class of elements of order @xmath0 _ , j. algebra . * 24 * ( 1973 ) , 591612 . j. assion , _ einige endliche faktorgruppen der zopfgruppen _ , math . z. * 163 * ( 1978 ) , 291302 . j. assion , _ a proof of a theorem of coxeter _ , c. r. math . canada * 1 * ( 1978/79 ) , 4144 . j. birman , b. wajnryb , _ markov classes in certain finite quotients of artin s braid group _ , israel j. math . * 56 * ( 1986 ) , 160178 . d. bullock , _ rings of @xmath501-characters and the kauffman bracket skein module _ , comment . * 72 * ( 1997 ) , 521542 . q. chen , _ the @xmath0-move conjecture for @xmath77-braids _ , master degree thesis of the george washington university ( 1999 ) , to appear in knots in hellas 98 ; the proceedings of the international conference on knot theory and its ramifications ; volume * 1 * ( the series on knots and everything , vol . coxeter , _ factor groups of the braid group _ , proc . fourth canadian math . congress , banff ( 1957 ) , 95122 . crowell , r.h . fox , _ an introduction to knot theory _ , ginn and co. ( 1963 ) . r. fenn , c. rourke , _ racks and links in codimension two _ , journal of knot theory and its ramifications , * 1 * ( 1992 ) , 343406 . p. freyd , d. yetter , j. hoste , w.b.r . lickorish , k. millett , a. ocneanu , _ a new polynomial invariant of knots and links _ , bull . ( n.s . ) * 12 * ( 1985 ) , 239246 . j. hoste , j.h . przytycki , _ a survey of skein modules of @xmath0-manifolds _ , proceedings of knots 90 ( ed . a. kawauchi ) , walter de gruyter ( 1992 ) , 363379 . t. kanenobu , _ kauffman polynomials for @xmath10-bridge knots and links _ , yokohama math . j. * 38 * ( 1991 ) , 145154 . t. kanenobu , _ genus and kauffman polynomial of a @xmath10-bridge knot _ , osaka j. math . * 29 * ( 1992 ) , 635651 . t. kanenobu , t. sumi , _ polynomial invariants of @xmath10-bridge knots through @xmath502 crossings _ , math . comp . * 60 * ( 1993 ) , 771778 . r. kirby , _ problems in low - dimensional topology _ , geometric topology ( athens , ga , 1993 ) ( ed . w.h.kazez ) , the american mathematical soc . ( 1997 ) , 35473 . k. murasugi , _ a study of braids _ , mathematics and its applications , kluwer academic publishers ( 1999 ) . przytycki , _ @xmath503 moves on links _ , contemporary mathematics * 78 * braids - proceedings of the santa cruz conference on artin s braid groups on july 1986 ( 1988 ) , 615656 . przytycki , _ skein modules of 3-manifolds _ , bull . pol . math . * 39 * ( 1991 ) , 91100 . przytycki , _ elementary conjectures in classical knot theory _ , in _ quantum topology _ , ed . l.j.kauffman , r.a.baadhio , series on knots and everything - vol.3 , world scientific , ( 1993 ) , 292320 . przytycki , _ a q - analogue of the first homology group of a 3-manifold _ , contemporary mathematics * 214 * perspectives on quantization ( proceedings of the joint ams - ims - siam conference on quantization , mount holyoke college , 1996 ) ; ed . l.a.coburn , m.a.rieffel , ams ( 1998 ) , 135144 . przytycki , _ @xmath0-coloring and other elementary invariants of knots _ , banach center publications * 42 * ; _ knot theory _ ( 1998 ) , 275295 . przytycki , _ algebraic topology based on knots : an introduction _ , proceedings of _ knots 96 _ ( edited by s. suzuki ) world scientific publishing co. ( 1997 ) , 279297 . przytycki , symplectic structure on colorings and lagrangian tangles , preprint , 2000 ( compare : _ abstracts of ams _ 21(4 ) ) . j.h . przytycki , a.s . sikora , _ on skein algebras and @xmath504-character varieties _ , topology * 39 * ( 2000 ) , 115148 . d. rolfsen , _ knots and links _ , mathematics lecture series * 7 * , publish or perish , inc . t. tsukamoto , _ the fourth skein module of @xmath0-dimensional manifolds _ , ph.d . thesis of the george washington university ( 2000 ) . t. tsukamoto , _ the fourth skein module for @xmath334-algebraic links _ , preprint ( 2000 ) . turaev , _ skein quantization of poisson algebras of loops on surfaces _ , ann . * 4 * ( 1991 ) , 635704 . b. wajnryb , _ a braid - like presentation of @xmath505 _ , israel j. math . * 76 * ( 1991 ) , 265288 . b. wajnryb , markov classes in certain finite symplectic representations of braid groups , in : braids ( santa cruz , ca , 1986 ) , 687695 , contemp . , 78 , amer . soc . , providence , ri , 1988 . b. westbury , _ quotients of the braid group algebras _ , special issue on braid groups and related topics ( jerusalem , 1995 ) , topology appl . * 78 * ( 1997 ) , 187199 .

soft gamma ray repeaters ( sgrs ) and anomalous x ray pulsars ( axps ) are young and radio - quiet x - ray pulsars which have been rapidly spun - down to slow spin periods clustered in the range s. most of these unusual pulsars also appear to be associated with supernova shell remnants ( snrs ) with typical ages kyr . by examining the sizes of these remnants versus their ages , we demonstrate that the interstellar media which surrounded the sgr and axp progenitors and their snrs were unusually dense compared to the environments around most young radio pulsars and snrs . we explore the implications of this evidence on magnetar and propeller - based models for the rapid spin - down of sgrs and axps . we find that evidence of dense environments is not consistent with the magnetar model unless a causal link can be shown between the development of magnetars and the external ism . propeller - driven spin - down by fossil accretion disks for sgrs and axps appears to be consistent with dense environments since the environment can facilitate the formation of such a disk . this may occur in two ways : 1 ) formation of a `` pushback '' disks from the innermost ejecta pushed back by prompt reverse shocks from supernova remnant interactions with massive progenitor wind material stalled in dense surrounding gas , or 2 ) acquisition of disks by a high velocity neutron stars , which may be able to capture a sufficient amounts of co - moving outflowing ejecta slowed by the prompt reverse shocks in dense environments .

astro-ph9912207

soft gamma ray repeaters ( sgrs ) are neutron stars whose multiple bursts of gamma rays distinguish them from other gamma ray burst sources ( e.g @xcite for a recent review ) . sgrs are also unusual x ray pulsars in that they have spin periods clustered in the interval @xmath2 s , and they all appear to be associated ( @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ; @xcite ) with supernova remnants ( snrs ) , which limits their average age to approximately @xmath3 kyr ( @xcite ) . the angular offsets of the sgrs from the apparent centers of their associated supernova remnant shells indicate that sgrs are endowed with space velocities @xmath4 km s@xmath5 , which are greater than the space velocities of most radio pulsars ( @xcite ) . anomalous x ray pulsars ( axps ) are similar to sgrs in that they are radio quiet x ray pulsars with spin periods clustered in the range @xmath6 s , and have similar persistent x ray luminosities as the sgrs ( @xmath7 ergs s@xmath5 , see e.g. @xcite for a recent review ) . most of the axps appear to be associated with supernova remnants , and therefore they are also thought to be young neutron stars like the sgrs . the spin periods of both axps and sgrs are increasing with time ( spinning - down ) , and show no evidence for intervals of decreasing spin period ( spin - up ) , although the spin - down rates of many of the sgrs and axps appear to be variable or `` bumpy '' ( e.g. @xcite ; @xcite ; @xcite ; @xcite ; although for a different viewpoint see @xcite ) . the lack of identified companions at non x ray wavelengths ( e.g. @xcite ) and doppler shifts associated with binary orbital motion ( @xcite ) , together with the problem of accelerating binaries to space velocities @xmath8 km s@xmath5 ( @xcite ) , imply that sgrs and axps are not members of high mass binary systems , although low mass systems with neutron star companions of @xmath9 are not constrained in most cases . if sgrs and axps spin - down primarily via the emission of magnetic dipole radiation ( mdr ) , as do radio pulsars , then they must have surface dipole fields of @xmath10 g or greater ( i.e. `` magnetars '' ; e.g. @xcite ) . observations of the sgrs 180620 ( @xcite ) and 1900 + 14 ( @xcite ; @xcite ) , however , indicate that the present - day spin - down rates of these sgrs are inconsistent with simple mdr , given the ages of their associated supernova remnants ( @xcite ; @xcite ) , and imply that the spin - downs are due to winds . magnetar - strength fields might still be possible in these sources , however , if alfvn wave wind emission is infrequent and intermittent ( @xcite , so that the presently observed spin - down rates are atypical . alternative scenarios for sgrs and axps involving typical pulsar magnetic fields ( @xmath11 g ) have been proposed ( @xcite ; alpar 2000 ; chatterjee , hernquist & narayan 2000 ; chatterjee & hernquist 2000 ) . in these models , the sgrs and axps have spun - down rapidly via magnetospheric accretion torques from outflowing `` propeller effect '' winds . the assumed sources of the magnetospheric material are either fallback accretion disks ( alpar 2000 ; chatterjee , hernquist & narayan 2000 ; chatterjee & hernquist 2000 ) , or fossil disks formed from expanding supernova ejecta intercepted by high velocity neutron stars ( @xcite ; @xcite ) . here we present a fresh look at evidence which suggests that sgrs and axps are born into unusually dense environments . we show that the environments of the sgr and axp progenitors into which their snrs expand are the dense , warm and cool phases of the interstellar medium ( ism ) , and not the hot tenuous phase of the ism where most of the neutron - star - producing , core collapse supernovae of massive o and b stars occur and where most young radio pulsars are found . this implies that there is an environmental factor influencing the development of sgrs and axps . the structure of this paper is as follows . we first discuss the typical environments of supernova progenitors in @xmath12 , and then supernova remnants associated with sgrs and axps in @xmath13 , followed by a discussion of the sgr and axp ages and distances are in @xmath14 . in @xmath15 the density of the sgr and axp progenitor environments is discussed , and in @xmath16 a similar analysis is done for the snrs associated with young radio pulsars . in @xmath17 we discuss the statistical significance of the results and the implications for magnetar and propeller - based models for sgrs and axps . finally , the appendix contains a short discussion of the ages , distances , and other information for each sgr and axp . what are the environments typical of neutron star progenitors ? observations clearly show that the majority of neutron stars are formed in `` superbubbles '' : evacuated regions of the ism which surround the ob associations in which the massive progenitors of most neutron stars live . this is because most o and b stars ( @xmath18 80% ) are observed ( e.g. mccray & snow 1979 ) to occur in clusters formed from giant molecular clouds ( @xmath19 ) ; much smaller clouds are disrupted by the radiation and winds from the first o star that forms . these massive ( @xmath20 m@xmath21 : @xcite ) and slow - moving ( @xmath22 km s@xmath5 : @xcite ) o and b star progenitors of type ii and ib / c supernovae do not travel far from their birthplaces during their relatively short ( @xmath1 myr : @xcite ) lives . the supernovae from these massive stars are therefore heavily clustered in space and time and form vast ( @xmath23 pc ) hii regions / superbubbles ( e , g . @xcite ) filled with a hot ( @xmath24 k ) and tenuous ( @xmath25 @xmath26 ) gas . there is also direct observational evidence that the great majority ( @xmath1880% ) of neutron star stars are born into superbubbles consisting of hot and diffuse ism gas . observations of a sample of 49 spectroscopically identified type ii and ib / c , core - collapse supernovae occurring in face - on late - type spiral galaxies by van dyk et al . ( 1996 ) found that @xmath27 of type ii and @xmath28 of type ib / c supernovae are in resolvable giant hii region superbubbles . we suggest that these fractions are , in fact , only a lower limit on the occurrence of core - collapse supernovae in superbubbles , because of the difficulty in detecting faint hii regions in distant galaxies . correcting for this effect using the h@xmath29 luminosity distributions of kennicutt , edgar , & hodge ( 1989 ) , we find that the h@xmath29 threshold used by van dyk et al . ( 1996 ) would have allowed them to resolve only @xmath30 of the hii regions in their sample galaxies . this clearly suggests that the great majority i.e. @xmath31 of neutron star progenitors reside in the hot and diffuse ( @xmath32 @xmath26 ) ism in superbubbles , with @xmath33 occurring in the denser phases of the ism . when one corrects for selection effects , the distribution of surface brightness of all detected galactic supernova remnants provides another measure of the fractions of supernovae occurring in the warm and hot phases of the ism . detections of radio remnants are limited by surface brightness , which is independent of their distance , and snrs in the denser ism have larger surface brightnesses than snrs of the same age in the hot / diffuse ism ( e.g. @xcite ) . a study ( @xcite ) of the age versus surface brightness of the remnants of historical supernovae , using gull s model , suggests that the maximum detectable ages of radio supernova remnants above a nominal surface brightness density of 10@xmath34 w m@xmath35 hz@xmath5 sr@xmath5 at 408 mhz are roughly 20 kyr in the warm ( @xmath36 @xmath26 ) gas and 6 kyr in the hot superbubble ( @xmath37 @xmath26 ) gas . combining this with an analysis of the number distribution of snrs as a function of surface brightness suggests that about @xmath38 of the detected radio snrs are in the warm denser ( @xmath39 @xmath26 ) phase of the ism ( e.g. @xcite ) . this implies that only about @xmath40 of the supernovae occur in the warm ( @xmath39 @xmath26 ) ism , and about @xmath41 in the hot phase , since , for a total ( including type ia ) sn rate of 1 sn every 30 yrs , there are then roughly 180 snr younger than 6 kyr that are detectable in the hot superbubble ism , and roughly 70 snr younger than 20 kyr that are detectable in the denser warm ism , or about @xmath38 of the detected radio snrs in the warm ism as the observed distribution implies . the environments of sgrs and axps are probed by the blastwaves of their associated supernova remnants , and from the size of the remnant shell as a function of the age we can constrain the external density . in table @xmath42 we have listed the @xmath43 known sgrs and axps and their associated supernova remnant shells . for more information on the individual sgrs and axps , see the recent reviews of hurley ( 2000a ) and mereghetti ( 1999 ) , and the appendix . we include the source axp 18450258 although no period derivative has been measured for this source because it is commonly cited as an axp ( see e.g. the discussion in @xcite ) . we also include the axp 07203125 for completeness , although its identification as an axp has been questioned as well ( see appendix ) . the new and tentative sgr 180123 has also been included on the basis of two soft gamma ray bursts consistent with a single location on the sky ( @xcite ) . the identification of the associated remnants are based on positional coincidences between the remnant and the sgr / axp , and in some cases on similar distances as implied by the hydrogen column densities measured from the x ray spectrum of the sgr / axp and its associated remnant . based on these criteria , associated snrs have been previously been identified for 7 out of the 12 sgrs and axps ( see table 1 and appendix ) . using these same criteria , we suggest three additional snr associations with sgr / axps ( also included in table 1 and appendix ) . we suggest two new axp / snr associations , axp 170940 with the snr g346.60.2 ( @xcite and references therein ) and axp 10485937 with g287.80.5 ( @xcite ) , based on the near coincident positions between each axp and a known supernova remnant , and also ( for axp 10485937/g287.80.5 ) on similar implied distances for the axp and the remnant . these two associations may have been discounted previously because they would imply larger than average neutron star velocities , but in view of the comparable velocities implied for three of the sgrs these possible associations should be considered . axp 10485937 lies @xmath44 shell radii from the apparent center of the snr g287.80.5 , which is associated with the carina nebula , a region of massive star formation at a distance @xmath45 pc ( @xcite ; @xcite ) . this relatively small distance of the snr is quite consistent ( see fig . 2 ) with the low @xmath46 cm@xmath35 found from the axp 10485937 x ray spectral fits ( @xcite ) . for an estimated snr age of @xmath4820 kyr , the displacement of the axp implies a relatively modest neutron star velocity of @xmath49 km s@xmath5 . a similar transverse velocity is indicated by the association of axp 170940 , which lies 1.7 shell radii from the center of the well defined shell remnant g346.60.2 , assuming a similar age and a distance of 3 to 5 kpc , based on galactic structure . such a snr distance is quite consistent with its association with axp 170940 , which has an x - ray absorption @xmath50 cm@xmath35 ( @xcite ) . we also propose that the new sgr candidate 180123 ( @xcite ) may be associated with the well studied snr w28 . although there is no distance estimate to the sgr ( @xcite ) , w28 is the only known snr through which the thin sgr error box passes . therefore we regard the tentative w28/sgr 180123 association as promising and encourage observations to search for x ray point sources ( e.g. @xcite ) within the remnant which may be the sgr counterpart . we include these three tentative snr associations because they are suggestive , but as we show in @xmath51 these data can be left out of the sample without affecting the overall conclusions of our analysis . using the snr catalog of green ( 2000 ) , we are unable to identify any possible remnants associated with axps 07203125 and 0142@xmath52615 , which may be the two oldest axps based on their pulsar timing ages @xmath53 , where @xmath54 is the pulsar period and @xmath55 is the period derivative , of @xmath56 kyr ( @xcite ) and @xmath57 kyr ( e.g. @xcite ) , respectively . for axp 07203125 the lack of an associated remnant is not surprising , because this source is thought to be so close ( @xmath58 pc ) that a neutron star with a velocity of @xmath48 1000 km s@xmath5 would have traveled @xmath59 pc in @xmath56 kyr and hence could have originated from a large area of the sky . in addition , the large scale radio surveys of the galactic plane appear to be incomplete for very large ( @xmath60 ) diameter remnants ( e.g. @xcite ) , indicating that such a remnant could easily go undetected . axp 0142@xmath52615 is situated in or behind a molecular cloud complex ( @xcite ; @xcite ) , and if its associated remnant expanded in high density material it may have already passed into the radiative phase and faded below the surface brightness detection threshold . therefore the detection of the remnants associated with both axp 07203125 and axp 0142@xmath52615 would be difficult , and we can not assign meaningful limits to the physical size of their ( unknown ) associated remnants , given the present data . we encourage new deep observations of the regions surrounding these objects to look for associated supernova remnants . we see from table 1 that 10 out of the 12 sgr / axps have proposed associations with radio shell snrs . the probabilities of a chance coincidence for some of the individual associations have been estimated ( e.g. @xcite ; @xcite ; @xcite ) at a few times 10@xmath35 or less ( see appendix ) . the chance probability for an association between an sgr / axp and an snr can be estimated by simply considering the spatial density of known supernova remnants in the galactic plane ( in the following we assume that the localization error box of the sgr / axp is small compared to the size of the snr ) . for the extreme case , we consider just the sgrs and axps located in the highly crowded inner galaxy . there all @xmath61 sgr / axps appear to have associated snrs . these have galactic longitudes between 286@xmath62 and 43@xmath62 and latitudes within @xmath63 1.2@xmath62 . within this area of 1.0@xmath6410@xmath65 arc min@xmath66 , there are 142 known snrs ( green 2000 ) covering a combined surface area of 3.4@xmath6410@xmath67 arc min@xmath66 , which is actually an overestimation due to overlap between the remnants . for a single association with the sgr or axp displaced from the center of the remnant by as much as @xmath68 times the remnant radius ( as implied by the sgr 162741/g337.00.1 association ) , the chance association probability is roughly @xmath69 , or @xmath70 correcting for overlap . thus there is a significant probability that a single association might be simply a chance coincidence , if we relied solely on position , but the probability of getting even 4 chance associations out of 8 tries is only 2% , and the probability that 8 out of 8 are spurious is just @xmath71 . in order to probe the snr environments , we need to know both the radii of the associated snr shells , which can be determined from their measured angular diameters , and their estimated distances . these distances are listed in table 2 , and discussed in the appendix . all but three of the distances were determined from galactic kinematic arguments based on interactions or associations with adjacent objects with known distances ( e.g. hii regions or molecular clouds ) , or on absorption line measurements . we did not use distances estimated from the standard snr surface brightness / diameter relationships , because we did not want to bias the snr sizes toward snrs in the dense ism , which is where most of the observed radio snrs are located ( e.g. higdon & lingenfelter 1980 ; kafatos et al . 1980 ) . snrs expanding in the tenuous hot ism become large and hard to detect ( e.g. @xcite ) much quicker than snrs in dense environments and are under - represented . the positions of the galactic snrs associated with sgrs and axps are shown in figure @xmath42 , projected onto a model of the galactic plane distribution of free electrons ( @xcite ) which traces star formation regions delineating the spiral arms ( e.g. georgelin & georgelin 1976 ) . we see that the estimated distances of the snrs are quite consistent galactic structure , placing them in or near the spiral arms where the bulk of recent massive star formation has occurred . we also see that the estimated distances of the snrs are not systematically underestimated in our analysis , since 5 out of the 9 snr - axp / sgr associations lie roughly at or beyond the distance of the galactic center as would be expected for a relatively unbiased sample of galactic sources . this also shows that they are not likely to be systematically much farther away , or most would lie on the other side of the galaxy or even outside of the galaxy . to further test both the axp / sgr snr associations and their estimated distances , we compare the electron column depth associated with the snr distance with the h column depth determined from the x - ray absorption of the axp / sgr emission . we calculate the dispersion measure to each source given the assumed distances and compare that to the measured @xmath72 value for each source . table @xmath73 lists the distance ranges , calculated dispersion measures using the taylor & cordes ( 1993 ) model , and observed @xmath72 values for each sgr / axp and associated snr . the dispersion measure ( dm ) vs. @xmath72 is plotted in figure @xmath73 . as seen from the figure , the data points are well correlated between lines representing free electron to hydrogen ratios @xmath74 , which is consistent with the range of @xmath75 , expected from the mean mass fraction of ionized gas in the galaxy of @xmath76 ( e.g. @xcite ) . thus we see that each of associations is consistent both in proximity of direction and in similarity of distance . this also shows that the estimated distances to the sgrs / axps and their associated remnants are not systematically underestimated . more information on the @xmath72 values for each source is given in the appendix . the ages of the most of the sgrs and axps are much more uncertain than their distances . this is because the usual pulsar timing age formula appropriate for pulsar spin - down due to magnetic dipole radiation ( mdr ) torque ( although @xcite ) does not seem to be applicable for sgrs and axps , in general , because it yields age estimates which are inconsistent with the likely ages of the associated supernova remnants and the implied neutron star velocities for at least three of the sources ( sgr 180620 , sgr 1900 + 14 , and axp 2259 + 58.6 : @xcite ; @xcite ; and @xcite , respectively ) . in addition , all of the sgr and axps for which pulsar braking indices have been measured axps 170940 and 2259 + 58.6 ( @xcite ) and sgr 180620 ( @xcite ) have braking indices which are inconsistent with simple mdr spin - down . the mdr timing age for axp 1841045 of 4000 yr , however , is consistent with the estimated age of the associated snr kes 73 ( @xcite ) , but the braking index for this axp has not yet been measured . we therefore adopt ages for the sgrs and axps based on the ages of their associated snrs . some of the associated remnants ( n49 , kes 73 , g29.6@xmath520.1 , and ctb 109 ) are relatively well - studied and have age estimates based on observed shock velocities and/or x ray temperatures . for these remnants we take the minimum and maximum values for the ages of each remnant from the literature . for the rest of associated snrs , we adopt a lower age limit of @xmath77 $ ] , where @xmath78 is the age of the remnant assuming free expansion , with @xmath79 the minimum estimated distance , @xmath80 the snr radius in radians , and @xmath81 km s@xmath5 is the maximum ejecta speed in free expansion . in addition , if we assume that the transverse velocity of the associated sgr or axp can not exceed @xmath82 , the minimum age is @xmath83 , where @xmath84 is angular offset of the pulsar from the center of the remnant . in this paper we will assume @xmath85 km s@xmath5 , which exceeds the observed velocity of any pulsar ( e.g. @xcite ) . for the upper age limit of the snrs we choose @xmath86 kyr , which is the maximum estimated age for the snr / radio pulsar associations shown in table 3 . for each snr , we assume that the most likely age is the arithmetic mean of the age range , which is equivalent to assuming that the supernova rate is uniform in time throughout the galaxy . in figure @xmath87 we have plotted the snr shell radii @xmath88 as a function of the estimated age @xmath89 of each remnant associated with an sgr or axp . overplotted with solid lines are simple approximations of the evolutionary tracks ( @xcite ) of supernova remnant expansion in a wide range of the external ism densities @xmath90 , assuming a total supernova kinetic energy of @xmath91 ergs . these snr evolutionary tracks move through three phases : the initial free expansion / ejecta - dominated phase of the remnant , where the mass of the sn ejecta is much greater than the mass of the swept - up ism and @xmath92 ; the sedov / adiabatic phase , which begins when the mass of the swept - up ism is roughly @xmath93 of the mass of the sn ejecta , and the remnant slows down with @xmath94 ; and finally the radiative / snowplow phase , where the shell of swept - up ism radiates away the energy of the remnant and it slows further with @xmath95 ( e.g. @xcite ) . for individual supernova remnants , the tracks may differ by less than @xmath96 for a factor of three variation in supernova kinetic energies ( @xcite ) . also overplotted as dotted lines are the tracks of neutron stars born at the origin of the supernova explosion with varying velocities , showing the times required for them to catch up with the outer ( radio ) shell of the expanding remnant . we see from figure @xmath87 , that in spite of the uncertainties in the remnant ages and distances , all of the supernova shell remnants associated with sgrs and axps seem to reside in the denser ( @xmath97 @xmath26 ) phases of the ism , where @xmath98 of the neutron star producing supernovae occur . although the ages of the the snrs are highly uncertain , information on the progenitor ism density can be extracted from the snrs because densities inferred from the sedov and radiative phase expansion formulae depend much more strongly on the snr radii than on their ages . as we discuss in more detail below , the inferred distribution of progenitor ism densities is not consistent with an origin of sgrs and axps that results from a purely intrinsic property of the neutron stars , because such stars should predominantly be born in the hot , diffuse ( @xmath99 @xmath26 ) phase of the ism , where @xmath100 of the neutron star and pulsar producing sn occur . the probability of finding 10 such snrs out of 12 possible occurring in the denser ( @xmath97 @xmath26 ) phases of the ism , when at most @xmath101 are expected is only @xmath102{(0.2)}^{10 } { ( 0.8)}^{2 } \sim 4\times 10^{-6}$ ] . even if we assume the maximum @xmath103 probability for chance associations with snr for all sgrs and axps _ and _ assume that all chance associations will be in denser ism , the combined probability of any single , typical snr being in the denser ism is 0.33 , equalling @xmath104 naturally occurring plus @xmath105 occurring by chance . thus the probability of finding 10 out of 12 possible occurring in the denser ism is still only @xmath102{(0.33)}^{10 } { ( 0.67)}^{2}$ ] or @xmath106 . there are two obvious ways that the distribution of these snrs might be made consistent with that of typical neutron stars and pulsars , i.e. @xmath107 residing in the hot diffuse ism and @xmath108 in the dense ism . one way would be if the snr distances were systematically underestimated by roughly a factor of 3 , but as we saw in figure 1 and 2 , that is not consistent with galactic structure and would place roughly half of the snrs well outside the galaxy . alternatively , exceedingly weak ( @xmath109 ergs ) explosive energies for the sgr and axp supernovae could in principle explain the unusually compact remnants associated with sgrs and axps , but this is inconsistent with dynamical requirements of supernova explosions ( e.g. woosley & weaver 1995 ) , and with the direct observations of sgrs and axps associated snrs ( e.g. @xcite ; @xcite ; @xcite ; @xcite ) . moreover , there is direct evidence that at least 8 out of the 10 sgrs and axps associated snrs are indeed in dense environments . oh maser emission , attributed to snr shock interactions with molecular clouds , has been detected ( frail et al . 1996 ; green et al . 1997 ) from molecular clouds thought to be associated with five of the snrs ( ctb 33 , kes 73 , w28 , g10.00.3 , and g346.60.2 ) . ctb 109 ( e.g. huang & thaddeus 1985 ; tatematsu et al . 1987 ) , g287.80.5 ( @xcite ) , and n49 ( @xcite ) also show evidence for molecular clouds associations . these associations with molecular clouds clearly show that the supernova remnant shells associated with sgrs and axps are expanding in high density environments . we can compare the observed properties of the snrs associated with sgrs and axps with the remnants associated with another population of young neutron stars the young radio pulsars listed in the catalogs of taylor , manchester , & lyne ( 1993 ) and taylor et al . we consider only the youngest pulsars with timing ages less than @xmath110 kyr , so as to be comparable to the age range in the sgr / axp sample . the mdr timing ages of radio pulsars , unlike those for the sgrs and axps , are generally though to be good measures of their true ages ( e.g. @xcite ) , and we assume that the remnant ages are consistent with the mdr timing ages of their associated pulsars . a notable exception is psr j18012451 , which has an estimated age much greater than its mdr timing age ( @xcite ) possibly indicating the presence of non - mdr spin - down torques in this pulsar . for the pulsar distances , we use the distances derived ( @xcite ; @xcite ) from the measured dispersion and the taylor & cordes ( 1993 ) model of the free electron distribution in the galaxy . the supernova remnant shells of many of the young pulsars in table @xmath87 have not been detected in large scale radio surveys of the galactic plane ( see e.g. @xcite ; @xcite ; @xcite ) , or in deep radio observations of the fields surrounding young pulsars ( e.g. @xcite ; @xcite ) . for these pulsars , we found no likely associated snrs in the green ( 2000 ) catalog within a search radius corresponding to a transverse velocity of as much as @xmath111 km s@xmath5 , assuming the estimated distances and timing ages for these sources listed in table @xmath42 . since the green ( 2000 ) snr catalog is more or less complete ( @xcite ) down to a limiting surface brightness , the missing snrs for these objects have probably expanded such that their surface brightnesses have faded below the limiting surface brightness of the radio surveys of the galactic plane . we therefore assume that the radio shells of the remnants associated with these pulsars ( except for the crab , for which we use a lower limit of @xmath112 pc for the remnant radius ; @xcite ) have expanded beyond this detectability threshold , and following braun , goss , & lyne ( 1989 ) we assign lower limits of @xmath86 pc to the undetected remnants corresponding to these pulsars . the observed and derived parameters for the young radio pulsars are listed in table @xmath87 , with references to the timing data , distances , and snr shell radii . a plot of the supernova remnant shell radii versus age for the radio pulsars is shown in figure @xmath113 . we clearly see from figure @xmath113 that most of the young pulsars appear to have been born in the hot , diffuse phase of the ism , as expected from other observations of their o and b star progenitors and their environments discussed in the preceding section . only @xmath114 of the @xmath115 young pulsars have snrs which may be expanding in the denser phases ( @xmath39 @xmath26 ) of the ism , which is consistent with the expected fraction of @xmath101 or less despite the small number statistics . these results are summarized in table @xmath113 . as seen from figures @xmath87 and @xmath113 , the sgrs and axps appear to form in denser regions of the interstellar medium than the sample of young radio pulsars . the significance of the apparent disparity between the sgr / axp and radio pulsar environments can be evaluated by using `` survival analysis '' ( @xcite ) methods , which are statistical techniques incorporating `` censored '' data ( e.g. upper limits ) into data analysis . we first convert the age and snr radii values of tables 1 and 3 to ambient ism densities @xmath90 using the standard formulae in shull ( 1983 ) and using values for each age and snr radius midway between the range of values listed in the tables . a total supernova kinetic energy of @xmath91 ergs was assumed for both the sgr / axps and radio pulsars . the lower limits on the radii of the undiscovered snrs in table 3 therefore became upper limits on the ambient densities for these snrs . the resulting distributions of @xmath90 for the sgr / axps and radio pulsars were then tested for consistency with a single parent sample using the statistical analysis package asurv rev 1.2 ( @xcite ) , which implements the methods presented in feigelson & nelson ( 1985 ) . the two - sample univariate nonparametric tests used to compare the distributions consisted of two versions of the gehan s generalized wilcoxon test ( permutation and hypergeometric variances ) , the logrank test , and the peto & peto and peto & prentice generalized wilcoxon tests ( see @xcite and references therein ; @xcite ; @xcite ) . these methods differ in their assumptions regarding the censoring process and in the ways which they weight the data , but we found that they gave results which were consistent to within two orders of magnitude . we use the statistical tests to compute the probability @xmath116 that the distribution of @xmath90 for the sgrs and axps is consistent with the same distribution for the young radio pulsars . for the full sample of snr associations in table @xmath42 we calculate @xmath117 . if we exclude from the sample the objects with tentative snr associations sgr 180123 and axps 170940 and 10485937 we find @xmath118 . finally , if we change the maximum possible remnant age from @xmath86 kyr to @xmath3 kyr ( @xcite ) , we obtain @xmath119 for the full set of snr associations in table @xmath42 , and @xmath120 if we exclude the tentative snr associations . the data therefore supports the conclusion that the progenitors of sgrs and axps exploded in denser environments than the progenitors of radio pulsars . this conclusion can be alleviated only if one assumes that a large fraction of the previously claimed snr associations with axps and sgrs are spurious _ and _ the remnants truly associated with these objects have large diameters and are currently undetected . as discussed in @xmath13 , however , it is unlikely that more than a couple of the associations in table 1 are spurious , given the distribution of supernova remnants in the galactic plane . at least five of the snr associations in table @xmath42 would have to be spurious before one could conclude that the evidence for dense sgr and axp progenitor was insignificant ( i.e. the probability @xmath121 ) . the association of sgr 1900 + 14 with snr g42.8 + 0.6 has recently been questioned after the recent discovery ( @xcite ) of a 40 kyr old pulsar psr j1907 + 0918 equally close @xmath122 to the snr , and the discovery ( @xcite ) of a compact cluster of massive stars at an estimated distance ranging from @xmath43 to @xmath123 kpc , and only @xmath124 from the sgr 1900 + 14 line of sight . the relatively low h column depth from absorption in the sgr s x ray spectrum ( @xcite ; @xcite ) clearly implies ( e.g. figure @xmath73 ) a distance closer to 5 kpc , so that the sgr and and snr may not be related to the cluster . if sgr 1900 + 14 is associated with the compact cluster at the more extreme distance , vrba et al.(2000 ) suggest an association with a possible compact ( @xmath125 pc diameter ) snr indicated by the local diffuse x ray emission . this new sgr / snr association would therefore imply a much greater sgr 1900 + 14 progenitor density than the g42.8 + 0.6 association , increasing the evidence for dense progenitor environments for sgrs . a similar compact ( @xmath126 pc diameter ) cluster of massive stars has also been observed ( @xcite ) to be coincident with sgr 180620 and g10.00.3 . in addition , axp 10485937 and its associated snr g287.80.5 are associated with the carina star - forming region ( @xcite ; @xcite ) . in these cases the associations with the clustered star - formation regions are quite consistent with the sgr / snr associations , since the distances of the star - forming regions and the snrs are comparable . the unusual properties of sgrs and axps may be due to the fact that they have unusually strong magnetic fields , as in the magnetar model ( @xcite ; @xcite ; @xcite ; @xcite ) . in the context of this model , it has been suggested that the association of sgrs and axps with dense progenitor environments may be explained by a selection effect in the following manner . magnetars are thought to form from progenitor stars with high angular momenta , such that the superstrong ( @xmath127 g ) magnetic field can be generated in the protoneutron star by dynamo action just after core collapse ( @xcite ) . since the total stellar angular momentum @xmath128 ( assuming rigid rotation , where @xmath129 is the observed rotational velocity ) , one might expect the largest and most massive stars to be the likely progenitors of magnetars . furthermore , since the main sequence lifetimes of stars decrease with increasing mass , the more massive stars would be the first to supernovae therefore exploding into more dense surroundings before the parent molecular cloud has been cleared by the successive supernovae of less massive stars . the two basic assumptions of this scenario do not appear to be consistent with observations . first , the observed stellar rotation velocities ( e.g. @xcite ) show that the angular velocity @xmath130 increases with stellar mass only up to about 2 m@xmath21 and then _ decreases _ with increasing mass , dropping to only @xmath108 of its maximum value for stellar masses @xmath131 . since the stellar evolution timescale even in massive stars is much longer than the convective timescale ( e.g. @xcite ) , the rotation of the star is effectively rigid and one would therefore expect the pre - collapse cores of the most massive stars to have less angular momenta than the cores of less massive supernova progenitors . this is supported by detailed calculations of the stellar evolution of rotating massive stars up to the time of core collapse ( @xcite ) , which indicate that the final angular momentum of the core region is less for a massive star than for a less massive star with the same initial rotation rate . this is due to the greatly increased mass losses of luminous massive stars , which shed angular momentum in their powerful stellar winds ( e.g. @xcite ) . second , the supernovae of massive stars would still not be expected to all explode in the denser medium , because star formation in massive ( @xmath18 10@xmath132 m@xmath21 ) clouds is episodic through several generations ( e.g. @xcite ) . these successive generations commence roughly every @xmath484 to 5 myr , over a period of @xmath4820 to 30 myr , each producing roughly 10@xmath133 o and b star supernova progenitors . each subsequent generation of star formation commences shortly after the onset of supernova explosions from the previous generation , at the end of their main sequence lifetimes of @xmath183 myr . therefore only the most massive stars in the first generation of star formation in a new star forming region would be expected to explode in denser environments , and the subsequent generations of massive stars should be born in the same low density superbubble environment as the population as a whole . thus even if magnetars were produced by the most massive stars , we still would not expect more than a small fraction of them to explode in the denser phases of the ism . the rapid spin - down rates , ages , clustered spin periods , and x ray luminosities of axps and/or sgrs can all be explained by models involving the propeller effect as the dominant spin - down torque ( @xcite ; alpar 2000 ; chatterjee , hernquist & narayan 2000 ; chatterjee & hernquist 2000 ) . in the context of these models , sgr bursts can be explained in terms of solid body accretion ( @xcite ; @xcite ) or crustal instabilities ( @xcite ) . because there are upper limits on the masses of the possible binary companions of at least some of the axps ( e.g. @xcite ) , the accreted material is probably ejecta from the neutron star s own supernova explosion ( @xcite ; @xcite ) . in addition , since some of the sgrs and axps are located outside their apparent supernova remnants , the accreted material must form an accretion disk which can store the angular momentum of the accreted material and spin - down the neutron star on time scales of @xmath134 kyr . upper limits on the disk emission from the nearby axp 2259 + 58.6 ( @xcite ; @xcite ) may rule out a standard hydrogen accretion disk around this axp , but a disk may still exist given the uncertainties in the spectra of disks formed from metal - rich supernova ejecta and dust ( e.g. @xcite ) . in addition , the noisy spin - down of sgr 180620 ( @xcite ) and the @xmath135 kev emission feature ( @xcite ) observed from an sgr 1900 + 14 burst are both consistent with accretion disks around these objects . there are two possible scenarios involving accretion from the ejecta of the supernova explosion which produces the neutron star : fallback disk accretion , `` pushback '' disk accretion , and accretion involving high velocity neutron stars . it was recently proposed that axps may be formed from neutron stars accreting material from `` fallback '' accretion disks ( chatterjee , hernquist & narayan 2000 ; chatterjee & hernquist 2000 ) . these disks may be formed from @xmath136 ( @xcite ; @xcite ) of inner ejecta material @xmath137 hours after the initial core collapse in a type ii supernova explosion ( e.g. @xcite ; @xcite ) . since a total accreted mass of only @xmath138 is required to explain the spin - down of sgrs and axps via the propeller mechanism ( e.g. alpar 2000 ; chatterjee , hernquist & narayan 2000 ; chatterjee & hernquist 2000 ) , only a very small fraction ( @xmath139 ) of the fallback material must be accreted into a disk for this model to explain spin - down rates and ages of the sgrs and axps . formation of such a fallback disk is limited to @xmath140 days after the core collapse because of heating of the ejecta by @xmath141ni ( @xcite ) . this is long before the remnant feels the external environment early in the sedov phase , and therefore formation of an early fallback disk may not be compatible with the evidence for dense sgr and axp progenitor environments . a model involving fallback disks which form later in the evolution of the snr , however , could be consistent with the evidence of dense sgr and axp environments . in particular , the expansion of slow - moving ejecta can be reversed by the sedov phase reverse shock resulting from the interaction between the blastwave and the external medium ( e.g. @xcite ; @xcite ) , and the subsequent implosion could result in the formation of a `` pushback '' disk so named because the snr ejecta is pushed - back onto the star because of the interaction with the dense environment surrounding the progenitor . pushback disk formation would probably be most likely for : 1 ) neutron star progenitors which experienced a large amount of mass loss prior to supernova , and 2 ) progenitors in the dense ism , which would provide the necessary pressure to confine the wind mass near the star . this is exemplified by the circumstellar environments surrounding the progenitor of sn 1987a , which is surrounded by @xmath142 in wind material and hii gas ( e.g. @xcite ) . the pushback process begins at the `` reversal time '' @xmath143 kyr ( @xcite ) , and the pushback mass is @xmath144 given a total ejecta mass of @xmath145 . as with the fallback disks discussed above , the formation of a pushback disk from only a small fraction of this matter would be required to explain the spin - down of sgrs and axps . since this later fallback occurs after the majority of @xmath141ni and @xmath141co decays , the formation of the accretion disk around the neutron star would not be limited by radioactive heating . such a model may be a plausible explanation for the dense sgr and axp environments , and needs to be explored in more detail . the final propeller - based scenario for sgrs and axps involves high velocity neutron stars ( hvnss ) capturing disk material from co - moving supernova ejecta , as first suggested van paradijs , taam , and van den heuvel ( 1995 ) . although the exact mechanism by which neutron stars are given substantial `` kick '' velocities at birth is not known , observations show that the kick velocities exceed @xmath146 km s@xmath5 in approximately @xmath108 of all neutron stars ( @xcite ) . in addition , this kick velocity appears to be independent of the dipole moment of the neutron star , as indicated by population studies of isolated radio pulsars ( @xcite ; @xcite ) and observations of extremely high velocity stars with canonical neutron star magnetic fields of @xmath11 g ( e.g. psr b2224@xmath5265 : @xcite ) . as implied by the ratio @xmath147 listed in table @xmath42 , many of the sgr / axp positions are significantly displaced from the apparent centers of their associated snrs . these displacements imply that the sgr / axps may have systematically large transverse velocities , although there is considerable uncertainty in the actual velocities , mainly because of uncertainties in the ages of the associated remnants . in addition , the actual space velocities of the sgr / axps are larger by an unknown factor dependent on the viewing angle . it has been estimated ( @xcite ) that a @xmath148 accretion disk may be acquired by a high velocity neutron star as it moves through nearly co - moving supernova ejecta . however detailed calculations of the time - dependent accretion rate for the range of ism densities , progenitor mass loss parameters , and neutron star magnetic fields , initial spin periods , and velocities are needed to properly constrain such a model . this is beyond the scope of the present paper , and will be left for future work . additional evidence in favor of the pushback disk and hvns models for sgrs and axps may be provided by the observed number of these sources . for the pushback disk model , the number of sgrs and axps less than @xmath89 years old is given by @xmath149 , where @xmath150 is the galactic neutron star birthrate , @xmath151 is the fraction of neutron star progenitors in the warm dense ism , and @xmath152 is the fraction of neutron star progenitors which experience mass loss sufficient to form a pushback disk in dense ism environments . since the rate of mass loss from stellar winds is an increasing function of the initial main sequence mass , the fraction @xmath152 can be estimated by considering the _ minimum _ stellar mass which undergoes pre - supernova mass loss sufficient to decelerate its supernova ejecta at early times . to first order , this should occur when the total wind mass emitted during the progenitor s life is approximately equal to the supernova ejecta mass . from the solar metallicity stellar models of schaller 1992 ) which include mass loss this occurs for stars of initial main sequence masses greater than @xmath153 . using @xmath154 and a salpeter imf with a maximum and minimum neutron star progenitor mass of @xmath155 and @xmath156 , respectively ( @xcite ) , we obtain @xmath157 . assuming @xmath158 ( as discussed in @xmath12 ) and @xmath159 yr@xmath5 ( @xcite ) , we obtain @xmath160 sgrs and axps with ages less than @xmath161 kyr . this is similar to the observed number ( @xmath43 ) of these sources , which can be taken as evidence supporting the pushback disk model . a similar calculation is possible for the hvns model . in this case , the expected number of sgrs and axps less than @xmath89 years old is given by @xmath162 , where @xmath163 of neutron stars with high space velocities and @xmath150 and @xmath151 are defined as before . using the same values of @xmath151 and @xmath150 , and assuming @xmath164 ( velocities @xmath4 km s@xmath5 ; @xcite ) , yields @xmath165 expected sgr and axp sources with ages less than @xmath86 kyr for the hvns model . again , these numbers are in the right ballpark for the observed numbers of sgrs and axps . we have shown that soft gamma ray repeaters ( sgrs ) and anomalous x ray pulsars ( axps ) are born in regions of the interstellar medium which are denser than the environments typical of young neutron stars . this suggests that the development of sgrs and axps may be related to their environments , and we examine the implications of this on magnetar and propeller - based models for sgrs and axps . the evidence of dense progenitor environments would be consistent with the magnetar model only if magnetars are born exclusively in dense environments , which does not appear to be the case if magnetars form only from the most massive stars . propeller - based models for sgrs and axps involving the formation of accretion disks from supernova ejecta appear to be consistent with the evidence for dense progenitor environments since these environments may induce the formation of such disks . this may occur in two ways . _ pushback _ disks may be formed from the infall of the innermost ejecta , pushed back towards the neutron stars by prompt reverse shocks from the interactions of the expanding remnants with massive progenitor winds confined close to the stars by dense surrounding gas producing rapid deceleration of the expanding ejecta and strong prompt reverse shocks ( truelove & mckee 1999 ) . fossil disks may also form around high velocity neutron stars accreting from nearly co - moving supernova ejecta , slowed by the strong prompt reverse shocks in such dense environments ( van paradijs et al . 1995 ) . we acknowledge helpful suggestions from anonymous referees which led to improvements in the paper . this research made extensive use of nasa s astrophysics data system abstract service . this work was performed while one of the authors ( dm ) held a national research council - gsfc research associateship . rer acknowledges support by nasa contract nas5 - 30720 , and rel support from the astrophysical theory program . here we discuss and reference the snr distances , radii , and ages listed in tables @xmath42 and @xmath73 . the snrs associated with all but three of the sgrs and axps have kinematic distances derived from associations with objects ( e.g. hii regions or molecular clouds ) having known distances . the distances of the other three snrs g29.6 + 0.1 , g346.60.2 , and g42.8 + 0.6 are derived from consideration of galactic spiral arms along the line of sight . the remnants kes 73 , n49 , w28 , and ctb 109 have published age estimates based on measurements of shock velocities and/or x ray temperatures , and for these remnants we adopt an age range corresponding to the minimum and maximum values quoted in the literature . the rest of the associated remnants have no published age estimates , other than those determined by assuming sedov expansion and distances determined from @xmath166 relations , which are biased toward denser phases of the ism . for these snrs we adopt a broad range of possible ages . we take a lower age limit of @xmath167 $ ] , where @xmath168 and @xmath169 are the minimum ages derived by assuming free expansion of the snr ( at @xmath170 km s@xmath5 ) and a maximum sgr / axp transverse velocity of @xmath111 km s@xmath5 , respectively . for an upper age limit , we take @xmath86 kyr , which is the maximum estimated age for a snr / radio pulsar association in table 3 . finally , the hydrogen column densities ( @xmath72 , listed in table @xmath73 ) are determined from the best - fit spectral models for the x ray spectra of the sgrs / axps . _ axp 1841045 _ : this axp is coincident with the snr g27.4 + 0.0 ( kes 73 ) , which is estimated from the x ray temperature to be @xmath171 kyr old ( @xcite ; @xcite ) . the diameter of the radio remnant is @xmath172 ( @xcite ) , and the distance from hi absorption is estimated to be @xmath173 kpc ( @xcite ) . a power law fit to the x ray spectrum of the axp yields a best - fit value of @xmath174 cm@xmath35 ( @xcite ) , which is consistent with such a distance ( see figure 2 ) , and with the @xmath72 of @xmath175 cm@xmath35 found ( e.g. @xcite ) of the x - ray spectrum of the snr . _ sgr 052666 _ : the error box of the sgr lies within the supernova remnant n49 ( @xcite ; @xcite ) , located in the lmc at a distance of @xmath176 kpc ( @xcite ; @xcite ) and interacting with a molecular cloud ( @xcite ) . the diameter of n49 is @xmath177 , and its estimated age is @xmath178 kyr ( @xcite ; @xcite ) . _ axp 2259@xmath5258.6 _ : the axp lies within the @xmath179 diameter ( @xcite ) snr ctb 109 ( g109.21.0 ) . the distance to the remnant , as determined by spectroscopy of stars in nearby hii regions , is in the range @xmath180 kpc ( @xcite ) . the estimated remnant age is @xmath181 kyr ( @xcite ; @xcite ) , and there is evidence for interaction between ctb 109 and surrounding molecular clouds ( e.g. @xcite ) . the best - fit model to the x ray spectrum of the axp yields @xmath182 cm@xmath35 ( @xcite ) , which agrees with the range of @xmath183 cm@xmath35 found for the x - ray spectra from various parts of the remnant ( @xcite ) . _ axp 18450258 _ : this axp is lies within the @xmath184 diameter remnant g29.6 + 0.1 ( @xcite ) . there are no kinematic distances , and the distance estimated from galactic structure ( see figure @xmath42 ) is @xmath185 kpc . this is consistent ( see figure @xmath73 ) with the axp x ray absorption of @xmath186 cm@xmath35 ( @xcite ) . there are no reliable snr age estimates , so we adopt an snr age range of @xmath187 kyr by taking the limits of free expansion and the maximum detected snr age from the radio pulsar sample . the chance association probability between the axp and the snr is estimated to be @xmath188 ( @xcite ) . _ sgr 162741 _ : this sgr was first associated with the snr g337.00.1 by hurley et al . ( 1999c ) . the distance to this remnant is estimated to be @xmath189 kpc ( @xcite ) , based on an association with a giant molecular cloud within the star - forming region ctb 33 . the angular diameter of snr g337.00.1 is @xmath190 ( @xcite ) . the snr age is unknown , so we adopt an age of @xmath191 kyr . the probability that the sgr / snr association is spurious was estimated to be @xmath192 ( @xcite ) . a marginal detection of a @xmath135 s pulsation from the sgr was reported ( @xcite ) but not confirmed by subsequent observations ( @xcite ) . the x ray spectrum of the sgr is equally well - fit by power law , blackbody , and thermal bremsstrahlung functions ( @xcite ) , and the mean @xmath72 for the power law spectral fit is @xmath193 cm@xmath35 ( @xcite ; @xcite ) , which is consistent with the snr distance ( see figure 2 ) . _ sgr 180123 _ : the most recently discovered sgr , this source has a long and thin error box ( @xcite ) which passes through the center of snr w28 ( g6.40.1 ) . w28 is associated with oh masers in a molecular cloud whose distance rules out its previously suggested association with psr b175823 ( @xcite ) . w28 has a radio diameter of @xmath194 ( @xcite ) , an estimated age of @xmath195 kyr ( @xcite ) , and is located at a distance of @xmath196 kpc ( @xcite ; @xcite ) . the low energy absorption in the x ray spectrum of w28 is @xmath197 cm@xmath35 ( @xcite ) , but there is no @xmath72 estimate for the sgr since its burst spectra were were not measured to low enough energy to determine a value , and the persistent low energy x - ray counterpart of the source has not yet been identified . _ axp 170940 _ : this axp lies @xmath198 form the center of the @xmath199 diameter ( @xcite ) remnant g346.60.2 . the chance association probability between the axp and this remnant is @xmath200 , from the method of @xmath13 . there are no published age estimates for the snr , so we derive and age range of @xmath201 kyr from the transverse velocity and maximum snr age limits . there is no kinematic distance estimate to the snr , so we adopt a distance of @xmath202 kpc , which agrees with galactic structure arguments placing it in or near spiral arms . this distance is also consistent ( figure @xmath73 ) with the axp x ray absorption of @xmath203 cm@xmath35 ( @xcite ) . _ sgr 180620 _ : this sgr lies within the @xmath204 diameter ( @xcite ) remnant g10.00.3 , which is located at a distance of @xmath205 kpc as determined from co line observations of maser - associated molecular clouds ( @xcite ) . this distance is consistent ( figure @xmath73 ) with the large column density @xmath206 cm@xmath35 found ( @xcite ) for the x ray spectrum of the sgr . the chance sgr / snr association probability has been estimated at @xmath207 ( @xcite ) . the age is unknown , so we adopt an age of @xmath208 kyr . this sgr and snr may be associated with a compact cluster of stars ( @xcite ) . _ sgr 1900@xmath5214 _ : this sgr has been associated ( e.g. @xcite ) with the @xmath209 diameter remnant snr g42.8 + 0.6 ( @xcite ) from whose center it is displaced by @xmath210 . no kinematic distances to the snr are available , so we adopt a distance of @xmath211 kpc from association with a galactic spiral arm ( figure 1 ) . this distance is also consistent ( see figure @xmath73 ) with the mean value of the sgr x ray absorption for the best - fit spectral model : @xmath212 cm@xmath35 ( @xcite ) . the age of the remnant is unknown , so we adopt an age range of @xmath213 kyr . the chance probability of the sgr / snr association by the method of @xmath13 is @xmath200 . this sgr has also been associated ( @xcite ) with a compact star cluster at a distant of anywhere from @xmath214 to @xmath123 kpc , although the relatively low value of @xmath72 from the absorption in the sgr x ray spectrum ( @xcite ; @xcite ) favors the near end of the range . _ axp 10485937 _ : this axp is situated near the @xmath215 diameter remnant g287.80.5 ( @xcite ) associated with the carina star - forming region at a distance of @xmath216 kpc ( @xcite ; @xcite ) . the geometrical chance association probability ( @xmath13 ) between the axp and the remnant is @xmath217 . there are no age estimates of the snr and therefore we derive an age of @xmath218 kyr from the transverse velocity limit and the maximum detectable snr age . the best - fit spectral model of the axp emission yields the low value of @xmath219 cm@xmath35 ( @xcite ) , which is consistent ( e.g. figure @xmath73 ) with the snr / carina distance . _ axp 07203125 _ : despite its low x ray luminosity ( @xmath220 ergs s@xmath5 , @xcite ) , this @xmath221 s pulsar has been included in the axp sample due to its spin period , lack of an optical counterpart ( @xcite ) , and fast spin - down rate ( @xmath222 , @xcite ) . although there is no adjacent snr in the catalog of green ( 2000 ) , the low x - ray absorption indicates a distance of only @xmath223 kpc ( @xcite ) to the source , which makes the detection of its associated snr unlikely ( see @xmath13 ) if its age is as great ( @xmath59 kyr ) as that indicated by its spin down rate . _ axp 0142@xmath52615 _ : this apparently old ( possibly @xmath224 kyr ) and well - studied ( e.g. @xcite ; @xcite and references therein ) axp has no remnant within @xmath225 of it in the green ( 2000 ) catalog . but there is evidence ( @xcite ) that this source is in or behind a giant molecular cloud . given the rapid evolution ( e.g. @xcite ) and possible self - absorption ( @xcite ) of the radio emission of snrs in dense environments , the associated snr may have faded below detectability , and therefore we are unable to constrain the radius of the associated remnant . green , d. a. 2000 , `` a catalogue of galactic supernova remnants ( 2000 august version ) '' , mullard radio astronomy observatory , cambridge , united kingdom ( available on the world wide web at `` http://www.mrao.cam.ac.uk/surveys/snrs/ '' ) ergs in a wide range of ism densities . the dotted lines denote the tracks of neutron stars born at the origin of the supernova explosion with varying space velocities . the data show that these objects are unusual in that they are all preferentially formed in the denser ( @xmath226 h @xmath26 ) phases of the interstellar medium ( ism ) , where @xmath22720% of all neutron - forming supernovae occur , as determined from observations of ob associations and galactic supernova remnants.,width=528 ] , except for young radio pulsars with timing ages less than @xmath228 kyr . the remnants for these pulsars reside primarily in the diffuse phase of the ism , as expected from observations of ob associations and galactic supernova remnants.,width=528 ] lccccr axp 1841045 & @xmath229 & kes 73 & @xmath230 & @xmath231 & @xmath232 sgr 052666 & @xmath233 & n49 & @xmath234 & @xmath235 & @xmath236 axp 2259@xmath5258.6 & @xmath237 & ctb 109 & @xmath238 & @xmath239 & @xmath104 axp 18450258 & @xmath240 & g29.6@xmath520.1 & @xmath241 & @xmath242 & @xmath232 sgr 162741 & @xmath135 ? & g337.0@xmath2430.1 & @xmath244 & @xmath245 & @xmath68 sgr 180123 & & w28 & @xmath244 & @xmath246 & @xmath232 axp 170940 & @xmath247 & g346.60.2 & @xmath248 & @xmath249 & @xmath250 sgr 180620 & @xmath251 & g10.00.3 & @xmath248 & @xmath252 & @xmath253 sgr 1900@xmath5214 & @xmath254 & g42.8@xmath520.6 & @xmath255 & @xmath256 & @xmath257 axp 10485937 & @xmath258 & g287.80.5 & @xmath255 & @xmath259 & @xmath44 axp 07203125 & @xmath221 & & @xmath56 & & axp 0142@xmath52615 & @xmath260 & & @xmath57 & & lccccccr kes 73 & @xmath261 & @xmath262 & @xmath173 & @xmath263 & axp 1841045 & @xmath264ctb 109 & @xmath265 & @xmath266 & @xmath180 & @xmath267 & axp 2259@xmath26858.6 & @xmath269g29.6@xmath520.1 & @xmath270 & @xmath271 & @xmath185 & @xmath272 & axp 18450258 & @xmath273g337.0@xmath2430.1 & @xmath274 & @xmath275 & @xmath276 & @xmath277 & sgr 162741 & @xmath278g346.60.2 & @xmath279 & @xmath280 & @xmath281 & @xmath282 & axp 170940 & @xmath283g10.00.3 & @xmath284 & @xmath285 & @xmath286 & @xmath287 & sgr 180620 & @xmath288g42.8@xmath520.6 & @xmath289 & @xmath290 & @xmath291 & @xmath292 & sgr 1900@xmath5214 & @xmath293g287.80.5 & @xmath294 & @xmath295 & @xmath216 & @xmath296 & axp 10485937 & @xmath297 lccccrr j0534@xmath522200 & @xmath298 & @xmath299 & @xmath300 & @xmath243 & @xmath301 & 1j15135908 & @xmath302 & @xmath303 & @xmath304 & msh 1552 & @xmath305 & 2j05406919 & @xmath306 & @xmath250 & @xmath307 & snr 0540693 & @xmath308 & 3j16145047 & @xmath309 & @xmath310 & @xmath311 & @xmath243 & @xmath312 & 4j16175055 & @xmath313 & @xmath314 & @xmath315 & @xmath243 & @xmath312 & 5j08354510 & @xmath316 & @xmath317 & @xmath297 & vela xyz & @xmath318 & 11 j13416220 & @xmath319 & @xmath43 & @xmath320 & g308.80.1 & @xmath321 & 7j18012451 & @xmath322 & @xmath323 & @xmath324 & g5.41.2 & @xmath325 & 8j18032137 & @xmath326 & @xmath115 & @xmath327 & @xmath243 & @xmath312 & 8j17094428 & @xmath328 & @xmath112 & @xmath329 & @xmath243 & @xmath312 & 2,10j1856@xmath520113 & @xmath330 & @xmath3 & @xmath331 & w44 & @xmath332 & 9j10485832 & @xmath333 & @xmath3 & @xmath334 & @xmath243 & @xmath312 & @xmath243j17403015 & @xmath335 & @xmath336 & @xmath337 & @xmath243 & @xmath312 & 6j18261334 & @xmath338 & @xmath339 & @xmath340 & @xmath243 & @xmath312 & 6j17303350 & @xmath341 & @xmath342 & @xmath343 & @xmath243 & @xmath312 & @xmath243j16464346 & @xmath309 & @xmath228 & @xmath344 & g341.2@xmath520.9 & @xmath345 & 10

we consider a graphical approach to index coding . while cycles have been shown to provide coding gain , only disjoint cycles and cliques ( a specific type of overlapping cycles ) have been exploited in existing literature . in this paper , we define a more general form of overlapping cycles , called the interlinked - cycle ( ) structure , that generalizes cycles and cliques . we propose a scheme , called the interlinked - cycle - cover ( ) scheme , that leverages structures in digraphs to construct scalar linear index codes . we characterize a class of infinitely many digraphs where our proposed scheme is optimal over all linear and non - linear index codes . consequently , for this class of digraphs , we indirectly prove that scalar linear index codes are optimal . furthermore , we show that the scheme can outperform all existing graph - based schemes ( including partial - clique - cover and fractional - local - chromatic number schemes ) , and a random - coding scheme ( namely , composite coding ) for certain graphs . shell : bare demo of ieeetran.cls for journals index coding problem , unicast , linear index codes , interlinked - cycle cover , optimal broadcast rate .

1603.00092

index coding ( introduced by birk and kol @xcite in 1998 ) , a sender broadcasts messages through a noiseless shared channel to multiple receivers , each knowing some messages a priori , which are known as side information . side information occurs frequently in many communication networks , e.g. , in a web browsers cache . knowing the side information of the receivers , the sender can send coded symbols , known as an index code , in such a way that all of the receivers can decode their requested messages using their side information and the received coded symbols . the aim is to find the shortest ( optimal ) index code . how to optimally design an index code for an arbitrary index - coding instance is an open problem to date . in the literature , various approaches have been adopted to solve the index - coding problem . we broadly classify these approaches into four categories : ( i ) numerical , ( ii ) shannon s random coding , ( iii ) interference alignment , and ( iv ) graph - based . numerical approaches include rank minimization over finite fields @xcite ( which is np - hard to compute in general @xcite ) , and mathematical optimization programming ( semi - definite programming @xcite , linear programming @xcite , and integer - linear programming @xcite ) . these approaches do not provide much intuition on the interaction between the side - information configuration and the index codes . shannon s random coding approaches @xcite require infinitely long message packets . interference - alignment approaches treat index coding as an interference - alignment problem @xcite , and construct index codes via two alignment techniques , namely one - to - one alignment and subspace alignment . these alignment techniques have no well - defined algorithms to construct index codes for arbitrary index - coding instances . graph - based approaches @xcite provide intuition on the side - information configurations and index codes . these approaches represent index - coding instances by graphs , and construct index codes as functions of the graphs . these graph - based schemes provide linear ( scalar and vector ) index codes . although linear index codes are not always optimal @xcite , they have simpler encoding and decoding processes . we classify graph - based approaches into two sub - categories : ( i ) maximum distance separable ( mds ) code based interference alignment approaches , and ( ii ) graph structure based approaches . the mds code based interference alignment approaches construct index codes by treating messages not known to a receiver as interference , and aligning all interference with the help of mds codes . these approaches include the partial - clique - cover scheme @xcite and its fractional version @xcite , the local - chromatic - number scheme and its fractional version @xcite , and the partitioned - local - chromatic - number scheme and its fractional version @xcite . graph structure based approaches exploit special graph structures , based on messages known to the receivers that can provide savings on index - coding instances . it has been shown that no structure in an acyclic graph can provide any savings @xcite . furthermore , if an arc does not belong to any cycle , then removing it does not change the optimal index code @xcite . these observations point to the importance of cycles on index coding . in the literature , only disjoint cycles and cliques , a specific combination of overlapping cycles , have been exploited so far . more precisely , disjoint cycles in digraphs are exploited by the cycle - cover scheme @xcite and its fractional version @xcite , and disjoint cliques in digraphs are exploited by the clique - cover scheme @xcite and its fractional version @xcite . overlapping cycles can provide more savings than disjoint cycles . we take a clique as an example . in a clique , every vertex forms a cycle with any other vertex , and we see overlapping of cycles at every vertex . if we consider only disjoint cycles in the clique , we get an index code strictly longer than that by considering the clique . however , not all forms of overlapping cycles are useful , in the sense that they provide more savings than considering only disjoint cycles and cliques . in this work , we consider a graph structure based approach , and propose structures of overlapping cycles that can be exploited in graphs to provide potentially more savings than the cycle - cover scheme , the clique - cover scheme , and other existing schemes . the proposed structures are called interlinked - cycle ( @xmath2 ) structures , and they generalize cycles and cliques . furthermore , we define a scheme , called the interlinked - cycle cover ( @xmath3 ) scheme , that constructs index codes based on @xmath2 structures . 1 . we propose a new index - coding scheme ( called the @xmath1 scheme ) that generalizes the clique - cover scheme and the cycle - cover scheme . the new scheme constructs scalar linear index codes . we characterize a class of digraphs ( with infinitely many members ) for which the @xmath1 scheme is optimal ( over all linear and non - linear index codes ) . this means scalar linear index codes are optimal for this class of digraphs . 3 . for a class of digraphs , we prove that the @xmath1 scheme performs at least as well as the partial - clique - cover scheme . we conjecture that the result is valid in general . furthermore , we present a class of digraphs where the additive gap between these two schemes grows linearly with the number of vertices in the digraph . 4 . for a class of digraphs , we prove that the @xmath1 scheme performs at least as well as the fractional - local - chromatic - number scheme . moreover , we present a class of digraphs where the additive gap between these two schemes grows linearly with the number of vertices in the digraph . 5 . we show that the @xmath1 scheme can outperform all of the existing graph - based schemes and the composite - coding scheme in some examples . we extend the @xmath1 scheme to the fractional-@xmath1 scheme . this modified scheme time - shares multiple @xmath0 structures , and constructs vector linear index codes that can be , for certain digraphs , shorter than the scalar linear index codes obtained from the @xmath1 scheme . consider a transmitter that wants to transmit @xmath4 messages @xmath5 to @xmath4 receivers @xmath6 in a _ unicast _ message setting , meaning that each message is requested by only one receiver , and each receiver requests only one message . without loss of generality , let each receiver @xmath7 request message @xmath8 , and possess side information @xmath9 . this problem can be described by a digraph @xmath10 , where @xmath11 , the set of vertices in @xmath12 , represents the @xmath13 receivers . an arc @xmath14 exists from vertex @xmath7 to vertex @xmath15 if and only if receiver @xmath7 has packet @xmath16 ( requested by receiver @xmath17 ) as its side information . the set of the side information of a vertex @xmath7 is @xmath18 , where @xmath19 is the out - neighborhood of @xmath7 in @xmath12 . let @xmath20 and all @xmath21 be ordered sets , where the ordering can be arbitrary but fixed . [ index code ] suppose @xmath22 for all @xmath7 and some integer @xmath23 , i.e. , each message consists of @xmath24 bits . given an index - coding problem modeled by @xmath25 , an index code ( @xmath26,@xmath27 ) is defined as follows : 1 . an encoding function for the source , @xmath28 , which maps @xmath29 to a @xmath30-bit index for some positive integer @xmath30 . 2 . a decoding function @xmath31 for every receiver @xmath7 , @xmath32 , that maps the received index code @xmath33 and its side information @xmath34 to the requested message @xmath8 . [ broadcast rate or index codelength ] the broadcast rate of an index code ( @xmath35 ) is the number of transmitted bits per received message bits at every user , or equivalently the number of transmitted coded symbols ( each of @xmath24 bits ) . this is denoted by @xmath36 , and also referred as the normalized length of the index code . [ optimal broadcast rate ] the optimal broadcast rate for a given index coding problem @xmath25 with @xmath37-bit messages is @xmath38 , and the optimal broadcast rate over all @xmath37 is defined as @xmath39 . [ maximum acyclic induced sub - digraph ] for a digraph @xmath12 , an induced acyclic sub - digraph formed by removing the minimum number of vertices in @xmath25 , is called a maximum acyclic induced sub - digraph ( mais ) . the order of the mais is denoted as @xmath40 . it has been shown that for any digraph @xmath25 and any message length of @xmath24-bits , @xmath40 lower bounds the optimal broadcast rate @xcite , @xmath41 in this sub - section , we describe the clique - cover , the cycle - cover and the partial - clique - cover schemes in detail . these schemes provide some basic intuitions about our proposed @xmath1 scheme . [ clique ] a clique is a sub - digraph where each vertex has an out - going arc to every other vertex in that sub - digraph . [ clique - covering number , @xmath42 the clique - covering number is the minimum number of cliques partitioning a digraph @xmath12 ( over all partitions ) such that if each clique is denoted as @xmath43 for @xmath44 , then @xmath45 , @xmath46 and @xmath47 ( here a vertex is a clique of size one ) . the clique - covering number is equal to the chromatic number of the underlying undirected graph of the complement digraph , i.e. , @xmath48 , where @xmath49 is the underlying undirected graph of @xmath50 , @xmath50 is the complement digraph of @xmath12 , and @xmath51 denotes the chromatic number of @xmath12 . [ clique - cover scheme ] the clique - cover scheme finds a set of disjoint cliques that provides the clique - covering number , and constructs an index code in which the coded symbol for each of the disjoint cliques is the bit - wise xor of messages requested by all of the vertices in that clique . the clique - cover scheme achieves the following rate : the optimal broadcast rate of an index coding instance is upper bounded by the clique cover number , i.e. , @xmath52 . [ path and cycle ] a _ path _ consists a sequence of distinct ( except possibly the first and last ) vertices , say @xmath53 , and an arc @xmath54 for each consecutive pair of vertices @xmath55 for all @xmath56 . we represent a path from the vertex @xmath57 to the vertex @xmath58 in a digraph @xmath12 as @xmath59 . here @xmath60 is the _ first vertex _ and @xmath58 is the _ last vertex_. a path with the same first and last vertices is a _ cycle_. [ cycle - covering number , @xmath61 the difference between the total number of vertices in @xmath12 and the maximum number of disjoint cycles in @xmath12 is the cycle - covering number . [ cycle - cover scheme ] the cycle - cover scheme finds a set of disjoint cycles in @xmath12 that provides the cycle - covering number , and constructs an index code that has ( i ) coded symbols for each disjoint cycle ( for a cycle @xmath62 , a set of coded symbols are @xmath63 ) , and ( ii ) uncoded messages which are requested by those vertices not included in any of the disjoint cycles in @xmath12 . the cycle - cover scheme achieves the following rate : the optimal broadcast rate of an index - coding problem is upper bounded by the cycle - covering number , i.e. , @xmath64 . a sub - digraph @xmath65 is a @xmath66-partial clique , where @xmath67 and @xmath68 is the minimum out - degree of @xmath69.[partialclique1 ] if @xmath70 , for some positive integer @xmath71 , partition a digraph @xmath12 such that @xmath72 and @xmath73 , @xmath74 and @xmath47 , then @xmath75= \sum_{i=1}^{m } ( \kappa(d_i)+1),\ ] ] and the partial - clique number of the digraph is @xmath76,\ ] ] where the minimum is taken over all partitions . the partial - clique - cover scheme finds a set of disjoint partial cliques in @xmath12 that provides the partial - clique number , and constructs an index code that has ( i ) @xmath77 coded symbols for each disjoint @xmath66-partial clique with @xmath78 ( a partial clique uses mds codes to generate coded symbols ) , and ( ii ) an uncoded message for each disjoint @xmath66-partial clique with @xmath79 . the partial - clique - cover scheme achieves the following rate : [ prepbrik ] the optimal broadcast rate of an index coding instance is upper bounded by the partial - clique number , i.e. , @xmath80 the partial - clique - cover scheme performs at least as well as the cycle - cover and the clique - cover schemes , i.e. , @xmath81 this is because the partial - clique - cover scheme includes the cycle - cover scheme or the clique - cover scheme as a special case . by definition , a clique is a @xmath82-partial clique , and a cycle with @xmath83 vertices is a @xmath84-partial clique . despite the fact that the partial - clique - cover scheme uses mds codes , which require sufficiently large message length @xmath37 in general , to construct index codes , one can find mds codes for any cycle and any clique for any @xmath37 . the clique - cover , the cycle - cover and the partial - clique - cover schemes provide scalar linear index codes . we can also construct vector linear index codes by time - sharing all possible cliques , cycles , partial cliques in their respective schemes , and these are called the fractional versions of those schemes . the fractional version can strictly decrease the broadcast rates ( over the non - fractional version ) for some digraphs , e.g. , a 5-cycle @xcite . we present an example that illustrates the importance of overlapping cycles on index coding . consider the digraph @xmath85 in fig . [ figmotivationa ] . in @xmath85 , the cycles @xmath86 and @xmath87 overlap at vertex @xmath57 , and some cycles similarly overlap at vertices @xmath88 and @xmath89 . note that @xmath90 . index codelengths for @xmath85 by existing graph - based schemes ( some schemes require a sufficiently large @xmath37 ) are depicted in table [ table:1 ] . .index codelengths for the digraph @xmath85 in fig . [ figmotivationa ] from existing schemes [ cols="^,^",options="header " , ] [ table:2 ] the index codelength provided by the aforementioned existing schemes are strictly greater than @xmath91 except the composite - coding scheme . there exist digraphs where the @xmath3 scheme outperforms the composite - coding scheme . for an example , the digraph shown in fig . [ 4a ] is a @xmath92-@xmath93 structure , and it is denoted @xmath94 . an index code from the @xmath95 scheme is @xmath96 , which is of length @xmath97 . from theorem [ theorem3 ] , @xmath98 . for @xmath99 , the index codelength provided by the composite - coding scheme is 3.5 , which is greater than @xmath100 . in this section , firstly , we extend the @xmath1 scheme using time - sharing to code over overlapping @xmath0 structures in a digraph , and to obtain vector linear index codes . secondly , we extend the definition of the @xmath0 structure in such a way that we can extend the @xmath1 scheme to code on extended @xmath0 structures . let @xmath101 denote a sub - digraph induced by a subset @xmath102 of the vertices in @xmath103 . we define a function @xmath104 the index codelength from the fractional @xmath1 scheme is represented as @xmath105 , and given by the following linear program : @xmath106,\ s\in s. \nonumber \end{aligned}\ ] ] here @xmath107 is the power set of @xmath108 . in the fractional @xmath1 scheme , each sub - digraph induced by the subset @xmath102 in @xmath109 is assigned a weight @xmath110 $ ] such that the total weight of each message over all of the subsets it belongs to is at least one . in this scheme , @xmath105 is the minimum sum of weights . the @xmath1 scheme is a special case of the fractional @xmath1 scheme where @xmath111 , so @xmath112 . we start with an example that provides an insight to the extension of the @xmath0 structure . consider a digraph @xmath12 that has three cliques @xmath113 , @xmath114 and @xmath115 each of size two . let @xmath116 and @xmath117 be vertex sets of those three cliques respectively . furthermore , for the clique pair ( @xmath118 ) , all vertices in @xmath119 have out - going arcs to all vertices in @xmath120 , and the result follows similarly for clique pairs ( @xmath121 ) and ( @xmath122 ) . this digraph is depicted in fig . [ fig2a ] . one can verify that @xmath12 is a @xmath92-@xmath0 structure with an inner vertex set @xmath123 . we can not get a @xmath124-@xmath0 structure in @xmath12 . suppose that we pick @xmath125 , there is no path from vertices @xmath57 and @xmath88 to vertex @xmath124 without passing through the inner vertex @xmath89 or @xmath92 . by symmetry , choosing any 5 vertices as an inner vertex set will have the same issue . now the @xmath1 scheme gives an index code @xmath126 of length @xmath127 . however , from the coding point of view of an @xmath0 structure , vertices @xmath124 and @xmath128 need not be separated during encoding because they have the same arc sets @xmath129 and @xmath130 , and have arcs to each other . this means we can get another index code by removing @xmath131 , and replacing @xmath132 with @xmath133 , i.e. , @xmath134 of length two . here due to the special connectivity of clique @xmath135 , we have treated it as a single vertex , and used @xmath133 in the code construction . in light of this , we will now extend the definition of an @xmath0 structure to capture cliques with such special configurations . to achieve this we define a term called a super - vertex . [ super - vertex ] in a digraph @xmath12 , let @xmath136 be a vertex set where ( i ) all vertices in @xmath136 have arcs to each other , i.e. , they form a clique , and ( ii ) every vertex @xmath137 has the same @xmath138 and the same @xmath139 . such a group of vertices ( all @xmath137 ) is called a super - vertex and denoted as @xmath140 . now we define extended @xmath0 structures and an index - coding scheme for them . [ extended @xmath0 structure ] the extended @xmath141 ( @xmath142 ) structure is defined as an @xmath0 structure that allows super - vertices in its non - inner vertex set . -15pt [ extended @xmath1 scheme ] for any digraph @xmath12 , the extended @xmath1 ( @xmath143 ) scheme finds a set of disjoint @xmath142 structures covering @xmath12 . it then codes each of these @xmath142 structures using the code construction described in the following : * each super - vertex ( non - inner vertices ) is treated as a single vertex during the construction and the encoding process of the @xmath142 structure . * we consider the message requested by the super - vertex to be the xor of all messages requested by the vertices forming the super - vertex . * each of these @xmath142 structures are treated as an @xmath0 structure , and an index code is constructed using the @xmath144 scheme . along with super - vertices , and taking their definition into account , one can prove the validity of the code constructed by the @xmath143 scheme similar to the proof of proposition [ propositionvalidic ] . denote the length of the index code produced by the @xmath143 scheme by @xmath145 . for a digraph @xmath12 , the index codelength obtained from the @xmath143 scheme is a better upper bound to the optimal broadcast rate than the codelength obtained from the @xmath1 scheme , i.e. , @xmath146 . it follows from the definition of @xmath142 structures that include @xmath0 structures as special cases . graph - based approaches have been shown to be useful for index coding , in which cycles play an important role . prior to this work , disjoint cycles and disjoint cliques ( including the timeshared version ) were used to construct index codes . in this work , we attempted to extend the role of cycles on index coding . we took a step further and showed the benefits of coding on interlinked - cycle structures ( one form of overlapping cycles ) . our proposed scheme generalizes coding on disjoint cycles and disjoint cliques . by identifying a useful interlinked - cycle structure , we were able to characterize a class of infinitely many graphs where scalar linear index codes are optimal . for some classes of digraphs , we proved that the @xmath1 scheme performs at least as well as some existing schemes such as the partial - clique - cover scheme and the fractional - local - chromatic - number scheme . furthermore , for a class of digraphs , we proved that the partial - clique - cover scheme and the @xmath1 scheme have linearly - growing additive gap in index codelength with the number of vertices in the digraphs . we proved a similar result for the fractional - local - chromatic - number scheme and the @xmath1 scheme for another class of digraphs . we extended the @xmath1 scheme , to allow time - sharing over all possible @xmath0 structures in digraphs . we also extended the @xmath0 structure to allow super vertices as its non - inner vertices . however , it remains an open problem to identify cycles overlapping in other useful ways . a @xmath147-@xmath0 structure @xmath148 has some properties captured in the following lemmas , which will be used to prove proposition [ propositionvalidic ] . here we consider @xmath149 and @xmath150 as any two distinct directed rooted trees present in @xmath148 with the root vertices @xmath7 and @xmath17 respectively . [ lemma1 ] for any vertex @xmath151 , and @xmath152 , the set of leaf vertices that fan out from the common vertex @xmath153 in each tree is a subset of @xmath154 . in a tree @xmath149 ( see fig . [ fig3 ] ) , for any vertex @xmath155 and @xmath152 , let @xmath156 be a set of leaf vertices that fan out from vertex @xmath153 . if vertex @xmath157 , then there exists a path from @xmath153 to @xmath17 in @xmath149 . however , in @xmath150 , there is a path from @xmath17 to @xmath153 . thus in the sub - digraph , a path present in any @xmath149 also present in @xmath148 . ] @xmath148 , we obtain a path from @xmath153 to @xmath17 ( via @xmath149 ) and vice versa ( via @xmath150 ) . as a result , a cycle including non - inner vertices and only one inner vertex ( i.e. , @xmath17 ) exists . this cycle is an i - cycle , and condition 1 ( i.e. , no i - cycle ) for @xmath148 is violated . hence , @xmath158 . in other words , @xmath159 similarly , @xmath160 . and @xmath150 with root vertices @xmath7 and @xmath17 respectively , and a non inner vertex @xmath153 in common . here we have used solid arrow to indicate an arc , and dashed arrow to indicate a path.,height=196 ] [ lemma2 ] for any vertex @xmath161 , and @xmath152 , the out - neighborhood of vertex @xmath153 is same in both trees , i.e. , @xmath162 . here the proof is done by contradiction . let us suppose that @xmath163 . for this proof we refer to fig . this proof has two parts . in the first part , we prove that @xmath164 , and then prove that @xmath165 in the second part . ( part 1 ) suppose that @xmath166 . from lemma [ lemma1 ] , @xmath167 is a subset of @xmath168 . now pick a vertex @xmath169 that belongs to @xmath168 such that @xmath170 but @xmath171 ( such @xmath169 exists since we suppose that @xmath166 , and we swap the indices @xmath7 and @xmath17 if @xmath172 ) . in tree @xmath149 , there exists a directed path from vertex @xmath7 , which includes @xmath153 to the leaf vertex @xmath169 . let this path be @xmath173 . similarly , in tree @xmath150 , there exists a directed path from vertex @xmath17 , which does nt include @xmath153 ( since @xmath171 ) , and ends at the leaf vertex @xmath169 . let this path be @xmath174 . in the digraph @xmath148 , we can also obtain a directed path from @xmath17 which passes through @xmath153 ( via @xmath175 ) , and ends at the leaf vertex @xmath169 ( via @xmath149 ) . let this path be @xmath176 . the paths @xmath174 and @xmath177 are different , which indicates the existence of multiple i - paths from @xmath17 to @xmath169 in @xmath148 , this violates the condition 2 for @xmath148 . consequently , @xmath164 . ( part 2 ) now we pick a vertex @xmath178 such that , without loss of generality , @xmath179 but @xmath180 ( such @xmath178 exists since we assumed that @xmath163 , and we swap the indices @xmath7 and @xmath17 if @xmath181 ) . furthermore , we have two cases for @xmath178 , which are ( case 1 ) @xmath182 , and ( case 2 ) @xmath183 . case 1 is addressed in the first part of this proof . on the other hand , for case 2 , we pick a leaf vertex @xmath184 such that there exists a path ( see fig . [ fig3 ] ) that starts from @xmath153 followed by @xmath178 , and ends at @xmath185 , i.e. , @xmath186 exists in @xmath149 . a path @xmath187 exists in @xmath150 . thus a path @xmath188 exists in @xmath148 . from the first part of the proof , we have @xmath164 , so @xmath189 . now in @xmath150 , there exists a path from @xmath17 to @xmath185 , which includes @xmath153 followed by a vertex @xmath190 such that @xmath191 and @xmath192 ( as @xmath193 ) , and the path ends at @xmath185 , i.e. , @xmath194 which is different from @xmath188 . note that in trees @xmath149 and @xmath150 , only the root and the leaf vertices are from @xmath195 , so multiple i - paths are observed at @xmath185 from @xmath17 . this violates condition 2 for @xmath148 . consequently , @xmath196 . [ lemma3 ] if a vertex @xmath197 such that @xmath198 , then its out - neighborhood is the same in @xmath149 and in @xmath148 , i.e. , @xmath199 . for any @xmath197 from lemma [ lemma2 ] , @xmath165 for all @xmath200 . since @xmath201 , vertex @xmath153 must have the same out - neighborhood in @xmath148 as well . [ proof of proposition [ propositionvalidic ] ] from , all @xmath202 which are non - inner vertices , can decode their requested messages . this is because the coded symbol @xmath203 is the bitwise xor of the messages requested by @xmath17 and its all out - neighborhood vertices , and any @xmath17 knows messages requested by all of its out - neighborhood vertices as side information . for an inner vertex @xmath7 , rather than analyzing the sub - digraph @xmath148 , we will analyze its tree @xmath204 , and show that it can decode its message from the relevant symbols in @xmath205 . we are able to consider only the tree @xmath149 due to the lemma [ lemma3 ] . now let us take any tree @xmath149 . assume that it has a height @xmath206 where @xmath207 . the vertices in @xmath149 are at various depths , i.e. , @xmath208 from the root vertex @xmath7 . the root vertex @xmath7 has depth zero , and any vertex at depth equal to the height of the tree is a leaf vertex . firstly , in @xmath149 , we compute the bitwise xor among coded symbols of all non - leaf vertices at depth greater than zero , i.e. , @xmath209 . however , in @xmath149 , the message requested by a non - leaf vertex , say @xmath210 , at a depth strictly greater than one , appears exactly twice in @xmath211 ; 1 . once in @xmath212 , where @xmath213 is parent of @xmath30 in tree @xmath149 , and 2 . once in @xmath214 . refer to for mathematical details . thus they cancel out each other while computing @xmath215 in the tree @xmath149 . hence , in the tree @xmath149 , the resultant expression is the bitwise xor of 1 . messages requested by all non - leaf vertices at depth one , and 2 . messages requested by all leaf vertices at depth strictly greater than one . refer to for mathematical details . secondly , in @xmath149 , we compute @xmath216 ( refer to for mathematical details ) which yields the bitwise xor of 1 . the messages requested by all non - leaf vertices at depth one , which are out - neighbors of @xmath7 , 2 . the messages requested by all leaf vertices at depth one , which are also out - neighbors of @xmath7 , and 3 . the message requested by @xmath7 , i.e. , @xmath8 . this is because the message requested by each leaf vertex at depth strictly greater than one in the tree @xmath149 is present in both the resultant terms of @xmath215 and in @xmath217 , thereby canceling out itself in @xmath216 . hence , @xmath216 yields the bitwise xor of @xmath8 and @xmath218 . as @xmath7 knows all @xmath218 as side - information , any inner vertex @xmath7 can decode its required message from @xmath216 . the mathematical computations of @xmath215 and @xmath219 in the tree @xmath149 are as follows : @xmath220 + where , @xmath221 , and @xmath222 here is obtained because each @xmath223 has only one parent in @xmath149 , and we exclude all @xmath223 whose parent is @xmath7 , and @xmath224 is bitwise xor of messages requested by all of the leaf vertices not in the out - neighborhood of @xmath7 . if we expand @xmath225 as per the group of vertices according to their depth , we get @xmath226 \hskip-2pt \right ] \hskip-3pt \dotsc \hskip-2pt \right ] \hskip-2pt \right ] \nonumber \\ & = \bigoplus \limits_{j\in n^+_{t_i}(i)\setminus v_{\mathrm{i } } } x_{j}. \end{aligned}\ ] ] note that the intermediate terms in @xmath225 cancel out ( we have used the same color to indicate the terms that cancel out each other ) . now substituting @xmath225 of and @xmath227 of in , we get @xmath228 @xmath229 let @xmath256 be a @xmath89-@xmath0 structure having an inner - vertex set @xmath257 . now we prove that any non - inner vertices of @xmath258 belonging to an i - path could not contribute to form a cycle including only non - inner vertices of @xmath256 . to show this , we start by picking two inner vertices @xmath259 and @xmath178 , and the i - path @xmath260 . we assume that the i - path includes @xmath261 non - inner vertices , i.e. , @xmath262 . further , let @xmath263 be denoted by @xmath264 . if @xmath265 , then @xmath266 for @xmath267 . now for @xmath268 , we have the following in @xmath256 : * @xmath269 , @xmath270 and @xmath271 can not contain any vertex in @xmath264 of @xmath260 because of the following : 1 . if @xmath269 or @xmath270 contains any vertex @xmath272 , then @xmath273 ( part of @xmath260 ) and @xmath274 ( part of @xmath269 or @xmath270 ) form an i - cycle at @xmath259 . the existence of the i - cycle in @xmath256 contradicts the definition of an @xmath0 structure . if @xmath271 contains any vertex @xmath272 , then @xmath275 ( part of @xmath271 ) and @xmath276 ( part of @xmath260 ) form an i - cycle at @xmath178 . the existence of the i - cycle in @xmath256 contradicts the definition of an @xmath0 structure . * one can verify that only the remaining i - paths @xmath277 and @xmath278 can contain vertices in @xmath264 without forming an i - cycle in @xmath256 . now for @xmath260 and @xmath277 , these two i - paths must form a directed rooted tree @xmath279 with the root vertex @xmath259 ( by the definition of an @xmath0 structure ) . thus these two i - paths alone could not form a cycle including only non - inner vertices . furthermore , if @xmath260 and @xmath277 contain some common non - inner vertices , then let a vertex @xmath280 , which is in @xmath279 , be the vertex from where @xmath279 branches to @xmath178 and @xmath169 ( refer to fig . [ abc1 ] ) . * if they contain some common non - inner vertices , then let @xmath281 be the first vertex where @xmath260 and @xmath278 meet each other ( refer to fig . [ abc2 ] ) . considering only these two i - paths , a cycle including only non - inner vertices can form only if a part of @xmath278 contributes to form a path from @xmath282 to @xmath280 for some @xmath283 . this path is not possible because multiple i - paths would be created from @xmath259 or @xmath169 to @xmath178 in @xmath256 ( contradiction of the definition of an @xmath0 structure ) . in fact , both the i - paths have the same destination , i.e. , @xmath178 , so there must be only one path from @xmath284 to @xmath178 common in both of them ( to avoid any multiple i - paths ) . for @xmath277 and @xmath278 , if these i - paths contain a non - inner vertex @xmath153 in common , then @xmath285 ( part of @xmath278 ) and @xmath286 ( part of @xmath277 ) forms an i - cycle at @xmath169 ( contradiction to the definition of an @xmath0 structure ) . thus @xmath277 and @xmath278 can not contain any vertex in common except @xmath169 . consequently , considering @xmath260 , @xmath277 and @xmath278 ( along with assumptions that there are some common non - inner vertices in ( i ) @xmath260 and @xmath277 , and ( ii ) @xmath260 and @xmath278 ) , we end up with the structure as shown in fig . [ abc3 ] , where @xmath287 and @xmath288 . this structure contradicts one of the necessary conditions ( let this vertex be @xmath289 ) and an out - going path from a vertex in @xmath264 ( let this vertex be @xmath290 ) , for some @xmath291 , where @xmath292 . one can easily verify this necessary condition . ] for any vertex in @xmath264 to contribute to form a cycle including only non - inner vertices in @xmath256 . thus there is no vertex in @xmath264 to contribute to form a cycle including only non - inner vertices in @xmath256 . due to symmetry , the result ( any non - inner vertices of @xmath258 belonging to @xmath277 could not contribute to form a cycle including only non - inner vertices of @xmath256 ) implies similarly for non - inner vertices belonging to any of the i - paths in @xmath256 . therefore , there is no cycle among the non - inner vertices in @xmath256 . we first prove one lemma that will help to prove the optimality of the @xmath1 scheme . [ proof of theorem [ theorem3 ] ] we will show that the mais lower bound is tight for all @xmath24 . we denote the digraph which is also an @xmath0 structure by @xmath230 , and consider that it has @xmath4 vertices . for @xmath231 , the digraph contains only one vertex , and @xmath232 . for @xmath233 , we have the following : ( case 1 ) from lemma [ lemma6 ] , any cycle must include at least two inner vertices , or no inner vertex , thus if we remove @xmath234 inner vertices , then the digraph @xmath148 becomes acyclic . thus @xmath235 from theorem [ theorem1 ] , we get @xmath236 it follows from , and that @xmath237 . thus @xmath238 . ( case 2 ) a @xmath239 can be viewed in two ways . the first way is considering the whole @xmath148 as a @xmath147-@xmath0 structure . the second way is considering induced sub - digraphs of @xmath239 which consist of 1 . @xmath58 disjoint cycles together consisting of a total of @xmath240 ( @xmath241 ) non - inner vertices ( if @xmath242 or @xmath57 , then @xmath243 , which is case 1 ) , 2 . @xmath244 disjoint @xmath0 structures each with @xmath245 vertices and @xmath246 inner vertices in such a way that @xmath247 , we consider that each @xmath0 structure is also disjoint from all @xmath58 cycles among non - inner vertices , and 3 . total remaining of @xmath248 non - inner vertices ( which are not included in @xmath58 cycles , or the @xmath244 @xmath0 structures ) . now we will show that both ways of looking at @xmath148 are equivalent in the sense of the index codelength generated from our proposed scheme , and both equal to @xmath249 . we prefer the second way of viewing @xmath148 for our proof since it is easier to find the mais lower bound . for the partitioned @xmath148 ( looking at in the second way ) , the total number of coded symbols is the summation of the coded symbols for ( i ) each of the @xmath58 disjoint cycles ( each cycle has saving equal to one ) , ( ii ) each of the @xmath244 disjoint @xmath0 structures ( each of the @xmath0 structures has savings equal to @xmath234 ) , and ( iii ) @xmath250 uncoded symbols for the remaining non - inner vertices , i.e. , @xmath251 from and , @xmath252 , thus from both perspectives the code length is the same . now for @xmath148 ( looking at in our second way ) , if we remove one vertex from each of the @xmath58 cycles among non - inner vertices ( @xmath58 removal in total ) , and remove @xmath253 vertices from each of the @xmath244 @xmath0 structures ( @xmath254 ) , i.e. , total removal of @xmath234 , then the digraph becomes acyclic . thus @xmath255 it follows from , , and that @xmath237 . thus @xmath238 . in this section , firstly , for every minimal partial clique @xmath294 with @xmath295 , we prove that there exists an @xmath0 structure within it such that both of the schemes ( partial - clique - cover and @xmath144 ) provide the same savings . secondly , we conjecture that the result is valid in general ( this is the main reason that results the conjecture [ conj1 ] ) . the summary is depicted in table [ table1 ] . finally , we prove the theorem . [ lemmaa1 ] a minimal partial clique @xmath294 with @xmath296 vertices and the minimum out - degree @xmath297 is a cycle , and both the partial - clique - cover scheme and the @xmath144 scheme achieve savings of one . the properties of @xmath294 with @xmath296 and @xmath297 are as follows : in @xmath294 , ( i ) as @xmath297 , the partial - clique - cover scheme provides the savings of one by proposition [ lemm1 ] , ( ii ) further partitioning could not provide any savings because it is a minimal - partial - clique , and ( iii ) there exists at least a cycle because an acyclic digraph should have a sink vertex ( a vertex with out - degree zero ) , and that does not exist since @xmath297 . thus @xmath294 , which is a minimal partial clique , having all of the above properties is a cycle . from theorem [ theorem2 ] , a cycle is a @xmath88-@xmath0 structure , and the @xmath144 scheme provides savings of one by . note that such @xmath294 with @xmath298 is also a clique of order two . [ lemmaa2 ] a minimal partial clique @xmath294 with @xmath303 vertices and the minimum out - degree @xmath304 , is a clique of order @xmath305 , and @xmath294 has the savings of @xmath306 from both the partial - clique - cover scheme and the @xmath144 scheme . in @xmath294 , if @xmath304 , then each vertex has an out - going arc to every other vertex , which is a clique by definition . thus the partial - clique - cover scheme provides savings of @xmath306 by proposition [ lemm1 ] . from theorem [ theorem2 ] , a clique is a @xmath305-@xmath0 structure , and the @xmath144 scheme provides savings of @xmath306 by . [ lemmaa3 ] in any minimal partial clique @xmath294 with @xmath307 vertices and @xmath308 , there exists a @xmath89-@xmath0 structure , and @xmath294 has the savings of two from both the partial - clique - cover scheme and the @xmath144 scheme . [ def1 ] a path in the forward direction indicates a path from a vertex @xmath7 to any vertex @xmath17 such that @xmath287 , and a path in the reverse direction indicates a path from a vertex @xmath17 to any vertex @xmath7 such that @xmath309 . for simplicity , we refer a _ forward path _ to a path in the forward direction , and a _ reverse path _ to a path in the reverse direction . [ farthest path ] [ def2 ] consider a sequence of vertices labeled in an increasing order such that there are multiple paths from a vertex @xmath7 in the sequence to other vertices in the sequence . among those paths from @xmath7 , the path to the vertex with the largest label is called the farthest path from vertex @xmath7 . for an example , let @xmath310 be a sequence of vertices in an increasing order , and @xmath57 has paths to vertices @xmath311 . the path from from @xmath57 to @xmath312 is the farthest path from @xmath57 . * a minimal partial clique @xmath294 with @xmath307 number of vertices and @xmath308 has at least one cycle by lemma [ lemmaa4 ] . without loss of generality , denote the cycle by @xmath313 , and its vertices by @xmath314 . for simplicity , vertices of @xmath313 are labeled in an increasing order as shown in fig . * for @xmath315 , @xmath316 denotes a path from @xmath259 to @xmath178 including only vertices and arcs of @xmath313 , and @xmath317 denotes a path from @xmath259 to @xmath178 consisting of arcs and vertices outside @xmath313 except vertices @xmath259 and @xmath178 . * for the vertices in @xmath318 ( see fig . [ 2 ] ) , vertex @xmath57 ( the first vertex of @xmath313 ) has only forward paths to other vertices of @xmath319 ( definition [ def1 ] ) , and vertex @xmath30 ( the last vertex of @xmath320 ) has only reverse paths to other vertices of @xmath320 ( definition [ def1 ] ) . let @xmath321 be the first vertex in the vertex sequence that has a reverse path to any vertex in @xmath318 , i.e. , there exist a @xmath322 such that @xmath323 and all vertices in @xmath324 have no reverse path . 1 . there exists at least one cycle , @xmath313 , and 2 . for a vertex @xmath326 , one of its out - going arcs ( beside the arc in @xmath313 ) contribute to form a path that always returns to some vertex in @xmath313 . in @xmath294 with @xmath307 vertices and @xmath308 , there exists at least one cycle , denoted @xmath313 . this is because if any @xmath327 is acyclic , then it must contain at least one sink vertex , and there can not be sink vertices in @xmath294 since @xmath308 . assume that vertices @xmath328 such that there is an arc from @xmath301 to @xmath329 in @xmath313 . now in @xmath294 , @xmath301 has out - degree @xmath330 , and the next out - going arc of @xmath301 ( other than the arc in @xmath313 ) consider the terminal vertex of the path @xmath332 to be @xmath333 such that @xmath334 . vertex @xmath333 can not be a sink vertex because @xmath335 , and it must contribute to form a path . since there are no disjoint cycles ( all cycles are connected , otherwise any two disjoint cycles in @xmath294 provides savings of two and such @xmath327 is not a minimal partial clique with @xmath308 ) , @xmath333 must have a return path to a vertex in @xmath318 . we had considered that vertex @xmath7 of @xmath313 is the first vertex having a reverse path . now if any path @xmath336 meets path @xmath322 at some vertices , then there exists a reverse path from vertex @xmath213 to @xmath17 . this is not possible because such @xmath213 would have been the first vertex in the vertex sequence that has a reverse path ( contradiction ) . if the farthest path from @xmath17 , @xmath340 , meets any path @xmath342 at some vertices , then there exists @xmath345 . this is not possible ; otherwise , path @xmath345 would have been the farthest path ( contradiction ) . if @xmath348 or @xmath349 , then we get a figure - of - eight structure at @xmath17 . two closed paths at @xmath17 will be ( i ) @xmath350 , and ( ii ) @xmath320 . if @xmath351 , then let @xmath352 be the common vertex which is nearest to @xmath17 in these two paths @xmath340 and @xmath346 . now we get a figure - of - eight structure at @xmath17 . two closed paths at @xmath17 will be ( i ) @xmath353 ( part of @xmath340 ) , @xmath354 ( part of @xmath346 ) , and ( ii ) @xmath320 . these closed paths are vertex - disjoint except @xmath17 . [ proof of lemma [ lemmaa3 ] ] the proof is done by the detailed structural analysis of the minimal partial clique @xmath294 . we divide the proof into two parts : in part - i , we prove that @xmath294 has a figure - of - eight structure ( see definition [ fig8 ] ) at a vertex , and in part - ii , we prove there exist a @xmath89-@xmath0 structure within @xmath294 having a figure - of - eight structure . ( part - i ) consider the cycle @xmath313 in @xmath294 has @xmath355 vertices . now based on @xmath313 , there are two cases in @xmath294 , and those are ( i ) @xmath313 with @xmath356 , i.e. , @xmath313 is a clique of size two , and ( ii ) @xmath313 with @xmath357 . we assume @xmath358 and @xmath359 are the two distinct vertices belonging to @xmath318 in such a way that there exist a directed arc from @xmath301 to @xmath153 . in @xmath294 , @xmath330 and the next out - going arc of @xmath301 goes out of @xmath313 , and contributes to form a path , say @xmath332 that returns to some vertex in @xmath318 by lemma [ lemmaa4 ] . ( case ( i ) ) the cycle @xmath313 is a clique of size two . therefore , the path @xmath332 returns to @xmath313 either at @xmath301 or @xmath153 . now we get a figure - of - eight structure at @xmath301 if the path returns to @xmath301 , otherwise it returns to @xmath153 . for the latter case , we get a new cycle that includes the path @xmath332 ( which includes @xmath301 , some vertices other than @xmath153 ) , vertex @xmath153 and the arc from @xmath153 to @xmath301 ( i.e. , arc of the cycle @xmath313 ) . the new cycle has more than two vertices , thus this ends up with case ( ii ) . on the basis where path @xmath332 returns , we have the following sub - cases : @xmath332 returns to ( ii - a ) @xmath301 ( see in fig . [ 1a ] ) providing a figure - of - eight structure , ( ii - b ) @xmath153 ( see in fig . [ 1c ] and [ 1d ] ) , and ( ii - c ) some vertex @xmath360 ( see in fig . [ 1b ] ) . for sub - case ( ii - b ) , we have a vertex @xmath361 such that there is a direct arc from @xmath352 to @xmath153 . since there are no disjoint cycles , the next out - going arc of @xmath352 ( besides the arc from @xmath352 to @xmath153 ) contributes to form a path to a vertex in @xmath362 . if @xmath352 has path to @xmath301 , then we get a figure - of - eight at @xmath301 ( shown in fig . [ 1c ] ) , otherwise we have the following : 1 . @xmath352 has a path to some @xmath360 , so this ends up with sub - case ( ii - c ) ( shown in fig . [ 1b ] ) , or 2 . @xmath352 has a path to @xmath153 , but for this case , we have another path ( shown in fig . [ 1d ] ) from @xmath301 to @xmath153 ( beside the direct arc from @xmath301 to @xmath153 , and path including the direct arc from @xmath352 to @xmath153 ) , and in this path , we can repeat sub - case ( ii - b ) by considering the predecessor of @xmath153 in place of @xmath363 . for sub - case ( ii - c ) , note that we have a path @xmath332 , which is vertex - disjoint from @xmath313 except the first and the last vertices , which starts from any vertex @xmath364 and returns to some vertex @xmath365 . we start analyzing the sub - case ( ii - c ) in @xmath294 considering vertices in @xmath318 . for the vertices in @xmath318 ( see fig . [ 2 ] ) , in a sequential order starting from vertex @xmath57 , we track their out - going paths ( which may include vertices not in @xmath366 ) to other vertices in @xmath318 . for this sub - case , we know that there exist a @xmath322 such that @xmath323 and all vertices in @xmath324 have only forward paths . now we consider out - going paths from @xmath17 ( the paths are always forward paths ) , and get the following subsub - cases : 1 . if the farthest path from @xmath17 is @xmath367 for some @xmath368 , then there exists a figure - of - eight at @xmath17 . two closed paths at @xmath17 will be ( i ) @xmath369 , @xmath322 , and ( ii ) @xmath367 , @xmath370 ( refer fig . if @xmath367 and @xmath322 are vertex - disjoint except @xmath17 , then by recalling the definition of @xmath371 and @xmath317 for any @xmath315 , one can show that the two closed paths are vertex - disjoint except @xmath17 . otherwise , @xmath367 and @xmath322 are not vertex - disjoint except @xmath17 , and by lemma [ prop4 ] , one can find a figure - of - eight at @xmath17 . if the farthest path from @xmath17 is @xmath367 for @xmath372 , then there exists a figure - of - eight at @xmath17 . two closed paths at @xmath17 will be ( i ) @xmath373 , @xmath322 , and ( ii ) @xmath374 , @xmath375 for @xmath376 ( one of the forward paths of @xmath377 ) , @xmath378 ( refer fig . [ 2b ] ) . using lemmas [ prop2 ] and [ prop3 ] , and recalling the definition of @xmath371 and @xmath317 for any @xmath315 , one can show that these two closed paths are vertex - disjoint except @xmath17 . 3 . otherwise , the farthest path from @xmath17 is @xmath340 for some @xmath379 . starting from @xmath17 , we have at least two forward paths to some vertices in @xmath318 such that these paths are vertex - disjoint except @xmath17 . we assume these paths are path - a ( the path from @xmath17 to @xmath380 ) and path - b ( the path from @xmath17 to @xmath213 ) . paths @xmath381 and @xmath367 will be path - a and path - b respectively ( refer fig . [ 3 ] ) . now considering the out - going paths from @xmath380 into account ( the paths are always forward paths ) , we get the following subsubsub - cases : 1 . if the farthest path from @xmath380 is @xmath382 for some @xmath376 , then there exists a figure - of - eight at @xmath17 . two closed paths at @xmath17 will be ( i ) path - a , @xmath382 , @xmath378 , and ( ii ) path - b , @xmath383 , @xmath322 ( refer fig . [ 3a ] ) . using lemmas [ prop2 ] and [ prop3 ] , and recalling the definition of @xmath371 and @xmath317 for any @xmath315 , one can show that these two closed paths are vertex - disjoint except @xmath17 . if the farthest path from @xmath380 is @xmath382 for @xmath384 , then there exists a figure - of - eight at @xmath17 . two closed paths at @xmath17 are ( i ) path - a , @xmath385 , @xmath322 , and ( ii ) path - b , @xmath386 , @xmath387 for @xmath388 ( one of the forward paths of @xmath377 ) , @xmath389 ( refer fig . [ 3b ] ) . if @xmath367 and @xmath322 are vertex - disjoint except @xmath17 , then by using lemmas [ prop2 ] and [ prop3 ] , and recalling the definition of @xmath371 and @xmath317 for any @xmath315 , one can show that the two closed paths are vertex - disjoint except @xmath17 . otherwise , @xmath367 and @xmath322 are not vertex - disjoint except @xmath17 , and by lemma [ prop4 ] , one can find a figure - of - eight at @xmath17 . 3 . otherwise , the farthest path from @xmath380 is @xmath382 for some @xmath390 . the union of path - b and @xmath391 give new path - a and we update @xmath392 . similarly , the union of path - a and @xmath382 give new path - b and we update @xmath393 ( refer fig . [ 3c ] ) . considering new path - a , new path - b , and updating @xmath394 , we repeat the subsubsub - cases of subsub - case 3 . during this iteration , if we get subsubsub - case ( a ) or ( b ) , then there is a figure - of - eight . otherwise , we have subsubsub - case ( c ) , where @xmath213 strictly increase to a value up to @xmath377 . however , when @xmath213 reaches @xmath377 , we must have either subsubsub - case ( a ) or ( b ) . now for the figure - of - eight structure at @xmath301 in @xmath294 ( see fig . [ fig4 ] ) , consider a vertex @xmath153 in a cycle @xmath299 ( indicated in blue color ) , and a vertex @xmath395 in another cycle @xmath300 ( indicated in red color ) in such a way that both @xmath153 and @xmath395 have direct arcs going to @xmath301 . one of the out - going arcs of @xmath153 must contribute to form a path , say @xmath332 , returning to a vertex in @xmath396 . this is because all other cases are not possible ; * @xmath332 can not return to any vertex in @xmath397 , otherwise a disjoint cycle to @xmath300 will be created . * @xmath332 can return to @xmath301 , but for this case , we have another path from @xmath153 to @xmath301 ( beside path including direct arc from @xmath153 to @xmath301 ) , and in this path , we can repeat the case by considering the predecessor of @xmath301 in place of @xmath153 . thus this case ends up with the same consideration as of the cycle @xmath299 with the vertex @xmath153 having direct arc to the vertex @xmath301 . * @xmath332 must return to some vertex in @xmath398 and @xmath399 due to lemma [ lemmaa4 ] . now in a similarly way , one of the out - going arcs of @xmath395 must also contribute a path to a vertex in @xmath398 other than @xmath301 . rearrange @xmath294 to get the structure in fig . now consider @xmath400 . we can see that any vertex in @xmath401 has only one i - path each to other two vertices in @xmath401 with no i - cycles . thus a @xmath89-@xmath0 exists in @xmath294 ( for an example , see fig . [ 123 ] ) . consecutively , the @xmath144 scheme provides savings of two by . again , for @xmath294 , the partial - clique - cover scheme provides savings of two by proposition [ lemm1 ] . for all minimal partial cliques that we have analyzed , there exists an ( @xmath402)-@xmath141 structure within each minimal partial clique having the minimum out - degree @xmath403 . we conjecture that this holds in general ( the following conjecture is not a part of the proof of the theorem [ theorem4 ] , but this provides an intuition about the conjecture [ conj1 ] ) . [ proof of theorem [ theorem4 ] ] given a digraph @xmath12 , if minimal partial cliques , partitioning the digraph to provide partial - clique number , have @xmath404 , then using the @xmath1 scheme on each of the minimal partial cliques , we achieve @xmath405 ( by using lemmas [ lemmaa1 ] , [ lemmaa2 ] and [ lemmaa3 ] , and considering zero savings for @xmath406 from any schemes ) . the @xmath1 scheme may produce a shorter index code by considering a different partitioning of the digraph ( for an example , see fig . [ ex1b ] and [ ex1a ] ) . [ eg3 ] this example illustrates that for the class of digraphs stated in theorem [ theorem4 ] , the @xmath1 scheme performs as least as well as the partial - clique - cover scheme . consider two digraphs that are depicted in fig . the digraph in fig . [ ex1b ] has more savings from the @xmath1 scheme than that obtained from the partial - clique - cover scheme , and the digraph in fig . [ ex1a ] has equal savings from both schemes . [ prop6 ] for any digraph @xmath12 of the class a , the index codelength obtained from the partial - clique - cover scheme and the @xmath1 scheme are @xmath147 and @xmath411 respectively , i.e. , @xmath412 and @xmath413 . if any minimal partial clique @xmath415 in @xmath12 of the class a includes a vertex @xmath416 for any @xmath417 , then @xmath418 . this is because the out - degree of any vertex @xmath416 in @xmath12 is two , i.e. , @xmath419 by construction . now any minimal partial clique without any vertex in @xmath420 has @xmath421 . this is because for any vertex in @xmath422 , only one out - neighbor is in this set , and the rest are in @xmath420 . so , any minimal partial clique in the digraph can have @xmath423 . now by proposition [ lemm1 ] , we know that @xmath424 . in @xmath12 , we try to construct a minimal partial clique @xmath294 that has @xmath425 ( we do not need to consider @xmath294 having @xmath426 because of lemma [ lemma10 ] ) by starting from @xmath294 with only one vertex and then adding vertices into its vertex set in such a way that we can obtain @xmath425 . we start from any vertex @xmath416 for an @xmath417 ( we will show a similar result if we start from some vertex @xmath7 ) . let @xmath427 be the vertex set of @xmath327 . now we include both of the two out - neighbor vertices of @xmath416 , i.e. , vertices in @xmath428 in @xmath429 . if we include only one out - neighbor vertex of @xmath416 in @xmath429 , then @xmath430 resulting @xmath431 . the new vertex - induced sub - digraph @xmath294 has @xmath432 , and @xmath431 because @xmath433 . to get a minimum out - degree of two , we must include another vertex in @xmath434 from @xmath435 . by symmetry , it does not make any difference which vertex to add . we arbitrarily choose @xmath436 , for some @xmath437 . now the new vertex - induced sub - digraph @xmath294 has @xmath438 for a @xmath17 . here @xmath439 because @xmath440 . we further include all vertices in @xmath441 in @xmath442 because @xmath443 , and if we include only one of its out - neighbor vertices , then @xmath431 . the new @xmath294 has @xmath444 and @xmath445 . further , including any vertex set in @xmath446 could not increase @xmath447 because any @xmath294 must have @xmath448 ( lemma [ lemma10 ] ) . if we start building @xmath294 with some @xmath449 , we will end up with a sub - digraph ( i ) that includes @xmath450 for some @xmath451 where @xmath452 , or ( ii ) that includes @xmath453 for some @xmath451 where @xmath454 and @xmath455 . altogether , any minimal partial clique with @xmath308 must contain @xmath456 for some @xmath451 where @xmath454 . however , @xmath294 is not the minimal because by simply considering two partial cliques among the vertices in @xmath294 , i.e. , partial cliques with vertex sets @xmath457 and @xmath458 , we get savings of two ( one in each ) . for a minimal partial clique @xmath294 in @xmath12 of the class a , lemma [ lemma10 ] provides @xmath423 , and lemma [ lemma11 ] proves that there exists no minimal partial clique with @xmath425 . thus the minimal partial cliques in a digraph @xmath12 of the class a are only cycles . for this case , the cycle - cover scheme and the partial - clique - cover scheme for @xmath12 are the same . since any cycle must include two vertices from @xmath422 , @xmath459 . the minimal partial cliques with vertex sets @xmath460 provide @xmath461 . consider a vertex set @xmath462 with @xmath147 number of vertices . in the digraph @xmath12 , any vertex @xmath463 has only one path each to all other vertices in @xmath464 such that only the first and the last vertices of each of these paths belong to @xmath465 ( i.e. , i - path ) . moreover , there is no i - cycle at any vertex in @xmath465 . thus this forms an @xmath0 structure with inner - vertex set @xmath465 . now from , @xmath466 . thus @xmath467 . a. blasiak , r. d. kleinberg , and e. lubetzky , `` broadcasting with side information : bounding and approximating the broadcast rate , '' _ ieee trans . inf . theory _ , 59 , no . 9 , pp . 58115823 , sept . 2013 . s. e. rouayheb , a. sprintson , and c. georghiades , `` on the index coding problem and its relation to network coding and matroid theory , '' _ ieee trans . inf . theory _ , 56 , no . 3187 3195 , jul .

we investigate shear thickening and jamming within the framework of a family of spatially homogeneous , scalar rheological models . these are based on the ` soft glassy rheology ' model of sollich _ et al . _ [ phys . rev . lett . * 78 * , 2020 ( 1997 ) ] , but with an effective temperature that is a decreasing function of either the global stress or the local strain . for appropiate , it is shown that the flow curves include a region of negative slope , around which the stress exhibits hysteresis under a cyclically varying imposed strain rate . a subclass of these have flow curves that touch the axis for a finite range of stresses ; imposing a stress from this range _ jams _ the system , in the sense that the strain creeps only logarithmically with time , . these same systems may produce a finite asymptotic yield stress under an imposed strain , in a manner that depends on the entire stress history of the sample , a phenomenon we refer to as _ history dependent jamming_. in contrast , when the flow curves are always monotonic , but we show that some generate an oscillatory strain response for a range of steady imposed stresses . similar spontaneous oscillations are observed in a simplified model with fewer degrees of freedom . we discuss this result in relation to the temporal instabilities observed in rheological experiments and stick slip behaviour found in other contexts , and comment on the possible relationship with ` delay differential equations ' that are known to produce oscillations and chaos . 2

cond-mat0109011

a wide variety of materials can be driven into a non equilibrium state that is either solid like and static , or fluid like but only relaxes on time scales that far exceed the experimental time frame , if at all . such states are often referred to as ` jammed , ' and are realisable in molecular liquids having undergone a rapid quench to low temperatures , or colloids at a high volume fraction , to cite just two examples @xcite . it has recently been postulated by liu and nagel that the nature of the jammed state may be independent of the manner in which it was formed @xcite . for example , a stress induced jamming transition may produce a qualitatively similar state to a temperature induced transition . this was expressed in the form of a ` jamming phase diagram , ' in which jammed configurations occupy a compact region near to the origin of a phase space comprised of three axes : the temperature @xmath12 , the volume @xmath13 and the load @xmath1 @xcite . in its simplest form , the liu nagel jamming phase diagram suggests that increasing the applied load can only increase the likelihood of flow . however , it is equally feasible for an applied load to induce a jammed state . for instance , if a pile of sand is formed and then gravity is switched off , the system unjams without any significant variation in volume . thus the the load ( here controlled by gravity ) jams the system . this concurs with the earlier claim that jammed systems may be classified as ` fragile ' that is , they can support only certain , _ compatible _ loads , and will rearrange or flow under an incompatible load @xcite . furthermore , one could also envisage a class of loadings that can alternately induce and destroy a jammed state as the magnitude of the load increases , a situation that could be referred to as ` re - entrant jamming . ' the purpose of this paper is to investigate the nature of transitions to or from jammed configurations that have , as their control parameter , the imposed shear stress @xmath1 . we do this by providing concrete examples of models that exhibit jamming transitions with @xmath1 , including instances of the re - entrant jamming scenario described above . these models are based on the ` soft glassy rheology ' ( sgr ) model of sollich _ et al . _ @xcite , which was originally devised to highlight the possible existence of glassy relaxation in a range of soft materials , such as foams , emulsions , pastes _ etc . _ it is parameterised by an effective temperature @xmath0 , which in in the context of soft matter is thought to represent some form of mechanical noise , but may refer to the true thermal temperature in other applications . as it was originally defined , the sgr model only exhibits shear thinning , which is clearly unsuitable for our purposes . therefore we consider variants of the model in which the effective temperature @xmath0 is no longer constant , but can vary with the state of the system . more precisely , @xmath0 is treated a function of both the global stress @xmath1 and the local strain @xmath2 , _ i.e. _ @xmath14 . by choosing suitable forms of @xmath15 , systems can be constructed that become ` colder ' as they become more stressed , which allows for the system to shear thicken and even ` jam . ' for clarity , we restrict our attention to two limiting forms of @xmath15 , namely @xmath5 and @xmath11 . in the former case , certain classes of @xmath5 are shown to exhibit a flow curve ( _ i.e. _ the curve of the stress @xmath1 versus the strain rate @xmath4 under conditions of steady flow ) that is non monotonic , and can touch the @xmath6 axis for range of finite values of @xmath1 . thus a ` jammed ' state with @xmath16 can be reached for some @xmath1 but not for others . furthermore , it is also shown that , for systems driven by an imposed @xmath17 , whether or not a jammed configuration is approached at late times depends on the entire strain history of the system , and not just the behaviour of @xmath17 as @xmath18 . we refer to this phenomenon as ` history dependent jamming . ' the behaviour of systems with @xmath10 is somewhat different , but no less remarkable . when driven by an imposed @xmath1 , certain forms of @xmath11 exhibit a regime in which the viscosity never reaches its steady flow value but oscillates in time , with a waveform that is approximately sinusoidal near to the transition point , but becomes increasingly sharp deep into the oscillatory regime . the models considered here are spatially homogeneous , and so this oscillation is purely temporal , having no spatial component . it is tempting to associate these oscillations with the stick slip behaviour observed in granular systems and plate tectonics @xcite , but we shall give reasons why we believe that the underlying physics might be rather different . we can not rule out the possibility of more complex oscillatory behaviour , maybe even chaotic , arising in as of yet unobserved regions of parameter space . the finding of a bifurcation to oscillatory behaviour for @xmath10 is all the more remarkable because the flow curve ( as already defined ) is everywhere monotonic increasing for this class of @xmath0 . by contrast , instances of rheological instabilities that have been found experimentally have tended to occur for ranges of parameters in which the flow curve has a negative gradient @xcite . this suggests that mechanism behind the instability observed here may be qualitatively different to those cited above , and may be realisable in some range of materials that has yet to be identified . this paper is arranged as follows . in sec . [ s : general ] the class of models to be studied is defined , with particular attention being paid to those aspects that differ from the sgr model . the known results of the sgr model that are relevant to the subsequent discussion are then briefly summarised in sec . [ s : sgr ] . systems with @xmath3 are described in sec . [ s : global ] , where it explained how the flow curves can be graphically interpreted as mappings from the sgr flow curves . dependent behaviour of the system has been found by numerical integration of the governing master equation . an example of history dependent jamming is presented , and explained in terms of the stability of the steady flow solutions . in sec . [ s : local ] we turn to consider the case @xmath10 , and analytically prove that the flow curve is everywhere monotonic . nonetheless , the simulations results presented here clearly show that @xmath19 can oscillate in time for a range of imposed stresses @xmath1 . a qualitative explanation of the emergence of the oscillatory phase is also given , in terms of the model s internal degrees of freedom . the intensive nature of the simulations has meant that only a small range of functional forms for @xmath11 have been investigated , and hence it has not been possible to fully characterise this range of models . therefore we consider a simplified model in sec . [ s : monodisperse ] , for which the simulation times are significantly shorter and a more complete picture has been realised . some results of this model have also been presented elsewhere @xcite . this reduced model clearly shows that the mean value of @xmath4 during the oscillatory motion deviates from the steady flow value by as much as an order of magnitude . finally , in sec . [ s : disc ] we discuss some outstanding issues raised by this work , before summarising our results in sec . [ s : summ ] . all of the models studied in this paper represent generalisations of the sgr model of sollich _ et al . _ it is therefore prudent to first describe those features that are common to this class of models , before considering each of the different generalisations in turn . this is the purpose of the current section . since the various assumptions that lie behind the construction of the sgr model have already been discussed at length elsewhere , we shall here give just a brief overview of the derivation and refer the reader to @xcite for more detailed physical arguments . only those aspects that differ from the sgr model will be discussed in full . our goal is to construct deliberately simplified models that exhibit shear thickening and jamming , but whose interpretation is more transparent than that of a detailed microscopic model . in this spirit , the models in this class all share a number of simplifications . only a single shear component of the strain and stress tensors are considered , which will be denoted by @xmath7 and @xmath1 respectively . this is therefore a class of _ scalar _ models . it is further assumed that the system can be coarse grained into a collection of mesoscopic subsystems , each of which are fully described by two scalar variables , namely a local strain @xmath2 and a stability parameter @xmath20 . these mesoscopic subsystems will be referred to as _ elements_. we simply assume here that such a coarse graining is possible , notwithstanding the significant practical challenges in constructing a suitable scheme for any given microscopic model . at any given instant , each element has a probability of yielding per unit time that is denoted by @xmath21 . when an element yields , it is assumed that its microscopic constituents are rearranged to such a degree that it loses all memory of its former configuration . its strain returns to zero , and it is assigned a new value of @xmath20 that is drawn from a _ prior _ distribution @xmath22 . suitable functional forms for @xmath23 and @xmath22 will be discussed below . the value of @xmath20 remains fixed until the element yields again , but @xmath2 follows the global strain @xmath7 according to @xmath24 . thus the elements are affinely deformed by the flow field , which is assumed to be homogeneous . let us define @xmath25 to be the probability density function of elements which have a local strain @xmath2 and a barrier @xmath20 at time @xmath8 . then @xmath25 evolves in time according to two distinct mechanisms : the homogeneous shearing at a rate @xmath4 , and the yielding of elements at a rate of @xmath23 per unit time . thus the master equation for @xmath25 is @xmath26 the second term on the left hand side represents the increase in local strains @xmath2 according to the uniform global strain rate @xmath4 . the right hand side describes the yielding of elements . the first term , which is negative , accounts for the loss of elements as they yield at a rate @xmath23 . conversely , the second term represents these same elements _ after _ they have yielded , which have a strain @xmath27 and a value of @xmath20 drawn from @xmath22 . the total rate of yielding @xmath28 is defined by @xmath29 here we have introduced the notation that the angled brackets ` @xmath30 ' represents the instantaneous average of the given function over @xmath25 . this will be used frequently in what follows . the master equation ( [ e : master ] ) only describes the evolution of the strain degrees of freedom . to characterise the rheological response of the system , some relation must be found between the local strains @xmath2 and the global stress @xmath1 . this involves two further assumptions . firstly , the elements are supposed to behave elastically between yield events , so that the local stress is just @xmath31 , where @xmath32 is a uniform elastic constant . secondly , @xmath1 is the arithmetic mean of the local stresses , or @xmath33 . other averaging procedures could be employed @xcite , but we focus here on the simplest non trivial option . although the local stress strain relationship is elastic , the global stress also incorporates the yielding of elements and , as will be seen below , the @xmath1@xmath7 relationship is not a linear one . we have been unable to find an analytic solution to the master equation ( [ e : master ] ) for any non trivial @xmath23 . instead it has been numerically integrated using the procedure summarised in appendix [ s : numerics ] . however , it will sometimes be necessary to refer to the steady state solution for a constant @xmath34 , which can be found exactly , @xmath35\quad . \label{e : st3}\ ] ] this is derived by setting @xmath36 in ( [ e : master ] ) and integrating the resulting first order ordinary differential equation with respect to @xmath2 . the asymptotic yield rate @xmath37 can be found by requiring that the integral of @xmath38 is unity . in the sgr model , the yield rate @xmath23 was assigned a functional form similar to that of an activated process @xcite . this was based on the idea that each element can be represented by a single particle moving on a free energy landscape , which remains confined within a well of depth @xmath20 until random fluctuations allow it to cross over this barrier into a different well , with a new barrier @xmath39 . thus yielding can be identified with barrier crossing . this description is similar to that of activated processes in thermodynamic systems ; however , the random fluctuations here are _ not necessarily _ due to the true thermodynamic temperature . they may arise rather from a form of homogeneous mechanical noise generated by non linear couplings between the elements ; see @xcite for a fuller discussion on this point . to avoid possible confusion , this effective temperature is denoted by the symbol @xmath0 rather than @xmath12 . although the energy barrier of an element is initially @xmath20 , as it becomes strained it will gain an elastic energy of @xmath40 and thus will have a smaller effective energy barrier @xmath41 . thus the yield rate will take the form @xmath42\quad , \label{e : gamma}\ ] ] where the attempt rate @xmath43 sets the time scale of the yielding . the effective temperature @xmath0 is constant in the sgr model , and essentially acts as a parameter of the model . however , since @xmath0 may be generated in part by internal couplings between the elements , it should be allowed to vary with the state of the system , _ i.e. _ @xmath44 . in this paper , we shall not consider the most general possible form for @xmath0 , but shall instead focus on a more restricted class for which @xmath14 . this corresponds to the realisation that , as an element is strained , it may become more or less susceptible to the noise , and hence its ` temperature ' @xmath0 may change . allowing @xmath0 to also depend on @xmath1 reflects that this change in susceptibility to noise may in part be a _ phenomenon , _ i.e. _ the @xmath0 of a given element may depend on the state of all of the elements around it . clearly , deriving the actual @xmath15 for any given material would be a highly complicated task , of comparable difficulty to the original coarse graining procedure described above . for further simplicity , we shall not consider @xmath14 but shall instead focus on two limiting cases : @xmath3 is described in sec . [ s : global ] , and @xmath10 is assumed in secs . [ s : local ] and [ s : monodisperse ] . the relevant results for the sgr model , which corresponds to the case of @xmath45(constant ) , are also summarised in sec . [ s : sgr ] . we have yet to specify @xmath22 , the distribution of energy barriers @xmath20 for elements that have just yielded . in the sgr model , @xmath22 is assumed to have an exponential tail , @xmath46 as @xmath47 , which gives rise to a finite yield stress and diverging viscosity for some values of @xmath0 , but not others @xcite . although it is possible to justify the occurrence of an exponential tail in some contexts @xcite , we prefer instead to treat this choice of @xmath22 , combined with the arrhenius form of @xmath23 ( [ e : gamma ] ) , as a recipe for generating a yield stress within this simple picture of activated yielding . this may seem _ ad hoc _ , but it should be realised that jamming is in reality a many body effect , involving collective behaviour between a large number of degrees of freedom . it should therefore come as no surprise that assumptions are required to describe jamming within this single particle picture . for much of this paper , we shall assume that @xmath22 has an exponential tail , just as in the sgr model . in fact , for the numerics the definite form @xmath48 has been used , although it is not expected that the precise shape of @xmath22 for small @xmath20 will alter the long time or steady state behaviour of the system . this is because small values of @xmath20 correspond to elements with short expected lifetimes . furthermore , it turns out that the oscillatory motion described in sec . [ s : local ] does _ not _ depend on the choice of @xmath22 , and in sec . [ s : monodisperse ] the simpler form @xmath49 is used , with qualitatively similar results . there remain three constants in the problem . these are the elastic constant @xmath50 , the attempt rate @xmath43 in ( [ e : gamma ] ) , and the constant @xmath51 in the definition of @xmath22 given above . however , these can all be scaled out of the model . for instance , @xmath50 sets the scale of the stress , and therefore can be removed by rescaling @xmath1 to @xmath52 . similarly , @xmath43 gives the only intrinsic time scale in the system , and can be scaled out through @xmath8 and @xmath4 . finally , @xmath51 sets the scale of the energy barriers , and can be scaled out through @xmath15 . in what follows , we have adopted _ natural _ units in which @xmath53 . this means that the actual values for @xmath8 , @xmath7 and @xmath1 given below should not be compared to experimental values without first rescaling with the appropriate @xmath50 , @xmath51 and @xmath43 for the material in question . for example , the typical yield strain according to the @xmath23 given in ( [ e : gamma ] ) is @xmath54 , which is when the effective energy barrier vanishes . in natural units , this is of @xmath55 ; however , for soft materials it will typically be of only a few percent , and will be even smaller for hard materials . thus the scale of the local strains @xmath2 , and therefore the scale of the global strain @xmath17 , will generally be significantly smaller in real materials than with natural units . as discussed in the previous section , the standard sgr model is realised when ( using natural units ) the yield rate @xmath23 is chosen to take an arrhenius form with an energy barrier @xmath56 and a constant effective temperature @xmath0 , and @xmath22 has an exponential tail @xmath57 . many results for this case are already known @xcite . the purpose of this section is to briefly describe those results that will be referred to in later sections . let us assume that the system is driven by a given strain @xmath17 , and that @xmath58 as @xmath59 , where @xmath34 . without loss of generality we take @xmath60 hereafter . under these conditions , the system will reach the steady state solution @xmath38 already given in ( [ e : st3 ] ) . the asymptotic stress @xmath1 is found by averaging the local strain @xmath2 over @xmath38 , _ i.e. _ @xmath61 , which defines the flow curve @xmath62 . example flow curves for different values of @xmath0 are shown in fig . [ f : sgr_flow_curves ] . in all cases the gradient @xmath63 decreases with increasing @xmath4 , indicating that the apparent viscosity @xmath64 decreases with @xmath4 and that the system is everywhere shear thinning . it should also be noted that there is a a one to one correspondence between @xmath4 and @xmath1 . this means that every point on the curve can be reached in one of two ways : either by applying a known strain @xmath58 , as described above , or by imposing a constant stress @xmath1 . the uniqueness of the steady state solution ensures that the same @xmath38 will be reached in both cases . of course , this assumes that the steady state is reached at all , for which it must both exist and be stable . for the standard sgr model , the steady state solution is stable as long as it exists @xcite , but this does not hold for all @xmath15 under an imposed stress , as will be discussed in later sections . the behaviour of @xmath1 as @xmath65 depends upon the choice of @xmath0 @xcite . for instance , for @xmath66 the stress scales as @xmath67 and the system is newtonian , whereas there is a power law fluid regime @xmath68 for @xmath69 . however , the most important regime for our purposes is @xmath70 , for which @xmath1 approaches a finite value . the yield stress @xmath71 , defined by @xmath72 smoothly tends to zero as @xmath73 and remains zero for all @xmath74 . if a system with @xmath70 has a stress @xmath75 applied to it , then it will never reach a steady state with a finite @xmath4 , simply because @xmath38 for these parameters does not exist . instead the strain will logarithmically creep according to @xmath9 @xcite . there may be a crossover to a different behaviour at late times if @xmath1 is close to @xmath71 , but it must always be true that @xmath76 . the logarithmic creep in @xmath17 under an imposed stress @xmath75 is an important result . if it is realised in an experimental situation , then when the strain rate @xmath77 drops below the resolution of the apparatus , which it must do at long times , it might be erroneously deduced that the system has stopped flowing altogether , _ i.e. _ @xmath78 . this is in accord with our intuitive notions about jamming : the sample initially flows when pushed , but later stops flowing , or ` jams . ' one might argue that , technically , a jammed system should have @xmath6 exactly , but it is not possible for any model in this class to sustain a finite stress without a finite @xmath4 ( unless the sample is allowed to age for an arbitrarily long period of time before being sheared ; see @xcite ) . this is because each individual element has a finite relaxation time until it yields and releases its stress , which can only be balanced by an increasing strain . with this insight , we now _ define _ a ` jammed ' state for this class of systems to be a configuration which has a finite limiting stress @xmath71 when it is driven by an arbitrarily small strain rate @xmath16 . for the sgr model , this corresponds to @xmath70 ; however , for @xmath3 it may also depend on the history of the sample , as will now be discussed . it is useful to recall the physical picture that underlies an @xmath3 . as the system becomes sheared , either by an imposed strain @xmath7 or an imposed stress @xmath1 , it will become distorted , and this may affect its susceptibility to noise . the precise manner in which this happens will depend on the microscopic composition of the material in question . for the systems we are interested in , _ i.e. _ those that can shear thicken or jam , the distorted configuration is _ less _ susceptible to noise that the undistorted one . one way to incorporate this effect is to regard the system as becoming ` colder ' as its shear stress increases , which corresponds to a decreasing @xmath5 . thus setting @xmath3 provides a mechanism for allowing the effective temperature @xmath0 to evolve with the global state of the system . in this section we shall consider @xmath5 that takes values greater than 1 for some ranges of @xmath1 , and less than 1 for others . according to the results of the previous section , this should allow the system to change from a jammed to a flowing state in response to the driving , or more precisely , in response to changes in @xmath79 . thus we may be able to observe a shear induced jamming transition ( something that is not possible in the sgr model , for which the existence of a yield stress depends purely on the choice of the external parameter @xmath0 ) . for clarity , we shall restrict our attention to the simplest choice of @xmath5 that has the potential to shear thicken and jam , namely @xmath80 for small @xmath1 and @xmath81 for large @xmath1 , with a monotonic ( possibly discontinuous ) behaviour for intermediate @xmath1 . although allowing @xmath0 to vary in time according to @xmath82 $ ] can complicate the transient behaviour of the system , as described below , the steady state is easy to analyse . this is because the very definition of a steady state means that @xmath1 asymptotically approaches a constant value , and thus @xmath0 also becomes constant . there is therefore a straightforward procedure for generating the @xmath3 flow curves from those of constant @xmath0 : for any given value of @xmath1 , one simply finds the sgr flow curve for the corresponding value of @xmath5 and reads off the required value of @xmath4 . this amounts to interpolating between the various sgr flow curves . some examples are given in fig . [ f : jamming_curves ] for various @xmath5 that change from 1.5 for small @xmath1 to 0.5 for large @xmath1 . if @xmath5 takes values greater than 1 for small @xmath1 , but smaller than 1 for larger @xmath1 , then it may jam under an applied stress . however , for this situation to be realised , the applied stress must be simultaneously large enough that @xmath81 , and also small enough that @xmath75 . since the yield stress @xmath71 also depends on @xmath0 , the precise criterion for the upper yield stress to be attainable is that there is a range of @xmath1 for which @xmath83\quad . \label{e : criterion}\ ] ] this corresponds to the region in which the @xmath5 curve falls below the @xmath84 curve when they are both plotted on the same axes , as in fig . [ f : roots ] . for example , of the flow curves already presented in fig . [ f : jamming_curves ] , only the third example obeys the criterion ( [ e : criterion ] ) for a finite range of @xmath1 . imposing a stress @xmath1 from within this range will result in a system that is jammed in the sense that the strain logarithmically creeps , @xmath9 and @xmath85 . two additional complications arise when the system is driven by an imposed strain rate @xmath4 rather than an imposed stress . firstly , if @xmath4 is increased at a sufficiently slow rate that the system is always arbitrarily close to its steady state , the stress may undergo a discontinuous jump from one branch of the flow curve to another , as marked in figs . [ f : jamming_curves](b ) and [ f : jamming_curves](c ) . if @xmath4 is decreased in a similar fashion , then @xmath79 will follow the upper branch until it again jumps discontinuously at a _ different _ value of @xmath4 . thus the system exhibits _ hysteresis due to non monotonic flow curves has also been observed experimentally @xcite . the complementary form of non monotonicity , which would allow a discontinuous jump in @xmath4 for a small change in @xmath1 , can not be realised in this class of models . this is because there is only one value of @xmath5 , and hence one steady state solution , for any given value of @xmath1 . the second complication concerns the stability of the jammed state . in fig . [ f : roots ] , the yield stress @xmath84 and an @xmath5 that obeys ( [ e : criterion ] ) for a range of @xmath1 have been plotted on the same axes . also plotted is the line of @xmath1 that can be reached for a small but finite @xmath60 , which converges to @xmath84 as @xmath86 . these are the @xmath1 that can be realised in practice , since the steady state solution ( [ e : st3 ] ) is not defined for @xmath87 . for this example there are 3 roots : a ` flowing ' root with @xmath88 , and two ` jammed ' roots with finite @xmath89 . the arrows in this diagram point in the direction in which the stress will be varying for a given point on the line @xmath5 . this direction is based on the assumption that , for time scales over which @xmath0 can be treated as a constant , which will certainly include the asymptotic regime , the system will behave like the sgr model and @xmath1 will evolve towards @xmath71 . it is clear from this that all points tend to flow away from the middle root , which is therefore unstable . if the system is initially placed near to this root , it will undergo a transient and converge to either the higher root , or to the ` flowing root ' with @xmath88 . [ f : roots ] can also be used to predict the stability of the steady state for finite @xmath4 . as @xmath4 increases , the asymptotic stress @xmath62 for the sgr model will move away from the yield stress curve , but will always remain a monotonic decreasing function of @xmath4 @xcite . thus a smoothly decreasing @xmath5 will only intersect with it 1 or 3 times ; when there are 3 roots , the middle one is always unstable , for precisely the same reasons as given in the previous paragraph . the range of @xmath1 which are unstable and can not be realised under an imposed @xmath4 have been labelled ` a ' in figs . [ f : jamming_curves](b ) and ( c ) . note that the unstable roots correspond to regions of the flow curve with a negative gradient , as in experiments and from other theoretical considerations @xcite . there are no stability issues for an imposed @xmath1 , for which there is always a single , stable root . there are in principle no difficulties in extending the results described above to a system that is driven by a time dependent imposed stress @xmath79 . as long as @xmath79 tends to a constant value @xmath90 at long times , then the same steady state behaviour as previously discussed will apply , with @xmath1 replaced by @xmath90 . the transient behaviour does not affect the final state . however , this is not the case when the system is instead driven by an imposed strain rate @xmath19 . in this case , @xmath79 can vary in a manner that is difficult to predict in advance , so that the final stress reached , and hence whether or not the system is jammed , will in general depend on the entire history of the system . a concrete example of the history dependence of a yield stress is given in fig . [ f : existence_of_yield ] . here , the system is first subjected to a step shear of magnitude @xmath91 , and is then continuously sheared at a rate @xmath4 , _ i.e. _ @xmath92 . as @xmath4 tends to zero , the stress approaches an asymptotic value @xmath71 ; however , whether or not @xmath71 is finite depends on @xmath91 . for @xmath93 the stress is seen to be converging to a finite value @xmath94 , but for @xmath95 it rather tends to a ` flowing ' state with @xmath88 ( more precisely , @xmath96 as @xmath76 ) . the system can be said to exhibit strong long term memory , where the use of the word ` strong ' means that the memory of an earlier perturbation of finite duration does not decay to zero with time @xcite . the choice of @xmath5 used in the previous example is in fact the same as that plotted in the stability diagram , fig . [ f : roots ] . this allows for a striking illustration of the instability of the middle root that lies near the point @xmath97 on the stability diagram . plotted in fig . [ f : missed_root ] are the @xmath79 curves for a fixed @xmath98 and a range of values @xmath99 . it is clear that the stress will always diverge away from the unstable root at late times , no matter how finely @xmath91 is tuned . thus only the roots at @xmath88 and @xmath94 are stable , as previously claimed . there are many other complications that can arise due transient behaviour in a strain controlled system . for instance , all of the @xmath79 curves plotted in fig . [ f : existence_of_yield ] reach their global minimum @xmath100 at a time @xmath101 . it can be shown that @xmath100 becomes arbitrarily small as @xmath76 . since this corresponds to a state with a high effective temperature @xmath102 in which the stress is rapidly dissipated , then if @xmath4 is sufficiently small , @xmath79 will remain low and the system will crossover to the flowing root , irrespective of @xmath91 . however , this effect can be reversed if the initial step shear is replaced by a smoothly varying @xmath19 , such as one that exponentially decays towards its final value of @xmath4 , which is more like what could be attained in an actual rheometer . we will not pursue these complications any further here . in summary , the sgr model generalised to allow the effective temperature @xmath0 to vary with the global stress @xmath1 can exhibit , for suitable choices of @xmath5 : _ hysteresis _ in @xmath79 for slowly varying @xmath4 , as seen from the flow curves in fig . [ f : jamming_curves ] ; shear induced _ jamming _ , as in fig . [ f : existence_of_yield ] ; and _ strong history dependence _ of the existence of a yield stress . the second limiting form for @xmath14 to be considered is one that depends only on the local strain , @xmath10 . this marks a more drastic departure from the sgr model than the @xmath5 considered in the previous section , since now every element will generally have a different effective temperature @xmath0 . the steady state will therefore be different to that of the sgr model , and it will not now be possible to map the flow curves for @xmath11 across from those for constant @xmath0 . one could also envisage regions of the parameter space for which the steady state can not even be reached . indeed , this is precisely what we find : for some finite region of parameter space , the system reaches an oscillatory regime under an imposed stress , but not under an imposed strain . this is the central finding of this section . the physical justification in choosing @xmath10 is similar to that already discussed for @xmath3 , _ i.e. _ it is assumed that the material can become more or less susceptible to noise in its strained state . the main difference is that this is now assumed to be a local effect , that can be described purely on the level of individual elements . just as in the previous section , we shall focus our attention on @xmath11 that decrease with @xmath2 , since it is these that have the potential to exhibit shear thickening . given @xmath11 , the steady state stress @xmath1 for a given @xmath4 can be found by calculating the mean strain @xmath103 for the steady state solution ( [ e : st3 ] ) . some example flow curves are plotted in fig . [ f : local_flow_eg ] . they are clearly monotonic : their gradient is everywhere positive , and there is a one to one relationship between @xmath1 and @xmath4 . in these figures , the @xmath11 were chosen to vary in a stepwise manner , taking a value @xmath104 for @xmath105 and a lower value @xmath70 for @xmath106 . this is typical of the @xmath11 employed in this section . however , the monotonicity result is entirely general and applies for any @xmath10 , as demonstrated in appendix [ s : monotonicity ] . the physical reason for the monotonicity of the flow curves is that the elements are uncoupled when @xmath10 and the system is driven by an imposed strain @xmath17 . that is , the expected strain reached before a given element yields does not depend on the state of any other elements . this is _ not _ true for the general case @xmath14 , when the value of @xmath0 for an element depends on the global stress @xmath1 , and thus on the state of the rest of the system . as long as independence holds , the average strain reached before each individual element yields , and hence the global stress @xmath1 , can only increase with increasing @xmath4 . thus the flow curves must be monotonic . the monotonicity of the flow curves is an important finding . as mentioned in the introduction , rheological instabilities often arise when the system has been driven to a point on its flow curve which has a negative gradient . fluctuations can then cause spatial inhomogeneities to arise , such as shear bands with either the same stress and different strain rates , or equal strain rates but differing stresses @xcite . alternatively , temporal oscillations may be observed @xcite . since the flow curves for @xmath10 have no regions of negative slope , one would not expect any instabilities to appear in this model . nonetheless temporal oscillations in @xmath4 do occur for a finite range of imposed @xmath1 , as we now describe . the monotonicity result described above was attributed to the independence of the elements , which is only true for a strain controlled system . by contrast , the elements become coupled under an imposed @xmath1 since , for example , a single element yielding causes the mean strain to decrease slightly , which must be immediately countered by an increase in @xmath4 , _ which affects every element in the system_. thus collective behaviour can now occur . in fact , this collective behaviour can not alter any system that has already reached steady flow , which we know is identical for strain and stress controlled systems . thus as long as the stress controlled system reaches steady flow , it will fall onto the same monotonic flow curve as before . however , the couplings between the elements can drastically alter the transient behaviour , to such an extent that steady flow may never be reached . an example of collective behaviour is given in fig . [ f : local_osc_eg ] , which shows @xmath19 for a system driven by an imposed stress @xmath107 . as before , this plot was generated by numerically integrating the master equation ( [ e : master ] ) from an initially unstrained state , using the procedure described in appendix [ s : numerics ] . for this example , the system undergoes a short transient before entering into an oscillatory regime , in which @xmath19 varies with a single period of oscillation . there is no suggestion of a decay in the amplitude of the oscillation in @xmath19 over the largest simulation times attainable , even when plotted against @xmath108 ( not shown ) , suggesting that this is the true asymptotic behaviour . for different choices of @xmath11 it is not possible to identify a single period of oscillation , as demonstrated in fig . [ f : local_trans_eg ] . it is not yet clear if this behaviour represents a distinct regime with more than one period of oscillation , possibly even a precursor to chaotic behaviour , or if it is merely a long lived transient that eventually crosses over to either steady flow or a single period oscillation at later times . the extensive simulation times means that we have so far been unable to map out the class of @xmath11 that give rise to an oscillatory or otherwise non steady state . nonetheless we can make the following observation . for all of the oscillatory cases that we have observed thus far , @xmath11 changes from a value in the range @xmath69 for small @xmath2 , to a value @xmath70 for larger @xmath2 . other choices of @xmath11 may produce an oscillatory transient , but its amplitude always seems to decay to zero in time . it is not clear to us why only this class of @xmath11 can produce a stable oscillatory regime , but it is tempting to note that the requirement that @xmath69 for small @xmath2 is also the range of @xmath0 for which the sgr model does not have a linear regime under an imposed @xmath1 @xcite . that is , a significant proportion of elements will eventually gain a finite strain , no matter how small @xmath1 is , and thus the variation in @xmath11 will eventually be ` felt . ' once a suitable @xmath11 has been found , it is possible to move in and out of the non steady regime by varying the applied stress @xmath1 . generally , we find that a high @xmath1 gives rise to steady flow , and low @xmath1 produces oscillations ; however , the excessive simulation times needed for small @xmath1 means that we have not been able to rule out another crossover to steady flow as @xmath109 . within the oscillatory regime , the mean strain rate averaged over a period of oscillation is generally much lower than that predicted by the flow curve . for the examples already given , the mean strain rate deviates from the flow curve by a factor of 2 for the oscillations shown in fig . [ f : local_osc_eg ] , and by two orders of magnitude for the ( possibly transient ) oscillations of fig . [ f : local_trans_eg ] . again , we have been unable to fully characterise this behaviour with the available simulation resources . to an extent , these problems will be alleviated by considering a simplified model considered in sec . [ s : monodisperse ] . before turning to consider this , we provide a qualitative description of the oscillatory behaviour in terms of the local strains @xmath2 . the mechanism behind the oscillatory behaviour can be qualitatively understood by inspecting the evolution of the @xmath25 distribution during a single period of oscillation . as an example , a suitable sequence of snapshots is given in fig . [ f : snapshots_poly ] . to aid in the interpretation of these figures , it is useful to recall some properties of the master equation ( [ e : master ] ) . firstly , the convective term @xmath110 means that any maxima or minima in @xmath25 will move in the direction of increasing @xmath2 at a rate @xmath19 . these extrema will eventually disappear when large values of @xmath2 are reached , and the yield rate @xmath23 becomes very high . also , there is a flux of newly yielded elements at @xmath27 , which have barriers @xmath20 weighted according to @xmath22 . since the oscillatory behaviour is by definition cyclical , it is somewhat arbitrary where one chooses to start the sequence of snapshots . in fig . [ f : snapshots_poly ] we have chosen times corresponding to before , during and just after the point when @xmath4 reaches its maximum value . in the first snapshot , @xmath25 is concentrated into two regions : a sharp peak at @xmath111 , and a broader peak with @xmath2 in the range @xmath112 . the first peak is ` hot ' in the sense that it has the higher value of @xmath11 , but since it has a low strain , it does not significantly contribute to the total stress @xmath1 . the second peak , although highly strained , lies in a region in which @xmath11 is small , and therefore has a low rate of yielding . as long as the yield rate is low , so too is the rate of stress loss from elements that belong to this peak . therefore the strain rate @xmath4 will also be low , and indeed this first example corresponds to a system in which @xmath4 is close to its minimum value . this state of affairs is not stable , however . although the yield rate of the highly strained elements is low , it is nonetheless non zero , and therefore so is @xmath4 . consequently the whole system is being convected at a finite rate , so the elements in the high@xmath2 region are becoming more strained . _ decreases _ their effective energy barrier @xmath113 , which _ increases _ their yield rate . but an increased yield rate means an increased @xmath4 , which in turn increases the rate at which the elements are becoming more strained , which increases their yield rate yet more , and so on . this description is that of a _ positive feedback loop _ , which causes @xmath4 to increase at ever faster rates until all of the heavily strained elements have been depleted . the second snapshot in fig . [ f : snapshots_poly ] show the state of the system shortly before this has happened . at the same time that the highly strained elements are being depleted , @xmath4 is close to its maximum value and consequently @xmath25 in the small @xmath2 region becomes somewhat flat , certainly much flatter than under a small @xmath4 . this can clearly be seen in the second and third snapshots of fig . [ f : snapshots_poly ] . when @xmath4 again falls to lower values , this flat part of the distribution will start to decay as the elements within it yield . however , the yield rate depends on @xmath11 , and since @xmath11 changes to a lower value at @xmath114 , @xmath25 will decay much more rapidly for @xmath2 just smaller than @xmath115 than for @xmath2 just greater than @xmath115 . thus a dip will occur around the point @xmath116 , which can clearly be seen in the final snapshot . as time increases , this dip will become more pronounced until the system can again be separated into a population of highly strained elements and a second population with @xmath111 . this is where we began , and thus the cycle is complete . it is possible to more completely describe the oscillatory regime for a simplified version of the model where every element has the same energy barrier @xmath117 . this formally corresponds to a ` monodisperse ' prior distribution @xmath49 , as opposed to the exponential @xmath118 which has been employed in all of the earlier sections . the reduction in the number of degrees of freedom that this entails significantly decreases the simulation times , and therefore allows for a more thorough numerical investigation of the model . it is also possible to derive an analytical criterion for the stability of the steady state , as described below . furthermore , the very fact that a monodisperse system also has an oscillatory regime clearly demonstrates that this phenomenon is robust . in particular , it does not require the usual sgr assumption of an exponential tail to @xmath22 . therefore the possible existence of a yield stress , which was so central to the history dependent jamming scenario at controlled @xmath4 described in sec . [ s : global ] , is not necessary . this robustness means that the mechanism behind the oscillations at controlled @xmath1 may be realisable in a much broader range of models than the restricted set considered here and , possibly , may also occur in real materials . the emergence of oscillations in this monodisperse model has been separately discussed elsewhere @xcite . here we supply extra details , such as the stability analysis of the steady flow state . for the sake of completeness , the overall behaviour of the model will also be briefly described , although we refer the reader to @xcite for a fuller discussion of many of the points raised below . perhaps the most immediate consequence of only allowing a single barrier @xmath117 is that the system can never have a finite yield stress . indeed , in the linear regime @xmath76 , the steady state stress is just @xmath119 , which vanishes with @xmath4 . there are no qualitative differences for @xmath70 , @xmath69 _ etc . _ , as in the sgr model . the complete lack of a yield stress means that a monodisperse system with @xmath3 will _ not _ exhibit a jamming transition , in contrast to the situation for a @xmath22 with an exponential tail described sec . [ s : global ] . one respect in which the monodisperse @xmath20 model is similar to the exponential @xmath22 case is that its flow curves for @xmath10 are also monotonic . in fact , the monotonicity proof given in appendix [ s : monotonicity ] holds for arbitrary @xmath11 and @xmath22 . thus there are no regions on the flow curve with a negative slope , and therefore there are no ranges of control parameters for which one might normally expect bulk shear banding to arise . just as in the previous section , however , oscillatory behaviour can be realised under an imposed constant stress . as in the polydisperse case , the monodisperse model exhibits an oscillatory regime under conditions of imposed stress , but not under an imposed @xmath4 . however , the oscillatory regime in the monodisperse model differs from the polydisperse case in that only single period oscillations have so far been observed . there is no suggestion of the more complex non steady behaviour hinted at earlier in fig . [ f : local_trans_eg ] , for example . some examples of the oscillatory behaviour in the monodisperse case are given in fig . [ f : osc_nonmon_eg ] . here , the strain @xmath17 is shown for 3 systems with @xmath120 and the same @xmath11 , but different imposed @xmath1 . remarkably , the mean strain rate @xmath121 , where the use of the angled brackets now denotes @xmath4 averaged over a single period of oscillation , is clearly a _ decreasing _ function of @xmath1 , in complete contrast to the monotonic flow curve . varying @xmath1 over a wider range of values reveals that steady flow is reached for either sufficiently small or sufficiently large imposed stresses ; only for intermediate @xmath1 are oscillations observed . the @xmath121 as read off from the simulations are plotted against @xmath1 in fig . [ f : osc_flowcurve ] , overlayed with the steady state flow curve . it is clear that , if the system reaches a steady state , it falls onto the flow curve and therefore the dependence of @xmath121 with @xmath1 is monotonic . within the oscillatory regime , however , the @xmath121 line deviates from the flow curve to an extent that it even becomes non monotonic for a broad range of @xmath1 . the shape of the oscillations varies with @xmath1 . close to either transition to the steady flow regime , the oscillations are approximately sinusoidal , indicating that there is only a single , finite frequency of oscillation at the transition , the amplitude of which vanishes with the onset of steady flow . further into the oscillatory regime , @xmath4 can no longer be decomposed into a single harmonic , but instead approaches a waveform in which most of the variation in @xmath4 is compressed into a small fraction of the total period of oscillation . consequently @xmath17 approaches an almost rectangular , staircase shape . the mechanism underlying the rapid increase in @xmath4 is the same positive feedback loop that has already been discussed for the polydisperse case , as can be seen by inspecting the evolution of the @xmath122 distribution with time @xcite . near to the transition points at @xmath123 , the amplitude of the oscillation appears to vanish as @xmath124 with @xmath125 , as demonstrated in fig . [ f : pow_law_div ] . data fitting suggests that the lower threshold lies at @xmath126 with a value of @xmath127 , and that the upper threshold is at @xmath128 and has a different value of @xmath129 . however , the range of values given in this figure is somewhat limited , constituting only an order of magnitude of variation in @xmath130 and an even narrower range of amplitudes . the reasons for this are purely technical : close to the transition points , the amplitude decays very slowly in time to its asymptotic value , rendering the required simulation times impractically long . hence it is conceivable that the true values of @xmath131 are significantly different from those found here . in particular , a value of @xmath132 for both cases , as expected for a hopf bifurcation @xcite , can not be ruled out . finally , plotted in fig . [ f : constant_gd_t ] is the product of the mean strain rate @xmath121 and the period of oscillation @xmath12 for different values of @xmath1 . to first order , @xmath133 is seen to be independent of @xmath1 , which suggests that the oscillatory regime can be characterised by a single strain @xmath134 . in this case @xmath135 , which is also the typical increase in @xmath7 during a single cycle of oscillation . a possible interpretation of @xmath136 is that it is the amount by which the system needs to be strained until the positive feedback loop discussed earlier starts to dominate the system behaviour , causing it to ` reset ' to the start of its cycle . if this is correct , then @xmath136 should correspond to the point at which highly strained elements have the same yield rate as unstrained elements , _ i.e. _ @xmath137 . rough calculations based on this assumption give @xmath138}$ ] , which for this example predicts @xmath139 , in fair agreement with the observed value . a second advantage of the monodisperse model is that it is possible to derive an analytical criterion for the stability of the steady state at fixed driving shear stress @xmath1 , although even in this simplified case the resulting expression is still somewhat unwieldy . suppose steady flow is reached with a strain rate @xmath140 . then the corresponding @xmath141 is @xmath142\quad,\ ] ] where the normalisation constant @xmath143 is fixed by the constraint @xmath144 . we now look for eigenfunctions @xmath145 and eigenvalues @xmath146 of the linearised relaxation operator by writing @xmath147 in these expressions , @xmath148 is a real constant , and the real functions @xmath149 remain to be found . the complex constant @xmath150 determines the stability of the steady state : it is unstable precisely when @xmath151 , since the amplitude of the perturbation will then increase exponentially in time . conversely , it is stable when @xmath152 . the frequency of the oscillatory part of the motion near to the steady state is @xmath153 @xcite . the unknown function @xmath149 can be found by inserting ( [ e : stabdefp ] ) and ( [ e : stabdefg ] ) into the master equation and neglecting terms of @xmath154 , which gives @xmath155 & = & -gp_{\infty}'(l)\\ & = & \frac{g\gamma(l)}{{\dot{\gamma}}_{\infty}}p_{\infty}(l)\quad . \label{e : linearised}\end{aligned}\ ] ] this is just a first order differential equation in @xmath2 and integrates to @xmath156\nonumber\\ & & \times \exp\left[-\int_{0}^{l}{\rm d}l'\,\frac{s+\gamma(l')}{{\dot{\gamma}}_{\infty}}\right ] \label{e : ptilde}\end{aligned}\ ] ] the constant @xmath157 can be found by imposing @xmath158 , _ i.e. _ @xmath159 . also , since we are considering a stress controlled system , @xmath61 must remain constant and hence @xmath160 this global constraint allows the value of @xmath150 to be specified . inserting ( [ e : ptilde ] ) into ( [ e : globcons ] ) gives , after some manipulations , the following equation for @xmath150 , @xmath161 although this equation can not be rearranged to give an expression for @xmath150 in terms of the system parameters , as would be desired , it is still enough to show that a purely exponential variation away from the steady flow can not be observed . this follows from the observation that the first line in ( [ e : s_eqn ] ) is an odd function in @xmath162 , but the second line is a strictly increasing function of @xmath163 when @xmath164 . hence the equation can not be obeyed for a purely real @xmath150 , and the transient behaviour close the the steady state must include an oscillatory component , which is consistent with the simulation results . one of the most striking findings of this work has been the observation of oscillations in @xmath19 under a constant imposed stress @xmath1 for some @xmath10 , as described in secs . [ s : local ] and [ s : monodisperse ] . some consequences of this phenomenon have already been discussed in @xcite . two further points will be discussed here : the identification of the dominant physical mechanisms , and the relationship to so called ` stick slip ' behaviour observed in other systems . both of these will be considered in turn . in sec . [ s : qual_desc ] , the mechanism behind the oscillatory behaviour was explained in terms of two separated populations of elements : a ` cold ' population of highly strained elements with a low @xmath0 , and a ` hot ' population of elements with @xmath111 and a high @xmath0 . it was explained how the cold elements can give rise to a positive feedback loop , causing @xmath19 to rapidly increase until the cold elements have yielded . at the same time , a new population of cold elements is produced from the hot one . known instances of rheological instability are often explained in terms of the spatial coexistence of subpopulations , or phases . for instance , the temporal oscillations in viscosity observed in wormlike micelles under an imposed stress was attributed to a slowly fluctuating interface between a fluid phase and shear induced structures @xcite . for surfactant solutions in the lyotropic lamellar phase , it was attributed to coexisting ordered and disordered phases @xcite . however , these instabilities occur in the vicinity of non monotonic regions of the flow curve . spatial instabilities , such as shear banding , can also be found near to where the flow curve has a negative gradient @xcite . by contrast , the temporal oscillations observed in our models arise even though they are by assumption spatially homogeneous , and the flow curves are everywhere monotonic . this suggests that the mechanism behind the oscillatory behaviour seen here has not yet been observed in a rheological context . it is therefore sensible to look further afield for analogous phenomena . one possibility comes from mathematical biology . the glass mackey equation describes the variation in time of the size of a population of white blood cells in response to a hormonal control system @xcite . it has a form similar to that of a first order differential equation , but includes a feedback term that depends on the state variable at an earlier time . this _ delay _ corresponds to the maturation time of white blood cells from stem cells . because of this delay , the population size can vary in time in a non trivial manner , including oscillatory behaviour and chaos @xcite . that some form of time delay may also play a role within our @xmath11 model is clear : once an element becomes ` cold , ' its yield rate remains very low until some time later , when it becomes sufficiently strained that its yield rate becomes comparable to that of ` hot ' elements . thus there is a delay between when an element first becomes cold and when it yields , although here the delay time is not constant but depends on @xmath17 . thus it is possible that the oscillatory behaviour observed for @xmath10 could be described by a simplified equation , similar in form to the glass mackey equation . this is a particularly enticing proposition , as the glass mackey equation is capable of producing chaos , and chaotic behaviour has also been observed in surfactant solutions @xcite , albeit in the signature of @xmath79 under an imposed @xmath4 . it is not yet clear to us if a meaningful mapping between the two models is possible . deep into the oscillatory regime , the waveform of @xmath19 throughout a single period of oscillation becomes increasingly ` sharp , ' with most of the variation in @xmath19 occurring in just a small fraction of the total period of oscillation . the ratio of the maximum value of @xmath19 to its minimum has been shown to exceed two orders of magnitude , and we see no reason to deny that greater separations may be attained for different parameter values . it is tempting to refer to this behaviour of @xmath19 as ` stick slip ' behaviour , in which the system is ` stuck ' until the short duration of time in which @xmath4 rapidly accelerates to its maximum value , corresponding to a ` slip ' event . this kind of response is also observed in earthquakes @xcite , ultra thin liquid films @xcite and granular media @xcite , for example . however , we have instead chosen to refer to the variation in @xmath19 as merely ` oscillatory , ' since the underlying physics seems to be somewhat different to the examples of stick slip behaviour mentioned above . in particular , the term stick slip is usually employed to refer to a surface friction phenomenon , whereas the models studied in this paper have no surface , or indeed any form of spatial definition . they are only intended to describe the bulk behaviour of a material . furthermore , we only find oscillations in @xmath19 under an imposed @xmath1 , and never _ vice versa _ , whereas stick slip seems to more usually ( although not exclusively ) refer to variations in the ( normal ) stress under a constant driving velocity . thus to avoid possible confusion , we choose not to refer to the behaviour observed in the models as stick we have introduced a range of schematic models that are capable of exhibiting a form of jamming under an imposed stress @xmath1 . the models are based on the sgr model of sollich _ et al . _ , but differ in that the effective temperature @xmath0 is no longer constant , and can instead vary with the state of the system through either the global stress @xmath1 or the local strain @xmath2 . we have considered choices of @xmath0 that decrease for increasing @xmath1 or @xmath2 , which is relevant to the study of shear thickening materials . these models have no spatial definition , and thus by construction can not exhibit any form of spatial heterogeneity . for @xmath3 , the flow curves can be extracted from the known curves for constant @xmath0 . many choices of @xmath5 produce flow curves with non monotonic regions , which exhibit hysteresis in @xmath79 under ramping the strain rate @xmath19 first upwards and then downwards . furthermore , a subset of these @xmath5 also give rise to a _ jammed _ state for a range of applied stresses , in that the strain @xmath17 creeps logarithmically , @xmath165 . the criterion for this to arise is that the curve of @xmath5 drops below the sgr yield stress curve @xmath84 when they are plotted on the same axes . for an imposed strain rate that decays to zero at late times , a jammed configuration was defined as one with a finite asymptotic stress , @xmath166 as @xmath167 and @xmath18 . it was found that whether or not a jammed configuration was reached depends on the entire strain history of the system , a situation that was referred to as _ history dependent jamming_. for @xmath10 , the flow curves are always monotonic , and steady flow is always reached under a constant imposed strain rate @xmath58 . however , for a range of imposed stresses and some choices of @xmath11 , steady flow is not realised . numerical integration of the master equation demonstrated that @xmath19 oscillated around a well defined mean with a single period of oscillation . the possibility of more complex non steady behaviour in some regions of parameter space could not be ruled out . a similar oscillatory behaviour occurs with a simpler model in which every element has the same energy barrier @xcite , which suggests that this phenomenon is robust . finally , we discussed the relationship between this oscillatory behaviour and that observed in experiments , and considered analogous phenomena from fields outside of rheology . the authors would like to thank suzanne fielding for stimulating discussions concerning this work . aa also wishes to thank the university of edinburgh for the hospitality that allowed this work to start . dah was funded by epsrc(uk ) grant no . gr / m09674 . direct numerical integration of the master equation ( [ e : master ] ) has proven to be unstable with respect to discretisation errors . instead , the results in this paper were generated from the transformed equation @xmath168 where @xmath169 . this removes the convective term and dramatically improves numerical stability . the discrete probability distribution @xmath170 was defined on a rectangular mesh of points @xmath171 and evolved according to ( [ e : dl_master ] ) by using a straightforward euler method . a further refinement was to average both the evolution equation ( [ e : dl_master ] ) and the initial conditions over the ranges @xmath172 and ( @xmath173 ) . this significantly reduces the number of mesh points required for the simulations to properly converge , without unduly increasing the algorithmic complexity . the delta function was treated as a triangle of base width @xmath174 and height @xmath175 , but any sufficiently narrow function gave the same results . two classes of initial conditions were employed , but the long time behaviour of the system was found to be identical in both cases . only the short term behaviour varied , and then only in a non essential manner . for the sake of completeness , the initial conditions were : ( i ) a ` quench ' @xmath176 , which corresponds to the unstrained equilibrium at @xmath177 , or ( ii ) @xmath178 with @xmath179^{-1}$ ] , which corresponds to an unstrained system that has been allowed to reached equilibrium . note that this second initial state is only defined for @xmath180 @xcite . the strain @xmath17 is only known _ a priori _ for a strain controlled system . when it is rather a known stress @xmath79 that is applied , @xmath17 must be chosen at every time step so that the actual stress does not differ from @xmath79 by more than a tolerance parameter @xmath181 . to ensure that this condition was satisfied in our simulations , a series of trial strain rates @xmath182 were generated and tested on a replica mesh @xmath183 . the @xmath184 were not updated until a suitable @xmath4 had been found . the trial values @xmath185 were generated as follows . for a continuous time variable , @xmath186 exactly , as seen by multiplying the master equation ( [ e : master ] ) by @xmath2 and integrating over all @xmath20 and @xmath2 . this is therefore the sensible choice for @xmath187 . however , integrating over a finite time step @xmath188 inevitably introduces errors of @xmath189 , and so the integrated stress will differ from the required value by some small amount @xmath190 . reintegrating with @xmath191 will therefore reduce the error to a smaller amount @xmath192 . this procedure can be repeated to generate a series of successively better estimates @xmath193 , @xmath194 _ etc . _ for our choice of @xmath195 , we have found that typically 35 such trials are needed at every time step . the purpose of this appendix is to show that the flow curves are monotonic for any yield rate @xmath196 that depends only on the local strain @xmath2 . this includes the thermally activated @xmath196 with @xmath10 , possibly constant , but not when @xmath0 also depends on the stress @xmath1 . it also applies for an arbitrary prior barrier distribution @xmath22 . where we have introduced the short hand notation @xmath199 . the stress is @xmath200 , where the angled brackets ` @xmath201 ' denote the average over @xmath38 . if @xmath202 depends on @xmath1 , then a @xmath1 must be chosen that is consistent with ( [ e : st3 ] ) . in general there can be more than one suitable @xmath1 . the final line in this equation is valid as @xmath210 is a constant , so @xmath211 . since @xmath212 is an increasing function of @xmath2 , the quantity in the angled brackets in ( [ e : mono_ans ] ) is positive both for @xmath213 and for @xmath214 , and vanishes smoothly at @xmath215 . thus the left hand side must be positive , _ i.e. _ @xmath1 always increases with @xmath4 , as claimed . the same conclusion does not hold when @xmath216 . since @xmath1 depends on @xmath4 , differentiating the steady state solution now gives rise to an extra term on the right hand side of ( [ e : mono_local ] ) , thus it is now possible for the gradient to diverge if @xmath220 is sufficiently negative , _ i.e. _ if @xmath221 decreases sufficiently rapidly with @xmath1 . for the particular case of @xmath222 , the gradient of the flow curve diverges at any point such that

we describe the spectral energy distributions ( seds ) of lyman break galaxies ( lbgs ) at z using deep mid - infrared and optical observations of the extended groth strip , obtained with irac and mips on board spitzer and from the ground , respectively . we focus on lbgs with detections at all four irac bands , in particular the 26 galaxies with irac 8 m band ( rest frame k - band ) detections . we use stellar population synthesis models and probe the stellar content of these galaxies . based on best fit continuous star - formation models we derive estimates of the stellar mass for these lbgs . as in previous studies , we find that a fraction of lbgs have very red colors and large estimated stellar masses ( m 10 m ) : the present spitzer data allow us , for the first time , to study these massive lbgs in detail . we discuss the link between these lbgs and submm - luminous galaxies . we find that the number density of these massive lbgs at high redshift is higher than predicted by current semi - analytic models of galaxy evolution .

astro-ph0605355

in recent years , great advances in our understanding of the nature and evolution of high redshift galaxies have been made , thanks to the availability of large samples . the techniques that have been developed rely either on colour selection criteria ( e.g. steidel et al . 2003 , 2004 , franx et al . 2003 , daddi et al . 2004 ) or detection in the submillimeter through blank field surveys using the submillimeter common user bolometer array ( scuba ) on the james clerk maxwell telescope ( jcmt , e.g. , hughes et al . , 1998 ) or the max - planck millimeter bolometer array ( mambo , e.g. bertoldi et al 2000 , eales et al . 2000 ) . all these surveys have unveiled large numbers of @xmath5 galaxies but the overlap between the various populations is by no means easy to investigate . among the various methods the lyman break dropout technique ( steidel & hamilton 1993 ) , sensitive to the presence of the 912@xmath6 break , is designed to select @xmath7 galaxies . the typical star formation rate ( sfr ) deduced from the uv continuum emission of lyman break galaxies ( lbgs ) is estimated to be moderate , around 2050 m@xmath8yr ( assuming h@xmath9=70 km s@xmath10mpc@xmath10 and q@xmath9=0.5 ) . the sfr value quoted is the `` mean '' sfr which is derived from the entire lbg sample which ranges from 10 to 1000 m@xmath8yr ( for the most massive lbgs ) . there is clear evidence , however , for the presence of significant amounts of dust in lbgs ( e.g. sawicki & yee 1998 , calzetti 2001 , takeuchi & ishii 2004 ) . when corrected for the presence of dust then this `` mean '' sfr values increase to @xmath0 100 m@xmath8yr . the high sfrs and co - moving density of lbgs , together with the results of the clustering analyses ( e.g. adelberger et al . 2004 ) makes them plausible candidates for the long - sought progenitors of present day elliptical galaxies ( e.g pettini et al . the nature of the relationship between lbgs and submm - luminous galaxies ( hereafter smgs ) has , however , remained unclear . one possible scenario is the one where lbgs and smgs form a continuum of objects with the submm galaxies representing the `` reddest '' dustier lbgs . central to such a hypothesis is of course the issue of dust in lbgs . a number of techniques have been used to deduce the dust content of lbgs ranging from studies of the optical line ratios ( pettini et al . 2001 ) to formal fits of the overall sed of lbgs based on various star forming scenaria ( e.g. shapley et al . 2001 , 2003 ) . both methods agree that the most intrinsically luminous lbgs , which have higher sfrs , contain more dust ( e.g. adelberger & steidel 2000 , reddy et al . more recently , x - ray stacking studies have shown that intrinsically more luminous lbgs also contain more dust ( e.g. nandra 2002 , reddy & steidel 2004 ) . another important issue in understanding lbg evolution and their connection to smgs is the study of their stellar mass content . the co - moving stellar mass density at any given redshift is the integral of past star - forming activity . thus , stellar mass is a robust tool to probe galaxy evolution and is subject to fewer uncertainties than the star formation rate ( sfr ) . until recently , stellar mass estimates for z@xmath03 lbgs have been based on ground based photometry which only samples out to rest frame optical band . the observed luminosity at these wavelengths is dominated by recent star - formation activity rather than the stellar population that has accumulated over the galaxy s lifetime . with the advent of the spitzer space telescope ( hereafter spitzer , werner et al . 2004 ) we now have access to longer wavelengths . in particular the availability of the irac instrument ( fazio et al . 2004 ) with imaging capabilities out to 8 microns allow us to probe rest frame k band ( for z@xmath03 galaxies ) where the light is sensitive to the bulk of the stellar content . preliminary results of adding irac photometry to stellar mass estimates have been presented in e.g. barmby et al . ( 2004 ) for z@xmath03 lbgs and shapley et al . ( 2005 ) for bx@xmath11bm objects . in this paper , we present sensitive mid infrared photometry for a sample of spectroscopically confirmed z@xmath03 lbgs detected as part of the irac guaranteed time observations ( gto ) program on the extended groth strip ( egs ) . we focus on the broad band spectral energy distributions ( seds ) of luminous lbgs detected with irac channel 4 ( 8 @xmath1 m corresponding to rest frame k band ) . the sample benefits from deep ground based optical photometry and spectroscopy ( steidel et al . our aim is to explore the range of stellar masses and assess the benefits of adding longer wavelength observations . we examine the correlation between mass and absolute luminosity ( magnitude ) from optical to the k band . the paper is organised as follows : in section 2 we present a brief account of the observations and data reduction while in section 3 we review the sample properties followed by a discussion of the detailed seds ( section 4 ) . we discuss the detailed models used to estimate stellar masses in section 5 while in section 6 we present the actual masses . we focus on the properties of massive lbgs in section 7 and we discuss their possible evolution and connection to submm - luminous galaxies . we estimate the number density of these massive lbgs in section 8 and summarize our conclusions in section 9 . the irac ( fazio et al . 2004 ) data presented here were acquired as part of the irac gto program . the extended groth strip was observed by irac at 3.6 , 4.5 , 5.8 and 8.0 @xmath1 m during january and july 2004 . irac observations of the egs cover an area of 2@xmath12@xmath1310@xmath14 . irac on board spitzer has the capability of observing simultaneously in the 4 bands , when scanning a region , with an effective f.o.v . of 5@xmath15 . the irac exposures consisted of 52@xmath13200 sec dithered exposures at each of the 3.8 , 4.5 , 5.8 @xmath1 m wavelengths and 50@xmath13208 sec exposures at 8@xmath1 m , in the 2@xmath12@xmath1310@xmath14 map . because the field was observed twice and at different position angles , removal of instrumental artifacts during mosaicing was significantly facilitated . the observations reach a ( 5@xmath16 ) point - source sensitivity limit of 24.0 , 24 , 21.9 , and 22.0 mag ( ab ) at 3.6 , 4.5 , 5.8 and 8.0 @xmath1 m , respectively . observations with mips ( rieke et al . 2004 ) were carried out in june 2004 in scanning mode . the mips 24@xmath1 m channel ( @xmath17 m ; @xmath18 m ) uses a 128@xmath13128 bib si : as array with a pixel scale of 255 pixel@xmath10 , providing a field of view of 54@xmath1354 . the scan map mode was used with the slow scan rate , which results in an integration time of 100 s pixel@xmath10 per scan pass ( 10 frames @xmath1310 s ) at 24 @xmath1 m . since the sky position angle of the scan direction is determined by the spacecraft roll angle , the observing dates were constrained such that the resulting scan map would extend along the position angle of the groth strip . the final scan maps cover a sky area of 2.4@xmath1310with integration times of @xmath19 s at 24 @xmath1 m . the 24@xmath1 m point source sensitivity ( 5@xmath16 ) is 70 @xmath1jy . the irac and mips basic calibrated data ( bcd ) delivered by the spitzer science center ( ssc ) include flat - field corrections , dark subtraction , linearity correction and flux calibration . the bcd data were further processed by each team s own refinement routines . these additional reduction steps include distortion corrections , pointing refinement , mosaicing and cosmic ray removal by sigma - clipping . source extraction was performed in the same way as described in ashby et al . ( 2006 ) . in brief , we used daophot to extract sources from both irac and mips images . with an fwhm of the point spread function ( psf ) of 1@xmath202@xmath21 for irac and 6@xmath22 for mips 24 @xmath1 m , virtually all objects at high redshifts are unresolved . we performed aperture photometry using a 3@xmath22 diameter apertures for both irac and mips sources . the aperture fluxes in each band were subsequently corrected to total fluxes using known psf growth curves from fazio et al . 2004 ; huang et al . the magnitudes presented in this work are all in the ab magnitude system . the egs lbg sample was constructed from the lbg catalogue of steidel et al . the catalogue was matched to the irac and mips source lists instead of doing direct photometry on the images . we searched for counterparts within a 2@xmath23 diameter separation centered on the lbg optical position . the typical size for an lbgs is about 1 and in most cases the lbg is clearly identified although in section 6.1 we discuss those cases where multiple counterparts were present . due to the resolution of 2 we do not anticipate finding many multiple component counterparts . steidel et al . ( 2003 ) have reported on the detection of 334 lbgs in the egs area . the spitzer egs survey covers 244 lbgs . about 200 lbgs are detected with irac at 3.6@xmath114.5 @xmath1 m , 53 lbgs are detected with irac at 5.8 @xmath1 m and 44 at 8.0 @xmath1 m . finally , 13 lbgs have counterparts in the mips 24 @xmath1 m survey of the field . of the initial sample of 244 lbgs observed by spitzer in the egs area 175 objects have confirmed spectroscopic redshifts and are identified as galaxies at z@xmath03 ( we stress that this number refers to objects classified as galaxies and that we exclude from our analysis those classified as agn , qso or stars ) . among the 13 lbgs with mips detections , 6 objects are classified as galaxies at z@xmath0 3 based on their optical spectra ( of the remaining mips detections 3 are classified as qso and 1 agn ) . in this paper we focus on the properties of 44 lbgs that have 8 detections ( referred to hereafter as `` the 8 lbg sample '' ) . of these , 26 have spectroscopic redshifts . the 8 lbg sample includes seven of the nine galaxies detected by mips and in particular the infrared luminous lbgs ( illbgs , huang et al . the criterion for an lbg to be classified as illbg is detection at 24 , which for the sensitivity of the egs survey ( @xmath070 @xmath1jy ( 5@xmath16 ) implies an infrared luminosity @xmath24 l@xmath4 . the two lbgs with mips 24 @xmath1 m detections but no 8 micron detection have not been added to the sample as their irac photometry is incomplete . in table 1 we list the ground based photometry for the 8 micron lbg sample . the ugrjk data and the spectroscopic redshifts come from the large compilation of steidel et al ( 2003 ) . in table 2 we list the spitzer photometry . the uncertainties in the irac @xmath11mips magnitudes were estimated from an analysis employing monte - carlo simulations . we added artificial sources to the images , extracted them in the same manner as real sources , and computed the dispersion between input and recovered magnitudes . the dispersion of the recovered magnitudes about the mean recovered magnitude , computed for each bin in input magnitude , were used as the uncertainty estimates . in figure 1 we plot the rest frame uv@xmath11optical@xmath11near - infrared seds of all lbgs in the egs area . the seds are normalised in r magnitude ( r=24 mag ) . frame uv ( ugr ) part of the spectrum shows relatively little variation . although k - band magnitudes are available for @xmath040% of the egs lbg sample we note a range of r k colors with a mean value of @xmath25r k@xmath26 1.5 . clearly , the addition of the spitzer irac@xmath11mips bands improves dramatically our understanding of the nature of lbgs . the rest frame near - infrared colors of lbgs are spread over 4 magnitudes . lbgs display a variety of colors ranging from those that are red with r-[3.6]@xmath272 to blue with r-[3.6]@xmath252 . the sed of those lbgs with blue colors is rather flat from the far uv to the optical with marginal ( in 1 band ) irac detections . on the other hand , the seds of the red lbgs are rising steeply towards longer wavelengths . it is worth noting , that a number of such red color lbgs display r-[3.6 ] values similar to the extremely red objects discussed by wilson et . 2004 ) . most of the illbgs display such extreme colors ( see discussion in section 7.2 and figure 1 of huang et al . the spitzer observations presented here show that a fraction of optically - selected lbgs ( @xmath030% ) have red colors and have a higher dust extinction and higher masses in contrast to the majority of the lbg population which comprises mostly of objects with blue colors modest extinction and masses . the seds of the spitzer detected z@xmath03 lbgs cover a wide range in wavelength from @xmath0900@xmath6 to 2@xmath1 m . in this section we discuss the models used to fit their seds in order to investigate and constrain the star formation histories , extinction , and masses of the lbgs . we use the bruzual & charlot code ( 2003 , hereafter bc03 ) to generate models in order to fit the lbg seds . the new bc03 models are based on a new library of stellar spectra with an updated prescription of agb stars . as suggested by the authors we adopted the padova 1994 stellar evolution tracks and constructed models with solar metallicity ( see discussion in shapley et al . 2004 ) and a salpeter intitial mass function ( imf ) extending from 0.1 to 100 m@xmath4 . we use the calzetti et al . ( 2000 ) starburst attenuation law to simulate the extinction . a major uncertainty in this type of analysis is the parameterization of the star formation history ( e.g. papovich et al . 2001 , shapley et al . 2001 , bundy et al , 2005 ) . for the present analysis we have considered mostly two simple single - component models : exponentially declining models of the form sfr(t ) @xmath28 exp(-t@xmath29 ) with e - folding times of @xmath30 = 0.05 , 0.1 , 0.5 , 2.0 , and 5.0 gyr and , continuous star formation ( csf ) models . although complex models , such as combinations of various star forming histories or single bursts on top of a maximally old underlying bursts ( as suggested by papovich et al . 2001 ) are probably more realistic , we do not consider them in this work as we can not constrain the model parameters easily . our main aim for each galaxy is to constrain the stellar population parameters . the fitted parameters are the following : dust extinction ( parameterised by e(b v ) ) , age ( t@xmath31 defining the onset of star formation ) , stellar mass ( m@xmath32 ) and star formation history ( @xmath30 ) . using bc03 we generate a grid of models with ages ranging between 1 myr and the age of the universe at the redshift of the galaxy in question and , varying extinction . for each set of e(b v ) , star formation history , and age we derive a model with the full set of colors , ugrjk plus the irac 3.6 , 4.5 , 5.8 and 8 @xmath1 m , placed at the redshift of the galaxy in question . the intrinsic model spectrum is then reddened by dust . the model sed is finally corrected for the intergalactic medium ( igm ) opacity . the predicted colors are then compared with the observed ones using @xmath33 minimization technique . the best - fit e(b v ) and age combination was chosen to minimize @xmath33 and the intrinsic sfr and stellar mass were determined from the normalization of the best - fit model to the measured sed . we estimated the confidence intervals for individual objects using monte carlo simulations . for each source we generated about 200 synthetic model seds of the data by varying the fluxes randomly ( random values where chosen according the the gaussian distribution of the measured uncertainties ) . we then repeated the fitting procedure for each new set and determined the 68% confidence intervals from the distributions of the best - fit values obtained with each set . the procedure used here is similar to the one used recently by papovich et al . ( 2006 ) for drgs , shapley et al . ( 2001 ) and papovich et al . ( 2001 ) for lbgs . we note a strong assymetry around best - fit values which we interpret as the presence of strong degeneracies especially between e(b v ) and t@xmath31 ( as we discuss in section 5.2 ) . we note however that , the inferred stellar mass is one of the best constrained quantities suffering much less from uncertainties involved with the specific star formation history used to describe a specific lbg . in what follows we discuss how varying properties of the model fits affect the best - fit sed parameters . the impact of different extinction laws has already been investigated by e.g. papovich et al ( 2001 ) , dickinson et al . ( 2003 ) who found the effect to be overall small . for the present work we have adopted the calzetti law since such a law reproduces the total sfr from the observed uv for the vast majority of lbgs ( e.g. reddy & steidel 2004 , reddy et al . , 2005 , nandra et al . 2003 ) . the choice of the calzetti law was also dictated by the desire to facilitate comparison with previous work in the field . as discussed in 4.1 we have chosen a salpeter imf for our stellar population models . in this section we investigate the effect on the models when using different imfs . a scalo ( scalo 1986 ) or a miller - scalo ( miller & scalo 1979 ) imf are also consistent with the data , as long as we keep the model lower and upper mass cutoffs fixed at 0.1 and 100 m@xmath4 , respectively . both the scalo and the miller - scalo imfs however , result in slightly younger ages than the ones obtained here using salpeter imf . a chabrier imf ( chabrier 2003 ) , also behaves very similarly to the salpeter imf . both of them have a very similar upper mass dependence . however , in the chabrier imf the low - mass end assumes a flatter behaviour , following a log - normal distribution . this imf results in m@xmath11l ratios that are a factor of @xmath01.5 smaller than those for a salpeter imf . overall the exact values of the upper@xmath11lower mass cutoffs have a more noticeable effect on the derived stellar masses . the major difference lies in the relative contribution of stars with m@xmath25m@xmath4 . a lower mass cutoff of 1m@xmath34 would , for instance , result in stellar mass values that are lower by about 3040% than those derived when the lower mass cutoff is set to 0.1m@xmath4 ( assuming a salpeter imf ) . the result would be more pronounced ( i.e. the estimated stellar mass will be lower ) if we use a chabrier ( 2003 ) imf instead . bell et al . ( 2003 ) seem to favour a `` diet '' salpeter imf ( a salpeter imf which is deficient in low mass stars ) which in the local universe provides a satisfactory explanation for the m@xmath11l ratios derived from colors . since the stellar content of @xmath35 galaxies and in particular the fraction of low - mass stars is presently unknown we do not investigate variations in the lower mass cutoff any further . so far , information on element abundances in lbgs is rather limited . pettini et al . ( 2002 ) determined element abundances in cb58 , a typical l@xmath32 galaxy which benefits from a factor of 30 magnification , and found it to be @xmath0 0.25 z@xmath4 . nagamine et al . ( 2001 ) suggested that near solar metallicities are in fact common in z@xmath03 galaxies with masses greater than 10@xmath3 m@xmath4 which is broadly consistent with the results for cb58 . shapley et al . ( 2004 ) also argued for solar metallicities for z@xmath03 lbgs . because 8 -detected lbgs are likely to be more massive than the typical lbg at @xmath36 , we used solar metallicity in the models . reducing metallicity to half solar would decrease the derived masses by 1020% . besides photometric uncertainties the parameters derived from sed fitting suffer from systematics and are subject to degeneracies simply because the models can not fully constrain the star formation history of a high redshift galaxy . the uncertainties are model - independent and plague even the simplest single - component models that we use in the present work . this is due to the fact that model parameters such as extinction and star formation are strongly dependent on the value of @xmath30 used to parameterise the star formation history . the extinction e(b - v ) is strongly dependent on @xmath30 and the exact prescription used to describe the star formation history . for example a decaying star formation model with a lower e(b - v ) will produce the same g - r and r - k colors as a model with a larger @xmath30 , smaller @xmath37t@xmath31 but a higher e(b - v ) . for constant star formation histories extinction is responsible for reddening the uv part of the spectrum which is made up of a mixture of stars of earlier types . likewise , the inferred age , t@xmath31 , for a given value of @xmath30 depends on the strength of the balmer break . the latter is sensitive to the relative number of b , a and f stars with respect to late type stars and the caii h and k absorption at 4000 @xmath6 ( which is determined by the relative number of late type stars ) . a larger value of @xmath30 corresponds to older ages for the o and a stellar populations . finally , the sfr also suffers from uncertainties since it is primarily derived from the uv slope which will correspond to different e(b - v ) and dust attenuation factor depending on the assumed stellar population and extinction law . the uncertainties we just discussed are of course likely to affect the inferred stellar masses but in a less dramatic way . even in the case of single star formation histories ( csf or exp ) the derived stellar masses vary by a factor of @xmath010% for different values of @xmath30 . uncertainties in extinction and age also influence the derived masses but , it is quite difficult to disentangle the effect each of these have on the masses . a highly extincted underlying stellar population will have a similar effect on the stellar mass as an older population . as we show in section 5 , the addition of rest - frame near - infrared photometry from spitzer has helped to constrain the dust properties of 8 micron selected lbgs . uncertainties in the mass estimates become more serious when one assumes more complex star formation histories . papovich et al . ( 2001 ) , for instance , introduced underlying maximally old bursts with t@xmath38@xmath30 and found an increase in the m@xmath11l by a factor of several without a noticeable effect on the uv colors . glazebrook et al . ( 2004 ) introduced random bursts in the range of star formation histories used to fit their multiband photometry and found that the mass increased by about a factor of 2 although the amount by which the stellar mass is underestimated depends on the specific galaxy sed . we conclude that use of the simple star formation histories ( csf or exp ) is likely to provide a lower limit on the stellar mass estimate . finally , it is worth noting that stellar population models including the thermally - pulsating asymptotic giant branch ( tp - agb ) phase ( maraston 2005 ) can also be used to fit the seds of z@xmath03 galaxies . the signature of the tp - agb phase is quite prominent for ages 0.22 gyrs and as van der wel et al . ( 2005 ) points out they provide a better fit for the seds of their z@xmath01 galaxies . we defer modelling of our z@xmath03 lbgs with models including the tp - agb phase for future work . using the stellar population synthesis models described in section 5.1 we have estimated the stellar masses for the entire sample of 181 lbgs with confirmed spectroscopic redshifts . figure 2 shows a histogram of the inferred stellar masses for the whole spitzer lbg sample . from this sample we selected lbgs with 8 micron detections and carried out a detailed study of their properties . in figure 3 we show the rest frame uv optical near - ir seds together with best - fit models for the 8 micron lbg sample ( 26 objects with confirmed spectroscopic redshifts ) . in table 3 we report the results from the best - fit stellar population models for each galaxy for the entire optical irac sed . parameters are listed for the best - fit model either csf or exponentially decaying models , whichever gives the lowest value of @xmath33 with @xmath30 as a varying parameter . the stellar mass inferred from the best fit model along with the fit parameters for each object are reported in table 3 . the median age for our sample is t@xmath31 = 700 myr . more than 50% of our galaxies have t@xmath39 500 myr while only 15% have t@xmath40 100 myr . as we discussed already in section 3 , the 8 micron lbg sample contains most ( all but two ) of the illbgs ( huang et al . illbgs appear to have red colors ( r-[3.6]@xmath273 ) and best - fit ages t@xmath411000myr . using best - fit csf models we compute the stellar masses of the illbgs and find them to be of the order of ( m@xmath2710@xmath42m@xmath4 ) . extinction for these massive lbgs ( parameterised by e(b v ) ) is around 0.3 . the fits of the illbgs are discussed in detail in section 7.1 . the inferred stellar masses quoted in table 2 do of course suffer from the uncertainties discussed in section 4.6 and are quantified with the error bars reported in the same table . the median stellar mass for the 8 micron lbg sample is ( 8.16@xmath431.04)@xmath1310@xmath3 m@xmath4 compared to ( 2.95@xmath431.51)@xmath1310@xmath3 m@xmath4 estimated for the entire egs - lbg sample ( and excluding the 8 micron selected lbgs ) . the fraction of lbgs with m@xmath275@xmath1310@xmath3m@xmath4 among the 8 micron lbgs is 40% compared to 25% for the entire egs - lbg sample . finally , it is also worth investigating the morphology of the 8 micron lbg sample . this investigation was carried out to assess at what level the irac fluxes might suffer from contamination from neighbouring sources . this is important as the resulting fluxes affect the modelling and the computation of stellar masses . for this reason we have compared the 8 micron irac images with deep r band subaru images ( miyazaki , priv . comm ) . as we discussed in section 2 the irac psf is @xmath02@xmath23 so we are looking for multiple sources within a 2@xmath23 area . out of the 26 sources in the 8 micron lbg sample we have identified two cases where a possible contamination of the irac 8 micron flux by a neighbouring source is possible . in figure 4 we show cutouts of the optical r - band images for westphal - m38 and westphal - md99 . while we include these sources in the discussion we caution that the reported fluxes might be contaminated . in this section we try to investigate the dust content of the 8 micron lbg sample based on the e(b v ) values derived from the best - fit models . in figure 5 we plot the estimated stellar mass ( based on best - fit csf models ) as a function of best - fit e(b v ) for the 8 micron lbg sample . we chose to plot e(b v ) against the stellar mass which , as we discussed extensively in section 4 , is relatively insensitive to the assumed star formation history . for comparison we also plot the same parameters for the lbg - egs sample ( i.e. all lbgs with detections in at least one irac band ) . the mean e(b v ) for the lbg - egs , the 8 micron lbg and the illbg samples is 0.156 , 0.232 and 0.354 , respectively . the more massive lbgs have , on average , higher e(b - v ) though with a considerable scatter . we confirm that 8 micron lbgs suffer higher extinction ( parameterised by the model value of e(b v ) ) when compared to the spitzer lbg sample ( ie those without 8 micron detections ) . moreover , extinction is higher for the illbgs although our sample is at present still small ( 6 objects ) to draw statistically significant conclusions . another way to investigate the amount of obscuration in lbgs is to look at the 24 micron detections ( ie the illbgs ) . out of the 188 lbgs in the lbg - egs sample only six of them are detected at 24 microns that is a fraction of @xmath03% . however , the percentage of 24 micron detections among the 8 micron lbg sample is higher 6 out of the 26 lbgs which makes up for 23% . we thus conclude that spitzer observations provide reliable means to probe the dust content of lbgs . in particular objects with spitzer 24 micron detections must be in the high extinction end of the dust distribution . the addition of longer wavelength photometric points ( ie the irac channels ) has an effect on the accuracy of stellar mass estimates ( e.g. labbe et al . 2005 , see also discussion in section 7 ) but also on our understanding of the properties of the lbg population as a whole . here , we investigate the distribution of stellar mass with rest frame wavelength as we move from optical to the near - infrared bands . in figure 6a we show the distribution of stellar masses as a function of absolute [ 3.6 ] @xmath1 m magnitude which at the median redshift of 3 corresponds to rest frame 0.9 @xmath1 m ( ie i band ) . although there is clearly a correlation between absolute i - band magnitude and stellar masses ( correlation coefficient r=0.53 ) there is considerable scatter in the values especially at the fainter end . at the brighter end ( which is where most of the massive galaxies are found ) the correlation becomes tighter but this might also be due to the smaller number of sources at these magnitudes . we suggest that the spread in the correlation is due to the wide range of star formation histories among lbgs . moreover , it is apparent that it is quite difficult to project a single i - band rest frame luminosity to a stellar mass . in figure 6b we show the distribution of stellar mass as a function of absolute [ 5.8 ] @xmath1 m magnitude which at z=3 corresponds to rest frame 1.4 @xmath1 m ( ie h band ) . the correlation between stellar mass and absolute rest frame h - magnitude improves significantly ( r=0.61 ) . the scatter in stellar mass decreases as one moves to longer wavelengths and probes the stellar luminosity due to recent@xmath11slightly older star formation activity . likewise , in figure 6c we plot the distribution of stellar masses as a function of absolute [ 8.0 ] @xmath1 m magnitude which would correspond to rest frame @xmath02.0 @xmath1 m ( i.e. k band ) . the correlation between stellar mass and magnitude becomes even tighter with r=0.77 . the scatter in m@xmath11l values decreases and is now a factor of @xmath012 . the values of m@xmath11l for the most massive galaxies show a spread of about 10 . from these simple comparisons , we conclude that the mid infrared bands and especially the irac 8 @xmath1 m channel ( which samples the rest frame near infrared wavelengths for z@xmath03 lbgs ) provides a more accurate estimate of the m@xmath11l ratio compared to that obtained when using optical bands . as we discussed already , irac 8 @xmath44 channel is sensitive to the light from the bulk of the stellar activity accumulated over the galaxy s lifetime ( see also bell & de jong 2001 ) . we finally examine the dependence of stellar mass on the r[3.6 ] color index . as we discussed already in section 5 , the massive lbgs tend to have red colors with r[3.6]@xmath272 . in figure 6d we plot stellar masses as a function of the r[3.6 ] color . the entire egs lbg sample ( with confirmed spectroscopic redshifts ) shows a wide range of r[3.6 ] color with the 8 micron lbg members showing colors in the range 1 to 4 . the most massive 8 micron lbgs ( which are also illbgs ) show the most extreme r[3.6 ] colors @xmath272.5 . these values are close to the values of extremely red objects presented by wilson et al . the correlation is tighter for the more massive@xmath11redder lbgs which have m@xmath11l close to that of present day galaxies . all these results stress the importance of obtaining longer wavelength ( rest frame near - infrared ) observations for the z=3 lbgs , as these allow us to probe the effect of recent star formation on the galaxy luminosities and masses . although proper knowledge of the current stellar content is crucial , as we have discussed , it is the luminosity of older stars , that dominates the k band light , that is found to correlate best with the derived stellar mass . in particular , this correlation becomes significant for the newly discovered illbgs and the massive 8 micron sample lbgs which have a significant dominant older stellar population . as we move to the other extreme of young and therefore less massive galaxies the current stellar content becomes perhaps more important as it dominates the luminosity and thus the stellar mass . all the above correlations have been based on the stellar masses inferred from the models described in section 3 . to some extent the results are dependent on the particular details of the models assumed , in our case a single exponentially decaying or continuous episodes of star formation . our models are , however , rather `` conservative '' in terms of the predicted stellar masses . if we use a two component model ( an underlying old stellar episode and a very young continuous episode ) this will result in a relative increase in the derived stellar masses which will be higher for the lower mass galaxies in comparison to the higher mass galaxies ( papovich et al . 2001 , shapley et al . this will likely make the trends seen in figure 6 shallower but will not significantly affect our conclusions . the original selection of lbgs was based purely on optical colors ( steidel & hamilton 1992 ) thus only probing the rest frame uv part of their spectrum . shapley et al . ( 2001 ) and papovich et al . ( 2001 ) obtained near - infrared photometry thus extending the available seds into the rest frame optical . with the present data we have extended the seds of lbgs into the rest frame near - infrared and have the opportunity to probe the bulk of the stellar emission . as we discussed already in section 3 about 1@xmath114 of the egs - lbg sample has been detected with irac [ 8.0 ] ( that probes the rest frame k band ) . in this section , we want to compare the properties of the 8 micron lbg sample to those of optical@xmath11near infrared selected samples such as the nirc sample ( shapley et al . 2001 ) . there are 16 lbgs in common between the nirc ( shapley et al . 2001 ) and the entire lbg - egs sample while five of these are part of the 8 micron lbg sample . of these five targets we chose westphal - d49 to carry out a detailed comparison . the best - fit csf model for d49 ( table 3 ) has an age of t@xmath31=1350 myr and an e(b v ) = 0.34 . in the absence of the spitzer photometry shapley fit d49 with a csf model with 1139 myrs and e(b v)=0.17 in figure 7 we want to compare the two models . it is obvious that in the absence of the spitzer data both models would provide a good fit to the rest frame uv optical photometry . it is only with the addition of the rest frame near infrared ( spitzer data ) that a model with higher extinction becomes necessary . one could of course argue for a more `` aged '' and less extincted stellar model . likewise , we have generated models where we varied the age of the stellar population ( figure 8) . it is obvious that stellar ages larger than 1.3 gyrs do not provide a satisfactory fit to the sed . a similar exercise was performed for the remaining objects that are in common between the nirc and our 8 micron lbg sample . for objects without 8 micron detections ( 11 lbgs ) we find a very good agreement in the models ( in particular age and extinction ) . the mean age of the nirc sample is 355 myrs compared to a mean age of 320 myrs derived including the irac detections . likewise the mean extinction derived for the nirc sample is 0.218 whereas our mean extinction is 0.208 . the differences are more pronounced however , for lbgs with 8 micron detections . out of the small sample of 5 lbgs we derive mean age and extinction of 400 myrs and e(b - v)=0.378 , while the corresponding values for the nirc fits are 293 myrs and 0.238 . for two of the five targets in common between nirc and 8 micron samples ( md109 and md115 ) shapley et al . could not constrain the best - fit ages and thus fixed the age at 10 myrs . figure 9 summarizes our results . the overlap between the nirc sample and the 8 micron lbg sample is small so the differences might not be very significant . we suggest , however , that it is likely that lbgs with 8 micron detections have higher extinctions ( see also discussion in 5.1 ) and slightly more aged population than lbgs undetected at 8 micron . the possibility for 8 micron contributions from a putative agn can not be constrained with the current data although optical spectroscopy ( steidel et al . 2003 ) or x - ray studies ( e.g. nandra et al . 2002 ) do not reveal any signs of such activity . a subset of the 8 micron lbg sample ( 12 objects out of 44 ) have masses m@xmath4510 @xmath42 m@xmath4 . from this subset 1 object is a confirmed qso and six objects do not have spectroscopic confirmation . in section 5 we discussed in details the various models used and the accuracy in the derived stellar masses . in this section we focus the discussion on these `` massive '' 8 micron lbg sample galaxies with confirmed spectroscopic redshifts . these massive 8 micron lbgs are also detected in the mips24 @xmath1 m band . there are two additional galaxies that do not have 8 micron counterparts but are detected in the 24 mips band , we will consider those as part of the massive luminous lbg sample . because of their luminosity and rather unusual properties ( compared to the rest of the lbgs ) huang et al . ( 2005 ) termed this subclass of objects as infrared luminous lbgs ( illbgs ) . the masses of the illbgs have been estimated using bc03 models and are reported in table 2 . all illbgs are best fit by a csf model with ages @xmath27 1000 myr with extinction that varies around a@xmath460.2 . although , as we discussed , the addition of the irac bands has significantly improved the accuracy of the derived stellar masses , the models do not constrain very well the star formation history of high - z and lbg objects ( as has been discussed already by papovich et al . 2001 , shapley et al . 2004 , 2005 ) . despite the uncertainties in the detailed properties of the star formation history the derived stellar mass is a robust estimate depending mostly on the sed shape rather than the details of the particular model . however , we will show that in the case of these `` massive '' luminous lbgs we can , for the first time , discriminate between the two `` simple '' models ( csf vs exp ) used to fit the seds of the lbgs . we caution that although the star formation history of galaxies is a more complicated process than a constant star formation activity or a a smooth exponential function the details of more complicated models ( short bursts of different strength superimposed on a constant star forming model ) can not be well constrained so we do not consider them here . as we have shown in section 6 the correlation between estimated stellar mass and absolute magnitudes becomes tighter as one moves to longer wavelengths especially for the brighter more massive objects . an accurate estimate of the total stellar mass can be obtained when the star formation of the past several hundred myrs is less than the integrated star formation over the lifetime of a galaxy . in a more realistic case , a good estimate of the stellar mass can be obtained when the light from the optical and more importantly near - infrared wavelengths far exceeds the uv light . the massive illbgs considered here represent exactly this phase in galaxy evolution : their light is dominated by the light of the older stars which output mainly in the near - infrared bands ( observed frame around 8 @xmath1 m covered by irac [ 8.0 ] channel ) and exceed the light in the uv ( due to the more recently formed stars ) . this is in line with the large m@xmath11l@xmath47 found for the massive lbgs . the last point to be addressed is the issue of the exact prescription of star formation history used to estimate masses for the illbgs . the discussion in section 6 is of particular relevance . there , we discussed how the luminous ( and massive ) lbgs show the most extreme ( very red ) r-[3.6 ] and g - r colors . through our detailed modelling we find that csf models with an extinction e(b v)@xmath00.3 , reproduce quite well the extreme colors of illbgs ( see also figure 2 in huang et al . 2005 ) . in models with declining star formation histories red g - r and r[3.6 ] colous can be produced either with t@xmath481 and little or no dust extinction or with smaller t@xmath49 and more dust extinction . in such models the reddening of the uv slope is caused by a mixture of stars of later spectral types . in continuous star formation models reddening of the rest - frame uv is due solely to extinction . the fact that illbgs are detected at 24@xmath1 m implies that they are quite dusty so that the constant star formation model provides a natural match to the extremely red colors observed . since ( observationally ) we can not constrain the spectral types of late type stars in lbgs we suggest that the csf model is a better ( simpler ) explanation for the extremely red colors displayed by illbgs . we now want to investigate the connection between the massive luminous z@xmath03 lbgs and the smgs . both lbgs and smgs have sufficient sfrs to form present - day elliptical galaxies . until recently , attempts to find a connection between the two populations had not been very successful ( webb et al . 2003 , chapman et al . 2000 ) although more recent studies ( e.g. chapman et al . 2005 , reddy et al 2005 ) have been exploring a possible link . naturally , most of the submm searches targeted lbgs with high predicted sfrs . these sfrs , however , have been estimated based on the relationship between far - infrared@xmath11far - uv flux and uv continuum for local starbursts ( adelberger & steidel 2000 ) ; this relationship however , does not apply to ultraluminous galaxies where a greater fraction of the total star formation is in optically - thick regions . another issue is of course the detailed properties of the dust distribution in lbgs which may differ from that of present - day evolved galaxies ( takeuchi 2003 ) . an obvious evolutionary scenario is one in which massive lbgs and smgs form a continuum of objects with the bright submillimeter - selected sources representing the highest star - forming lbgs . the current sample of massive lbgs has many properties in common with smgs . for one , as we discussed in section 6 their r k colors are extreme ( @xmath273 ) similar to those of smgs . indeed the only lbg detected in the submm so far , westphal -mmd11 , by chapman et al . ( 2000 ) shows such extremely red colors . the spitzer colors of massive lbgs are also very similar to those of smgs . egami et al . ( 2004 ) reported on spitzer detections of smgs in the lockman hole . based on the observed seds of the smgs and from comparisons with local templates they classified smgs as warm ( mrk231-like sed ) and cold ( arp220-like sed ) . they found that the majority of smgs are cold . more recently , ashby et al . ( 2006 ) presented a detailed study of the mid - colors of smgs detected in the egs area . using these measurements as a reference point , we plot in figure 10 ( a and b ) the r-[3.6 ] and [ 8.0]-[24.0 ] colors for smgs and illbgs and find that they are indeed quite similar . for the smgs we used the 17 candidates with secure 8 @xmath1 counterparts ( see note in table 1 of ashby et al . finally , it is worth looking at the inferred stellar masses of smgs and lbgs : both populations appear to have large estimated stellar masses ranging from several @xmath1310@xmath3 to 10@xmath42 m@xmath4 ( e.g. borys et al . 2005 , tecza et al . we caution that stellar mass estimates for smgs are relatively uncertain due to the likely presence of an agn component in some members . the values we quote are based on the estimates by borys et al . ( 2005 ) for relatively young and gas - rich smgs as well as the dynamical masses measured by tecza et al . based on the similarities of the colour indices , mid - infrared colors and inferred stellar masses we propose that a link must exist between illbgs and smgs . to further investigate the possible connection between the two populations let us also take a look at some sample statistics . huang et al . ( 2005 ) found that illbgs represent about 15% of the total lbg population , excluding objects with possible contamination from agn . this number most likely represents a lower limit to the true fraction of illbgs among the general lbg population for two reasons:(a ) the mips depth is not uniform over all of the steidel et al . ( 2003 ) fields and ( b ) we do not have spectroscopic confirmations for the whole illbg sample . recently , chapman et al . ( 2005 ) presented spectroscopy of 73 smgs with faint radio emission . the median redshift of the chapman et al . sample is 2.2 with 19 objects with redshifts z@xmath502.5 . although the redshift distribution of the smgs does not match that of the z@xmath03 lbgs we will use the chapman sample as is the largest compilation of measured spectroscopic redshifts for smgs ( though we have to make some adjustments to allow for the slight difference in the mean redshifts ) for our comparisons . chapman et al . find that 30 of the 73 smgs with confirmed spectroscopic redshifts have rest - frame uv - characteristics similar to those of star forming galaxies . among the 19 z@xmath272.5 smgs , 6 of them have starburst - like uv colors ( similar to those of lbgs ) . thus , we estimate that @xmath030% of smgs are star - forming galaxies ( with uv colors matching those of lbgs ) with this number likely to be a lower limit ( due to unidentified objects which are unlikely to show agn signatures ) . among the egs - lbg sample , on the other hand , huang et al . found that 15% of lbgs ( the so - called illbgs ) show evidence for increased dust extinction ( via their 24 @xmath1 m detections ) . very recently we confirmed detection of illbgs at mm wavelengths with iram@xmath11mambo ( rigopoulou et al . 2006 , in prep ) at a flux level similar to that of faint smgs . bringing these statistics together 30% of smgs display lbg - like uv - colors while 15% of lbgs show evidence for higher extinction and submm emission . thus , it is likely that illbgs ( with red colors , higher a@xmath51 , submm emission ) and a sub - fraction of smgs ( those with uv colors similar to starburst galaxies ) appear to be identical in their uv and@xmath11or submm properties , this is likely to happen at the 23 mjy level . it is possible that smgs fainter than the current survey limits ( 2 - 3 mjy ) will have lower star formation rates and lower levels of dust obscuration , such that the overlap fraction between smgs and lbgs would become increasingly larger as one goes to fainter flux levels . we suggest that an evolutionary or physical link between these two subsamples is likely to exist . as a last point , we want to address the connection ( if any ) of illbgs with the drgs ( franx et al . foerster - schreiber et al . ( 2004 ) and labbe et al . ( 2005 ) compared the properties of drgs and blue lbgs and found that the former are , not surprisingly , on average dustier and more massive . when examining the m@xmath11l@xmath52 mass - to - light ratio they find that drgs display higher values ( the mass - to - light ratios remains of course critically dependent on the assumed star formation history , metallicity and imf values ) than the lbgs . the mean age and extinction of drgs is @xmath25t@xmath27=1.3 gyr and @xmath25a@xmath531.5 mag . from the sed fits in figure 2 of labbe et al . ( 2005 ) we conclude that illbgs resemble more the `` old and dusty '' drgs rather than the `` dead '' drgs . recently , it has become apparent that the number densities of high - redshift galaxies with large stellar masses exceed the theoretical predictions of semi - analytical models of galaxy evolution ( e.g. saracco et al . 2004 , daddi et al . 2004 , tecza et al . 2004 ) . in the widely accepted hierarchical merging scenario for galaxy formation , massive galaxies are the end points of mergers which increase with cosmic time . in the semi - analytical models of kauffmann et al ( 1999 ) , the predicted number density of @xmath54 @xmath55 ellipticals at @xmath56 is about @xmath57 mpc@xmath58 . a similar conclusion is reached by moustakas & somerville ( 2002 ) . the semi - analytic models of galaxy formation in the hierarchical clustering scenario by baugh et al . ( 1998 , 2002 ) predict , in fact , no such massive galaxies at redshift z@xmath272 . in figure 11 we plot the evolution of the number density @xmath59 based on the massive lbgs ( assuming an @xmath60 , @xmath61 cosmology with h@xmath9=70 km s@xmath10 mpc@xmath10 ) as a function of redshift . if we assume an effective volume for the u - dropouts v = 450 h@xmath58 mpc@xmath62 arcmin@xmath63 ( taken from from steidel et al . 1999 ) then , the effective co - moving volume becomes v = 1400 mpc@xmath62 arcmin@xmath63 ( the volumes have been weighted according to the number of objects per r - magnitude bin ) . for the 8 micron lbgs with m @xmath2710@xmath42 m@xmath4 in the 227 sq . arcmin westphal field , we derive a co - moving density of @xmath64 mpc@xmath58 at the average redshift @xmath65 . the derived number density is an actual underestimate of the real number density of galaxies with @xmath66 @xmath55 at @xmath67 since our calculation refers only to the 8 micron selected massive lbgs . although we are fairly confident that the 8 micron lbg sample contains _ all _ massive lbgs it does not , of course , account for the optically faint massive population at z@xmath03 . however , deriving @xmath68 for other @xmath67 objects ( e.g. the drgs of franx et al . ) is a difficult task due to the very limited availability of measured spectroscopic redshifts for such targets . to date lbgs consitute by far the largest @xmath69 sample with measured spectroscopic redshifts and thus we confined our calculations of the number density to this sample . the current value of @xmath68 is about a factor 3 higher than the predictions of hierarchical models by e.g. kauffmann et al . ( 1999 ) . in figure 11 we also plot the co - moving densities of galaxies at z@xmath00 ( cole et al . 2001 ) , z@xmath01 ( drory et al . 2004 ) , z@xmath02 ( fontana et al . 2004 ) . in the same plot we show the theoretical predictions from semianalytical models of kauffmann et al . ( 1999 ) , and baugh et al . it is evident from figure 11 that the evolution of the number density of massive ( @xmath70 @xmath55 ) galaxies with redshift is slower than the prediction of the current hierarchical models , at least in the redshift range 0@xmath25z@xmath253 . we note , however , that the number densities shown in figure 11 are based on a variety of surveys with different selection techniques . this , undoubtedly , introduces selection biases . although disentagling the various biases is beyond the scope of the present work , we stress that despite the selection effects the main conclusion regarding the number density @xmath68 of z@xmath03 galaxies remains unaffected . the fact that these galaxies have already assembled this mass at z@xmath03 , places the possible merging event of their formation at z@xmath713.5 assuming a dynamical time scale of 3@xmath1310@xmath72 yr ( e.g. mihos & hernquist 1996 ) . the inferred mass weighted age of the stellar populations places the formation of their bulk at z@xmath73 3.5 implying substantial activity at such high redshifts . ferguson et al . ( 2002 ) postulated that the assembly of massive systems at z@xmath03 can happen if star formation in lbgs is episodic and proceeds with a top heavy imf . another possibility is that the bulk of the stellar mass was formed during a single star formation episode with large @xmath30 ( modeled as a single burst with large @xmath30 resembling conditions similar to a csf model ) . the addition of long wavelength spitzer@xmath11irac data reveals , for the first time , that lbgs are rather inhomogeneous in their rest frame near - infrared properties . while very little scatter is observed in the optical ( rest frame uv properties ) , irac has revealed the existence of a distinct class of ( rest frame ) near infrared luminous lbgs whose properties deviate from those of typical blue less massive lbgs . using the bc03 stellar population synthesis code we have generated simple models with exponentially decaying and@xmath11or constant star formation histories to fit the lbg seds and estimate stellar masses . while in most cases both star formation models provide a good fit to the lbg seds , there is a specific sub - group of lbgs ( luminous at 8 microns ) that are best fit by a constant star formation history . this subgroup of lbgs are also luminous in the mid ir and are detected by mips 24 micron channel . incidentally , these are among the most massive z@xmath03 lbgs with estimated stellar masses in excess of 10@xmath42m@xmath4 . we have compared the properties of the 8 micron lbg sample to the near - ir selected sample of shapley et al . the 8 micron lbgs form a slightly different subclass of `` redder '' lbgs . the results of their sed modelling reveals that 8 micron lbgs are dustier ( with marginally older ages ) and also have higher masses . the need for a higher dust extinction is only apparent when the irac data are included in the fit . while the addition of the irac data has improved on the accuracy of the estimated stellar masses the biggest gain has come from the tight correlation between the absolute [ 8.0 ] micron magnitude and the infrerred stellar mass . the [ 8.0 ] micron magnitude ( rest frame k band ) correlates well with the stellar mass estimates and that in turns implies that the observed mid infrared absolute magnitudes `` trace '' the stellar mass much better than the optical magnitudes . we infer an average m@xmath74l of @xmath0 0.2 although the values for the most massive galaxies show a spread of about 10 . it is interesting to note that the massive lbgs also display `` red colors '' ( index r[3.6 ] ) similar to those found by wilson et al . ( 2004 ) and to the colors of submm - galaxies ( e.g. egami et al . 2004 ) . with such high masses and star formation rates lbgs along with submm galaxies could well be the progenitors of today s ellipticals . it is thus natural to investigate the connection between the two population from a mid infrared point of view . the connection between lbgs and submm galaxies has been addressed indirectly by chapman et al . ( 2005 ) . however , it is only with the availability of the irac observations ( see figure 1 ) that a proper comparison can be made . huang et al . ( 2005 ) have addressed this issue at length . given the ever increasing number of high redshift galaxies with high masses ( stellar or dynamical ) it is of interest to investigate the number density for these massive , high - redshift systems . in the semi - analytical rendition of most hierarchical models for structure formation the predicted number density of massive m@xmath75 galaxies decreases with redshift . the number density we derived based on the 8 micron lbg sample ( with m@xmath75 ) is @xmath76 mpc@xmath58 is at least a factor of 3 higher than semi - analytical predictions . if we compare the density of local l@xmath71l@xmath32 galaxies with our estimate we find that their density can not decrease by more than a factor of 3 from z=0 to z=3 thus we conclude that a significant fraction of the stellar mass in local massive galaxies must have been in place at z@xmath03 . we thank the anonymous referee for his / her suggestions on improving the manuscript . we thank s. miyazaki for making the subaru r - band image available . this work is based on observations made with the spitzer space telescope , which is operated by the jet propulsion laboratory , california institute of technology under nasa contract 1407 . support for this work was provided by nasa through contract number 1256790 issued by jpl . dr acknowledges support from the leverhulme trust via a research fellowship grant . baugh , c.m . , benson , a. j. , cole , s. , frenk , c. s. , lacey , c. g. , 2003 , in the mass of galaxies at low and high redshift . proceedings of the eso workshop held in venice , italy , 24 - 26 october 2001 , p. 91 . rrrrrrr c10 & 28.22 & 25.55 & 25.33 & & & 3.053 + m30 & 28.09 & 25.62 & 24.57 & & & 3.380 + md9 & 26.65 & 25.18 & 24.9 & & & 3.035 + d29 & 27.43 & 25.14 & 24.82 & 24.12 & 23.78 & 3.245 + d49&25.95 & 23.67 & 23.5 & 23.07 & 21.976 & 2.808 + m63 & 27.8 & 25.78 & 24.82 & & & 3.101 + c49 & 28.03 & 25.7 & 24.93 & & & 2.932 + c69 & 27.68 & 25.08 & 24.24 & & & 2.948 + c15 & 27.84 & 25.13 & 24.07 & & & 3.024 + c47&27.67 & 25.02 & 24.44 & & & 2.973 + c50 & 27.63 & 24.63 & 23.96 & & & 2.910 + c58 & 27.97 & 25.00 & 24.51 & & & 2.747 + c76&27.54 & 24.12 & 23.33 & & & 2.876 + d40&27.5 & 24.44 & 23.67 & & & 2.958 + d55&26.54 & 24.1 & 23.66 & & & 2.994 + m28&27.42 & 25.49 & 24.65 & & & 2.903 + m38 & 27.9 & 25.75 & 24.86 & & & 2.928 + m41&28.14 & 26.09 & 25.4 & & & 3.396 + m43&27.96 & 25.77 & 24.78 & & & 3.351 + md23&26.91 & 25.04 & 24.22 & & 23.23 & 2.862 + md91&26.23 & 24.33 & 23.84 & 23.63 & 22.93 & 2.740 + md97 & 25.3 & 23.81 & 23.43 & & & 2.496 + md98 & 27.73 & 25.57 & 24.64 & & & 3.119 + md99&27.55 & 25.06 & 24.02 & & & 3.363 + md109 & 26.62 & 24.78 & 23.94 & & 22.71 & 2.719 + md115 & 26.27 & 24.62 & 23.97 & & 24.17 & 3.208 + rrrrrrrrrrrr c10 & 24.83@xmath430.31 & 24.68@xmath430.39 & & 23.95@xmath430.12 & + m30 & 23.04@xmath430.14 & 22.67@xmath430.12 & & 22.19@xmath430.12 & 19.35@xmath430.45 + md9 & 22.40@xmath430.07 & 22.30@xmath430.09 & 22.42@xmath430.31 & 22.77@xmath430.12 & & + d29 & 23.49@xmath430.20 & 23.08@xmath430.16 & & 21.75@xmath430.12 & + d49 & 21.03@xmath430.03 & 20.72@xmath430.05 & 20.53@xmath430.07 & 20.39@xmath430.06 & 18.53@xmath430.46 & + m63 & 22.79@xmath430.11 & 22.79@xmath430.12 & & 22.69@xmath430.12 & + c49 & 22.33@xmath430.07 & 22.16@xmath430.08 & & 21.78@xmath430.12 & + c69 & 22.68@xmath430.10&22.63@xmath430.11 & & 22.63@xmath430.12 & + c15 & 22.78@xmath430.11 & 22.97@xmath430.14 & & 22.13@xmath430.12 & + c47 & 22.79@xmath430.11 & 22.68@xmath430.11 & & 22.48@xmath430.12 & + c50 & 20.90@xmath430.03 & 20.62@xmath430.02 & 20.97@xmath430.11 & 20.49@xmath430.07 & 20.98@xmath430.65 + c58 & 21.88@xmath430.06 & 21.79@xmath430.05 & 22.05@xmath430.27 & 21.52@xmath430.12 & 18.95@xmath430.06 + c76 & 22.49@xmath430.12 & + d40 & 22.13@xmath430.06 & 21.99@xmath430.07 & 22.07@xmath430.27 & 21.90@xmath430.12 & + d55 & 22.81@xmath430.12 & 22.71@xmath430.12 & & 22.98@xmath430.12 & + m28 & 20.50@xmath430.02 & 20.22@xmath430.01 & 20.51@xmath430.07 & 20.41@xmath430.06 & 18.24@xmath430.30 + m38 & 23.40@xmath430.11 & 23.54@xmath430.22 & & 22.70@xmath430.12 & + m41 & 22.55@xmath430.09 & 22.43@xmath430.10 & 22.54@xmath430.31 & 22.23@xmath430.12 & + m43 & 23.70@xmath430.21 & 23.36@xmath430.19 & & 26.56@xmath430.12 & + md23&22.45@xmath430.08 & 22.52@xmath430.11 & 21.92@xmath430.14 & 22.74@xmath430.12 & + md91&22.63@xmath430.09&22.44@xmath430.10 & & 22.31@xmath430.12 & + md97 & 21.59@xmath430.05 & 21.50@xmath430.04 & 21.07@xmath430.11 & 21.29@xmath430.12 & + md98 & 23.10@xmath430.15 & 23.09@xmath430.16 & & 22.67@xmath430.12 & + md99 & 21.64@xmath430.05 & 21.41@xmath430.04 & 21.21@xmath430.12 & 20.72@xmath430.08 & 17.39@xmath430.09 + md109 & 22.55@xmath430.09 & 22.52@xmath430.11 & & 22.68@xmath430.12 & + md115 & & 23.53@xmath430.23 & & 22.80@xmath430.12 & + rrrrr c10&0.214 & 65 & 9.56@xmath430.11 & 9.82@xmath430.09 + m30 & 0.201 & 2076 & 10.83@xmath430.18 & 10.91@xmath430.22 + md9 & 0.123 & 1542 & 10.48@xmath430.05 & 10.62@xmath430.09 + d29 & 0.184 & 402 & 10.45@xmath430.11 & 10.37@xmath430.14 + d49 & 0.340 & 1350 & 11.06@xmath430.09 & 11.10@xmath430.06 + m63 & 0.104 & 551 & 10.71@xmath430.13 & 10.84@xmath430.18 + c49 & 0.211 & 351 & 10.62@xmath430.24 & 10.71@xmath430.19 + c69 & 0.111 & 641 & 10.53@xmath430.08 & 10.61@xmath430.11 + c15 & 0.251 & 850 & 10.68@xmath430.12 & 10.84@xmath430.14 + c47 & 0.187 & 1250 & 10.69@xmath430.22 & 10.78@xmath430.13 + c50 & 0.223 & 1009 & 11.58@xmath430.15 & 11.69@xmath430.08 + c58 & 0.224 & 1351 & 10.62@xmath430.06 & 10.85@xmath430.09 + c76&0.378 & 346 & 10.92@xmath430.28 & 10.78@xmath430.22 + d40 & 0.118 & 910 & 10.41@xmath430.07 & 10.53@xmath430.11 + d55 & 0.092 & 989 & 10.27@xmath430.14 & 10.32@xmath430.11 + m28&0.254 & 1140 & 11.48@xmath430.13 & 11.51@xmath430.19 + m38 & 0.322 & 48 & 9.76@xmath430.09 & 9.84@xmath430.05 + m41 & 0.100 & 1933 & 11.06@xmath430.04 & 11.12@xmath430.08 + m43 & 0.258 & 267 & 10.69@xmath430.16 & 10.77@xmath430.10 + md23&0.34 & 96 & 10.48@xmath430.25 & 10.56@xmath43-.19 + md91&0.224 & 375 & 10.49@xmath430.13 & 10.52@xmath430.26 + md97 & 0.331 & 104 & 10.46@xmath430.07 & 10.54@xmath430.12 + md98 & 0.208 & 114 & 10.46@xmath430.11 & 10.51@xmath430.22 + md99 & 0.264 & 2112 & 11.06@xmath430.10 & 11.12@xmath430.14 + md109 & 0.66 & 67 & 10.43@xmath430.19 & 10.67@xmath430.11 + md115 & 0.100 & 1551 & 10.48@xmath430.09 & 10.61@xmath430.05 + vs. [ 3.6 ] color - magnitude plot for 8 micron lbgs ( open squares ) , illbgs ( filled squares ) and smgs ( stars ) . ( upper panel ) : [ 8.0]-[24.0 ] vs. [ 8.0 ] color - magnitude for illbgs ( filled squares ) , smgs ( asterisks ) . the data for the smgs are from ashby et al . all magnitudes are in ab . , width=604 ]

we summarize the relations among three classes of laws : infinitely divisible , selfdecomposable and stable . first we look at them as the solutions of the central limit problem ; then their role is scrutinized in relation to the lvy and the additive processes with an emphasis on stationarity and selfsimilarity . finally we analyze the ornstein uhlenbeck processes driven by lvy noises and their selfdecomposable stationary distributions , and we end with a few particular examples . pacs numbers : 02.50.cw , 02.50.ey , 05.40.fb msc numbers : 60e07 , 60g10 , 60g51 , 60j75 key words : selfsimilarity , selfdecomposability , lvy processes , additive processes .

0708.1239

selfsimilarity is a very popular research topics since a few years , and it has been approached from many different standpoints producing an unavoidable level of confusion @xcite . the present paper is devoted to a short summary of the properties of some important and well known families of laws : infinitely divisible , selfdecomposable and stable ( for details see for example @xcite ) . first of all we will recall their role in the formulation and in the solutions of the central limit problem : as we will see in the next section this amounts to a quest for all the limit laws of sums of independent random variables . our families of distributions will then be analyzed by means of both their possible decompositions in other laws , and the explicit form of their characteristic functions : the celebrated lvy khintchin formula . in particular it will be discussed the intermediate role played by the selfdecomposable laws between the more popular stable , and infinitely divisible distributions . we will then explore these laws in connection with the additive and the lvy processes , looking for their importance with respect to both the properties of stationarity and selfsimilarity . in particular it will be recalled how from selfdecomposable distributions it is always possible to define both stationary and selfsimilar additive processes which with the exception of important particular cases will in general be different . a few remarks are also added to show the differences between this selfsimilarity and that of the well known fractional brownian motion . we will also analyze the ornstein uhlenbeck processes driven by these lvy noises , and their stationary distributions which are always selfdecomposable . we will finally elaborate a few examples to illustrate these results , and to compare the behavior of our processes . in what follows we will adopt the following notations ( for details see for example @xcite ) : @xmath0 will denote the random variables , namely the measurable functions @xmath1 defined on a probability space @xmath2 where @xmath3 is a sample space , @xmath4 is a @xmath5algebra of events and @xmath6 a probability measure . we will then respectively write @xmath7 for their cumulative distribution functions @xmath8 and @xmath9 for their characteristic functions @xmath10 where the symbol @xmath11 denotes the expectation value of a random variable according to the probability @xmath6 , namely for example @xmath12 when they exist , @xmath13 will be the probability density functions , and in that event we will have @xmath14 a law will be indifferently specified either by its cumulative distribution function ( or density function ) , or by its characteristic function . to say that a random variable @xmath15 is distributed according to a given law we will also write either @xmath16 or @xmath17 ; if a family of laws is denoted by a specific symbol , say @xmath18 , then we also write @xmath19 to say that @xmath15 is distributed according to one of these laws . when two random variables @xmath15 and @xmath20 are identically distributed we will adopt the notation @xmath21 if a property is true with probability 1 we will also say that it is true @xmath6_almost surely _ and we will adopt the notation @xmath22 a _ type of laws _ ( see @xcite section 14 ) is a family of laws that only differ by a centering and a rescaling : in other words , if @xmath23 is the characteristic function of a law , all the laws of the same type have characteristic functions @xmath24 with a centering parameter @xmath25 , and a scaling parameter @xmath26 ( we exclude here the sign inversions ) . in terms of random variables this means that the laws of @xmath15 and @xmath27 ( for @xmath26 , and @xmath25 ) always are of the same type , and on the other hand that @xmath15 and @xmath20 belong to the same type if it is possible to find @xmath26 , and @xmath25 such that @xmath28 . we will also say that a random variable @xmath17 and its law are _ composed _ of @xmath29 and @xmath30 when @xmath31 and @xmath32 are independent and @xmath33 , or equivalently when @xmath34 ; then @xmath31 and @xmath32 are also called _ components _ of @xmath15 . of course , in terms of distributions , a composition amounts to a _ convolution _ ; then , for example , if @xmath35 denotes a normal law with expectation @xmath36 and variance @xmath37 , the composition of two normal laws will also be indicated as @xmath38 . for the stochastic processes @xmath39 we will say that @xmath40 and @xmath41 are _ identical in law _ when the systems of their finite dimensional distributions are identical , and in this case we will write @xmath42 a process @xmath40 is said to be _ stochastically continuous _ if for every @xmath43 @xmath44 where @xmath45 denotes the probability for the increment @xmath46 of being larger that @xmath43 . in this paper all our random variables and processes will be one dimensional . in this exposition we do not pretend neither rigor , nor completeness : we just list the results and the properties that are important to compare , and we refer to the existing literature for proofs and details , and for a few hints about possible recent applications . in fact the aim of this paper is just to draw an outline showing without embarrassing the reader with excessive technical details the deep , but otherwise simple ideas which are behind the properties of our families of laws and the simplest procedures to build from them the most important classes of processes . we hope that this , with the aid of some telling examples , will be helpful to approach this field of research by clarifying the roles , the differences and the subtleties of these laws and processes . it is well known that there are three kinds of _ classical limit theorems _ ( all along this paper wherever we speak of convergence it is understood that we speak of _ convergence in law _ ) which are characterized by the form of the respective limit laws : * the _ law of large numbers _ with limit laws of the degenerate type @xmath47 with characteristic function @xmath48 * the _ normal central limit theorem _ whose limit laws are of the gaussian type @xmath35 with @xmath49 * and the _ poisson theorem _ whose limit laws are of the poisson types @xmath50 with @xmath51 the exact statements of these theorems in their traditional formulations are reprinted in every handbook of probability ( see for example @xcite p. 323 and following ) , and we will not reproduce them once more . remark however that in our list , while speaking of _ the type _ for the degenerate and the gaussian laws , we also referred to _ the types _ for the poisson laws . in fact all the normal laws @xmath35 with expectation @xmath25 and variance @xmath52 constitute a unique type , and the same is true for the family of all the degenerate laws . on the other hand the standard poisson laws @xmath53 with different parameters @xmath54 belong to different types : indeed we can not recover a law @xmath55 from another @xmath56 ( with @xmath57 ) just by means of a centering and a rescaling . in other words we could say that every poisson law @xmath55 with a given @xmath54 generates by centering and rescaling a distinct type @xmath50 with characteristic functions ( [ poisschf ] ) . the particularities of the poisson laws with respect to the other two families of limit laws are also apparent from their properties of _ composition and decomposition_. for a family of laws ( not necessarily a type ) to be closed under composition means that the composition ( convolution ) of two laws of that family still belongs to the same family . on the other hand closure under decomposition means that if a law of the family is decomposed in two laws , these two components necessarily belong to the same family . the types of limit laws appearing in the classical limit theorems show an important form of closure under composition and decomposition summarized in the following result ( see @xcite p. 283 ) : _ the degenerate and normal types are closed under compositions and under decompositions ; the same is true for every family of poisson laws @xmath50 with the same @xmath58 . _ the closure under composition is in fact an elementary property ; not so for the closure under decomposition : the proofs for the normal and the poisson case were given in 1935 - 37 by h. cramr and d.a.raikov respectively . the normal and poisson composition and decomposition properties can then be stated by saying that @xmath59 hence it is also apparent that a type of poisson laws never is closed under composition and decomposition : we are always obliged to switch from a poisson type to another while composing and decomposing them . it is important to remark , however , that the classical limit theorems are not the embodiment of these composition and decomposition properties only , but they are more far reaching and profound statements . in fact not only these theorems deal with limits of sums of independent random variables of a type which in general is different from that of the eventual limit laws the poisson law is the limit of sums of bernoulli 01 random variables , while the normal and degenerate laws are limits of sums of still more general random variables but also the distribution of the sum of the @xmath60 random variables does not coincide with the limit law at every step @xmath60 of the limiting process , as happens instead in a simple decomposition . to formulate the central limit problem ( _ clp _ ) let us look at it first of all in terms of sequences of random variables : usually we take a sequence @xmath61 of random variables , and then the sequence of their sums @xmath62 . in this case when we go from @xmath63 to @xmath64 we just add another random variable @xmath65 without changing the previous sum @xmath63 . however this is not the more general way to produce sequences of sums of random variables . consider indeed a triangular array of random variables @xmath66 @xmath67 with @xmath68 and @xmath69 , and suppose that 1 . in every row @xmath70 the random variables @xmath71 are independent , 2 . the @xmath66 are _ uniformly , asymptotically negligible _ , namely that @xmath72 define now the _ consecutive sums _ @xmath73 then the central limit problem _ for consecutive sums of independent random variables _ ( _ clp_@xmath74 ) reads : _ _ find the family of all the limit laws of the consecutive sums _ _ ( [ consecsum ] ) _ and the corresponding convergence conditions _ ( see @xcite p.301 - 2 ) . three remarks are in order here : * for a given @xmath70 the random variables @xmath71 are independent , but in general they are neither identically distributed , nor of the same type ; * going from the row @xmath60 to the row @xmath75 the random variables and their laws change : in general @xmath66 and @xmath76 are neither identically distributed , nor of the same type ; as a consequence going from @xmath63 to @xmath64 we not only add the @xmath75-th random variable , but we also are obliged to adjourn the law of @xmath63 ; * the uniformly , asymptotically negligible condition is an important technical requirement added to avoid trivial answers to the _ clp_@xmath74 ( see for example @xcite p. 302 ) ; in fact without this condition it is easy to show that any law @xmath77 would be limit law of the consecutive sums ( [ consecsum ] ) : it would be enough for every @xmath60 to take @xmath78 , and @xmath79 for @xmath80 . the uniformly , asymptotically negligible condition will tacitly be assumed all along this paper . important particular cases of the _ clp_@xmath74 are then selected when we specialize the sequence @xmath66 in the following way ( see @xcite p. 331 ) : let us suppose that there is a sequence @xmath61 , @xmath81 , of independent ( but in general not identically distributed ) random variables , and two sequences of numbers @xmath82 and @xmath83 , @xmath68 , such that for every @xmath84 and @xmath60 @xmath85 it is apparent that now going from @xmath60 to @xmath86 we just get a centering and a rescaling of every @xmath61 , so that @xmath66 and @xmath87 always are of the same type . then the consecutive sums take the form of _ normed sums _ ( namely centered and rescaled sums ) of independent random variables @xmath88 where we adopt the notation @xmath89 then the central limit problem _ for normed sums of independent random variables _ ( _ clp_@xmath90 ) reads : _ _ find the family of all the limit laws of the normed sums _ _ ( [ normsum ] ) _ and the corresponding convergence conditions_. finally there is a still more specialized formulation of the central limit problem when we add the hypothesis that the random variables @xmath61 are not only independent , but also identically distributed ( see @xcite p. 338 ) : in this case we speak of a central limit problem _ for normed sums of independent and identically distributed random variables _ ( _ clp_@xmath91 ) . the answers to the different formulations of the central limit problem need the definition of several important families of laws that are much more general than the gaussian type , and that can be defined by means of the properties of their characteristic functions . suppose that @xmath23 is the characteristic function of a law : we will say that this law is _ infinitely divisible _ ( see @xcite p. 308 ) if for every @xmath70 we can always find another characteristic function @xmath92 such that @xmath93^n.\ ] ] apparently the name comes from the fact that our law can always be decomposed in an arbitrary number of identical laws ; remark however that for different values of @xmath60 we get in general laws @xmath92 of different types . in terms of random variables if @xmath17 is infinitely divisible , then for every @xmath70 we can find @xmath60 independent and identically distributed random variables @xmath94 all distributed as @xmath92 and such that @xmath95 namely @xmath15 is always decomposable ( in distribution ) in the sum of an arbitrary , finite number of independent and identically distributed random variables . let us call @xmath96 the family of all the infinitely divisible laws . many important distributions are infinitely divisible : degenerate , gaussian , poisson , compound poisson , geometric , student , gamma , exponential and laplace are infinitely divisible . on the other hand the uniform , beta and binomial laws are not infinitely divisible : in fact no distribution ( other than the degenerate ) with bounded support can be infinitely divisible(see @xcite p. 31 ) . remark that if @xmath23 is an infinitely divisiblecharacteristic function , then also @xmath97 is an infinitely divisible characteristic function for every @xmath98 ( see @xcite p. 35 ) : we will see that this is instrumental to connect the infinitely divisible laws to the lvy processes . a second important family of laws selected by their decomposition properties is that of the _ selfdecomposable _ laws ( see @xcite p. 334 ) : a law @xmath23 is selfdecomposable when for every @xmath99 we can always find another characteristic function @xmath100 such that @xmath101 in terms of random variables this means that if @xmath17 is selfdecomposable , then for every @xmath99 we can always find two independent random variables , @xmath102 and @xmath103 , such that @xmath104 in other words for every @xmath99 @xmath15 can always be decomposed into two independent random variables such that one of them is of the same type of @xmath15 . it can be shown that every selfdecomposable law , along with all its components , is also infinitely divisible ( see @xcite p. 335 ) , so that if we call @xmath105 the family of all the selfdecomposable laws , then @xmath106 . the gaussian , student , gamma , exponential and laplace laws are examples of selfdecomposable laws ( see @xcite p. 98 ) . on the other hand the poisson laws are not selfdecomposable : they only are infinitely divisible . finally we will say that a law @xmath23 is _ stable _ ( see @xcite p. 338 ) if for every @xmath107 and @xmath108 we can find @xmath26 and @xmath25 such that @xmath109 this means now that if @xmath17 is stable , then for every @xmath107 and @xmath108 we can find two independent random variables @xmath110 and @xmath111 , and two number @xmath26 and @xmath36 such that @xmath112 in other words we can always decompose a stable law in other laws which are of the same type as the initial one . our definition can also be reformulated in a slightly different way ( see @xcite p. 69 ) : for every @xmath113 we can always find @xmath26 and @xmath25 such that @xmath114^c,\ ] ] and this apparently means that for every @xmath113 the law @xmath115^c$ ] is of the same type as @xmath23 . a law is also said _ strictly stable _ if for every @xmath113 always exists @xmath26 such that @xmath116^c.\ ] ] all the stable laws are selfdecomposable , and hence if @xmath117 is the family of all the stable laws we will have @xmath118 . in fact our classification of laws in only three families ( infinitely divisible , selfdecomposable and stable ) is an oversimplification of a much richer structure explored for example in @xcite , chapter 3 . among the classical laws only the gaussian and the cauchy laws are stable . remark that if a law @xmath77 belongs to one of our families , then also all its type belongs to the same family . hence it would be more suitable to say that @xmath96 , @xmath105 and @xmath117 are families of types of laws ; however in the following , for the sake of simplicity , this will be understood without saying . the relevance of our three families of laws lies in the fact that they represent the answers to the three formulations of the central limit problem discussed in the previous section . in fact it can be shown that @xmath119 _ and _ @xmath117 _ exactly coincide with the families of the limit laws sought for respectively in _ _ clp_@xmath74 , _ clp_@xmath90_and _ _ clp_@xmath91 ( see for example @xcite p. 321 , p. 335 and for the three statements ) . to summarize these results a few examples will be shown in the section [ exampleclt ] we can then say that : * the laws of the normed sums ( [ normsum ] ) of sequences @xmath61 of independent and identically distributed random variables converge toward stable laws ; in particular : when the @xmath61 have finite variance their normed sums converge according to the classical theorem to normal laws , while the non gaussian , stable distributions are limit laws only for sums of random variables with infinite variance ; a well known example of the non gaussian limit laws is the cauchy distribution ; * the laws of the normed sums ( [ normsum ] ) of sequences @xmath61 of independent , but not necessarily identically distributed , random variables converge toward selfdecomposable laws ; the special case of the stable laws is recovered when the @xmath61 are also identically distributed ; in other words when a selfdecomposable law is not stable it can not be the limit law of normed sums of independent and identically distributed random variables ; * finally the laws of the consecutive sums ( [ consecsum ] ) of triangular arrays @xmath66 converge to infinitely divisible laws : a classical example of non selfdecomposable convergence is represented by the poisson limit theorem recalled in the section [ classicalth ] ; of course the selfdecomposable case is obtained when the triangular array has the form ( [ specialseq ] ) and the consecutive sums become normed sums ( [ normsum ] ) ; however infinitely divisible laws which are not selfdecomposable ( as the poisson law ) can not be limit laws of normed sums of independent random variables . nothing forbids , of course , that in this scheme a gaussian law be also the limit law for consecutive sums of triangular arrays that do not reduces to normed sums . to complete the picture we will hence just recall here that there are also necessary and sufficient conditions for the convergence to normal laws of triangular arrays of independent , but not necessarily identically distributed , random variables with finite variances ( see @xcite p. 326 ) . since the stable distributions are limit laws of normed sums ( [ normsum ] ) of independent and identically distributed random variables we can also introduce the notion of _ domain of attraction of a _ stable _ law _ @xmath120 : we will say that a law @xmath77 belongs to the domain of attraction of @xmath121 when we can find two sequences of numbers , @xmath82 and @xmath122 , such that the normed sums ( [ normsum ] ) of a sequence @xmath61 of random variables _ all distributed as _ @xmath77 , converge to @xmath121 . remark that this definition can not be immediately extended to the non stable distributions which are not limit laws of normed sums of independent and identically distributed random variables , so that we can not speak of a _ unique _ distribution @xmath77 being attracted by the limit law . every law with finite variance belongs to the domain of attraction of the normal law ( see @xcite p. 363 ) . it is also important to stress here that , while all stable laws are attracted by themselves ( see @xcite p. 363 ) , a non stable , infinitely divisible law @xmath77 which always is in itself the limit law of a suitable consecutive sum ( [ consecsum ] ) of some triangular array of random variables also belongs to the domain of attraction of some stable law @xmath121 : normed sums ( [ normsum ] ) of random variables _ all distributed as _ @xmath77 will converge toward some stable law @xmath121 . for instance the poisson law which is an infinitely divisible limit law , as the poisson theorem shows apparently also is in the domain of attraction of the normal law since it has a finite variance : a normed sum of random variables all distributed according to the same poisson law will converge to the gauss law . finally we recall , without going into more detail , that for a given law @xmath77 it is always possible to find if it belongs to some domain of attraction , and then it is also possible to find both the stable limit law @xmath121 , and the admissible numerical sequences @xmath82 and @xmath122 entering in the normed sums ( [ normsum ] ) converging to @xmath121 ( see @xcite p. 364 ) . it is not easy to find out if a given law @xmath23 belongs to one of the families defined in the previous section just by looking at the definitions introduced up to now . it is important then to recall the explicit form of the characteristic functions of our families of laws given by the _ lvy khintchin formula_. it can be proved ( see @xcite p. 343 ) indeed that the _ logarithmic characteristic _ @xmath123 of an infinitely divisible law is uniquely associated , through the formula @xmath124 to a _ generating triplet _ @xmath125 where @xmath126 , and the _ lvy function _ @xmath127 is defined on @xmath128 , is non decreasing on @xmath129 and @xmath130 , with @xmath131 and @xmath132 for @xmath133 finite . in other words a law will be infinitely divisible if and only if its logarithmic characteristic satisfies the relation ( [ lkformula ] ) for a suitable choice of @xmath125 . the lvy function @xmath127 also defines the _ lvy measure _ ( for details see @xcite section 8) @xmath134 for every measurable set @xmath135 of @xmath136 and @xmath137 , so that often we will refer to @xmath138 as the generating triplet . when the lvy measure @xmath139 is absolutely continuous we will also denote by @xmath140 its density . remark that the lvy khintchin formula ( [ lkformula ] ) can be given in a variety of equivalent versions ( see @xcite p. 37 ) by suitably choosing the integrand functions , but we will not go into such details . for a given law even the verification of the formula ( [ lkformula ] ) is not in general an easy task . in the case of stable laws , however , the lvy khintchin formula is considerably simpler since it no longer explicitely involves integrals on the lvy measure . a stable law is characterized by a parameter @xmath141 ( see @xcite p. 76 ) and it is then also said @xmath142stable ( @xmath142-stable ) : the case @xmath143 corresponds to the gaussian laws , and to a vanishing lvy measure . in the non gaussian @xmath142stable cases ( @xmath144 ) on the other hand the lvy measure is not zero , it is absolutely continuous and we have ( see @xcite p. 80 ) @xmath145 with @xmath146 and @xmath147 . finally in both cases gaussian and non gaussian the logarithmic characteristic must satisfy the following relation ( see @xcite p. 86 ) @xmath148 where @xmath149 $ ] , @xmath150 , @xmath151 and @xmath152 . the gaussian case simply corresponds to @xmath143 . when the law is also symmetric the characteristic function is real and the formula reduces itself to the quite elementary expression @xmath153 the form of the lvy khintchin formula , or equivalently of the triplet @xmath125 , of the selfdecomposable laws , on the other hand , is not so simple . they in fact play in some sense a sort of intermediate role between the generality of the infinitely divisible laws and the special properties of the stable laws . it can be proved indeed ( see @xcite p. 95 ) that a law is selfdecomposable if and only if its lvy measure is absolutely continuous and its density is @xmath154 where the function @xmath155 is non negative , is increasing on @xmath129 and is decreasing on @xmath130 . by the way this also show why a poisson law ( whose lvy measure is not absolutely continuous ) can not be selfdecomposable . as a consequence the lvy khintchin formula ( [ lkformula ] ) of the selfdecomposable laws is more specialized than that of the general infinitely divisible laws , but it still contains a non elementary integral part . an _ additive process _ @xmath40 is a stochastically continuous process with independent increments and @xmath156 ( namely the probability of not being zero vanishes ) ; on the other hand a _ lvy process _ is an additive process with the further requirement that the increments must be stationary ( for further details see @xcite section 1 ) . the _ stationarity of the increments _ means that the law of @xmath157 does not depend on @xmath158 . processes with independent increments are also markov processes , and hence the entire family of their joint laws at an arbitrary , finite number of times can be deduced just from the one and the two times distributions . in other words it is enough to know the laws of the increments to have the complete law of the process . this of course is a very good reason to be interested in markov , and in particular in additive processes , but it must be recalled here that there are also non markovian processes which can still be defined by means of very simple tools . an important example that will be briefly mentioned later is the _ fractional brownian motion _ which takes advantage of being a gaussian process to make up for its lack of markovianity . it is also important to recall here that with the exception of gaussian processes the additive processes trajectories can make jumps . this does not contradict their stochastic continuity because the jumping times are random , and hence , for every @xmath159 , the probability of a jump occurring exactly at @xmath159 is zero . in the following the law of the process increment @xmath160 will be given by means of its characteristic function @xmath161 , so that the stationarity of the lvy processes simply entails that @xmath161 only depends on the difference @xmath162 : in this case we will use the shorthand , one time notation @xmath163 . as for every markov process the laws of the increments of an additive process must satisfy the chapman kolmogorov equations which for the characteristic functions are @xmath164 for lvy ( stationary ) processes these equations take the form @xmath165 there is now a very simple and intuitive procedure to build a lvy process : take the characteristic function @xmath23 of a law and define @xmath166^{t / t}.\ ] ] it is immediate to see that @xmath167 satisfies ( [ chkstat ] ) , so that it can surely be taken as the characteristic function of the stationary increments of a lvy process . here @xmath168 plays the role of a dimensional time constant ( a _ time scale _ ) introduced to have a dimensionless exponent . to have a consistent procedure , however , we must be sure that when @xmath23 is a characteristic function , also @xmath167 in ( [ stchf ] ) is a characteristic function for every @xmath169 , but unfortunately this is simply not true for every characteristic function @xmath23 . we are led hence to ask for what kind of characteristic functions ( [ stchf ] ) is again a characteristic function . we know , on the other hand , that if @xmath77 is an infinitely divisible characteristic function , then also @xmath170 with @xmath98 is a characteristic function , and an infinitely divisible one too . in fact it is possible to show that ( [ stchf ] ) is a characteristic function if and only if @xmath23 is infinitely divisible . in other words there is a one - to - one relation between the class @xmath96 of the infinitely divisible laws and that of the lvy processes ( see @xcite section 7 ) . remark however that in general for a lvy process defined by ( [ stchf ] ) the infinitely divisible law of the increments at a generic time @xmath159 is neither @xmath23 , nor of the same type of @xmath23 . only at @xmath171 the law is necessarily @xmath23 , while for @xmath172 it can be rather different and but for few well known cases its explicit cumulative distribution function ( or density function ) could be quite difficult to find . on the other hand when @xmath23 is a stable law it is easy to see from the very definition ( [ stable ] ) of stability that at every time @xmath159 the law of the increments ( [ stchf ] ) will always belong to the same type ( this is famously what happens for the gauss and cauchy laws ) . in this case we speak of a stable process , and its evolution can be summarized just in the time dependence of the law parameters which will produce a trajectory inside a unique type . a different situation arises instead when @xmath23 only belongs to a family of infinitely divisible laws closed under composition and decomposition . as we have already remarked these families do not in general constitute a type ( as the family of the poisson laws @xmath55 ) : if however they are closed under composition ( as are both the poisson and the compound poisson processes ) the law ( [ stchf ] ) of the increment of the lvy process stays in the same family of laws all along an evolution which is described by the time dependence of the law parameters ; this however does not amount to the stability of the process since our family is not a single type . since every lvy process is associated to an infinitely divisible law @xmath173 , and since every infinitely divisible law is associated to a generating triplet @xmath138 we will also speak of the logarithmic characteristic @xmath174 and of the generating triplet @xmath138 of a lvy process . in this case however the lvy measure @xmath139 has also an important probabilistic meaning w.r.t . the lvy process ( see for example @xcite pp.75 - 85 ) : _ for every borel set @xmath175 of @xmath176 , @xmath177 represents the expected number , per unit time , of ( non - zero ) jumps with size belonging to @xmath175_. it can also be proved that for every compact set a such that @xmath178 we have @xmath179 , namely the number of jumps per unit time of finite ( neither infinite , nor infinitesimal ) size is finite . remark however that this does not mean that @xmath139 is a finite measure on @xmath136 : in fact the function @xmath127 associated to @xmath139 can diverge in @xmath180 so that the process can have an infinite number of infinitesimal jumps in every compact @xmath181 $ ] . in this case , when @xmath182 , we speak of an _ infinite activity _ process , and the set of the jump times of every trajectory will be countably infinite and dense in @xmath183 $ ] . for the sake of simplicity we will not introduce here the important _ lvy it decomposition _ of a lvy process into its continuous ( gaussian ) and jumping ( poisson ) parts : the readers are referred to @xcite , section 3.4 for a synthetic treatment . a process @xmath40 ( possibly neither additive , nor stationary ) is said to be _ selfsimilar _ when for every given @xmath26 we can find @xmath151 such that @xmath184 namely when every change @xmath58 in the time scale can be compensated _ in distribution _ by a corresponding change @xmath36 in the space scale . in terms of the increment characteristic functions this means that for every @xmath26 we must have a @xmath151 such that @xmath185 in fact it can be proved more about the form of this space time compensation : given a selfsimilar process we can always find @xmath186 such that @xmath187 ( see @xcite p. 73 ) . this number @xmath188 is called the _ exponent of the process _ or _ hurst index _ , and we will also speak of @xmath188__selfsimilar _ _ processes . for further details about selfsimilar , additive processes see also @xcite . since a lvy process is completely specified by ( [ stchf ] ) as characteristic function of its increments , then in this case the selfsimilarity means that for every @xmath26 it exists @xmath151 such that @xmath189^{at / t}=[\varphi(bu)]^{t / t}.\ ] ] from the definition ( [ stable ] ) of the strictly stable laws and from ( [ statssim ] ) it is easy to understand then that the unique selfsimilar lvy processes must be strictly stable . for instance in a wiener process we have @xmath190 , namely @xmath191^{t / t}=e^{-u^2dt/2},\qquad d=\frac{\sigma^2}{t}\,,\ ] ] and hence @xmath191^{at / t}=e^{-u^2dat/2}\ ] ] so that @xmath192 ( namely @xmath193 ) is the required compensation . this means that , insofar as the coefficient @xmath194 remains the same , we can change the space and time scales @xmath5 and @xmath168 ( namely we can change the units of measure ) without changing the wiener process distribution . more precisely than these simple remarks , it can be proved that _ a lvy process @xmath40 is selfsimilar if and only if it is strictly stable _ ( see @xcite p. 71 ) . things are rather different , however , when we consider only additive ( not necessarily lvy ) processes , namely when we can also live without stationarity . now we must stick to the general selfsimilarity equation ( [ ssim ] ) , and we must remark again that there is another simple , intuitive procedure producing additive ( but not necessarily stationary ) , selfsimilar processes : simply consider a characteristic function @xmath23 , a real number @xmath186 and take @xmath195 it is now apparent that the characteristic functions of this family satisfy the equation ( [ chk ] ) and are also selfsimilar according to the definition ( [ ssim ] ) with the space time scale compensation produced by @xmath196 . namely ( [ ssimchf ] ) produces an @xmath188selfsimilar process . of course we must ask here the same question surfaced w.r.t . equation ( [ stchf ] ) in the case of stationary processes : when can we be sure that the function defined by the ratio ( [ ssimchf ] ) of two characteristic function still is the bona fide characteristic function of a law ? even in this case , however , the answer can be hinted to by looking at the definition ( [ sdlaw ] ) of a selfdecomposable characteristic function@xmath197 more precisely it can be proved that ( [ ssimchf ] ) is a characteristic function if and only if @xmath23 is selfdecomposable ( see @xcite p. 99 ) : this ultimately brings out the intimate relation connecting selfsimilarity and selfdecomposability . since selfdecomposable laws are also infinitely divisible the previous remarks show that from a given selfdecomposable @xmath23 we can always produce two different kinds of processes : a lvy process whose stationary increments follow the law ( [ stchf ] ) ; and a family of additive , selfsimilar process one for every value of @xmath186 whose ( possibly non stationary ) increments follow the law ( [ ssimchf ] ) . all these processes generated from the same @xmath23 are in general different with one exception : when @xmath23 is an @xmath142stable law the associated lvy process coincide with the @xmath188selfsimilar one with hurst index @xmath198 . in this last case indeed the characteristic functions ( [ stchf ] ) and ( [ ssimchf ] ) are identical ( to see it take for example the symmetric form ( [ symmstable ] ) of a stable characteristic function ) . remark also that the index of an @xmath142stable law always satisfies @xmath141 ( @xmath143 for the gaussian law ) , and that this is coherent with the limitation @xmath199 for the hurst index of the stable , selfsimilar processes ( see @xcite p. 75 ) . on the other hand , in every other case ( either non stable , or @xmath142stable with @xmath200 ) , from a selfdecomposable law @xmath23 we can always build a lvy , non selfsimilar process from ( [ stchf ] ) , and a family of additive , selfsimilar processes with non stationary increments from ( [ ssimchf ] ) . finally from a infinitely divisible , but not selfdecomposable law we can only get a lvy process from ( [ stchf ] ) , but no selfsimilarity is allowed . for more details about present interest of the selfdecomposabledistributions and selfsimilar processes in the applications see for instance @xcite and @xcite . remark that selfsimilarity is not tied to the dependence or independence of the increments : we have seen here that among independent increment processes we find both selfsimilar and non selfsimilar processes ; and on the other hand a process can be selfsimilar without showing independence of the increments . a celebrated example of this second case is the so called _ fractional brownian motion _ : this is a centered , gaussian , @xmath188selfsimilar ( for @xmath201 $ ] ) process @xmath202 with stationary increments , and covariance function @xmath203}=\frac{|t|^{2h}+|s|^{2h}-|t - s|^{2h}}{2}\,,\qquad\quad t , s>0.\ ] ] it is apparent that this is nothing else than a generalization of the well known covariance function of the usual brownian motion @xmath204}=\min(t , s)$ ] that is recovered when @xmath193 . since @xmath202 is centered and gaussian , this covariance function is all that is needed to define the process also if it is not markovian . in fact a fractional brownian motion coincides with the usual brownian motion ( and hence is markovian with independent increments ) only for @xmath193 , while for @xmath205 it is non markovian , has correlated increments and for @xmath206 shows long range dependence . in other words an @xmath188selfsimilar fractional brownian motion with @xmath205 is neither additive , nor markovian : in fact it is not even a semimartingale , and hence few results of stochastic calculus can be used . from another standpoint ( see @xcite p. 230 ) we can say that the selfsimilarity can have different origins : it can stem either from the length of the distribution tails of independent increments , or from the correlation between short tailed , gaussian increments , and the two effects can also be mixed . for more information about the fractional brownian motion see @xcite and @xcite selfdecomposable laws appear also in another important context : they are the most general class of stationary distributions of _ processes of the ornstein uhlenbeck type_. take a lvy process @xmath207 with generating triplet @xmath208 and logarithmic characteristic@xmath209 , and for @xmath151 consider the stochastic differential equation ( for simplicity we take @xmath210 ) @xmath211 whose exact meaning is rather in its integral form @xmath212 when @xmath207 is a wiener process the equation ( [ ousde ] ) coincides with the stochastic differential equation of an ordinary , gaussian ornstein uhlenbeck process ; but the equation ( [ ousde ] ) keeps the meaning of a well behaved stochastic differential equation even if @xmath207 is a generic , non gaussian lvy process , and its solution @xmath213 will be called a process of the ornstein uhlenbeck type . of course to give a rigorous sense to this solution we should define our stochastic integrals for a generic lvy process @xmath207 : since all lvy processes are semimartingales ( see @xcite p. 255 ) , this can certainly be done , but we will skip this point referring the reader to the existing literature ( see @xcite and @xcite , or @xcite for an extensive treatment ) . we will rather shift our attention to the possible existence of _ stationary distributions _ for a process of the ornstein uhlenbecktype . in fact it is possible to show ( see @xcite p. 108 , and @xcite p. 485 ) that if @xmath214 then the ornstein uhlenbeck process @xmath40 solution of ( [ ousde ] ) has a stationary distribution @xmath173 which is selfdecomposable with logarithmic characteristic@xmath215 and generating triplet @xmath216 where @xmath217 , @xmath218 , and according to the equation ( [ sdlevymeas ] ) the absolutely continuous lvy measure @xmath139 has a density @xmath219\ } , & \hbox{if $ x<0$. } \end{array } \right.\ ] ] conversely for every selfdecomposable law @xmath23 there is a lvy process @xmath207 such that @xmath23 is the stationary law of the ornstein uhlenbeckprocess driven by @xmath207 . remark also that by the simple change of variable @xmath220 the relation ( [ oulch1 ] ) takes the form @xmath221 which is well suited to the inverse problem of finding the lvy noise of an ornstein uhlenbeck process for a prescribed selfdecomposable stationary distribution . we will consider now several families of distributions ( for further details see for example @xcite and references quoted therein ) , all absolutely continuous , centered and symmetric , with a space scale parameter @xmath26 which will of course span the types since the centering parameters always vanish : * the type of the normal laws @xmath222 with density function and characteristic function@xmath223 and with variance @xmath37 ; * the types ( one for every @xmath98 ) of the variance gamma laws @xmath224 with @xmath225 where @xmath226 are the modified bessel functions and @xmath227 is the euler gamma function @xcite ; their variance @xmath228 is always finite ; * the types ( one for every @xmath98 ) of the student laws @xmath229 with density function and characteristic function@xmath230 where @xmath231 is the euler beta function @xcite . their variance is finite only for @xmath232 and its value is @xmath233 . important particular types within the variance gamma and the student families are respectively the laplace ( double exponential ) laws @xmath234 with density function and characteristic function@xmath235 and finite variance @xmath236 , and the cauchy laws @xmath237 with @xmath238 and divergent variance . finally we will also consider in the following another type of student laws @xmath239 with density function and characteristic function@xmath240 and finite variance @xmath37 . all the laws of our families are selfdecomposable ( and hence infinitely divisible ) , but only @xmath222 and @xmath241 are types of @xmath142stabledistributions : more precisely @xmath222 laws are 2stable , and @xmath241 are 1stable . on the other hand the family @xmath224 is closed under convolution , while @xmath229 is not . of course this does not mean that the variance - gamma laws are stable since @xmath224 is not a unique type , and a convolution will mix different types with different @xmath54 values . the infinitely divisible laws @xmath222 , @xmath242 , @xmath241 and @xmath243 are of course entitled to their characteristic triplets @xmath216 . since they are all centered and symmetric we have @xmath244 for all of them . as for @xmath245 it can be seen that for @xmath242 , @xmath241 and @xmath243 we have @xmath246 ( in fact , in terms of the lvy it decomposition , they generate so called _ pure jump processes _ ) , while for @xmath222 we have @xmath247 . as for the lvy measures , on the other hand , we first of all have @xmath248 for the laws @xmath222 : from the point of view of the sample path properties this simply means that at variance with the other three cases under present investigation the lvy processes generated by gaussian distributions never make jumps . the lvy measures of the other three cases are instead all absolutely continuous and have the following densities @xmath249 @xmath250 \qquad\qquad\qquad \hbox{for $ \mathcal{s}(a)$}\end{aligned}\ ] ] where the sine and the cosine integral functions for @xmath133 are @xcite @xmath251 of the lvy measures for the selfdecomposablelaws @xmath242 ( black ) , @xmath241 ( red ) and @xmath243 ( blue ) . to make the plots comparable we have chosen @xmath252 for all the three densities , and to make the differences more visible we have plotted only the positive @xmath253axis since the curves are exactly symmetric on the negative axis.,width=340 ] examples of these three densities are plotted in figure [ fig01 ] . to give examples of consecutive sums converging to our laws let us first of all recall what happens in the case of the poisson laws @xmath55 . let @xmath254 represent the binomial laws for @xmath60 independent trials of verification of an event occurring with probability @xmath255 , and take the triangular array of bernoulli 01 random variables @xmath256 with @xmath257 and @xmath258 apparently they are uniformly , asymptotically negligible because for @xmath259 we have @xmath260 it is very well known that the consecutive sums are binomial random variables , namely @xmath261 and that , according to the classical poisson theorem , the limit law of these @xmath63 is @xmath55 . remark that , since in passing from @xmath60 to @xmath86 the random variables @xmath66 change type , it will not be possible to put @xmath63 in the form of a normed sum as ( [ normsum ] ) . the poisson laws , however , are also limit laws in a still legitimate , but rather trivial sense due to the composition properties of the family @xmath55 : take for instance a triangular array of poisson random variables @xmath262 with @xmath257 and @xmath258 which are again uniformly , asymptotically negligible because for @xmath259 we have @xmath263 now at variance with the previous example of the bernoulli triangular array at every step @xmath60 we exactly have @xmath264 and hence , albeit in a trivial sense , the limit law again is @xmath55 . also in this case @xmath66 and @xmath87 belong to different types , so that it will be impossible to put @xmath63 in the form of a normed sum as ( [ normsum ] ) . of course both these examples show in what sense the poisson laws are infinitely divisible but not selfdecomposable : they are limit laws of consecutive sums ( [ consecsum ] ) of uniformly , asymptotically negligible triangular arrays , but not of normed sums ( [ normsum ] ) of independent random variables . at the other end of the gamut of the infinitely divisible laws we find the 2stablenormal laws . to see in what sense they are limit laws take an arbitrary sequence @xmath61 of centered , independent and identically distributed random variables with finite variance @xmath265 and define the triangular array @xmath266 which always turns out to be uniformly , asymptotically negligible because of the chebyshev inequality : @xmath267 the consecutive sums are now also normed sums of independent and identically distributed random variables since @xmath268 and according to the classical , normal central limit theorem their limit law is @xmath269 . also in this case , however , it is possible to exploit the composition properties of the normal type to find another , more trivial form of the consecutive sums : take the sequence of normal independent and identically distributed random variables @xmath270 and define the triangular array @xmath271 which again apparently is uniformly , asymptotically negligible . now the consecutive sums are also normed sums which at variance with the previous example are all normally distributed @xmath272 and hence , in a trivial sense , the limit law is @xmath269 . finally let us remark that , since the normal laws are infinitely divisible , nothing will forbid them to be also limit laws of consecutive sums of triangular arrays that _ do not _ reduce to normed sums of independent and identically distributedrandom variables . however we will not elaborate here examples in this sense . we have introduced the trivial forms of the consecutive sums in the case of the poisson and normal laws ( see also the remarks at the end of the section [ classicalth ] ) only because in our other subsequent examples this will be the unique explicit form available for our @xmath63 . remark also that these trivial forms essentially derive from the fact that our laws are all infinitely divisible . in fact if @xmath77 is infinitely divisible , then also @xmath273 is a characteristic function , and of course @xmath274 for every @xmath60 . in general with the exception of the stable laws the @xmath275 are not of the same type for different @xmath60 , and hence the sums can not take the form of normed sums . that notwithstanding , in a trivial sense , every infinitely divisiblelaw @xmath77 is the limit law of the consecutive sums of independent and identically distributed random variables all distributed according to @xmath275 . what is less trivial , however , is to give an explicit form to the cumulative distribution function or density function of the component laws @xmath275 : as the subsequent examples will show this can be easily done only when we deal with families of laws closed under composition and decomposition . let us consider first the cauchy laws introduced in the previous section [ examplelaws ] : they are 1stable and , by taking advantage of the fact that the family @xmath241 is closed under composition and decomposition , it will be easy to show how they are limit laws of suitable sums of random variables . take for instance a sequence of independent and identically distributed cauchy random variables @xmath276 and define the triangular array @xmath277 . since now there is no variance to speak about , to show that this sequence is uniformly , asymptotically negligible we can not use the chebyshev inequality . if however @xmath278 is the common cumulative distribution function of the @xmath61 , it is easy to see that the sequence is uniformly , asymptotically negligible because @xmath279{\stackrel{n}{\longrightarrow}}0\,.\ ] ] now , as for the gaussian case , the consecutive sums are also normed sums of independent and identically distributed random variables and are all distributed according to the cauchy law @xmath241 @xmath280 so that , in a trivial sense , the limit law is @xmath241 . what forbids here the convergence to the normal law is the fact that the variance is not finite , so that the normal central limit theorem does not apply . at variance with the poisson and gaussian previous examples , however , we do not know non trivial forms of a cauchy limit theorem embodying the stability of the cauchy law . in other words we have neither explicit examples , nor general theorems characterizing the form of the normed sums of independent and identically distributed random variables whose laws converge to @xmath241 , without being coincident with @xmath241 at every step @xmath60 of the limiting process . this last remark holds also in the case of the laplace selfdecomposable , but not stable laws @xmath242 . in fact , since the @xmath224 family is closed under composition and decomposition , we can always take a triangular array @xmath281 for @xmath257 and @xmath68 , and remark first that they are uniformly , asymptotically negligible by virtue of the chebyshev inequality ( the @xmath66 have finite variance @xmath282 ) , and then that for every @xmath60 @xmath283 so that the limit law trivially is @xmath242 . it must also be said that in this example the random variables of the triangular array change type with @xmath60 so that the corresponding consecutive sums @xmath63 can not be recast in the form of normed sums of independent random variables . since however the laplace laws are not only infinitely divisible , but also selfdecomposable we would expect to find normed sums of independent ( albeit not identically distributed , because the laplace laws are not stable ) random variables whose laws converge to @xmath242 . unfortunately we do not have general theorems characterizing the needed sequences of independent random variables , and we can just show an example slightly more general than the previous one . take for instance the triangular array @xmath66 with laws @xmath284 and variances @xmath285 . they are uniformly , asymptotically negligible because from the chebyshev inequality we have @xmath286 while , for a known property of the harmonic numbers , the consecutive sums are @xmath287 now the sums @xmath63 are not trivially distributed according to @xmath242 at every @xmath60 , but again they can not be put in the form of normed sums as they should since @xmath242 is selfdecomposable . finally similar remarks can be done for the selfdecomposable , but not stablestudent laws @xmath243 introduced in the section [ examplelaws ] , but in this last case it is not even possible to give a simple form to the trivial consecutive sums because the student family @xmath288 is not closed under composition and decomposition . in other words if @xmath77 is the characteristic function of a law @xmath243 we are sure that @xmath273 again is the characteristic function of a infinitely divisible law such that @xmath274 for every @xmath60 , but these component laws @xmath275 no longer belong to the @xmath288 family as happens for the variance gamma family , and in fact the form for their density functionis rather complicated @xcite . since all the laws of our examples are infinitely divisible we can use all of them to generate the corresponding lvy processes by using ( [ stchf ] ) to give the law of the increments on a time interval of width @xmath159 . in particular we will explicitly do that for the laws @xmath222 , @xmath242 , @xmath241 and @xmath243 . we then get as stationary increment characteristic functions @xmath167 respectively @xmath289 it is then apparent that the laws of the increments for the @xmath142stable wiener and cauchy processes are simply @xmath290 and @xmath291 , while for the laplace process the law of the increments is actually a laplace law only for @xmath171 , while in general it is a @xmath292 at other values of @xmath159 . for our student process , on the other hand , the situation is less simple because the student family is not closed under convolution , and the increment law no longer is in @xmath288 for @xmath172 . in this case it is not easy to find the actual distribution from its fourier transform @xmath167 , and only recently it has been suggested that the increments are distributed according to a mixture of other student laws ( for further details see @xcite ) . the wiener and the cauchy processes are @xmath188selfsimilar with @xmath193 and @xmath293 respectively . this can also be seen by looking at the interplay between the two spatial and temporal scale parameters @xmath58 and @xmath168 . in fact in the wiener and cauchy processes these two scale parameters appear in two combinations respectively @xmath294 and @xmath295 such that a change in the time units can always be compensated by a corresponding , suitable change in the space units ; as a consequence the distribution of the process is left unchanged by these twin scale changes . this , on the other hand , would not be possible in the laplace and student processes since @xmath58 and @xmath168 no longer appear in such combinations . that notwithstanding we can achieve selfsimilarity in additive , non stationary processes produced by all our selfdecomposable laws . from ( [ ssimchf ] ) in fact we can give the characteristic function @xmath161 of the increments in the interval @xmath296 $ ] for our four types of law . first of all from the normal type @xmath222 we get @xmath297 namely @xmath298 these laws define processes which coincide with the usual wiener process if and only if @xmath193 . for @xmath205 , on the other hand , our process is additive , @xmath188selfsimilar , and gaussian with non stationary increments , and hence does not even coincide with a fractional brownian motion which has stationary and correlated increments . in a similar way from the laws of the cauchy type @xmath241 we get @xmath299 and the process will coincide with the stationary ( lvy ) cauchy process when @xmath293 , while for @xmath300 we have an additive , @xmath188selfsimilar process with non stationary increments . from the laplace type @xmath242 on the other hand we obtain an @xmath188selfsimilar ( with @xmath186 ) , additive process when we take @xmath301 \frac{1}{1+a^2\left(\frac{t}{t}\right)^{2h}u^2}\,;\ ] ] so that , for @xmath302 , the law of the increment on an interval @xmath296 $ ] actually is a mixture with time dependent probabilistic weights of a law degenerate in @xmath180 and of a laplace law : @xmath303 \mathcal{l}\left(\frac{a^2t^{2h}}{t^{2h}}\right)\,.\ ] ] in other words this means that for @xmath302 there is always a non zero probability that in @xmath296 $ ] the process increment will simply vanish . finally from the student type @xmath243 we get the additive , @xmath188selfsimilar , non stationary process with @xmath304 so that @xmath305 , while nothing simple enough can be said of the independent increment laws . all the lvy processes introduced in the section [ exampleproc ] can now be used as driving noises of ornstein uhlenbeck processes according to the discussion of the section [ ouproc ] . here we will only list the essential properties of the corresponding stationary distributions by analyzing their logarithmic characteristics ( [ oulch ] ) . first of all , if @xmath36 is the parameter of the process as in ( [ ousde ] ) ( remember that we took there @xmath210 for simplicity ) , for the wiener and the cauchy driving noises we immediately have from ( [ oulch ] ) that the logarithmic characteristics @xmath174 of the stationary distributions are respectively @xmath306 namely that the stationary distributions simply are @xmath307 and @xmath308 . in the case of the wiener noise ( namely in the case of the ordinary , gaussian ornstein uhlenbeckprocess ) this means that the stationary distribution has a variance @xmath309 while the gaussian law generating the wiener process had a variance @xmath37 . for the cauchy noise , on the other hand , there is no variance to speak about . for the other two ornstein uhlenbeck lvy noises ( laplace and student ) an important role is played by the so called _ dilogarithm _ function @xcite @xmath310 in fact a direct calculation of the integrals ( [ oulch ] ) gives for the logarithmic characteristics @xmath174 @xmath311 from the characteristic functions @xmath173 we can also calculate the stationary variances as @xmath312 and we get @xmath313 and @xmath309 respectively in the laplace and in the student case . remark that when they exist finite the variances of the stationary distributions always are in the same relation with the variance of the law generating the noise : the stationary variance is the generating law variance divided by @xmath314 . the form of the corresponding density functions is not known analytically , but it can be assessed by numerically calculating the inverse fourier transforms of the characteristic functions : the results of these calculations for a couple of particular cases are shown in the figures [ fig04 ] and [ fig05 ] . finally , since the types of laws analyzed in this section are all selfdecomposable , by reversing the previous procedure we can also add a few remarks about the ornstein uhlenbeck driving noises required to have @xmath222 , @xmath242 , @xmath241 and @xmath243 as stationary distributions . in fact we have already said that for our two @xmath142stable cases the stationary distributions are of the same type of the laws generating the driving noise , so that there is essentially nothing to add for the @xmath222 and @xmath241 stationary distributions . as for the other two cases on the other hand we will use the second equation ( [ oulch ] ) to calculate the noise logarithmic characteristics @xmath209 : generating the driving noise ( red lines ) . here @xmath315 so that both the variances ( that of the stationary distribution , and that of @xmath243 ) are equal to 1.,width=491 ] generating the driving noise ( red lines ) . here @xmath315 so that both the variances ( that of the stationary distribution , and that of @xmath243 ) are equal to 1.,width=491 ] @xmath316 in both cases however the characteristic functions can not be elementarily inverted so that we do not have an explicit expression for the increment density functions of the the lvy noises that produce these two stationary ornstein uhlenbeckdistributions . we can only add a few remarks about the laplace case : here the characteristic function @xmath173 does not vanish at the infinity since @xmath317 . as a consequence the lvy noise increment characteristic function can be better written as @xmath191^t = e^{-2bta^2u^2/(1+a^2u^2)}=e^{-2bt}+(1-e^{-2bt})\,\frac{e^{2bt/(1+a^2u^2)}-1}{e^{2bt}-1}\ ] ] so that the independent increments of an ornstein uhlenbeck process with the laplace type @xmath242 as stationary law are distributed according to a time dependent mixture of two laws , one of which is degenerate in @xmath180 . as for the second law of this mixture , it has a density function given by @xmath318 but this integration can not be analytically performed . since many years selfsimilarity is a fashionable subject of investigation , in areas ranging from fractals to long range interactions in complex systems : to have an idea just ask for the papers with the word `` self similarity '' either in their title or in their abstract present on ` arxiv.org ` and you will find @xmath319 articles , and almost 200 of them only in the first six months of 2007 . on the other hand this is a subject that has been approached from many different standpoints producing an unavoidable level of confusion @xcite , while in fact it would be better discussed by placing the reader in the perspective of the general theory of the infinitely divisible ( even non stable ) processes . in the field of mathematical finance the use of non stable lvy processes is widespread , and several families of selfdecomposable laws and processes have been intensively studied in recent years : see for example the case of the generalized hyperbolic family @xcite , of the student family @xcite , and of the variance gamma family @xcite . considerable interest has also been elicited by the use of selfdecomposable laws in connection with the ornstein uhlenbeck processes @xcite , in particular for the stochastic volatility modelling . however , while in econophysics some non stable lvy laws are recognized as possible candidates for a consistent modelling of the underlying processes @xcite , they remain less popular in the field of statistical mechanics and only recently their use has been proposed in connection with applications to the technology of accelerator beams @xcite in this paper we have tried to elucidate just a few points in the framework of the theory of stochastic processes . in particular we focused our attention on the relation between on the one hand the selfsimilarity , and on the other the independence and the stationarity of the increments . this has led our inquiry toward the analysis of the laws of the process increments , and we have stressed the connection between the selfsimilarity of the process and the selfdecomposability of the increment laws . selfdecomposable laws naturally arise in the study of the central limit problem and of its solutions : in fact they are an intermediate ( and more elusive ) class of distributions between the more general infinitely divisible , and the more particular ( and more popular ) stable distributions . we found then that , in the case of selfdecomposable generating laws , both the stationarity of the increments and the selfsimilarity are always possible , but are not always present in the same process . on the other hand selfsimilarity can also be a property of ( non markovian ) processes with non independent increments as in the case of the fractional brownian motion . we finally stressed the connection between the selfdecomposability and the stationary laws of generalized ornstein uhlenbeck processes with non gaussian , lvy noises . all that has also be elucidated by means of a few particular examples , and some kind of application from physics to finance has also been pointed out . y hu and b ksendal , _ fractional white noise calculus and applications to finance _ , preprint 10/1999 in pure mathematics of the department of mathematics of the oslo university ; available at ` http://www.math.uio.no/eprint/pure_math/1999/10-99.html `

we report on the identification of the new galactic center ( gc ) transient as a likely low mass x - ray binary ( lmxb ) located at only 16 arcsec from . this transient was detected on 2016 february during the gc monitoring , and it showed long - term spectral variations compatible with a hard to soft state transition . we observed the field with on february for 35 ks , detecting the source in the soft state , characterised by a low level of variability and a soft x - ray thermal spectrum with a high energy tail ( detected by up to kev ) , typical of either accreting neutron stars or black holes . we observed : i ) a high column density of neutral absorbing material , suggesting that is located near or beyond the gc and ; ii ) a sub - solar iron abundance , therefore we argue that iron is depleted into dust grains . the lack of detection of fe k absorption lines , eclipses or dipping suggests that the accretion disc is observed at a low inclination angle . radio ( vla ) observations did not detect any radio counterpart to . no evidence for x - ray or radio periodicity is found . the location of the transient was observed also in the near - ir with grond at mpg / eso la silla 2.2 m telescope and vlt/ pre- and post - outburst . within the error region we find multiple objects that display no significant variations . [ firstpage ] x - rays : binaries , methods : observational , techniques : spectroscopic

1606.01138

( genzel et al . 2010 ) , the supermassive black hole ( bh ) at the galactic center ( gc ) , generates a deep gravitational potential that is expected to bind tens of thousands of stellar remnants , such as neutron stars ( ns ) or bh that have settled dynamically into the central parsec of the milky way ( morris 1993 ; lee 1995 ; miranda - escud & gould 2000 ; muno et al . deep radio surveys were so far unsuccessful in detecting either ordinary or millisecond pulsars , leading to the `` missing pulsar problem '' ( wharton et al . 2012 ; dexter & oleary 2014 ) . surprisingly a few years ago the magnetar sgrj1745 - 2900 was discovered via its outburst activity ( mori et al . 2013 ; rea et al . 2013 ; eatough et al . 2013 ; kaspi et al . 2014 ; coti zelati et al . 2015 ) , representing the first radio and x - ray pulsar discovered in the galatic center region ( at less than a few parsecs from ) . many millisecond radio pulsars are expected in the region , but currently not a single one has been detected . due to the three body interaction between this concentration of degenerate stars and multiple stellar systems , a large population of x - ray binaries is also expected . the monitoring of x - ray transients allows us to study this population ( pfahl & loeb 2004 ; muno et al . 2005 ; hopman 2009 ; faucher - giguere & loeb 2011 ) . the majority of accreting x - ray binaries spend most of their time in a quiescent state ( below the detection limit ) , sporadically interrupted by outbursts during which the x - ray luminosity rises by many orders of magnitudes ( fender & belloni 2012 ) . pioneering works on monitoring of gc x - ray sources have been performed with and ( muno et al . 2005 ; wijnands et al . 2006 ; campana et al . 2009 ; degenaar et al . 2012 ; 2015 ) . was discovered by the satellite on feb . 6@xmath3 2016 ( reynolds et al . 2016 ) during the first x - ray observation after the gc exited the solar constraint window . the x - ray telescope ( xrt ) data showed a bright ( @xmath4 erg @xmath5 s@xmath6 , in the @xmath7 kev range ) transient located at less than 16 arcsec from ( reynolds et al . 2016 ) . was detected also in the 20 - 80 kev energy range ( esposito et al . 2016 ) , while observations with the karl g. jansky very large array ( vla ) and giant metrewave radio telescope ( gmrt ) did not detect the radio counterpart ( however the limits still leave open the possibility of a hard state accreting bh or ns ) or any pulsation ( maan et al . 2016 ; bower et al . 2016 ) . within a week from this discovery , observed the field and refined the transient position to : ra@xmath817:45:40.66@xmath90.34 , dec@xmath8 - 29:00:15.61@xmath90.33 , where no previous x - ray sources were ever detected ( baganoff et al . therefore , is a new transient source . in this paper we report the identification of the source as a low - inclination x - ray binary hosting a neutron star or a black hole , most probably with a low mass companion . we discuss the temporal and spectral characteristics of the transient using the long - term daily x - ray monitoring of the field performed by the satellite , the high energy long term monitoring provided by the satellite , a 35 ks observation obtained in director s discretionary time ( to study the dust scattering halo ) performed about 20 days after the onset of the outburst as well as vla observations . finally , we report on the search for the near - ir counterpart within vlt/ and grond data . ( jansen et al . 2001 ) observed on february 26@xmath3 for 35 ks . we processed this data set starting from the observation data files . the data have been treated using the latest version ( 15.0.0 ) of the science analysis system ( sas ) with the most recent ( 13 march 2016 ) calibrations . we used all epic ( strder et al . 2001 ; turner et al . 2001 ) cameras equipped with the medium filter . the arrival times of the events were corrected to the solar system barycentre , applying the barycen task of sas . we screened the periods of enhanced particle - induced activity by inspecting the epic - pn 10 - 15 kev light curves , binned with 5 s resolution , extracted from a circular region , with 5 arcmin radius , close to the border of the field of view . we then filter out all the periods with more than 0.85 cts s@xmath6 . since the main focus of the observation was the study of the full extension of the dust scattering halo around the source , the epic - pn camera was in full frame mode . having a frame time of 73.4 ms , the inner core of the point spread function shows pile up . on the other hand , both epic - mos1 and mos2 were in timing mode , with a frame time of 1.75 ms . this assures that no pile up is present in these datasets . a bad column is falling in the center of the extraction region in the epic - mos1 camera , inducing lower statistics . therefore these data have been used for consistency checks only . to reduce pile up , the source photons in the epic - pn camera were extracted from an annular region . the outer radius of the annulus was limited to 40 arcsec , to avoid contribution from the low mass x - ray binary ( ponti et al . 2015 ) and the inner radius has been fixed to 15 arcsec . by comparing this spectrum from the wings of the point spread function with the pile - up free epic - mos2 spectrum , we verified that it is not affected by pile up . we note that the magnetar sgr j1745 - 2900 ( rea et al . 2013 ) is located inside the source extraction region . however , the magnetar is currently observed at a 2 - 10 kev flux of @xmath10 erg s@xmath6 @xmath5 , about three orders of magnitude lower than the instantaneous 2 - 10 kev flux of ( @xmath11 erg s@xmath6 @xmath5 ) . therefore , sgr j1745 - 2900 gives a negligible contribution to the light curves and spectra of . a strong contribution from the emission of the supernova remnant sgr a east ( maeda et al . 2002 ; ponti et al . 2015 ) is expected within the epic - pn extraction region of , therefore we extracted the background plus diffuse emission from a long archival observation ( obsid 0505670101 ) during which had flux and spectral properties similar to the ones during this observation ( being in the soft state and at similar flux ) , the magnetar was in quiescence and did not show any flare ( ponti et al . 2015 ) . in the epic - mos2 camera , the source photons were extracted between rawx @xmath12 and @xmath13 . the larger extraction region , imposed by the timing mode , implies a larger contribution from and diffuse emission , compared to the epic - pn spectrum . we converted the source extraction region ( from rawx coordinates into sky coordinates through the sas task ecoordconv ) and we extracted the background plus diffuse emission from the epic - mos2 camera , during the same long observation ( obsid 0505670101 ) with in quiescence . because of the higher contamination by we use the epic - mos spectra for comparison only . we note that the same best fit parameters ( with all values consistent with the ones obtained by fitting the epic - pn spectrum ; [ specxmm ] ) were obtained by fitting the epic - mos spectra . the combination of high column density of absorbing material ( @xmath14 @xmath5 ) and the intense diffuse emission in the direction towards the galactic center , prevents us from obtaining a good characterisation of the source emission below @xmath15 kev ( e.g. , the low energy absorption edges ) . for this reason in this work we do not consider the data from the reflection grating spectrometer ( rgs ) instruments . spectral fitting was performed using xspec version 12.8.2 . the spectra were grouped so that each bin contains at lest 30 counts . the _ swift _ x - ray telescope ( xrt ) has been monitoring the central @xmath16 arcmin of the galaxy daily since year 2012 ( degenaar et al . we downloaded all the _ swift _ pc mode datasets between 5 - 2 - 2016 and 10 - 3 - 2016 from the heasarc data archive . this comprises 31 pc mode xrt observations . to characterise the contribution from the background plus complex diffuse emission within the extraction region of ( that contains a contribution from sgr j1745 - 2900 and ) , we used the data before the outburst of . since there was no _ swift _ observation of this region between 03 - 11 - 2015 and 05 - 02 - 2016 , we retrieved 15 _ swift _ observations between 15 - 10 - 2015 and 02 - 11 - 2015 in xrt pc mode when the source was quiescent . these 15 observations are used to measure the background plus diffuse emission underneath . we applied the standard data reduction following the threads described on the _ swift _ website . because of the high absorption column density towards the source , we considered only data in the 2 - 10 kev band . images and spectra were produced using the xselect ( v2.4c ) package . since is very bright , significant pile up is affecting all the pc mode observations . we followed the standard xrt pile - up thread to determine the pile - up level . we extracted the products from an annulus with inner radius of 15@xmath17 , to ensure that spectra , fluxes and light curves are not affected by pile up . a 40@xmath17 outer radius is used to include most source flux and less contamination from . we manually adjusted the astrometry of all datasets by visually inspecting the images of all the observations . this has been obtained by adjusting the relative position between and ax j1745.6 - 2901 . this allows us to correct the astrometry shifts , which would otherwise affect the psf correction and the correct determination of the source flux . by comparing the light curve with that derived without psf correction , we ensured that the uncertainties in the psf correction are not the cause of the observed source variability . we also note that the background flux underneath is only 3% of the source flux in the 2 - 10 kev band , so it does not significantly contaminate the light curve of , either . in addition to the pc mode data , there are also 12 _ swift _ xrt windowed timing ( wt ) mode observations of . compared to the pc mode , the wt mode provides higher time resolution and more counts , but the 1-dimensional image in the wt mode makes it difficult to determine the background underneath the source region from previous observations as we did in the pc mode . however , since the background flux is only 3% of the source flux , we use anyway the wt mode observations as an independent check of the pc mode results . and @xmath18 , respectively , and binned at 200 s. no bursts ( not even in 1 s binned light light curves ) or dips are observed . the red dashed line shows the best fit with a constant ( that results in 43.9 cts s@xmath6 and @xmath19 for 170 dof ) . ] is not piled up in the wt mode , so we defined a circular source region of 20@xmath17 radius , centred on the pixel which is the perpendicular projection of the source position on the 1-dimensional wt mode image ( again we consider only data in the 2 - 10 kev band ) . the choice of 20@xmath17 radius was a compromise between including more source flux and less contamination from . for eleven wt mode observations , there were also pc mode observations on the same day , for which we could compare the spectra between the two modes . for nine observations , both the spectral shape and normalization were consistent between the two modes , except that wt mode provided more counts and thus higher s / n . the remaining two wt mode spectra ( obsid : 00035063134 , 00035063136 ) have the same spectral shape as in the pc mode but with different normalisation . we found that this is caused by the bad pixels within the source extraction region in these two wt mode observations , which affected the source flux and psf correction , but did not affect the spectral shape . we excluded these two observations from further analysis . since the source spectra from wt mode observations were consistent with the pc mode , and the background underneath was much weaker than the source itself , we decided to apply the same background flux and spectra for the wt mode data as for the pc mode , and this should not affect our results . the exposure maps and ancillary files for the spectra were computed with the xrtexpomap and xrtmkarf tools provided in the heasoft ( 6.18 ) software . [ [ section ] ] the outburst of the source was observed by starting from 2016 february 11 , i.e. during satellite revolution 1643 . at the time of writing , only consolidated data up to revolution 1652 were made available and thus we were able to observe the evolution of the source hard x - ray emission until 2016 march 6 . as the source is located right into the galactic center , which is a complex and crowded region , we also made use of the last data collected toward this region before the outburst of . these included data during revolutions 1597 , 1603 , and 1604 , covering the period from 2015 october 11 to 19 . observations are divided into `` science windows '' ( scws ) , i.e. , pointings with typical durations of @xmath202 - 3 ks . we considered only scws where the source was within a maximum off - axis angle of 12 deg from the satellite aim point , in order to reduce the uncertainties on the energy calibrations . all data were analyzed by using version 10.2 of the off - line scientific analysis software ( osa ) distributed by the isdc ( courvoisier et al . 2003 ) . we first extracted the ibis / isgri ( ubertini et al . 2003 ; lebrun et al . 2003 ) mosaics in the 20 - 100 kev energy band and the jem - x ( lund et al . 2003 ) mosaics in the 3 - 10 kev energy band by summing up all data in revolutions 1597 to 1604 . no source was detected at a position consistent with in the 20 - 100 kev energy band and we estimated a 3@xmath21 upper limit on the corresponding energy range of 3 mcrab ( corresponding to roughly 5.4@xmath2210@xmath23 erg @xmath5 s@xmath6 ) . from the jem - x mosaic , we noticed that the well known transient lmxb axj1745.6 - 2901 ( ponti et al . 2015 ) was detected at a position consistent with that of . the limited spatial resolution of the jem - x instrument does not allow us to disentangle the emission of axj1745.6 - 2901 from the one of . therefore , we decided to extract a jem - x spectrum of axj1745.6 - 2901 from observation taken before the outburst of . we then used this file as a background for the spectrum of , that we obtained from the later observations ( we checked in the data that did not go through major flux and/or state changes during the considered period ) . the jem - x spectrum of axj1745.6 - 2901 could be well fit ( @xmath24/d.o.f.=0.5/5 ) with a power - law model of photon index @xmath25 ( @xmath26 erg @xmath5 s@xmath6 ) . for all revolutions from 1643 to 1652 , we built the ibis / isgri and jem - x mosaics and extracted the source spectra in 13 energy bins for isgri and 8 energy bins for jem - x ( we used data from both the jem - x1 and jem - x2 units ) . the jem - x spectrum obtained during revolution 1643 revealed a 3 - 20 kev flux of @xmath2710@xmath28 erg @xmath5 s@xmath6 , therefore axj1745.6 - 2901 contaminated the emission from by no more than 30% ( see also fig . [ fig : mosaic ] ) . as axj1745.6 - 2901 was not detected in the previous ibis / isgri mosaic , the recorded 20 - 100 kev x - ray flux of 7.4@xmath2210@xmath29 erg @xmath5 s@xmath6 is to be completely attributed to . this object remained bright in all data we used , and was detected at high significance in the ibis / isgri and jem - x mosaics of all revolutions . vla observations of were carried out on two epochs , 25 feb 2016 and 25 mar 2016 . as we will discuss later ( see [ secswift ] ) , was in the soft state during at least the first of these vla observations . in each epoch , data were obtained in two frequency bands , c and ku , with center frequencies of 6.0 and 15.0 ghz , respectively . the vla correlator was simultaneously configured for imaging and a phased array beam centered on . the vla was in its c configuration for both observations with a maximum baseline length of 3 km . imaging data were obtained over frequency ranges of 3.976 8.072 ghz and 12.952 17.048 ghz , respectively . the correlator was configured to produce cross - correlations that were split into 64 frequency windows each with 128 channels in dual circular polarization . these imaging data were reduced using the casa vla pipeline . the radio galaxy 3c 286 was used to set the flux calibration and the compact source j1744 - 3116 was used as a phase calibration source for . additional steps of self - calibration with as the reference were performed using only visibilities with baselines longer than 50 @xmath30 and 20 @xmath30 in ku and c bands , respectively . images of the field using only these long baseline visibilities were produced . these long baselines filtered out much of the substantial emission associated with the northern arm of sgr a west . details of the epochs and images are summarized in table [ tab : vla ] . no radio source was identified at the location of in either epoch at either frequency . [ cols= " < , < , > , < , > " , ] we also recorded high time resolution data from a phased array beam at the position of . in both frequency bands , these data were taken with 4.096 ghz total bandwidth divided into 1024 frequency channels at 250 @xmath31s time resolution . we searched for dispersed , periodic signals over dispersion measures ranging from 0 to 10,000 pc/@xmath32 and pulse frequency drift ( source acceleration ) up to @xmath33 hz / s . at both frequencies , pulses from the bright , nearby 3.76-s magnetar psr j1745 - 2900 were detected with high signal to noise ratio , but no other significant periodicities were found . assuming 10% pulse duty cycle , the 10-sigma flux density limits for the periodicity searches were 30 @xmath31jy at c - band and 45 @xmath31jy at ku - band . throughout the paper we assume a distance to the source equal to 8 kpc ( genzel et al . 2010 ; gillessen et al . error - bars are quoted at the 90 per cent confidence level for a single parameter of interest . we also assume that the accretion disc is observed face on . this is in agreement with the indication ( 3.1 ) that the system is observed at low inclination ( @xmath34 ; frank et al . 1987 ; diaz - trigo et al . 2006 ; ponti et al . would the inclination be @xmath35 , the inner disc radii and luminosities would be 1.4 and 2 times larger , respectively . figure [ cluce ] shows the psf corrected epic - pn light curve in the 2 - 10 kev band with 200 s bins . no variability is observed and the data are well fit with a constant that is shown by the red dashed line ( @xmath19 for 170 degrees of freedom ; dof ) . no bursts or dipping events are observed during the observation ( white & mason 1985 ; frank et al . 1987 ; diaz - trigo et al . 2006 ) . to refine our capability to detect bursts and dips we investigated a variety of energy bands and time bins . no bursts are detected in light curves with time bins , as short as sub - second . we also investigated soft energy bands , where the effect of dipping is stronger ( diaz - trigo et al . 2006 ; ponti et al . 2016 ) , and their hardness ratio , but no dipping events were found . we extracted mos and pn light curves in the energy bands 2 - 10 kev , 2 - 6 kev , and 6 - 10 kev , with a time resolution of 3.5 ms and 200 ms , respectively . the power spectral density function ( psd ) was computed in each energy band and for each instrument separately . the poisson noise level was estimated from the mean power at frequencies @xmath36 hz , where counting noise variability dominates . the relatively long frame time of the epic pn instrument in full frame observing mode prevents us from sampling these frequencies . therefore , the poisson noise level can not be accurately estimated , thus we did not consider the epic - pn psd further . figure [ psd ] shows the psd of the mos data in the 2 - 10 kev band . for comparison we also plot the psd of the bh x - ray binary gx 339 - 4 in a high - luminosity hard state ( de marco et al . 2015 ) . swiftj174540.7 - 290015 displays significantly less power than observed in a typical hard state of a bhxrb . indeed , the 2 - 10 kev fractional root - mean - square ( rms ) variability amplitude ( e.g. nandra et al . 1997 ; vaughan et al . 2003 ; ponti et al . 2004 ) in the 0.1 - 64 hz frequency interval is estimated to be @xmath37 . this value is consistent with what is typically observed in either a bh or ns xrb in a soft / soft - intermediate state ( muoz - darias et al . 2011 ; 2014 ) . we searched for periodic signals using the pn and moss data , as well as wt - mode data ( being sensible to signals between the nyquist frequency , and the frequency resolution of each dataset ; van der klis 1988 ) . we used a modified version of the xronos analysis software to search for periodicities following the prescriptions described in vaughan et al . ( 1994 ) , and we did not find any periodic or quasi - periodic signal in any of the reported x - ray observations ( we had accounted for the number of bins searched , and the different d.o.f . of the noise power distribution in the non - detection level ) . we have searched all dataset performing fast fourier transforms over the total length of the observations , but also over small intervals of 0.5 , 1 , 3 , and 5ks in order to search for signals that might have been possibly smeared by doppler shifts due to hour - long orbital periods . unfortunately , the dataset with by far the largest number of counts , the epic - pn , had a timing resolution of @xmath38ms , making our searches rather insensitive to fast spinning pulsars ( i.e. with spin periods of @xmath20 twice this value ) . we computed the 3@xmath21 upper limits on the sinusoid semi - amplitude pulsed fraction ( @xmath39 ) according to vaughan et al . ( 1994 ) and israel & stella ( 1996 ) . the deepest limits at frequencies @xmath40hz were derived from the epic - pn data , having the larger number of collected photons . note that given the nature of this source , if a very short orbital period causes doppler smearing of the putative signal ( i.e. if the signal is not strong enough to be detected in our small @xmath201ks chunks that might be free of doppler smearing ) , these pf limits are not constraining . to have a handle on the pf limits on faster periodicities ( anyway smaller than 300hz ) , we used the moss data ( taken in timing mode : 1.75ms timing resolution ) as well as the wt data ( 1.7ms timing resolution ) , although having to cope with a much reduced number of counts , and again using the full length of the observation and searching in small consecutive time intervals . we show the resulting 3@xmath21 upper limit on the pf of a periodic signal ( if not doppler smeared in the presence of a short orbital period ) for both pn and mos2 data in fig . [ pf ] . ^2 $ ] ) of swift j174540.7 - 290015 ( red squares ) , computed from mos2 data in the 2 - 10 kev energy band . the 2 - 10 kev psd of gx 339 - 4 in a high - luminosity hard state ( de marco et al . 2015 ) is reported for comparison ( gray dots ) . ] upper limits as derived for the 2 - 10kev epic - pn . the fractional pf is in units of 100 % . _ ( upper inset ) _ mos2 power density spectra . ( _ lower inset ) _ fractional pulsed fraction 3@xmath21 upper limits as derived for the 2 - 10kev mos2 . ] we fit the epic - pn spectrum by always including an absorption component from neutral matter plus the contribution from the dust scattering halo . the neutral absorption is fitted with the model tbnew ( see wilms et al . 2000 ; 2011 ) with cross sections of verner et al . ( 1996 ) and abundances of wilms et al . the dust scattering halo is fitted with the model dust ( predehl & schmitt 1995 ) . we assume that the dust scattering optical depth at 1 kev is tied to the x - ray absorbing column density by the relation : @xmath41 ; as in nowak et al . 2012 ) . however , we note that a more proper modelling of the dust scattering effect on the observed spectrum requires better knowledges about the location and properties of the dust along the line of sight ( smith , valencic & corrales 2016 ) , as well as a better constrained location of , which would then require detailed analysis of the extended emission from the dust scattering halo . this is already beyond the scope of this paper , but will be included in our further study . we first fit the spectrum with a series of single component models . we start with an absorbed power law model ( dust*tbnew*powerlaw in xspec ) . large residuals remain at high energy ( @xmath42 for 1189 dof ; see tab . [ single ] ) . moreover , the best fit power law spectral index is unrealistically steep ( @xmath43 ) , clearly indicating the thermal nature of this emission . therefore , we fit the spectrum with either an absorbed black body ( dust*tbnew*bbody ) or an absorbed multi - temperature disk black body component ( dust*tbnew*diskbb ) . none of these two models provides an acceptable fit , all leaving significant residuals at high energies ( see tab . [ single ] ) . -0.2 cm c c c c c c + + @xmath44 & @xmath45 + @xmath46 & @xmath47 + @xmath48 & @xmath49 + @xmath50 & 1530.9 / 1189 + + @xmath44 & @xmath51 + @xmath52 & @xmath53 + @xmath54 & @xmath55 + @xmath50 & 2897.9 / 1189 + + @xmath44 & @xmath56 + @xmath57 & @xmath58 + @xmath59 & @xmath60 + @xmath50 & 2479.3 / 1189 + c c c c c + & & & + model & ( 1 ) & ( 2 ) & ( 3 ) & ( 4 ) + @xmath44 & @xmath61 & @xmath62 & @xmath63 & @xmath64 + @xmath52 & @xmath65 & & @xmath66 & @xmath67 + @xmath54 & @xmath68 & & @xmath69 & @xmath70 + @xmath57 & & @xmath71 & @xmath72 & @xmath73 + @xmath59 & & @xmath74 & @xmath75 & @xmath76 + @xmath46 & @xmath77 & @xmath78 & + @xmath79&@xmath80 & @xmath81 & + @xmath50 & 1255.2 / 1187 & 1272.8 / 1187 & 1246.2 / 1187 & 1262.7 / 1187 + the large residuals at high energy clearly indicate the presence of a second emission component . therefore , we add to the thermal emission model the contribution provided by a thermal comptonisation component ( reproduced by the nthcomp model ; zdziarski et al . 1996 ; zycki et al . 1999 ) . being the high energy cut off outside of the energy band , we fix the electron temperature to a high value of @xmath82 kev ( consistent results are obtained by fixing the electron temperature to e.g. 20 or 100 kev ) . the fit with an absorbed black body plus thermal comptonisation model provides a satisfactory description of the data ( see tab . [ double ] ) . the best fit black body temperature is @xmath83 kev produced from a surface area with radius @xmath84 km . the fit with an absorbed multi - temperature disc black body plus comptonisation emission , provides a temperature of @xmath85 kev and a very reasonable inner disc radius of @xmath86 km ( or @xmath87 r@xmath88 , with @xmath89 , where @xmath90 is the gravitational constant , @xmath91 is the mass of the primary , assumed to be @xmath92 m@xmath93 and @xmath94 is the speed of light ) , however the fit is slightly worse ( see tab . [ double ] ) . nevertheless , in both these cases , we observe a rather steep spectral index of the comptonisation component , with values of @xmath95 . therefore , we substituted the comptonisation component with a second thermal component . on a statistical ground we obtained an improved solution with the sum of an absorbed black body ( with @xmath96 kev ) plus a multi - temperature disk black body emission ( @xmath97 kev ) . however , we note that the best fit inner radius of the disc black body component is @xmath98 km , therefore inside the ns surface ( see tab . [ double ] ) . for this reason we think that this solution is unlikely . we note a second minimum in the @xmath99 distribution ( @xmath100 for 1187 dof ) , leading to a solution with @xmath101 kev plus a multi - temperature disk black body emission ( @xmath102 kev ) . the best fit inner radius of the disc black body component now results to be @xmath103 km ( corresponding to @xmath104 @xmath105 ) , while the black body component is produced from a region with radius of @xmath106 km ( tab . [ double ] ) . in conclusion , the shape of the spectrum is dominated by a thermal component ( either black body or disc black body , with temperature of @xmath107 , @xmath108 kev ) plus an additional component at higher energy that could be reproduced either by a steep comptonization component ( perhaps associated with the corona seen in accreting systems ) or a black body emission ( possibly generated by the hotter parts on the surface of a ns ) . any of these possibilities appear viable , based uniquely on the data . in [ secswift ] and [ secinte ] , we will investigate further the source radiative process and we will break this model degeneracy by studying the and long term spectral evolution . high inclination accreting bh and ns are known to display ionised absorption features , whenever they are observed in the soft state ( neilsen et al . 2009 ; ponti et al . 2012 ; 2016 ) . no narrow fe k absorption or emission line is evident in the spectra of . fitting the spectrum with additional narrow gaussian absorption lines , we compute upper limits on the line equivalent width ( ew ) as stringent as @xmath109 ev to the presence of fe xxv and fe xxvi lines , therefore excluding the presence of either a disc wind or an ionised atmosphere ( diaz - trigo et al . 2006 ; ponti et al . 2012 ; 2014 ; 2015 ; 2016 ; miller et al . 2015 ) . similar upper limits are valid for narrow emission lines between 6.4 and 7 kev . we also note that the addition of a broad fe k emission line is not required . the observation of this bright state of allows us to investigate the properties of the ism along a line of sight placed at less than @xmath110 arcsec from ( baganoff et al . 2016 ) . to perform the investigation of the x - ray properties of the ism , we reproduced the source emission with the absorbed disc black body plus black body model ( model 4 in [ sec - xmm - meanspec ] and tab . [ double ] ) . the absorption is fitted with the tbnew model and the effect of the dust scattering halo was taken into account ( see tab . [ double ] ) . we observed that using the phabs absorption model , we measure a column density of absorbing material ( @xmath111 @xmath5 ) consistent with the values observed using tbnew or tbabs . on the other hand , a significantly lower value is observed if either the wabs model ( @xmath112 @xmath5 ) or the anders & grevesse ( 1989 ) solar abundances ( @xmath113 @xmath5 ) are used . starting again from the best fit baseline model we explored the possibility to constrain the abundances of the various elements . we did so by leaving the abundance of any metal producing edges in the 2 - 10 kev band free to vary ( see fig . [ spec ] ) . we observe that the abundances of si , cl , cr , ca , and co are unconstrained , therefore we left their abundances fixed at the solar value . on the other hand , iron ( @xmath114 solar ) and argon ( @xmath115 solar ) are under abundant , compared to solar , while sulphur ( @xmath116 solar ) appears to be slightly overabundant . we note that a larger column density of absorbing material is required ( @xmath117 @xmath5 ) , once the iron abundance is left free to vary . the photon indices are derived by fitting an absorbed power law to the 2 - 10 kev spectrum of every observation , separately , assuming @xmath118 from fitting the spectra . the square symbols indicate the hard state observations ( filled blue - pc mode , empty purple - wt mode ) , while the circular symbols indicate the soft state observations ( filled red - pc mode , empty orange - wt mode ) . the black star indicates the _ xmm - newton _ observation . ] fixed at @xmath119 ) . the square symbols indicate the hard state observations ( filled blue - pc mode , empty purple - wt mode ) , while the circular symbols indicate the soft state observations ( filled red - pc mode , empty orange - wt mode ) . the black star indicates the _ xmm - newton _ observation . ] and [ fig - spec ] ) . the orange empty circles are in the wt mode . panel - d shows the hardness ratio between the hard band ( 5 - 8 kev ) and soft band ( 2 - 5 kev ) . in all panels , the black star and the dashed line indicate the _ xmm - newton _ observation on 26 - 02 - 2016.,scaledwidth=49.0% ] , @xmath120 kev ) , while the hard state by an absorbed power law component ( @xmath121 , @xmath122 ) . ] we first fitted all spectra with an absorbed power law model dust*tbnew*powerlaw . due to the lower statistics of the single @xmath123 ks spectra , we assumed a column density of @xmath124 ( see tab . [ single ] ) as found in fitting the higher resolution _ xmm - newton _ epic spectra . this model can not satisfactorily reproduce all spectra , especially when enters softer states . nevertheless , by applying this simple model we can probe the evolution of the spectral shape by tracking the variations of the power law photon index . figure [ fig - specpl ] shows the best fit power law photon index to the 2 - 10 kev band . we observe a gradual steepening of the spectral shape with time . in particular , very steep spectral shape ( with @xmath125 ) are observed in the later observing period . such steep spectral indices might indicate a thermal nature of the emission mechanism . therefore , we then fitted all the spectra with an absorbed multi - temperature disk black body model dust*tbnew*diskbb . [ fig - tflux ] shows the best fit inner disc temperature vs. the inner disc radius . a kendall s rank correlation test provides a correlation coefficient of @xmath126 and null hypothesis probability of @xmath127 . despite its significance , we think that this anti - correlation is induced by fitting the data with the wrong spectral model . indeed , we note that : i ) the spectra accumulated before mdj=57442 ( blue filled and violet open squares in fig . [ fig - specpl ] & [ fig - tflux ] ) are better fit by a power law component ; ii ) the best fit inner disc radii of these spectra is significantly smaller than 10 km , implying that the accretion disc extends either inside the ns surface or inside the event horizon of a stellar mass bh during these observations . on the other hand , all the _ swift _ spectra after mjd=57442 ( red filled and orange open points in fig . [ fig - specpl ] & [ fig - tflux ] ) are well fitted by this absorbed thermal emission model . however , these relatively low s / n spectra can not rule out the presence of an additional component in the hard x - ray as observed by _ xmm - newton_. in fig . [ fig - tflux ] the orange open and red filled points cluster around a small parameter space , indicating that these spectra also have similar shape . moreover , this model can not reproduces the spectra when there is clearly a hard component , as represented by the blue and purple squares in fig . [ fig - tflux ] . fitting these spectra requires a very high temperature ( @xmath128 kev ) of the thermal component and variations of the normalisation that imply unlikely variations of the thermal emitting area ( see fig . [ fig - tflux ] ) . finally we adopt model 2 of tab . [ double ] that is composed by the sum of a thermal multi - temperature disc black body component plus a power law dust*tbnew*(diskbb+powerlaw ) . this model can well produce all the spectra . from top to bottom , fig . [ fig - variability ] shows the source flux , in the 2 - 5 , 5 - 8 and 2 - 8 kev band as well as the hardness ratio plotted against the observation date . during the first few days of observations , showed a clear and significant flux increase , by a factor of 3 - 4 . very significant variability , stronger in the hard band , and a hard x - ray spectrum are observed up to mjd=57442 ( see fig . [ fig - variability ] ) . afterwards , the spectrum softens , with an hardness ratio @xmath129 and no strong variability is observed anymore ( see fig . [ fig - variability ] ) . to enhance the signal to noise , we combined the spectra obtained before and after mjd=57442 . the combined spectra of these two periods are shown in fig . [ fig - spec ] with blue and red symbols , respectively . to capture what are the bulk of the spectral variations and to avoid model degeneracies ( as demonstrated by the higher statistic spectrum ) , we consider here only a single emission component model . the spectral degeneracy will be broken by the addition of the data , above 10 kev . the hard state spectrum is better fit ( @xmath130 for 538 dof ) by a simple absorbed power law model with spectral index of @xmath131 ( @xmath132 @xmath5 ) . a fit with either a single black body or multi - temperature disk black body component provides a significantly worse fit ( @xmath133 and 56 , respectively ) . the combined soft state spectrum is fit by a simple absorbed multi - temperature disc black body component with best fit temperature @xmath134 ( @xmath135 @xmath5 , @xmath136 for 329 dof ) and inner disc radius of @xmath137 km . this value is fully compatible with what is expected by an accreting neutron star or a black hole . note that the best fit temperature is slightly lower than the mean value in fig . [ fig - tflux ] , which is because the column density of neutral material ( @xmath44 ) is allowed to vary when fitting the combined spectra . for the soft state combined spectrum , a fit with an absorbed power law provides a slightly better fit ( @xmath138 for 329 dof ) , however the suspiciously large best fit spectral index ( @xmath139 , @xmath140 @xmath5 ) disfavours this interpretation and leaves the disc black body model as the more physical interpretation . the marked spectral and variability difference before and after mjd=57442 is reminiscent of spectral transitions in accreting x - ray binaries ( fender et al . 2004 ; remillard & mcclintock 2006 ) and it suggests that might have undergone a state transition during the monitoring campaign . radius . the bright candidate counterpart ( 1 ) as well as a fainter one ( 2 ) are indicated . see text for details . north is up , and east is left . ] to break the degeneracies in the and data , induced by the limited energy band considered , we investigated the source behaviour in the energy band . figure [ fig : lcurve ] shows the jem - x and ibis / isgri light - curves of , since the start of the outburst . the source is detected in all energy bands . in particular , a significant drop of hard x - ray emission ( @xmath141 kev ) is observed after day 20 , in correspondence to the state transition inferred from the data ( fig . [ fig : lcurve ] ) , therefore further confirming this scenario ( see [ secswift ] ) . we accumulated all data that are quasi - simultaneous to observations and we created a combined `` hard state '' and `` soft state '' spectrum ( summing all isgri and jem - x spectra from revolution 1643 - 1648 and 1649 - 1652 , corresponding to mjd 57429 - 57441 and 57443 - 57453 , respectively ) . we then performed a broad - band fit by considering these and spectra . as already clear from the analysis of the and data , a single component emission model is not an adequate description of the emission of ( see [ sec - xmm - meanspec ] and [ secswift ] ) . therefore , we began by applying the best fit double component models fitting the spectrum ( see tab . 3 ) to the soft state cumulative and spectra . in all models that we explored , we applied an inter - normalisation factor ( see tab . [ integr ] ) between the different instruments . we first considered the absorbed double thermal component ( cons*abs*(diskbb+bbody ) , see model 4 in tab . [ double ] ) . this model ( see [ sec - xmm - meanspec ] and tab . [ double ] ) , being composed by two thermal components peaking below 2 kev , predicts very little emission above @xmath142 kev . on the contrary , even in the soft state , emits significant radiation above 20 kev . to try to reproduce the hard x - ray emission , this model increases the best fit temperature of the disc black body component and lowers the disc inner radius to unrealistic values . indeed , the best fit temperature is @xmath143 kev and the inner disc radius is @xmath144 km , the latter being too small for either a ns or a bh . for this reason we rule out this model . we then tested the absorbed black body plus comptonisation component ( cons*abs*(bbody+nthcomp ) , model 1 in in tab . [ double ] ) as well as the absorbed disc black body plus comptonisation component ( cons*abs*(diskbb+nthcomp ) model 2 in tab . [ double ] ) . incidentally , we note that even adding the data , the energy of the cutoff in the electron distribution is unconstrained ( @xmath145 kev ) . therefore we fix this value to 50 kev , as in the previous fits ( [ sec - xmm - meanspec ] and tab . [ double ] ) . both these thermal plus comptonisation models provide a good description of the data ( @xmath146 for 331 dof , @xmath147 for 331 dof , respectively ) . we note that the best fit parameters of the absorber , black body and disc black body components are consistent with the ones obtained by fitting the higher statistics spectrum alone ( see tab . [ double ] ) . however , the addition of the data allows us to obtain now more reliable constraints on the parameters of the comptonisation component . in particular , the photon index of the comptonisation component ( @xmath148 ) is now consistent with the values typically observed in accreting compact objects . moreover , the detection of x - ray emission up to @xmath2 kev , allowed us to rule out that the soft state emission is only composed by a combination of pure thermal components . we then also fitted the combined hard state spectrum ( see fig . [ fig : spectra ] ) . the broad band spectra can be fit by an absorbed comptonisation component plus either a black body or a disc black body ( see models 1h and 2h in tab . [ integr ] ) . none of these two models provide a completely acceptable fit ( @xmath149 for 552 dof and @xmath150 for 552 dof , respectively ) . we attribute this to the strong spectral and flux evolution ( e.g. , variation in photon index ) observed during the hard state ( see [ secswift ] ) , that can not be properly taken into account by the fit of these combined spectra . therefore , we refrain the reader from extracting too detailed information from these spectra . it is clear , however , that the hardness ratio changed by a factor of @xmath151 between the hard ( @xmath152 erg @xmath5 s@xmath6 , @xmath153 erg @xmath5 s@xmath6 ) and soft ( @xmath154 erg @xmath5 s@xmath6 , @xmath155 erg @xmath5 s@xmath6 ) state ( see fig . [ fig : spectra ] ) . this reinforces the interpretation that the source underwent a state change during the monitored period . c c c c c c + & & + & & & & & + model & ( 4s ) & ( 1s ) & ( 2s ) & ( 2h ) & ( 1h ) + @xmath44 & @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath160 + @xmath52 & @xmath161 & @xmath162 & & & @xmath163 + @xmath54 & @xmath164 & @xmath165 & & & @xmath166 + @xmath57 & @xmath167 & & @xmath168 & @xmath169 & + @xmath59 & @xmath170 & & @xmath171 & @xmath172 & + @xmath46 & & @xmath173 & @xmath173 & @xmath174 & @xmath175 + @xmath79 & & @xmath176 & @xmath177&@xmath178 & @xmath179 + @xmath180 & @xmath181 & @xmath182 & @xmath182 & @xmath183 & @xmath184 + @xmath185 & @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath190 + @xmath191 & @xmath192 & @xmath193 & @xmath194 & @xmath195 & @xmath196 + @xmath50 & 369.5 / 331 & 357.0 / 331 & 355.9 / 331 & 648.9 / 552 & 659.2 / 552 + the location of was observed with the gamma - ray burst optical near - infrared detector ( grond ; greiner et al . 2008 ) mounted at the mpg / eso 2.2 m telescope in la silla , on 2016 febuary 19.35 ut . observations were obtained simultaneously in the j , and h photometric bands for 640 s integration time each , r@xmath197 , i@xmath197 , z@xmath197 , j , h , k. due to the extreme galactic foreground reddening affecting the optical bands and saturation affecting the k - band , only the j and h band data are discussed here . ] . the data were reduced with the standard tools and methods as described in krhler et al . the photometry was measured from circular apertures with radii corresponding to the image full width at half maximum ( @xmath198 in j and @xmath199 in h ) and calibrated against 2mass stars in the same field . a single bright source , reported also by masetti et al . ( 2016 ) from archival vvv imaging , is seen at the x - ray position . the apparent aperture magnitudes ( in the vega system ) are @xmath200 and @xmath201 . the location of the new transient was also contained in a series of 15 grond observations obtained on 2015 april 26.35 - 26.42 ut when monitoring another galactic center transient . no variability was found for the bright near - ir source in the j or h band during this set of observations and the median magnitudes of @xmath202 ( seeing @xmath198 ) and @xmath203 ( @xmath204 ) are consistent within errors with the post - outburst measurements . the near - ir source density at the position of the transient is very high and grond photometry corresponds with the co - addition of the flux of all sources within the aperture . therefore , we can only conclude , that the brightest source within the error circle , dominating the total emission , remained constant between the pre- and post - outburst observations and none of the fainter objects brightened to a level where it would increase the overall emission within the aperture by more than 0.1 magnitude . the gc is routinely observed with adaptive optics using at the vlt ( see gillessen et al . 2009 for a description of the program and data reduction ) . figure [ naco ] shows a stacked k band image from archival ( pre - outburst ) observations . besides the bright source dominating the grond photometry , multiple fainter objects are resolved either inside or in the close proximity to the error region ( black circle ) . for the bright source ( see fig . [ naco ] ) we measure magnitudes of @xmath205 and @xmath206 while , for comparison , the fainter star ( indicated with 2 in fig . [ naco ] ) has @xmath207 and @xmath208 . the mpe - ir group obtained new data on 2016 march 16 . no significant variation in brightness of any of the sources within the error circle is observed ( gillessen private communication ) . the probability of having such a bright star within the error circle of @xmath209 arcsec@xmath210 can be computed from number of stars with @xmath211 mag within an annulus centred on and with inner and outer radii of @xmath212 arcsec and @xmath213 arcsec , respectively . this annulus includes the position of the putative near - ir counterpart . the observed surface density within the annulus is 0.032 stars arcsec@xmath214 implying a chance coincidence probability of @xmath123 % of finding a source with @xmath211 mag within the error circle , i.e. that the bright source is not associated with the the x - ray transient . using the average reddening towards the galactic center of @xmath215 mag ( fritz et al . 2011 ) , the absolute k - band magnitude is @xmath216 . the star spectroscopically shows the co band heads in the k - band , which excludes stars hotter than spectral class f ( wallace & hinkle 1997 ) . the absolute magnitude would suggest it to be a late type giant or super giant . we report the identification of a new accreting x - ray binary , and results from the early phases of its x - ray emission evolution . reached a peak flux at an absorbed value of @xmath217 erg @xmath5 s@xmath6 in the 2 - 10 kev band . during the observation , shows an absorbed and un - absorbed 2 - 10 kev flux of @xmath218 and @xmath219 erg @xmath5 s@xmath6 , respectively , that corresponds to @xmath220 erg @xmath5 s@xmath6 in the 0.1 - 10 kev band . during the observation , showed an observed 2 - 10 kev luminosity of @xmath221 erg s@xmath6 and an un - absorbed luminosity of @xmath222 erg s@xmath6 in the 0.1 - 10 kev band . therefore , if is a ns with @xmath223 , the observed luminosity is @xmath224 % of the eddington limit during the observation ( @xmath225 % if is a bh ) . given these luminosities , does not belong to the class of faint x - ray transients ( wijnands et al . 2006 ) . we point out that is located at less than 16 arcsec from ( the supermassive bh at the gc ) , less than 17 arcsec from ( rea et al . 2013 ; mori et al . 2013 ; coti - zelati et al . 2015 ) and less than 1.7 arcmin from the neutron star low mass x - ray binary ax j1745.6 - 2901 ( ponti et al . 2015 , 2016 ) , therefore it lies in the field of view of all observations pointed to one of these sources ( ponti et al . 2015a , b , c ; 2016 ) . no evidence for previous outbursts of is present in the , and archive ( degenaar et al . 2012 ; ponti et al . 2015a , b ) . we compute the upper limit on the x - ray emission of during quiescence by stacking all acis - i observations between 1999 and 2011 , for a total exposure of @xmath226 ms . we observe 54 counts within @xmath227 from the position of , in the 0.5 - 10 kev band . this translates in a @xmath228 % upper limit of @xmath229 cts s@xmath6 ( gehrels 1986 ) . assuming the quiescent spectrum of is described either by an absorbed ( @xmath14 @xmath5 ) power law ( with spectral index @xmath230 ) or a black body ( @xmath231 kev ) this can be translated into an un - absorbed luminosity of @xmath232 erg s@xmath6 or @xmath233 erg s@xmath6 . this upper limit on the quiescent x - ray luminosity is consistent with both a ns or bh origin of ( narayan et al . 1997 ; garcia et al . 2001 ; rea et al . 2011 ; armas padilla et al . 2014 ) . radio observations did not detect either a steady or periodic counterpart to . steady emission is expected under two scenarios . first , the fundamental plane of bh activity can be used to predict the radio luminosity under the assumption that the system is in the low hard state ( plotkin et al . for a 10 @xmath234 black hole with the unabsorbed x - ray luminosity observed by , we predict a radio flux density of 6 mjy . the scatter in the fundamental plane is approximately an order of magnitude so that values as high as 100 mjy are possible . the imaging non - detection with an rms of 6 mjy , thus , can neither rule out nor confirm the presence of a jet in this system . we note that by the time the first vla observation has been taken , already transitioned to the soft state . second , bright radio emission can be produced through interaction of a jet with the dense interstellar medium of the galactic center . the low - mass x - ray binary cxogc j174540.0 - 290031 , located only 0.1 pc in projection from , reached a peak flux density of 100 mjy that was resolved into two lobes on either sides of the x - ray ( bower et al . 2005 ) . the proximity of to sgr a west suggests that it could produce similar interactions but we find no evidence for these . the emission of , during the observation is dominated by a thermal component that can be well fitted either by multi - temperature disc black body , or black body emission . in addition , emission at higher energies ( up to 20 - 50 kev ) is clearly detected by . the high energy emission is a clear sign of a comptonization component , ruling out a double thermal component consistent with the fits limited to energies below 10 kev . the disc black - body plus comptonisation scenario is perfectly compatible either with a ns or bh primary . in this framework , the comptonization component might be associated with the coronal emission typically observed in accreting sources and the multi - temperature disc black body could be produced by a standard accretion disc extending down to @xmath235 r@xmath88 for a neutron star ( @xmath236 m@xmath93 ) or @xmath15 r@xmath88 for a bh with @xmath237 m@xmath93 . in the second scenario ( model 1 and 1s in tab . [ double ] and [ integr ] , respectively ) , the presence of the black body emission suggests that the primary is a ns . also in this scenario , the comptonisation might be due to coronal emission , but the soft radiation would be dominated by the boundary layer emission . however , the best fit black body radius appears to be too large ( @xmath84 km ) to be associated with a conceivable ns . nevertheless , we can not exclude that the large observed area , attributed to the black body emission , might be induced by the contribution from a disc emission component not considered in our modelling . alternatively , the black body component might be in part produced by hot spots on the surface of the ns . would this indeed be the case , then a pulsation should be present . however , this pulsation might be undetected either because of the weakness of the black body emission or because is a fast spinning millisecond pulsar ( e.g. , faster than our timing resolution , or a weak signal possibly also smeared by a doppler orbital shift ) . from the spectral behaviour of we can rule out a high mass x - ray binary ( hmxb ) nature of the source . the spectra of hmxbs are characterised by a dominant hard power law in the 0.2@xmath23810.0 kev band with typical photon indices of 1.0@xmath2381.5 for supergiant systems and around 0.9 for be / x - ray binaries ( e.g. haberl et al . 2008 ; bozzo et al . 2012 ; bodaghee et al . 2012 ; walter et al . hmxbs also do not show spectra state transitions as observed from . was initially identified as a possible magnetar due to its fast outburst rise . however , many observational characteristics of this new transient are at variance with what is usually observed in magnetar outbursts ( see rea & esposito 2011 and turolla , zane & watts 2015 for a review on outbursts and on magnetars in general , respectively ) . the strong variability of the outburst flux decay ( see fig.[fig - variability ] ) is unseen in the outburst decay of magnetars , which are usually observed to decay smoothly toward quiescence . this is also in line with the interpretation of these outbursts as being due to a large energy injection in the magnetar crust triggered by the instability of the internal or external magnetic field ( possibly by a strongly twisted magnetic bundle ) , with a consequent crustal heating and subsequent cooling ( beloborodov 2009 ; pons & rea 2012 ) . furthermore , if placed at 8kpc the peak luminosity of the outburst is too high with respect to what is observed for magnetars , @xmath239erg / s , which is a limit believed to be regulated by the strong dependence on temperature of the crustal neutrino emissivity ( pons & rea 2012 ) . the deep upper limits on the pf of this new transient ( see fig.[pf ] ) are rather unlikely ( although a few exceptions exists ) compared with the usual strength ( @xmath2020 - 80% ) of the periodic signals ( around 2 - 10s ) observed for magnetars . therefore , we believe the magnetar interpretation for this new transient is extremely unlikely . observations of the x - ray reflection nebulae ( irradiated molecular clouds within a few hundred parsecs from ) suggest an either solar or supersolar iron abundance of the gc ism ( revnitsev et al . 2004 ; terrier et al . 2010 ; ponti et al . 2010 ; 2013 ; 2014b ; zhang et al . the observed low iron abundance ( @xmath114 solar ) of the neutral absorbing material towards is therefore rather surprising also with respect to the accepted iron gradient in the galactic disk ( friel et al . 2002 ; pedicelli et al . we note that the measured value is consistent with the iron abundance towards 4u 1820 - 303 , an x - ray binary located at a distance of @xmath240 kpc , at less than 1 kpc from the gc ( towards @xmath241 and @xmath242 ; kuulkers et al . 2003 ; pinto et al . the observed under abundance is also in line with the values measured towards a small sample of bright galactic x - ray binaries ( juett et al . we attribute this to depletion of iron into dust grains in the ism ( see juett et al . 2006 ) . indeed , depletion of elements in dust grains can reduce the effective cross section for absorption . in grains , the optical depth can be higher than one , therefore absorption will occur primarily on the surface . moreover , iron atoms located deeper into the grain will be shielded and they will not contribute to the absorption , effectively reducing the strength of the element absorption edge ( e.g. juett et al . the similarity of the observed iron abundance to the ones measured along many other line of sights , is consistent with the idea that a significant fraction of the absorption is produced in the ism along the galactic arms and therefore it does not carry information about the metal abundances characteristic of the gc region . future detailed modelling of the dust scattering halo of and will better constrain this . given the long - term x - ray variability and spectral decomposition , we then identify as an accreting x - ray binary hosting a neutron star or a black - hole , most likely with a low mass star companion . the discovery and identification of this new gc transient as well as the continuous x - ray monitoring of the central parsec is improving the determination of the population of stellar remnants orbiting . the authors wish to thank n. schartel and the team for performing the observation and the `` mpe ir - group galactic center team '' for sharing the reduced , cleaned , calibrated images and photometry as well as for help with the interpretation and discussion . the gc monitoring project is supported by the bundesministerium fr wirtschaft und technologie / deutsches zentrum fr luft- und raumfahrt ( bmwi / dlr , fkz 50 or 1408 ) and the max planck society . the national radio astronomy observatory is a facility of the national science foundation operated under cooperative agreement by associated universities , inc . nr and fcz acknowledge financial support from a dutch nwo vidi grant . nr also acknowledges support from spanish grants aya2015 - 71042 and sgr2014 - 1073 this research has made use primarily of data obtained with , an esa science mission with instruments and contributions directly funded by esa member states and nasa , and on data obtained from the data archive . part of the funding for grond ( both hardware as well as personnel ) was generously granted from the leibniz - prize to prof . g. hasinger ( dfg grant ha 1850/28 - 1 ) . we thank p. wiseman , j. greiner ( both mpe ) and a. hempel ( puc ) for their support with the grond observation .

social relationships are a natural basis on which humans make trust decisions . online social networks ( osns ) are increasingly often used to let users base trust decisions on the existence and the strength of social relationships . while most osns allow users to discover the length of the social path to other users , they do so in a centralized way , thus requiring them to rely on the service provider and reveal their interest in each other . this paper presents social pal , a system supporting the privacy - preserving discovery of arbitrary - length social paths between any two social network users . we overcome the bootstrapping problem encountered in all related prior work , demonstrating that social palallows its users to find all paths of length two and to discover a significant fraction of longer paths , even when only a small fraction of osn users is in the social palsystem e.g. , discovering 70% of all paths with only 40% of the users . we implement social palusing a scalable server - side architecture and a modular android client library , allowing developers to seamlessly integrate it into their apps . [ distributed applications ]

1412.2433

the ability to learn the social path lengthto other social network users can often help individuals make informed trust and access control decisions . for instance , if attendees at a large convention could easily find other attendees with whom they share social links ( e.g. , a linkedin connection ) , this could help them decide who to chat or meet up with . similarly , travelers and commuters could more consciously decide with whom to interact , share rides , etc . in general , discovering the social path lengthbetween users is beneficial in many interesting scenarios , such as estimating the _ familiarity _ to a location ( which can in turn be used for context - based security @xcite ) , as well as for routing in delay - tolerant ad - hoc mobile networks @xcite and anonymous communications @xcite . * problem statement . * the widespread adoption of online social networks ( osns ) makes it appealing to measure the length of a path between two nodes , e.g. , to use this information as a signal of reciprocal trust and/or social interest . today , a facebook user can see the number of common friends with another user , while linkedin also displays the social path length . however , as popular osns are centralized systems , so are the features they offer to discover social paths . as such , they do not particularly adapt well to mobile environments where social interactions are tied to physical proximity , thus severely limiting the feasibility of many applications scenarios users may not always be able to connect to the internet or willing to reveal their location and/or interests to the provider . relying on centralized systems to learn social path lengths implies that , whenever alice queries a server for the social path lengthto bob , the server learns that alice is interested in bob , their frequency of interactions , and their locations . this prompts the need for decentralized and privacy - preserving techniques for social path lengthestimation . users should only learn if they have any common friends , without having to reciprocally reveal the identities of friends that they do not share , and discover the length of the social path between them ( for paths longer than two ) , without learning which users are in the path . * technical roadmap . * our work builds on common friends @xcite , a system supporting privacy - preserving common friend discovery on mobile devices : building on a cryptographic primitive called private set intersection ( psi ) @xcite , it allows mutual friends to be discovered by securely computing the intersection of friendship _ capabilities _ , which are periodically distributed from a common friendsuser to all the friends who are also using it . however , besides being limited to social paths of length two , common friendsonly discovers the subset of the mutual friends who are _ already _ in the system , thus suffering from an inherent bootstrapping problem . this paper introduces social pal , the first system that supports the privacy - preserving estimation of the social path lengthbetween any two social network users . we introduce the notion of _ ersatz nodes _ , created for users that are direct friends of one or more users of social palbut are not part of the system . we guarantee that two users of social palwill be able to discover _ all _ their common friends in the osn ( i.e. , all paths of length two ) . we then present a hash chain - based protocol that supports the ( private ) discovery of social paths longer than two , and demonstrate its effectiveness by means of extensive simulations . our work is not limited to designing protocols : we also fully implement social paland deploy a scalable server architecture and an android client library enabling developers to seamlessly integrate it into their applications . * contributions . * in summary , we make the following contributions : 1 . we present an efficient privacy - preserving estimation of social paths of arbitrary length ( section [ sec : social ] ) . 2 . we state and prove several properties of social palincluding the fact that , for two users and : ( i ) social palwill find all common friends of and , including those who are not using it , and ( ii ) for each discovered path between and , social palallows each party to compute the _ exact _ length of the path ( section [ sec : analysis ] ) . 3 . using samples of the facebook graph , we empirically show that even when only 40% of users use the system , social palwill discover more than 70% of all paths , and over 40% with just 20% of the users ( section [ sec : evaluation ] ) . 4 . we support facebook and linkedin integration and release the implementation of a scalable server - side architecture and a modular android client library , allowing developers to easily integrate social palin their applications ( section [ sec : implementation ] ) . we build two android apps : a friend radar displaying the social path lengthto nearby users , and a tethering app enabling users to securely share their internet connection with people with whom they share mutual friends ( see section [ sec : apps ] ) . we start by discussing the problem of privately discovering common friends , i.e. , social paths of length two . we argue that minimal security properties for this problem include both _ privacy _ and _ authenticity _ , as users should neither learn the identity of non - shared friends nor claim non - existent friendships . * private set intersection ( psi ) @xcite . * a straightforward approach for privately discovering common friends is to rely on psi , a primitive allowing two parties to learn the intersection of their respective sets ( and nothing else ) . if friend lists are encoded as sets , then psi could be used to privately find common friends as the set intersection . one could also limit disclosure to the _ number _ , but not the identity , of shared friends , using private set intersection cardinality ( psi - ca ) @xcite , which only reveals the size of the intersection . however , using psi ( or psi - ca ) guarantees privacy but not authenticity , as one can not prevent users from inserting arbitrary users in their sets and claim non - existent relationships . * bearer capabilities . * in order to guarantee authenticity , nagy et al . @xcite combine bearer capabilities @xcite ( aka bearer tokens ) with psi , proposing the common friendsservice , whereby users generate ( and periodically refresh ) a random number the `` capability '' and distribute it to their friends via an authentic and confidential channel . as possession of a capability serves as a proof of friendship , users can input it into the psi / psi - ca protocol , thereby only learning the identity / number of common and authentic friends . since capabilities are large random values , a simpler variant of psi for private common friend discovery that only relies on cryptographic hash functions and does not involve public - key operations can be used . parties can hash each item in their set and transfer the hash outputs : since the hash is one - way , parties can not invert it , thus they only learn the set intersection upon finding matching hashes . this can be further optimized using the * _ bloom filter based psi ( bfpsi ) _ * primitive ( for high - entropy items ) outlined in @xcite . on the other hand , it is not clear whether it is possible to do so for psi - ca , i.e. , to only count the number of common friends . * bloom filters . * a bloom filter ( bf ) is a compact data structure for probabilistic set membership testing @xcite . specifically , a bf is an array of @xmath0 bits that can be used to represent a set of @xmath1 elements in a space - efficient way . bfs introduce no false negatives but can have false positives , even though the probability of a false positive can be estimated ( and bounded ) as a function of @xmath1 and @xmath0 . formally , let @xmath2 be a set of @xmath1 elements , and bf be an array of @xmath0 bits initialized to 0 . bf@xmath3 denotes the @xmath4-th item in bf . then , let @xmath5\}_{i=1}^\gamma$ ] be @xmath6 independent cryptographic hash functions , salted with random periodically refreshed nonces . for each element @xmath7 , set @xmath8 for @xmath9 . to test if @xmath10 , check if bf@xmath11 . an item appears to be in a set even though it was never inserted in the bf ( i.e. , a false positive occurs ) with probability @xmath12 . the optimal size of the filter , for a desired probability @xmath13 , can be estimated as : @xmath14 . [ [ subsec : system - model ] ] in figure [ fig : architecture ] , we illustrate the common friends @xcite system : it involves a server , a set of osn servers ( such as facebook or linkedin ) , and a set of mobile users , members of one or more of these osns . is implemented as a social network app ( i.e. , a third - party server ) , which stores the bearer capabilities uploaded by the common friendsapplication running on users devices . it also allows a user s common friendsapplication to download bearer capabilities uploaded by that user s friends in the osns . common friendsconsists of three protocols : ( 1 ) user authentication , ( 2 ) capability distribution , and ( 3 ) common friend discovery . [ fig : architecture ] * osn user authentication * enables the osn server to authenticate a user , provide s osn identifier ( @xmath15 ) to , and to authorize to access information about s friends in the osn , which can be done using standard mechanisms , such as oauth @xcite . * capability distribution * involves and , communicating over a secure channel with server authentication provided by certificate @xmath16 , and client authentication based on the previous osn user authentication process . user generates a random capability @xmath17 ( taken from a large space ) and uploads it to over the established channel . stores @xmath17 along with the social network user identifier @xmath15 , and sends back @xmath18 , which contains pairs of identifiers and corresponding capabilities of each friend that has already uploaded his capability . the protocol needs to be run periodically to keep @xmath19 up - to - date , as capabilities are periodically refreshed . * common friend discovery * is a protocol run between two users , and , illustrated it in figure [ fig : bffof ] , allowing and to privately discover their common ( authentic ) friends , based on bfpsi . first , and exchange their public keys ( @xmath20 and @xmath21 , respectively ) and generate a shared key ( @xmath22 ) used to encrypt messages exchanged as part of the protocol . to prevent man - in - the - middle attacks , ( resp . , ) cryptographically binds the dh channel to the protocol instance : ( resp . , ) extends each item in the capability set @xmath23 ( resp . , @xmath24 ) by appending dh public keys @xmath20 , @xmath21 , building effectively a new set @xmath25 ( resp . , @xmath26 ) . inserts every element of the @xmath25 set into a bloom filter @xmath27 and sends it to . discovers the set of friends ( @xmath28 ) in common with by verifying whether each item of his input set @xmath26 is in @xmath27 . since bloom filters introduce false positives , the set @xmath28 may contain non - common friends . thus a simple challenge - response protocol is run , where requires to prove knowledge of items available in @xmath28 . at the end of the protocol , and output the set of their common friends @xmath29 . [ fig : bffof ] we now present the design and the instantiation of the social pal , the system to compute the social path lengthbetween two osn users in a decentralized and privacy - preserving way . * limitations of @xcite . * before introducing social pal s requirements , we discuss two main limitations of common friends @xcite , as addressing them constitute our starting point : 1 . _ bootstrapping _ : users and can discover a mutual friend ( say ) , only if has joined the common friendssystem and uploaded his capability to . that is , common friendswill only discover a subset of the mutual friends between and until all of them start using it . longer social paths _ : common friendsonly allows its users to learn whether they are friends or have mutual friends . if two users have a longer social path between them , common friendscannot detect it . to illustrate common friends s bootstrapping problem , we plot , in figure [ fig : bootstrapping ] , a simple social network with 27 nodes ( i.e. , users ) and 34 edges ( i.e. , friendship relationships ) . black circles represent users who are using common friends , and white circles those who are not . purple / solid edges represent direct friend relationships ( i.e. , social paths of length 1 ) that are discoverable by common friends . when the user base is only 40% of all osn users ( figure [ subfig:0.25-users ] ) , only 7 out of the 34 direct friend relationships are discoverable ( i.e. , coverage is approximately @xmath30 ) . when it increases to 60% , coverage increases to about 50% ( figure [ subfig:0.5-users ] ) . 0.35 and @xmath31 of users using the system . black ( resp . , white ) nodes denote users ( resp . , non - users ) of the system . purple / solid edges denote a direct friendship discoverable by common friends.,title="fig:",scaledwidth=85.0% ] 0.35 and @xmath31 of users using the system . black ( resp . , white ) nodes denote users ( resp . , non - users ) of the system . purple / solid edges denote a direct friendship discoverable by common friends.,title="fig:",scaledwidth=85.0% ] * system model . * social pal s system model is the same as that of common friends(presented in section [ subsec : system - model ] ) . it involves a server ( which we design as a social network app ) , a set of osn servers ( such as facebook or linkedin ) , and a set of mobile users members of one or more of these osns ( cf . figure [ fig : architecture ] ) . * functional requirements . * ideally , social palshould allow any two users to always compute the exact length of the social path between them , even when social palis being used only by a fraction of osn users . in order to characterize how well social palmeets this requirement , we use a measure of the likelihood that any two users would discover , using social pal , an existing social path of a given length between them . we denote this measure as social pal s _ coverage_. we define social pal s functional requirements as follows : * _ ( correctness ) . _ users and can determine the exact length of a social path between them ( if any ) . * _ ( coverage maximization ) . _ social palshould maximize coverage , in other words , the ratio between the number of social paths ( of length @xmath32 ) between and _ discovered _ by social paland the number of _ all _ social paths ( of length @xmath32 ) between and . * privacy requirements . * from a privacy point of view , social palshould satisfy the following requirements . let and be two social palusers willing to discover the length of the social path existing between them : 1 . and discover the set of their common friends but learn nothing about their non - mutual friends ; 2 . and do not learn any more information other than what it is already available from standard osn interfaces . in other words , social palshould allow two users to learn the social path length between them ( if any ) , but not the nodes on the path , without reciprocally revealing their social link . if a path between the users exists that is of length two ( i.e. , users have some common friends ) , then they learn the identity of the common friends ( and nothing else ) . this only pertains to interacting users , as ensuring that no eavesdropping party learn any information about users friends can be achieved by letting users communicate via confidential and authentic channels . * threat model . * we assume that the participants in social palare honest - but - curious . the osn server is trusted to correctly authenticate osn users and not to attempt posing as any osn user . the social palserver is trusted to distribute social palcapabilities only to those social palusers authorized to receive them . social palusers use the legitimate social palclient , but they might attempt to learn as much information as possible about friends of other social palusers with whom they interact . we aim to guarantee the privacy requirements discussed above in this setting , and prevent the osn server or the server to learn any information about interactions between social palusers . before presenting the details of the system , in table [ tab : fofprotocol - notation ] , we introduce some notation used throughout the rest of the paper . [ tab : fofprotocol - notation ] * ersatz nodes . * one fundamental building block of social palare ersatz nodes , which we introduce to overcome the bootstrapping problem faced by services like common friends . recall from section [ subsec : system - model ] that , in the original common friendsdesign , the server stores bearer capability @xmath17 , uploaded by a user , together with his social network identifier @xmath15 . the pair @xmath33 constitutes s _ user node _ in the social graph maintained by . the set of s friends @xmath34 is the set of edges incident in the user node . in social pal , we let create an _ ersatz node _ for all users who have not joined the system but who are friends with a user who has . an ersatz node is identical to a standard user node , but its capability is generated by , instead of the user . figures [ subfig:0.25-users - ersatz ] and [ subfig:0.5-users - ersatz ] show how coverage improves when ersatz nodes are added , e.g. , with only @xmath35 joining the system , coverage reaches @xmath36 . 0.38 and @xmath31 of users using the system and with the addition of ersatz nodes ( in grey ) . purple / solid edges here denote a direct friendship discoverable by social pal.,title="fig:",scaledwidth=83.0% ] 0.37 and @xmath31 of users using the system and with the addition of ersatz nodes ( in grey ) . purple / solid edges here denote a direct friendship discoverable by social pal.,title="fig:",scaledwidth=83.0% ] * ersatz node creation . * adding ersatz nodes requires a few changes in the capability distribution protocol , compared to that from section [ subsec : system - model ] . we highlight these changes in figure [ fig : ersatz - capability - updates ] , specifically , in the blue - shaded box . before returning @xmath19 to , first computes the set @xmath37 which contains the social network identifier of each `` missing user '' @xmath38 . then , @xmath39 , it creates @xmath38 s an ersatz node as follows : 1 . create an _ ersatz capability _ @xmath40 ( where @xmath41 is the length of a capability ) for @xmath38and store @xmath42 . 2 . create an initial friend set @xmath43 , which at this stage contains only @xmath15 . after the successful creation of all needed ersatz nodes , returns @xmath19 , which includes the capabilities from the nodes of all of s friends , including the ersatz nodes . [ fig : ersatz - capability - updates ] * active social graph updates . * users of social palexplicitly authorize the server to retrieve their friend lists from the osns . since an ersatz node @xmath38is not a user of social pal , can not learn the full set of @xmath38 s friends @xmath43 . instead , it maintains an estimate of @xmath43 based on the events it can observe from users of social pal . for example , when a user adds @xmath38as a friend , learns that @xmath44 is added to @xmath34 and can infer that @xmath15 should be added to @xmath43 . each @xmath45 corresponds to a real user who has explicitly authorized to learn about the edge -@xmath38 in the social graph . * turning ersatz nodes into `` standard '' nodes . * if a user @xmath38for whom has created an ersatz node later joins social pal , he can simply upload his capability @xmath46 to , who ( 1 ) overwrites the old ersatz capability with @xmath46 , ( 2 ) queries the osn for @xmath38 s friend list @xmath47 , and ( 3 ) updates the existing , possibly incomplete , list of @xmath38 s friends with @xmath47 , turning an ersatz node into a standard node . note that this operation is transparent to all users . we now present the full details of our social palinstantiation : besides addressing the bootstrapping problem ( using ersatz nodes ) , it also allows two arbitrary users to calculate the social path lengthbetween them . we denote with @xmath48 the social path lengthbetween two users and , i.e. , the minimum number of hops in the social network graph that separates and . * intuition . * we set to allow social palto discover the social path lengthbetween users in the osn by extending the capability distribution to include further relationships beyond friendship ( e.g. , friend - of - a - friend ) and rely on capability matches for estimating the social path length . by using cryptographic hash functions , we can generate and distribute capabilities of higher order that serve as a proof of a social path between users . * notation . * in the rest of the paper , we use the following notation : the hash chain @xmath49 of item @xmath50 ( of length @xmath51 ) corresponds to the evaluation of a cryptographic hash function @xmath52 performed @xmath51 times on @xmath50 . when @xmath53 , @xmath54 . specifically : @xmath55 @xmath56is a @xmath57-degree capability and is defined as @xmath58 @xmath59 denotes the set of social contacts that are @xmath57-hops from user @xmath60 . * capability distribution : * in figure [ fig : distance - capability - protocol ] , we detail social pal s protocol for capability distribution . interaction between and is identical to the capability distribution protocol from figure [ fig : ersatz - capability - updates ] , up until the creation of missing ersatz nodes is completed . in the updated protocol returns two sets , namely @xmath61and @xmath62 , where @xmath62denotes the set of higher order capabilities provided to by other osn members that are at least @xmath63-hops from . it is composed of a number of subsets @xmath64 with each subset @xmath65containing @xmath66 order capabilities of users in @xmath67 . formally , @xmath68 { \ensuremath{\mathit{c_{i}}}\xspace}=\ { ( i-1,c^{i-1}_{j } ) : \exists ( { \ensuremath{id_{j}}\xspace } , { \ensuremath{\mathit{c_{j}}}\xspace } ) \land { \ensuremath{id_{j}}\xspace}\in { \ensuremath{\mathit{f^{i}}}\xspace}({\ensuremath{id_{{\textsf{u}\xspace}}}\xspace } ) \}\vspace{-0.1 cm } \end{array}\ ] ] consequently , the total cardinality of @xmath62and @xmath61is : @xmath69 finally , generates missing higher order capabilities . for every received capability @xmath70of degree @xmath51 , hashes it @xmath71 times to generate a sequence of higher order capabilities of the form : @xmath72 all elements of such sequences are combined into one set of derived higher order capabilities @xmath73 . finally all capability sets are combined to form @xmath74 : @xmath75 the resulting set @xmath76 will be used to derive the input sets for psi during the social path lengthdiscovery protocol as explained below . the cardinality of the input set to psi is therefore : @xmath77 [ fig : distance - capability - protocol ] to construct @xmath78 of user , tracks changes in friend lists of users by using the following logical implication : if @xmath79 represents a friend of @xmath80 and @xmath80 is @xmath81 hops from @xmath60 and @xmath79 was not previously identified at less than @xmath57 hops from @xmath60 , then @xmath79 is @xmath57 hops from @xmath60 . formally : @xmath82 finally , as capabilities are meant to be short - lived ( i.e. , they should expire within a couple of days ) , the protocol needs to be run periodically in order to keep @xmath83 up - to - date . * social path discovery : * in figure [ fig : distance - client - protocol ] , we illustrate the social paldiscovery protocol . the protocol involves two users and , who are members of the same osn . it begins with establishing a secure channel ( via diffie - hellman key exchange ) , followed by cryptographic binding of the diffie - hellman channel to the protocol instance , which is needed to avoid man - in - the - middle attacks . , ) appends both public keys to each capability @xmath84 in @xmath85 ( resp . , @xmath86 ) set to form @xmath87 . the resulting sets are : @xmath88 @xmath89 note that the @xmath90 symbol in the above equations indicate that , while constructing @xmath91 and @xmath92 , the first element of each pair contained in @xmath85 and @xmath86 is ignored . next , both users execute the steps of common friends s discovery protocol , on the above input sets . specifically , they interact in a bloom - filter based psi execution and run the challenge - response part of the protocol needed to remove potential false positives ( as discussed in section [ subsec : system - model ] ) . the interactive protocol ends with parties outputting the intersection of the sets . from this point on , both users perform identical actions to calculate the social path length between them . all operations are done locally , i.e. , with no need to exchange data . this process consists of two phases : ( 1 ) calculating the social path length input set @xmath93 ( i.e. , the set containing lengths for all discovered paths between and ) , and ( 2 ) selecting the shortest length among all lengths contained in @xmath93 . to this end , ( ) builds set @xmath93 by performing following actions on every item @xmath7 : 1 . finding a capability @xmath94 such that @xmath95 2 . calculating path length @xmath96 via matching capability @xmath50 ( which was obtained from some user , say ) and inserting it into @xmath93 : @xmath97 l.insert(l_{x})\vspace{-0.1cm } \end{array}\ ] ] at the end , and learn the final path length @xmath98 between them by finding the lowest value of items included in @xmath93 : @xmath99 if @xmath100 , then and have common friends between them , thus social palreturns identifiers of all these common friends as in the original common friendsservice . while we could use social palto reveal the first hop identifiers for @xmath101 , we do not due to the privacy requirements outlined in section [ subsec : privacy - consideration ] . this section presents the analysis of social pal , showing that it fulfills functional and privacy requirements from section [ subsec : requirements ] . lemma 1 . _ if @xmath102 , then : _ 1 . there exists a path @xmath103 , for @xmath104 , between and , in the social graph . 2 . receives the @xmath105 order capability @xmath106 from every @xmath107 in @xmath29 . proof . when @xmath102 , by using the logical implication for the social graph building ( see section [ sec : distance ] ) , it must hold that , in order to include @xmath108 in @xmath109 , @xmath110 must be included in @xmath111 . therefore , we can recursively argue : + note that , for every @xmath112 , there exists a connection to @xmath113 and @xmath114 , thus , @xmath115 form a path @xmath29 between and . considering @xmath116 , since @xmath117 , then receives @xmath106 . theorem 1 . _ let there be a path @xmath118 between and in the social graph . if path @xmath29 is discovered by the social paldiscovery protocol , then both and can estimate the exact length @xmath119 of path @xmath29 . _ proof . let @xmath32 denote the highest degree of capabilities . + if @xmath120 : * the set of capabilities of and are @xmath121 , and @xmath122 , respectively . ( see figure [ subfig : proof-1 ] for a graphic illustration of the distribution of capabilities for and . ) * if gets a matching capability for @xmath123 , then it must corresponds to @xmath124 for . * substitutes @xmath125 in @xmath48 and receives : @xmath126 * gets multiple capability matches , and sets @xmath48 to be the minimum @xmath127 , which is for @xmath53 , and @xmath128 . ( similar argument holds for . ) if @xmath129 : * the set of capabilities of and are @xmath130 and @xmath131 , respectively see figure [ subfig : proof-2 ] . ( capabilities for which and obtains matches are marked in green . ) * if gets a capability match for : @xmath132 , substitutes @xmath133 in @xmath48 and receives : @xmath134 * gets multiple capability matches , and sets @xmath48 to be the minimum @xmath127 , which is for @xmath135 , and @xmath128 . ( similar argument holds for . ) 0.75 0.75 [ fig : proofs ] as discussed in section [ subsec : requirements ] , our social palinstantiation needs to provide users with strong privacy guarantees , i.e. , interaction between two users and does not reveal any information about their non - mutual friends or any other information than they could discover by gathering information from the standard osn interface . * capability intersection . * first , we review the security of the common friends discovery protocol from common friends @xcite , since it constitutes the basis of our work . its security , in the honest - but - curious model , reduces to the privacy - preserving computation of set intersection . that is , privacy stems from the security of the underlying private set intersection ( psi ) protocol that common friendsinstantiates to privately intersect capabilities and discover common friends . this is proven by means of indistinguishability between a real - world execution and an ideal - world execution where a trusted third party receives the inputs of both parties and outputs the set intersection . @xcite uses bloom filter based psi ( bfpsi ) and , as discussed in section [ subsec : pcd ] , this does not impact security since sets to be intersected are random capabilities , thus , high - entropy , non - enumerable items . * discovery . * now observe that the interactive part of the social pal s ( social path ) discovery protocol i.e. , the part where information leakage might occur mirrors that of common friends s discovery protocol . during the protocol execution , users and engage in a bfpsi interaction , on input , respectively , @xmath91 and @xmath92 , i.e. , the sets of their capabilities , and obtain @xmath29 , which is used to reconstruct the social path between and . if and are friends with each other ( or have mutual friends ) , they can find out the identity of the user(s ) corresponding to matching capabilities , thus , learning that there exists a social path of length 1 ( or 2 ) _ and _ the identity of their mutual friends , but nothing else . in fact , if an adversary could learn the identity of non - mutual friends , then , we could build an adversary breaking the common friends discovery protocol from @xcite based on bfpsi . similarly , if there exists no social path between and , then the bfpsi interaction does not reveal any information to each other . on the other hand , if there exists a social path between and of length @xmath136 , then the matching capabilities are for user nodes for which has removed identifiers @xmath137 . therefore , and do not learn the identity of the users yielding a social path between them , but only how many . * trust in server . * each user explicitly authorizes social palto retrieve the set @xmath138of s friends . requesting users to disclose their friend lists is a common practice in social network and smartphone applications . social paluses this information to have the server maintain , distribute , and , in the case of ersatz nodes , create capabilities attesting to the authenticity of friendships . this implies that gradually learns the social graph from social palusers , however , what learns is a small subset of what the osn already knows . neither nor the osn learn any additional information , e.g. , as opposed to centralized solutions , user locations or interactions between users . * authenticity of capabilities . * in section [ subsec : requirements ] , we assumed the use of legitimate social palclient applications : as all mobile platforms provide application - private storage , it is reasonable to assume that an adversary on a client device can not steal the capabilities downloaded on that device by the legitimate client application or otherwise manipulate the input to the protocol . alternatively , the integrity of the bloom filter could be ensured by letting sign the bloom filter along with the public key of the corresponding social palclient . the bfpsi protocol would then need to be modified accordingly so that each party checks the signature on the other party s bloom filter is valid and that the same keypair is used to establish the secure channel . we now present an empirical evaluation of social pal s coverage , using three publicly available facebook sample datasets . specifically , we analyze how _ coverage _ attained by the social path discovery depends on the fraction of osn users who join the system , i.e. , the probability that two users discover an existing path between them in the social network . we use three datasets derived from a single dataset , created by gjoka et al . @xcite , using three different sampling techniques : \(1 ) the * social filter dataset * @xcite is our primary dataset . it contains @xmath139 users , a connected component derived using the forest fire " sampling method @xcite from the original dataset @xcite . as forest fire sampling does not preserve node degree , each node in this dataset has an average node degree of @xmath140 , which is significantly less than in the original dataset . to investigate the effect of the reduced node degree on coverage , we also use the two more datasets . ( 2 ) the * mhrw dataset * @xcite is built using the _ metropolis - hastings random walk _ ( mhrw ) method with @xmath141 independent random walks . it contains the friend lists of @xmath142 users . we call this the set of _ sampled users_. each of them has an average of @xmath143 friends , including both other sampled users and those who were part of the original dataset but that were not sampled we call them _ outside users_. the mhrw datasetcontains a total of @xmath144 million _ outside users _ ( who are friends of one or more _ sampled users _ ) . because of the nature of the mhrw sampling , the average number of connections between two _ sampled users _ in this set is only 3 , thus it is used to evaluate social pal s coverage among poorly connected users . \(3 ) the * bfs dataset * @xcite is built using _ breadth first search _ ( bfs ) from @xmath141 independent bfs traversals . it consists of @xmath145 million _ sampled users _ , with an average of @xmath146 friends . the number of _ outside users _ is @xmath147 million . bfs sampling results in highly connected subgraphs , and the average number of connections among _ sampled users _ is @xmath148 . thus , we use the bfs datasetto measure social pal s coverage among well connected users . * procedure . * to evaluate social pal s coverage on each of the three datasets , we used the following simulation procedure . first , we chose , at random , a subset of _ sampled users _ , which we call the _ test set_. for the social filter dataset , we used the whole set as the _ sampled users _ set . we chose four different sizes for the _ test set _ : @xmath149 of the _ sampled users_. note that the _ test set _ represents the fraction of the users of an osn who use social pal . then , for a given social path length @xmath32 ( @xmath150 ) , we randomly selected @xmath151 pairs of users from the _ test set _ in such a way that at least one path of length @xmath32 exists . finally , we computed the fraction of user pairs for which social paldiscovers an existing path between them . we did this for two cases : social palwith support for ersatz nodes , and without it . each simulation was repeated 10 times . in total , we conducted 720 different simulations . * results . * we now present the results of our simulations for each of the three datasets . social pal s discovery coverage is presented in figure [ fig : user - coverage-1 ] andtable [ tab : user - coverage-1 ] for the the social filter dataset , figure [ fig : user - coverage-2 ] and table [ tab : user - coverage-2 ] for the mhrw dataset , and figure [ fig : user - coverage-3 ] andtable [ tab : user - coverage-3 ] for the bfs dataset . additional graphs on coverage results are available from the full version of the paper @xcite . each graph shows how the coverage ( for paths of different length ) of social palrelates to fraction of osn users who use social pal . red dotted lines indicate the performance of social palwithout ersatz node support , while black solid lines correspond to the user of ersatz nodes . without ersatz nodes , coverage increases linearly as more users start using social pal . the rate of growth is highest for the social filter datasetand lowest for the mhrw dataset . in general , the coverage figures are low . for instance , even if 80% of osn users have social pal , the coverage for paths of length 4 ranges between 0.19% ( the mhrw dataset ) and 68.71% ( the social filter dataset ) . the introduction of ersatz nodes results in a remarkable improvement across the board in all datasets . as expected , the coverage for paths of length 2 is 100% . when 80% of osn users are in the social palsystem , the coverage is well above 80% in all cases . even when only 20% of users have social pal , coverage is still above 40% in all cases , except for the mhrw dataset . [ tab : user - coverage-1 ] [ tab : user - coverage-2 ] [ tab : user - coverage-3 ] * ersatz nodes dramatically improve coverage . * with ersatz nodes , social paldiscovers 100% of social paths of length 2 , thus addressing one of the major limitations of the common friendssystem @xcite . the coverage for paths of length 3 and 4 always increases , between @xmath152 and @xmath153 , depending on the fraction of osn users in social paland the dataset used for the simulations . * variation of coverage across different datasets . * we observe better coverage results with the bfs datasetthan with the mhrw dataset . as the bfs datasetrepresents coverage among well connected users , the density of ersatz nodes between random users is higher than in the mhrw dataset , thus yielding better overall coverage . the bfs datasetmodels societies , such as most of the western societies , where the penetration of osns is high . the high coverage results with the bfs datasetsuggests that social palwill do well in this context . on the other hand , the mhrw datasetmodels societies where osn connectivity is poor and , although social palis not as effective here , it may still perform reasonably well , detecting the majority of social paths even before the number of users joining social palreaches 50% . in this section , we present our full - blown implementation of the social palsystem . we aim to support scalability for increasing number of users ( in terms of cpu performance and memory ) and to enable developers to easily integrate it into their applications . * server components . * on the server side , the social palsystem extends the ` peershare`server @xcite , which allows two or more users to share sensitive data among social contacts , e.g. , friends in a social network . we use the following basic functions of ` peershare ` : ( 1 ) osn interfaces to retrieve social graph information , ( 2 ) the oauth @xcite component for user authentication , and ( 3 ) the data distribution mechanism . on top of these components , we develop a new server architecture that supports the addition of server - based applications via an extension mechanism . this design choice allows us to implement the social palfunctionality in such a way that the system can efficiently scale ( in terms of memory and cpu performance ) to support an increasingly large number of users . as illustrated in figure [ fig : server - arch ] , the server architecture consists of the following components : the _ common apps server _ , a group of applications ( e.g. , the social palapp ) , the _ osn communicator module _ , and the _ bindings database _ ( bindings db ) . the common apps serverprovides the basic functionality that is common for all applications : ( 1 ) storage of data uploaded by users in the bindings db , ( 2 ) distribution of users data to other authorized users , and ( 3 ) retrieval of basic social graph information , which is needed for enforcing the appropriate data distribution policy . the osn communicator module is a plugin - based service responsible for querying osns for social graph information . its plugin - based structure allows us to easily add support for new osns . the bindings db stores data uploaded by users and information on how to distribute them among social contacts . the components interact with each other using a number of different interfaces : _ server - osn query _ , _ bindings protocol _ , _ app event _ , _ app - db updates _ and _ app - osn query _ ( cf . fig [ fig : server - arch ] ) . the server - osn query interface is used by the common apps serverto retrieve social graph information from the osn . the bindings protocol interface specifies communication between the common apps serverand the bindings db . these two interfaces provide basic data distribution functionality to all applications . the other three interfaces are used only by applications that need to perform specific modifications on distributed data on the server side . each application that requires logic on the server side has to be registered in the common apps servervia a uniform resource identifier ( uri ) . this uri is used by the app eventcallback interface to notify the application about incoming application - specific events ( e.g. , an upload of new data items ) . in addition , this interface may be used to modify data read from the bindings db before returning them via the common apps serverto the user ( e.g. , generation of higher order capabilities ) . the server application itself has access to the two remaining interfaces . it uses the app - db updatesto modify data stored in the bindings db or to create new data items . the app - osn query interface is used when the application needs to obtain social graph information from the osn . [ fig : server - arch ] * social palserver implementation . * the social palserver application is notified via a the app eventcallback interface about new capabilities uploaded by users . it uses a the app - db updatesinterface to create any required ersatz nodes and properly update the recipient sets of capabilities during the social graph building process . the social palapplication also uses the app eventinterface to handle capability download requests . the common apps server , instead of immediately returning data it has read from the bindings db , passes them to the social palapplication that generates any missing higher order capabilities . finally , the social palapplication returns the complete set of capabilities back to the common apps server , which completes the request handling . note that our implementation of the osn communicator module supports interactions with both linkedin and facebook . * implementation details . * our core social palserver is written in php . we support capabilities of @xmath154 and @xmath155 order , allowing to discover social paths between users that are up to @xmath156 hops from one another . based on relevant prior work @xcite , a @xmath156-hop distance is enough for most practical use cases . as the bindings db needs to store the capabilities of users and information about how to share those , the necessary amount of persistent storage will substantially grow if social palbecomes widely used . thus , to limit the data storage overhead , social palserver does not store any higher order capabilities in the bindings db , but generates them when requested by the requesting client . tests on our server show that generating higher order capabilities has a negligible impact on the social pal__capability distribution protocol _ _ performance ( i.e. , the server generates 1 million higher order capabilities in about 500ms using the hardware described in section [ sec : server - eval ] ) . finally , in order to implement the linkedin oauth module for the osn communicator module , we use the oauth pecl extension for php , while , for the bindings db , postgresql database server . * system scaling . * since social palmay generate a large number of server requests if used by a large number of users , we can take following steps to ensure that the system can scale . our proposed scaling architecture is illustrated in figure [ fig : server - scaling ] . it includes a powerful http front - end server ( such as nginx ) acting as load balancer , which terminates incoming secure https connections and forwards server requests upstream to @xmath32 instances of social palservers acting as request handlers . each social palserver instance will run the _ hiphop virtual machine _ ( hhvm ) daemon that handles http requests . hhvm usage can massively improve server performance , as it uses just - in - time compilation to take advantage of the native code execution instead of the interpreted one @xcite . each instance of social palserver runs , locally , a database query router ( pgpool ) providing access to the actual database cluster including multiple postgresql servers . the query router enhances the overall database access performance by keeping open connections to the database cluster , load - balancing the stored data among multiple instances of the database servers , and temporarily queuing requests for database access in case of cluster overload . note that there are no cross - dependencies between the social palserver instances for the database read access , thus , no complex control mechanism is needed to support this parallelism . * server code . * the source code of the server implementation is available from https://github.com/secures-aalto/sopal . * performance testbed . * we evaluated our server implementation in a testbed consisting of two machines : the first played the role of a single social palserver instance ( cf . figure [ fig : server - arch ] ) , while the second simulated a group of client devices . the server ran on a 4-core machine with a 2.93ghz cpu on each core and 128 gb of ram . it hosted nginx ( version 1.1.19 ) , postgresql server ( version 9.1 ) , and php5-fpm for the php 5.6 engine . to improve the overall server performance , we adjusted the default settings for nginx , postgresql and php5-fpm ( see table [ tab : server - settings ] ) . .details of server implementation settings . [ cols="^,^ " , ] inter - process communication was implemented via unix sockets . the machine running the clients had 8 cpu cores ( at 2.93 ghz ) and 64 gb of ram . to eliminate the unpredictability of network latency , we modified the server implementation by replacing the osn communicator module with the local service that provided social graph information based on the mhrw dataset . we populated the bindings db with capabilities generated for the 120,000 _ sampled users _ from the mhrw dataset . the capabilities of _ sampled users _ together with capabilities generated for ersatz nodes constituted about 10 million data items that were stored in the bindings db . finally , to minimize impact of the client - server transmission delay , we kept the server and client machines in the same network and connected them using a 1 gbit / s ethernet link via the single switch . * experiments . * we evaluated server performance by sending bursts of @xmath32 requests per second , for @xmath157 , from the client machine to fetch capabilities from the server ( i.e. , download of @xmath61and @xmath73 in figure [ fig : distance - capability - protocol ] ) for @xmath158 seconds . fetching capabilities involves many read operations on the bindings db , thus yielding the highest load on the server among all operations of the social palcapability distribution protocol . in each experiment , which we repeated ten times , we measured the number of received responses per second together with the latency of each response on the client machine , and cpu usage together with memory consumption on the server machine . [ fig : server - eval ] * results . * figure [ fig : server - eval ] illustrates the results of our experiments . we observe that @xmath159 requests per second yields a saturation point for the server . below @xmath159 requests / second , the number of responses per second and the response latency grow linearly . whereas , as depicted in figure [ subfig : recv - latency ] , above @xmath159 , we observe an exponential growth of the response latency and the constant number of received responses per second . figure [ subfig : cpu - ram ] also shows that cpu usage reaches more than @xmath160 above @xmath159 requests / second . peak memory consumption is about @xmath161 mb , which also shoots up significantly when the number of requests crosses the saturation point . the @xmath159 requests / second saturation point shows that the performance of our server implementation is in line with that emerging from studies of systems equipped with similar hardware @xcite . we also looked at the server performance when only handling @xmath162 request per second and observed that the average response latency is about @xmath163ms , and the average bindings db interaction time is around @xmath164ms . since client - server network latency is negligible , the vast majority of request handling takes place in the php interpreter , which highlights that php is the server s bottleneck . therefore , in order to improve the server performance , php5-fpm should be replaced with hhvm , which is reported to be significantly more performant @xcite . further gains could also be obtained by migrating the postgresql server to a separate machine connected over a fast link ( cf . fig . [ fig : server - scaling ] ) . we leave these as part of future work . assuming that the server handles @xmath159 requests per second , a total of @xmath165 million requests can be processed daily by a single instance of social palserver with comparable hardware capabilities . assuming that each user executes the social palcapability distribution protocol around @xmath156 times a day , about @xmath166 million social palusers can be handled by one social palserver instance . since user requests are independent of each other , and because the scalable architecture of the social palserver allows adding further instances easily ( as described in section [ sec : server - arch ] ) , the total capacity of social palsystem amounts to the cumulative number of users that can be handled across all social palserver instances . finally in order to avoid making postgresql become the bottleneck of the system ( which may be caused if many social palserver instances are added ) , the bindings db should be turned into a database cluster with data sharding and replication enabled . this guarantees that the data kept in the bindings db is synchronized and accessible with high enough availability . * client architecture . * the client - side architecture of social palis depicted in figure [ fig : client - arch ] . it consists of the _ common apps client _ , the set of osn plugins , and the _ social palclient_. the first two components are responsible for the communication with the social palserver , while the last one provides the interface for the applications . together , these components form a mobile platform library that can be easily imported by developers into their applications . to facilitate support for multiple osns , similar to our server design , we have decoupled osn - specific functionality from the common apps client and made it a plugin - based solution . we have considered two possible design choices for the client architecture : ( 1 ) designing it as a stand - alone service with applications connecting to it , using available inter - process communication mechanisms , or ( 2 ) as a service integrated into the application . we choose the latter as it supports application - private storage for capabilities ( i.e. , not accessible by other applications ) and enables each application to have its own social palserver . this choice provides additional protection against capability leakage to a malicious application and removes the requirement to deploy the global social palserver . on the other hand , if multiple instances of social palapplication runs on the same device , we would incur increased network traffic and require more storage space in comparison to the stand - alone service approach . we argue that this tradeoff is acceptable , as the social palcapability distribution is run no more than a few times a day . also , the amount of data to store is likely to be limited in the order of tens of megabytes , which is justifiable given the clear usability and deployability advantages . * implementation details and performance . * we have implemented the client library on android , operating as an android lightweight service . to evaluate the performance on the client , we measured running times of the social paldiscovery protocolon a samsung galaxy tablet gt - p3100 running android 4.1.2 api 16 and a zte blade s6 running android 5.0 api 21 connected via bluetooth . we assumed that both parties have the same number of input items , ranging from 1000 to 35000 ( with 5000 increments ) . we also fixed the intersection of the sets to be @xmath167 of the set size . table [ tab : perf ] shows average computation and communication times . * social palclient api . * the social palapplication interface is used by applications to run the social paldiscovery protocol . it has been designed to be readily usable by application developers that are not cryptography experts , but are nonetheless interested in implementing privacy - preserving discovery of social paths . this allows developers to delegate the responsibility of this process to the social palclient , and requires them to integrate only four basic methods into the application code , which we present below . applications running the social paldiscovery protocol act as social palmessage relays between the two social palclient instances . the application starting the social paldiscovery session calls the ` startsopalsession ` method . this returns an opaque social palobject , which is forwarded to the remote party . from this point onward , both parties invoke ` handlesopalmessage ` for every message received . this method processes the received message , and if needed , creates a response . it also returns a flag indicating if the protocol execution is completed . if so , the application uses the ` getresult ` method to get the social path length it has to a remote party . finally , the application must call ` endsopalsession ` to let the social palclient release all resources from the session . besides these four basic methods , the social palclient also provides three advanced methods : ( 1 ) ` rejectsopalsession ` creates a social palmessage that can be sent by the application to the remote party if it does not want to run the discovery protocol ; ( 2 ) ` updatecapabilities ` and ( 3 ) ` renewcapability ` can be used by the application to force fetching the most recent capabilities from the server , and to generate and upload a new capability to the server , respectively . table [ tab : sopal - iface ] summarizes all the methods available in social palclient . [ fig : apps ] to illustrate social pal s relevance and practicality , we integrate it into two android apps , ` spotshare`and ` nearbypeople ` , supporting both facebook and linkedin . * ` spotshare ` * is an extension of tetheringapp , presented in @xcite . it allows a user to provide tethered internet access to other ` spotshare`users ( where access to the tethering hotspot is protected by a password ) so that access control policies can be based on social relationships . for instance , the user can decide to allow tethered access only to friends of friends : to this end , ` spotshare`uses social palin order to determine , in a privacy - preserving way , if the specified social relationship holds . if so , the password is securely and automatically sent to the requesting device . in the current version of ` spotshare ` , we do not enable discovery of social paths beyond two hops , as we assume that most users would not want to allow people with whom they have no common friends to tether off their smartphone , but removing this constraint is trivial . in figure [ subfig : spotshare ] , we present two screenshots of the app . * ` nearbypeople ` * is a `` friend radar '' app allowing users to interact with people around them and discovering common friends shared with users of nearby devices , as well as social path lengths , without having to broadcast their social profiles or rely on a central server . it relies on the privacy guarantees of social paland the scampi opportunistic router @xcite for device - to - device communication . a preliminary version of the app was successfully tested at the acm ccs workshop on smart energy grid security workshop . in figure [ subfig : nearbypeople ] , we also show two screenshots . * privately discovering social relationships . * nagy et al . @xcite introduce common friends , reviewed in section [ subsec : system - model ] , combining bearer capabilities with bfpsi / psi - ca to allow osn users to discover , respectively , the identity or the number of their common friends in a private , authentic , and decentralized way . while we build on the concept of capabilities and rely on bfpsi , recall that common friendssuffers from an inherent bootstrapping problem and is limited to the discovery of social paths to osn users that are two hops away . our work does not only address common friends s limitations via a novel methodology , but also presents the full - blown implementation of a scalable server architecture and a modular android client library enabling developers to easily integrate social palinto their applications . mezzour et al . @xcite also describe techniques for decentralized path discovery in social networks . they use a notion similar to capabilities to represent friendships and hashing to derive higher - order capabilities , however , their scheme distributes a _ different _ capability on behalf of a given user to every other user , while social paldistributes the same capability to all users at a given distance . @xcite s computational / communication overhead is significantly higher than that of social pal : the former requires two psi runs , with sets of size equal to the total number of paths from a node up to the maximum supported path length , whereas , the latter only requires a single bfpsi run , with input sets as big as the number of paths that have length equal to half the maximum supported path length .. ] length . table [ tab : input - study ] compares the expected number of input items for social paland @xcite based on the datasets introduced in section [ sec : evaluation ] . furthermore , @xcite incurs the same bootstrapping problem as @xcite : if a friend of user does not participate in the system , can not detect paths to some other user that go through . finally , @xcite aims to build a decentralized social network , while we aim to bootstrap the system based on existing centralized social networks . [ tab : input - study ] liao et al . @xcite present a privacy - preserving social matching protocol based on property - preserving encryption ( ppe ) , which however relies on a centralized approach . li et al . @xcite then propose a set of protocols for privacy - preserving matching of attribute sets of different osn users . similar work include @xcite , @xcite , and @xcite . private friend discovery has also been investigated in @xcite and @xcite , which do not provide authenticity as they are vulnerable to malicious users claiming non - existent friendships . while @xcite addresses the authenticity problem , it unfortunately comes at the cost of relying on relatively expensive cryptographic techniques ( specifically , a number of modular exponentiations linear in the size of friend lists and a quadratic number of modular multiplications ) . smokescreen @xcite , smile @xcite , and pike @xcite support secure / private device - to - device handshake and proximity - based communication . lentz et al . introduce sddr @xcite , which allows a device to establish a _ secure encounter _ i.e. , a secret key with every device in short radio range , and can be used to recognize previously encountered users , while providing strong unlinkability guarantees . the encore platform @xcite builds on sddr to provide privacy - preserving interaction between nearby devices , as well as event - based communication for mobile social applications . * building on social relationships . * prior work has also focused on building services on top of existing social relationships . cici et al . @xcite use osns to assess the potential of ride - sharing services , showing that these would be very successful if users shared rides with friends of their friends . sirivianos et al . @xcite propose a collaborative spam mitigation system leveraging social networks of administrators , while @xcite and @xcite use osns to verify the veracity of online assertions . freedman and nicolosi @xcite describe a system using social network for trust establishment in the context of email white - listing , by verifying the existence of common friends . besides not discovering paths longer than two , @xcite also does not address the issue of friendships authenticity unlike social pal . daly et al . @xcite present a routing protocol ( called simbet ) for dtn networks based on social network data . their protocol attempts to identify a routing bridge node based on the concept of centrality and transitivity of social networks . li et al . @xcite design another dtn routing protocol ( called social selfishness aware routing ) which takes into account user s social selfishness and willingness to forward data only to nodes with sufficiently strong social ties . other work @xcite also propose adjusting message forwarding based on some social metrics . * osn properties . * another line of work has studied properties of osns . ugander et al . @xcite and backstrom et al . @xcite study the structure of facebook social graph , revealing that the average social path length suggested by the small world experiment " @xcite ( i.e. , six ) does not apply for facebook , as the majority of people are separated by a 4-hop path . gilbert et al . @xcite define the relationship between tie strengths ( i.e. , the importance of a social relationship between two users ) and various variables retrieved from the osn social graph . in @xcite , arnaboldi et al . investigate the link between the tie strength definition ( given by granovotter @xcite ) and a composition of factors describing the emotional closeness in online relationships . they demonstrate the existence of the _ dunbar number _ ( i.e. , the maximum number of people a user can actively interact with ) for facebook . in follow - up work @xcite , they also show the existence of four hierarchical layers of social relationships inside ego networks . existence of the dunbar number is also shown for twitter in @xcite . finally , saramki et al . @xcite find an uneven distribution of tie strengths within ego networks that is characterized by the presence of a few strong and a majority of weak ties . this paper presented social pal a system geared to privately estimate the social path length between two social network users . we demonstrated its effectiveness both analytically and empirically , showing that , for any two osn users , social paldiscovers all social paths of length two and a significant portion of longer paths . using different samples of the facebook graph , we showed that even when only 20% of the osn users use the system , we discover more than 40% of all paths between any two users , and 70% with 40% of users . we also implemented a scalable server - side architecture and a modular client library bringing social palto the real world . our deployment supports facebook and linkedin integration and allows developers to easily incorporate it in their projects . social palcan be used in a number of applications where , by relying on the ( privacy - preserving ) estimation of social path length , users can make informed trust and access control decisions . * acknowledgments . * we thank minas gjoka and michael sirivianos for sharing the facebook datasets , swapnil udar for helping with the ` spotshare`implementation , and jussi kangasharju , pasi sarolahti , cecilia mascolo , panos papadimitratos , and narges yousefnezhad for providing feedback on the paper . simon eberz suggested the idea of signed bloom filters discussed in section [ subsec : privacy - consideration ] . this work was partially supported by the academy of finland s `` contextual security '' project ( 274951 ) , the ec fp7 precious project ( 611366 ) , and the eit ict labs .

we study the discrimination of mixed quantum states in an optimal measurement that maximizes the probability of correct results while the probability of inconclusive results is fixed at a given value . after considering the discrimination of states in a-dimensional hilbert space , we focus on the discrimination of qubit states . we develop a method to determine an optimal measurement for discriminating arbitrary qubit states , taking into account that often the optimal measurement is not unique and the maximum probability of correct results can be achieved by several different measurements . analytical results are derived for a number of examples , mostly for the discrimination between qubit states which possess a partial symmetry , but also for discriminating equiprobable qubit states and for the dicrimination between a pure and a uniformly mixed state in dimensions . in the special case where the fixed rate of inconclusive results is equal to zero , our method provides a treatment for the minimum - error discrimination of arbitrary qubit states which differs from previous approaches .

1504.03917

the discrimination between a finite number of quantum states , prepared with given prior probabilities , is essential for many tasks in quantum information and quantum cryptography . since nonorthogonal states can not be distinguished perfectly , discrimination strategies have been developed which are optimal with respect to various figures of merit . based on the outcome of a measurement , in these strategies a guess is made about the actual state of the quantum system . several strategies admit a certain probability , or rate , @xmath2 , of inconclusive results , where the measurement outcomes do not allow to infer the state . in the strategy of minimum - error discrimination @xcite inconclusive results are not permitted , and the overall probability @xmath3 of making a wrong guess is minimized with @xmath4 , which corresponds to maximizing the overall probability @xmath5 of getting a correct result . apart from studying some general features of the optimal measurement @xcite , in the beginning mainly the minimum - error discrimination of states obeying certain symmetry properties or of two mixed states was investigated , see e. g. @xcite . the minimum - error discrimination of more than two states that are arbitrary has gained renewed interest only recently @xcite . generalizing the concept of minimum - error discrimination , an optimal discrimination strategy has been studied which maximizes the total rate of correct results , @xmath5 , with a fixed rate @xmath2 of inconclusive results @xcite . this strategy also maximizes the relative rate @xmath6 for the fixed value @xmath2 , that is it yields the maximum achievable fraction of correct results referred to all conclusive results , which in most cases is larger for @xmath7 than for @xmath8 necessary and sufficient operator conditions for the optimal measurement have been derived @xcite , and it was shown @xcite that under certain conditions the strategy for maximizing @xmath5 with a fixed rate @xmath2 contains the optimal strategies for unambiguous discrimination @xcite or for discrimination with maximum confidence @xcite as limiting cases for sufficiently large values of @xmath2 . moreover , it was found @xcite that when the maximum of @xmath5 is known as a function of the fixed value @xmath2 , then from this function one can also obtain the maximum of @xmath5 in another optimal discrimination strategy where the error probability is fixed @xcite . optimal discrimination with a fixed rate @xmath2 has been also investigated for a modified optimization problem where in contrast to the usual assumption the prior probabilities of the states are unknown @xcite . solutions for optimal state discrimination with a fixed rate @xmath2 of inconclusive results and with given prior probabilities of the states have been recently derived in three independent papers : starting from the operator equations for the optimality conditions , in ref . @xcite we obtained analytical solutions for the discrimination of symmetric states and of two states occurring with arbitrary prior probabilities , where the two states are either pure or belong to a certain class of mixed qubit states . in refs . @xcite and @xcite the respective authors showed that the optimization problem with fixed @xmath2 can be solved by reducing it to a resulting minimum - error problem the solution of which is known and to an additional optimization . they derived analytical solutions for discriminating between two pure states occurring with arbitrary prior probabilities and between the trine states @xcite , and also for the discrimination of geometrically uniform states @xcite . the present paper goes beyond these previous investigations in several respects . based on the ideas of our earlier work @xcite we study general properties of the optimal measurement for discriminating with fixed @xmath2 between @xmath0 arbitrary qudit states in a @xmath1-dimensional hilbert space . specializing on the case @xmath9 , we develop a method for treating the optimal discrimination with fixed @xmath2 between @xmath0 arbitrary qubit states , occurring with arbitrary prior probabilities . in contrast to the previous papers @xcite we take into account that often the optimal measurement is not unique , which means that the maximal probability of correct results with fixed @xmath2 can be obtained by a number of different measurements . in the special case @xmath4 , where the problem corresponds to minimum - error discrimination , our method differs from the approaches developed previously for treating the minimum - error discrimination of @xmath0 arbitrary qubit states @xcite . we obtain explicit analytical results for several problems that have not been solved before , mostly for the discrimination with fixed @xmath2 between qubit states which posses a certain partial symmetry , but also for discriminating @xmath0 equiprobable qubit states and for the discrimination between a pure state and a uniformly mixed state in a @xmath1-dimensional hilbert space . the paper is organized as follows . in sec . ii we start by considering the optimal discrimination with fixed @xmath2 for @xmath0 qudit states . after this we specialize on the discrimination of @xmath0 qubit states in a two - dimensional hilbert space and develop a method for solving the problem . sec . iii is devoted to the discrimination of @xmath0 partially symmetric qubit states . we conclude the paper in sec . iv with discussing the relation of our method to previous studies of the minimum - error discrimination of arbitrary qubit states and with a brief summary of results . the detailed derivations referring to sec . iii are presented in the appendix . we consider the discrimination between @xmath0 qudit states , given by the density operators @xmath10 ( @xmath11 ) and prepared with the respective prior probabilities @xmath12 , where @xmath13 . a complete measurement performing the discrimination is described by @xmath14 positive detection operators @xmath15 with @xmath16 where @xmath17 is the identity operator in the @xmath1-dimensional hilbert space @xmath18 jointly spanned by the @xmath0 density operators @xmath10 . the conditional probability that a quantum system prepared in the state @xmath19 is inferred to be in the state @xmath10 is given by @xmath20 , while @xmath21 is the conditional probability that the measurement yields an inconclusive outcome which does not allow to infer the state . the overall probability of inconclusive results is then given by @xmath22 and the probability of correct results , @xmath5 , reads @xmath23 where @xmath3 denotes the total rate of errors . our task is to maximize @xmath5 , or minimize @xmath3 , respectively , under the constraint that @xmath2 is fixed at a given value . upon introducing a hermitian operator @xmath24 and a scalar real amplifier @xmath25 the necessary and sufficient optimality conditions take the form @xcite @xmath26 ( @xmath27 ) . for @xmath28 and @xmath29 these conditions refer to minimum - error discrimination @xcite , where @xmath4 . provided that eqs . ( [ optz1 ] ) and ( [ optz2 ] ) are fulfilled , the rate of correct results takes its maximum value , given by @xmath30 the detection operators @xmath31 ( @xmath27 ) satisfying the optimality conditions need not always be unique , which means that different measurements can be optimal , yielding the same value @xmath32 . this will be outlined later in this section and will be illustrated by examples in sec . it is useful to introduce the ratio @xmath33 of the maximal rate of correct results , @xmath34 and the rate of conclusive results , @xmath35 by taking the derivative we find with the help of eq . ( [ pcmax- ] ) that @xmath36 here we used the fact that due to the positivity constraint @xmath37 the condition @xmath38 is equivalent to @xmath39 since @xmath40 for @xmath41 , the operator @xmath42 must turn into a full - rank operator when @xmath2 approaches unity . ( [ optz1 ] ) therefore implies that for sufficiently large values of @xmath2 the condition @xmath43 is necessarily fulfilled @xcite . when for the given states the optimality conditions can be satisfied with @xmath44 then due to eq . ( [ lim3 ] ) the ratio @xmath33 grows with increasing @xmath2 if @xmath45 and stays constant for @xmath46 . ( [ lim3a ] ) defines @xmath47 as the smallest value of @xmath2 at which further increasing @xmath2 does not yield any advantage , since the relative rate of correct results , given by @xmath33 , remains constant . if @xmath48 the operator @xmath42 can not be a full - rank operator , due to the equality condition in eq . ( [ optz1 ] ) . for @xmath49 however , this condition does not put any restriction on the rank of @xmath42 . ( [ lim3a ] ) takes into account that for @xmath50 the expressions for @xmath24 and @xmath25 determining the optimal solution are not unique , see the example at the end of sec . ii a. let us first review the limiting case of large @xmath51 where @xmath33 does not change when @xmath2 is increased , which means that @xmath52 after substituting this expression for @xmath24 into eq . ( [ optz2 ] ) we obtain the conditions @xmath53 and @xmath54 for @xmath27 , where @xmath55 and @xmath56 . these conditions require that @xcite @xmath57 the constant @xmath58 , that is the largest eigenvalue of @xmath59 , characterizes the maximum confidence that can be achieved for the individual measurement outcome @xmath60 . in other words , @xmath58 is equal to the maximum achievable value of the ratio @xmath61 which denotes the conditional probability that the outcome @xmath60 is correct given it was obtained @xcite . due to eq . ( [ a - opt ] ) the conditions for optimality following from eq . ( [ optz2 ] ) require that @xmath62 for states where @xmath63 that is where the support of @xmath64 spans the whole hilbert space @xcite . this means that only the states for which the maximum confidence @xmath58 is largest , @xmath65 , are guessed to occur in an optimal measurement where @xmath43 , that is where @xmath66 according to eq . ( [ lim3a ] ) . using eq . ( [ pcmax- ] ) with @xmath43 and @xmath67 we arrive at @xmath68 where @xmath47 is defined by eq . ( [ lim3a ] ) . the actual value of @xmath2 depends on the given states . in special cases where @xmath69 the maximum relative probability of correct results @xmath70 can not be increased at all by admitting inconclusive results but stays always equal to @xmath71 , see for instance the example of equiprobable states resolving the identity operator discussed in ref . @xcite . if @xmath72 the explicit expression for @xmath24 can be easily determined with the help of eq . ( [ a - opt ] ) . the equality conditions in eq . ( [ optz2 ] ) then restrict the supports of the non - zero operators @xmath31 to certain known subspaces . @xmath47 is the smallest value of @xmath73 that satisfies @xmath74 on the condition that the operators @xmath31 have their supports in the given subspaces . @xmath42 is a full - rank operator for @xmath75 when @xmath2 decreases starting from 1 , at a certain value @xmath76 one of the @xmath1 positive eigenvalues of @xmath42 may become zero . if an eigenvalue immediately turns negative when @xmath2 is decreased beyond @xmath76 , then @xmath76 is equal to @xmath47 . in this case @xmath42 is a full - rank operator in the whole region @xmath77 , see the example at the end of sec . ii a , but this need not be valid in general . now we exploit the optimality conditions without the restriction to the limiting case of large @xmath2 . if @xmath78 , the equality in eq . ( [ optz1 ] ) can be only fulfilled when the support of the operator @xmath79 does not span the full hilbert space . this implies that at least one of its eigenvalues is equal to zero and the determinant therefore vanishes , yielding the condition @xmath80 without lack of generality we suppose that in the optimal measurement the detection operators @xmath31 are different from zero for @xmath81 states ( @xmath82 ) and vanish for the remaining @xmath83 states , which means that the remaining states , if any , are never guessed to occur , @xmath84 in analogy to eq . ( [ det1 ] ) we then arrive at the condition @xmath85 the requirement that the determinants in eqs . ( [ det1 ] ) and ( [ det2 ] ) vanish yields a system of @xmath86 real equations with only @xmath87 real variables , namely the parameter @xmath25 and the @xmath88 real quantities determining the matrix elements of the hermitean operator @xmath24 acting in @xmath18 . when the quantum states and their prior probabilities are completely arbitrary , the system of equations therefore does not have a solution for @xmath89 . in special cases , however , a solution can exist where more than @xmath88 states are guessed to occur , for instance when the states have a certain symmetry . in these cases we obtain the same operator @xmath24 and the same parameter @xmath25 when for some of the states we drop eq . ( [ det2 ] ) , keeping it for @xmath88 or less states , and when we put the detection operators corresponding to the dropped states equal to zero , see the examples treated in sec . the measurement then differs from the one described by eq . ( [ op ] ) , but yields the same value @xmath90 when an optimal measurement exists where more than @xmath88 states are guessed , this measurement is therefore not unique . the above considerations lead to the conclusion that we never need to make a guess for more than @xmath88 states in order to discriminate optimally between @xmath0 states in a @xmath1-dimensional hilbert space with a fixed rate @xmath2 of inconclusive results . this conclusion generalizes previous results @xcite that have been obtained in a different way in the context of minimum - error discrimination , where @xmath4 . it follows that for discriminating optimally with fixed @xmath2 between more than @xmath88 arbitrary states we have to consider all possible subsets @xmath91 containing @xmath88 states separately , where @xmath92 , and we have to find the maximal rates of correct results @xmath93 for discriminating the states within each subset . the largest of these rates then determines the optimal solution for discriminating the @xmath0 states , in analogy to the findings obtained for the minimum - error discrimination of qubit states @xcite . the operator @xmath24 and the parameter @xmath25 satisfying eqs . ( [ det1 ] ) - ( [ det2 ] ) only determine the optimal solution if the optimality conditions , eqs . ( [ optz1 ] ) and ( [ optz2 ] ) , can be fulfilled under the constraint that @xmath94 . if for @xmath95 this is not the case , we have to try whether for @xmath96 a solution exists , and so on until @xmath97 where always the same state is guessed to occur when a conclusive outcome is obtained . the described procedure will be applied in sec . ii b for the case of qubit states . an interesting special case arises when one of the states , say the first , has the property that @xmath98 for @xmath99 , implying that @xmath100 . it is obvious that eqs . ( [ optz1 ] ) and ( [ optz2 ] ) and also the completeness relation , eq . ( [ compl-0 ] ) , are then fulfilled if @xmath101 the explicit expressions for @xmath42 and @xmath25 follow from replacing @xmath24 by @xmath102 in eq . ( [ optz1 ] ) , taking into account that @xmath103 and that the positivity constraint @xmath104 has to be satisfied . the condition @xmath105 restricts the validity of this solution to a certain region , and for this region it follows from eq . ( [ pcmax- ] ) that @xmath106 , see the example treated below . clearly , for minimum - error discrimination , where @xmath28 , eq . ( [ ma ] ) yields @xmath107 , which means that the maximum probability of correct results can be obtained without performing any measurement , simply by always guessing the state @xmath108 with the largest prior probability to occur @xcite . in general , it is very hard to obtain analytical solutions of the optimization problem with fixed @xmath2 for @xmath109 in our previous paper @xcite we derived the solution for a special class of linearly independent symmetric pure qudit states . in the following we treat another example , which contains also the special case discussed above . let us consider the optimal discrimination between a uniformly mixed qudit state @xmath110 and a pure qudit state @xmath111 , both living in the same hilbert space @xmath18 and occurring with the prior probabilities @xmath112 and @xmath113 , respectively . taking into account that the spectral representation of @xmath114 is given by @xmath115 where @xmath116 is a @xmath117-dimensional projector , we find that the optimality conditions , eqs . ( [ optz1 ] ) - ( [ optz2 ] ) , are satisfied for @xmath118 , or , equivalently , for @xmath119 , if @xmath120 while for @xmath121 , corresponding to @xmath122 , we obtain in agreement with eq . ( [ ma ] ) the solution @xmath123 in both cases we get @xmath124 . since the detection operators have to be positive , this solution , where @xmath125 , only holds true if @xmath126 with the help of eq . ( [ pcmax- ] ) we find that @xmath127 for @xmath4 this result agrees with the solution derived earlier for minimum - error discrimination @xcite . note that for @xmath128 where @xmath129 the optimal measurement is not unique , as becomes obvious from comparing eqs . ( [ umix2 ] ) and ( [ umix3 ] ) . it is easy to check that the relative rate of correct results , @xmath130 , grows for both lines of eq . ( [ umix1 ] ) when the fixed rate @xmath2 is increased , until it reaches unity for @xmath50 . since according to eq . ( [ a - opt ] ) the maximum confidences for discriminating @xmath108 and @xmath131 are given by @xmath132 and @xmath133 respectively , eq . ( [ umix1 ] ) yields @xmath134 hence for @xmath50 the optimal measurement can be alternatively obtained when @xmath135 with @xmath136 , as becomes obvious from eq . ( [ lim2 ] ) . using eqs . ( [ optz1 ] ) - ( [ optz2 ] ) we find that the optimality conditions are satisfied if @xmath137 since the detection operators have to be positive , this result only holds true if @xmath138 that is @xmath139 we therefore arrive at @xmath140 which because of eq . ( [ pc ] ) expresses the fact that errors do not occur in the optimal measurement for @xmath141 hence in our example @xmath47 is the smallest rate of inconclusive results for which an unambiguous discrimination is possible . ( [ umix2 ] ) or ( [ umix3 ] ) , as well as eq . ( [ umix5 ] ) , show that for @xmath50 the optimal measurement is a projection measurement with @xmath142 , @xmath143 , and @xmath144 , which discriminates @xmath108 unambiguously , that is with the maximum confidence @xmath132 , and yields an inconclusive result when the state @xmath131 is present . while the operator @xmath42 is a rank - one operator for @xmath145 , it follows from eq . ( [ umix5 ] ) that @xmath42 has the rank @xmath1 if @xmath146 that is in this example @xmath42 turns into a full - rank operator at @xmath50 . as mentioned after eq . ( [ lim3a ] ) , for @xmath147 the solution does not have any practical relevance since the relative rate of correct results remains constant with growing @xmath2 and its absolute rate @xmath148 decreases . in the rest of the paper we do not provide the explicit expressions of the detection operators for @xmath149 from now on we specialize our investigations on the optimal discrimination of qubit states in a two - dimensional joint hilbert space @xmath150 here it is sufficient to treat the case where the optimality conditions are satisfied with @xmath151 ( @xmath27 ) and @xmath152 when the first condition does not hold , the optimal measurement is determined by eq . ( [ ma ] ) , while the violation of the second condition corresponds to the limiting case of large @xmath2 where eq . ( [ lim2 ] ) applies . following our earlier paper @xcite we conclude that for @xmath153 and @xmath154 the equalities in the optimality conditions , eqs . ( [ optz1 ] ) and ( [ optz2 ] ) , imply that the non - zero optimal detection operators @xmath42 and @xmath31 are proportional to the projector onto the eigenstates @xmath155 and @xmath156 belonging to the eigenvalue zero of the operators @xmath157 and @xmath158 , respectively . hence if @xmath78 and @xmath159 for @xmath160 with @xmath82 , see eq . ( [ op ] ) , the optimality conditions require that @xmath161 @xmath162 , where @xmath163 and @xmath164 . since we supposed that @xmath165 for @xmath166 the completeness relation , eq . ( [ compl-0 ] ) , is given by @xmath167 due to eqs . ( [ detec0 ] ) and ( [ detec1 ] ) the projectors onto the normalized states @xmath168 and @xmath169 that are orthogonal to @xmath170 and @xmath171 , respectively , are determined by the operator @xmath24 and the parameter @xmath25 via the relations @xmath172 @xmath162 , where @xmath173 . ( [ compl ] ) we find that the completeness relation takes the alternative form @xmath174 for solving the optimization problem we have to determine the parameter @xmath25 and the operator @xmath24 which satisfy eqs . ( [ detec0 ] ) - ( [ compl ] ) , or , equivalently , eqs . ( [ perp ] ) - ( [ compl1 ] ) , for a certain value of @xmath81 on the condition that the constants @xmath175 are positive . for this purpose we can proceed in two steps : in the first step we use eq . ( [ det1 ] ) together with the positivity constraint @xmath176 in order to express @xmath25 and @xmath42 in dependence of the matrix elements of @xmath177 taken with respect to an orthonormal basis in @xmath150 . it is advantageous to choose the particular basis that is the eigenbasis of @xmath114 . introducing the spectral representation @xmath178 where @xmath179 , we represent the operator @xmath24 as @xmath180 the requirement @xmath181 see eq . ( [ det1 ] ) , then leads to @xmath182= |z_{01}|^2.$ ] it turns out that from the two possible solutions only the result @xmath183 ^ 2+\frac{|z_{01}|^2}{r(1-r)}}\ ] ] satisfies the inequality @xmath184 which implies that also @xmath176 since the determinant vanishes . after substituting this result for @xmath25 into eq . ( [ perp ] ) , we can express the completeness relation of the detection operators , eq . ( [ compl1 ] ) , as a function of @xmath2 and of the matrix elements of @xmath24 . in the second step we apply eqs . ( [ det2 ] ) and ( [ compl1 ] ) , together with the positivity constraints @xmath185 the requirement @xmath186 valid for @xmath187 , leads to @xmath188=| z_{10}-\eta_j \rho_{10}^{(j)}|^2 \quad(j=1,\ldots , m),\ ] ] where we used the representation @xmath189 after introducing the abbreviations @xmath190 and after decomposing the non - diagonal matrix elements into their real and imaginary parts , eq . ( [ m - det ] ) can be rewritten as @xmath191 which yields for @xmath192 the equations @xmath193 when the completeness relation , given by eq . ( [ compl1 ] ) together with eq . ( [ perp ] ) , is written in matrix form we arrive at four real equations . our task is to find the matrix elements of @xmath24 , determined by four real parameters , and the @xmath81 constants @xmath175 which together fulfill the @xmath194 equations resulting from eqs . ( [ m - det ] ) and ( [ compl1 ] ) . provided that the solution satisfies the positivity constraints given by eq . ( [ constr ] ) , the optimization problem is solved . when the states are completely arbitrary , the system of @xmath81 equations given by eq . ( [ m - det ] ) can only have a solution if @xmath195 . therefore , when more than four completely arbitrary qubit states are to be optimally discriminated with fixed @xmath2 , we need to separately investigate all possible subsets containing four states , as described in sec . ii for the discrimination of qudit states in @xmath196 we now focus on one such subset containing @xmath197 qubit states and discuss the different possible cases . _ ( i ) guessing of all four states is optimal_. first we assume that @xmath198 . ( [ arb4 ] ) with @xmath199 then represents a system of three linear equations for determining the variables on their left - hand sides , @xmath200 , @xmath201 and @xmath202 , as linear functions of the variable @xmath203 occurring on the right - hand sides . if we insert the resulting expressions into eq . ( [ arb2 ] ) with @xmath204 we arrive at a quadratic equation for @xmath205 . after solving this equation we get explicit expressions for the matrix elements of @xmath24 and , due to eq . ( [ a - sol ] ) , also for the parameter @xmath25 . by inserting these expressions into the four algebraic equations resulting from the matrix form of the completeness relation , we arrive at a linear system of equations for explicitly determining the constants @xmath206 . provided that the conditions in eq . ( [ constr ] ) are satisfied , we have succeeded in determining the optimal solution . if one of the conditions is violated , we next investigate the case where @xmath207 . _ ( ii ) guessing of three states is optimal . we now assume that @xmath197 and @xmath207 , which means that in the optimal measurement one of the four states in the set , say the state with @xmath208 , is never guessed to occur , corresponding to @xmath209 . using eq . ( [ arb4 ] ) with @xmath210 together with eq . ( [ arb2 ] ) we can determine three of the four matrix elements of @xmath24 in dependence of the remaining one , say @xmath205 . after inserting the resulting expressions into the matrix form of eq . ( [ compl1 ] ) we again arrive at four equations . using three of them , the constants @xmath211 , and @xmath212 are obtained as functions of @xmath205 , and after inserting the results into the fourth we obtain an equation for determining @xmath205 which is , however , highly nonlinear for @xmath213 . provided that the positivity conditions given by eq . ( [ constr ] ) with @xmath214 are satisfied , the solution has been obtained . _ ( iii ) guessing of two states or of one state is optimal . _ if the positivity conditions can not be satisfied for @xmath215 or @xmath207 , a corresponding procedure has to be applied for @xmath216 . it may happen that also in this case the positivity conditions can not be fulfilled and therefore @xmath97 in the optimal measurement , which means that it is a projection measurement where either a particular state is guessed to occur or an inconclusive result is obtained . we mention at this place that in our previous paper @xcite the explicit solution for discriminating two mixed qubit states with fixed @xmath2 has been derived for cases where simple algebraic expressions result . the density operators of the qubit states can be represented in the general form @xmath217 where @xmath218 is a normalized state vector , and where the parameters @xmath219 with @xmath220 depend on the purity of the respective states . when the qubit states have equal prior probabilities and equal purities , that is when @xmath221 it is easy to check that the operator @xmath222 leads to @xmath223 . hence the optimality condition given by eq . ( [ optz2 ] ) is satisfied if @xmath224 for @xmath27 . in order to exploit also the first optimality condition , eq . ( [ optz1 ] ) , we use the spectral representation of @xmath114 , given by eq . ( [ rho - diag ] ) , where we denote its eigenstates in such a way that @xmath225 . with @xmath226 it follows that eq . ( [ optz1 ] ) is fulfilled if @xmath227 and @xmath228 . taking the completeness relation in the form of eq . ( [ compl ] ) into account and using eq . ( [ pcmax- ] ) , we find that the maximal probability of correct results is given by @xmath229 with @xmath230 . here @xmath231 denotes the largest eigenvalue of @xmath232 , and @xmath233 is the corresponding eigenstate . we note that for @xmath4 and @xmath234 our result agrees with the result obtained previously for the minimum - error discrimination of equiprobable pure states @xcite . when eq . ( [ compl - rel ] ) can not be satisfied with non - negative values of @xmath175 and eq . ( [ pmax ] ) therefore does not apply , @xmath235 depends on the given states @xmath218 not only via the eigenvalue @xmath231 , but also in a direct way , as will become obvious from the example investigated in sec . iii c. moreover , even if for @xmath4 a solution with non - negative coefficients @xmath175 does exist , at least one of these coefficients may become negative if @xmath2 exceeds a certain critical value @xmath236 , which means that for @xmath237 eq . ( [ pmax ] ) is not valid anymore , see sec . iii c. we mention that in our previous paper @xcite we considered the special case where @xmath238 , that is where @xmath239 and @xmath240 , for which we obtained the complete solution @xmath241 , valid for arbitrary values of @xmath2 . we wish to study a discrimination problem that yields simple analytical solutions but is nevertheless sufficiently general to demonstrate the basic features of the method developed in sec . ii . for this purpose we consider the discrimination of @xmath0 mixed qubit states , represented by eq . ( [ rhoj - gen ] ) , which fall into two groups containing @xmath242 and @xmath243 states , respectively . we suppose that within each group the qubit states have equal prior probabilities and purities , @xmath244 and that @xmath245 with @xmath246 where @xmath247 . the diagonal density matrix elements @xmath248 of @xmath249 are then given by @xmath250 in addition , we assume that @xmath251 , and we consider the cases where @xmath252 eqs . ( [ rho - jj ] ) - ( [ zero - i ] ) imply that the non - diagonal density matrix elements of @xmath253 vanish and yield the spectral representation @xmath254 the state vectors @xmath218 in eq . ( [ psi - j ] ) can be visualized with the help of a bloch sphere where @xmath233 and @xmath255 are the north and south pole , respectively , that is @xmath256 with @xmath257 and @xmath258 . clearly , in each group the end points of the bloch vectors representing the underlying states @xmath218 have the same latitude , lying in the upper hemisphere if @xmath259 and in the lower hemisphere if @xmath260 , see fig . the phases can be represented as @xmath261 where @xmath262 is an arbitrary phase shift . clearly , the density operators obey the symmetry conditions @xmath263 where the unitary operator @xmath264 with @xmath265 characterizes the symmetry in the first group of states , and @xmath266 with @xmath267 refers to the second group . in the case ( ii ) we put @xmath268 . the cases ( i ) and ( ii ) belong to the specific kind of partially symmetric states introduced in our earlier paper , ref . @xcite , while in the case ( iii ) the @xmath0 states are symmetric . for later purposes we still calculate the maximum confidence @xmath58 for discriminating an individual state @xmath60 , using eq . ( [ a - opt ] ) . we get @xmath269 for @xmath270 , and @xmath271 for @xmath272 , where @xmath273 ^ 2}}\right],\ ] ] and where @xmath274 follows from eq . ( [ maxconf1 ] ) when @xmath275 , and @xmath276 are replaced by @xmath277 , and @xmath278 , respectively . in the special case where @xmath279 , meaning that @xmath280 and @xmath281 the expression for @xmath282 is in agreement with our earlier result @xcite for the maximum - confidence discrimination of symmetric mixed qubit states . in order to solve our optimization problem we apply the method outlined in sec . ii . due to the partial symmetry of the states , the treatment can be considerably simplified . in fact , assuming that at least from one of the two groups , say the first , two or more states are guessed to occur , we get for these states from eq . ( [ m - det ] ) the conditions @xmath283=| z_{10}-\eta \rho_{10}^{(j)}|^2 $ ] with @xmath284 since these conditions can only be simultaneously satisfied for two or more different phases @xmath285 if @xmath286 , we have to search for an operator @xmath24 where @xmath287=0.\ ] ] we note that the requirement @xmath288=0 $ ] was derived already in our earlier paper @xcite , using the fact that there always exists an optimal measurement where the detection operators @xmath31 obey the same symmetry as the density operators @xmath10 of the partially symmetric states . as mentioned in the beginning of sec . ii b , it is sufficient to perform the derivation for the case where @xmath38 and @xmath289 @xmath290 . in the first step we make use of eq . ( [ det1 ] ) . since @xmath286 , the condition @xmath291 is equivalent to @xmath182= 0 $ ] . taking the positivity constraint @xmath292 and the relation @xmath103 into account , we find that the optimality condition in eq . ( [ optz1 ] ) is satisfied if @xmath293 where the expression for @xmath25 is in accordance with eq . ( [ a - sol ] ) . note that if @xmath294 both eigenvalues of @xmath79 vanish which means that @xmath135 and which corresponds to the case @xmath295 , yielding eq . ( [ lim2 ] ) . due to eq . ( [ pcmax- ] ) the maximum probability of correct results for @xmath48 that is for @xmath296 takes the form @xmath297 in the second step we apply eq . ( [ det2 ] ) and the completeness relation , represented by eq . ( [ compl1 ] ) , together with the positivity constraints given by eq . ( [ constr ] ) . ( [ m - det ] ) we arrive at the conditions @xmath298=\eta^2 p^2\ , b(1\!-b)\,\;\;\;{\rm if}\;a_1>0,\qquad\\ ( z_{00}\!-\!\eta^{\prime } s^{\prime})[z_{11}\ ! -\!\eta^{\prime}(1\!-\!s^{\prime})]=\eta^{\prime 2}p^{\prime 2}c(1\!-\!c)\;\;\,\,{\rm if}\;a_2>0,\quad \end{array}\ ] ] where we introduced @xmath299 because of eq . ( [ detec1 ] ) the condition @xmath300 is equivalent to requiring that at least one state from the first group is guessed , and @xmath301 means that at least one state from the second group is inferred to occur . next we make use of eqs . ( [ compl1 ] ) and ( [ perp ] ) . because of the explicit form of the operator @xmath42 , given by eq . ( [ pcmax1 ] ) , the diagonal matrix elements of eq . ( [ compl1 ] ) lead to the conditions @xmath302 where the constants @xmath303 and @xmath304 read @xmath305 since @xmath306 the non - diagonal elements of eq . ( [ compl1 ] ) yield the condition @xmath307 due to the symmetry properties of the partially symmetric states , eq . ( [ compl2c ] ) is certainly fulfilled if @xmath308 that is if in the optimal measurement the detection operators for the individual states obey the same symmetry as their density operators , as can be seen with the help of eqs . ( [ detec11 ] ) and ( [ perp ] ) . however , other choices of the coefficients @xmath175 and therefore of the detection operators are also possible . taking into account that @xmath309 for @xmath310 see eq . ( [ phi ] ) , we find for instance that @xmath311 if for even numbers @xmath242 @xmath312 and for odd numbers @xmath242 @xmath313 while all other coefficients @xmath175 @xmath314 vanish . similarly , various solutions with @xmath315 exist for the coefficients @xmath175 that belong to the states in the second group and depend on @xmath316 . moreover , if @xmath317 eq . ( [ compl2c ] ) may be also solved in a way where the terms referring to the two groups of states do not vanish separately . in order to find an optimal measurement we have to determine the values of @xmath205 , @xmath200 , @xmath318 and @xmath316 that satisfy eqs . ( [ two1a ] ) - ( [ f ] ) with @xmath319 and @xmath320 , where according to eq . ( [ optz2 ] ) the positivity constraints @xmath321 have to be fulfilled . for @xmath317 , where both lines of eq . ( [ two1a ] ) apply , the positivity constraints are satisfied provided that @xmath322 while for @xmath300 and @xmath323 the additional condition @xmath324\geq \eta^{\prime 2 } p^{\prime 2 } c(1-c)\ ] ] has to be taken into account . a similar condition arises when @xmath301 and @xmath325 since in general the resulting solutions are rather involved , we treat the general case only in implicit terms and provide explicit derivations of special solutions in the appendix . two different cases have to be considered : _ ( i ) guessing states from both groups is optimal . _ for @xmath326 eq . ( [ two1a ] ) represents a system of two coupled equations from which we obtain expressions for @xmath205 and @xmath200 after solving a quadratic equation . when these expressions are inserted into eqs . ( [ compl2a ] ) and ( [ compl2b ] ) , the resulting linear system of equations yields the values of @xmath318 and @xmath316 , which depend on @xmath2 due to eq . ( [ f ] ) . provided that these values are indeed positive and that eq . ( [ constr2 ] ) is satisfied , we have obtained the optimal solution . if this is not the case , a measurement where states from both groups are guessed can not be optimal . _ ( ii ) guessing states from only one group is optimal . _ to be specific , let us assume that only the states from the first group are guessed , which means that @xmath300 and @xmath323 . from eqs . ( [ compl2a ] ) and ( [ compl2b ] ) with @xmath323 we obtain the two equations @xmath327.\ ] ] using in addition the first line of eq . ( [ two1a ] ) and taking eq . ( [ f ] ) into account , we arrive at three equations for determining @xmath328 and @xmath318 in dependence of @xmath2 . provided that indeed @xmath300 and that eqs . ( [ constr2 ] ) and ( [ constr2a ] ) are satisfied , the optimal solution has been determined . for the case where only the states from the second group are guessed we can proceed in an analogous way . in the next two subsections we present the complete solutions for two simplified but non - trivial discrimination problems for partially symmetric states . in our first problem we suppose that the purity of the states in the two groups is the same , @xmath329 , and that the states occur with the same prior probability , @xmath330 . we denote the eigenstates of @xmath114 in such a way that @xmath231 , given by eq . ( [ rhodiag ] ) , is the largest eigenvalue , @xmath331 \geq \frac{1}{2}.\ ] ] to be specific , we suppose that @xmath332 which implies that @xmath333 if @xmath334 . with the help of eq . ( [ maxconf1 ] ) this yields the relation @xmath335 for the maximum achievable confidences for discriminating the states in the two groups , where the equality sign holds if @xmath240 . as derived in the appendix , the maximum probability of correct results with fixed @xmath2 is then given by the solution @xmath336 where @xmath337 and @xmath338 , in agreement with eq . ( [ pmax ] ) , while @xmath47 and @xmath339 are defined by eqs . ( [ solq3 ] ) and ( [ a-2 ] ) in the appendix , respectively . three different regions of @xmath2 have to be distinguished : for @xmath340 , corresponding to the full lines in fig . 2 , states from both groups are guessed in the optimal measurement . clearly , this only applies if @xmath341 , that is if @xmath342 . @xmath32 is then determined by @xmath343 , in agreement with the general result derived in eq . ( [ pmax ] ) . in this region the optimum detection operators are @xmath344 and @xmath345 for @xmath346 , where the states @xmath218 are given by eq . ( [ psi - j ] ) . the values of the constants @xmath175 can be determined by applying the method described in sec . iii b , using the explicit expressions for @xmath318 and @xmath316 given in the appendix . in the region where @xmath347 , corresponding to the dashed lines in fig . 2 , only the states of the first group are guessed in the optimal measurement . the optimal detection operators follow from the expression for @xmath24 , see eq . ( [ z0 ] ) , and from eqs . ( [ perp ] ) and ( [ detec11 ] ) . the constants @xmath175 can be obtained as described sec . iii b , with the help of the value of @xmath318 given by eq . ( [ a1 ] ) . if @xmath348 and therefore @xmath349 the second line of eq . ( [ solq ] ) holds true in the whole region @xmath350 . since @xmath332 , the bloch vectors of the states are then confined to the upper hemisphere of the bloch sphere . the region where @xmath295 corresponds to the limiting case of large @xmath2 described in sec . ii a. in this region the ratio @xmath351 , does not increase anymore with growing @xmath2 but stays equal to @xmath282 , see the dotted lines in fig . 2 . the squares in fig . 2 indicate the points where @xmath50 for different numbers of @xmath242 and @xmath352 . two special cases are worth mentioning . for @xmath4 , that is for minimum - error discrimination , we obtain from eq . ( [ solq ] ) @xmath353 when the first line applies , states from both groups are guessed in the optimal measurement , while otherwise only states from the first group are guessed to occur . interestingly , the solution only depends on the total number @xmath354 of the states and is independent of their distribution over the two groups . the second special case refers to @xmath279 , that is to the discrimination of @xmath355 equiprobable symmetric mixed qubit states . ( [ r1 ] ) then reduces to @xmath356 and requires that @xmath357 . this means that the first line of eq . ( [ solq ] ) does not apply , and that @xmath339 is determined by the first line of eq . ( [ a-2 ] ) . this solution coincides with our earlier result @xcite that was derived for symmetric detection operators . in contrast to this , the present derivation does not impose any restriction on the choice of the detection operators . in particular , it shows that for @xmath0 symmetric mixed qubit states an optimal discrimination with fixed @xmath2 can be always accomplished when not more than three ( for @xmath0 odd ) or two ( for @xmath0 even ) states are guessed to occur , as follows from eqs . ( [ even ] ) and ( [ odd ] ) . for @xmath4 , that is for minimum - error discrimination , the optimum measurement therefore can be always realized by a simple projection measurement if @xmath0 is even . now we assume that the states are pure , @xmath358 , and that the respective prior probabilities @xmath276 and @xmath278 in the two groups of states can be different , where @xmath359 . in order to obtain the complete solution for arbitrary values of @xmath276 we have to consider both the cases where the relations @xmath360 and @xmath361 hold true between the maximum confidences for the states in the two groups , see eqs . ( [ pure4a ] ) and ( [ pcmax2 ] ) in the appendix . we present explicit results for the simplest case , where @xmath362 , @xmath363 and @xmath364 , that is where the states are given by latexmath:[\[\label{mirror } and occur with the prior probabilities @xmath276 for @xmath366 and @xmath367 , and @xmath368 for @xmath369 respectively . these states correspond to the three mirror - symmetric pure states the minimum - error discrimination of which has been previously investigated @xcite . the condition @xmath360 holds true provided that @xmath370 , where for @xmath371 we get @xmath372 as shown in the appendix , for @xmath360 the maximum probability of correct results reads @xmath373 where @xmath47 is determined by eqs . ( [ solq3 ] ) and ( [ solq2a ] ) with @xmath234 , and where @xmath236 , @xmath343 , and @xmath339 are given by eqs . ( [ crit ] ) - ( [ mirror1a ] ) with @xmath374 . on the other hand , for @xmath375 we obtain @xmath376 where @xmath377 , @xmath378 and @xmath379 are determined by eqs . ( [ mirror1a ] ) - ( [ mirror3 ] ) . in the regions of @xmath2 where the solution is given by @xmath339 , corresponding to the dashed lines in fig . 3 , only the states @xmath380 and @xmath381 are guessed to occur in the optimal measurement . on the other hand , all three states are guessed if @xmath382 for @xmath383 and if @xmath384 for @xmath375 , which corresponds to the full lines in fig . 3 . for @xmath385 we find that @xmath375 if @xmath386 as outlined in the appendix , for @xmath387 the first line of eq . ( [ pure3prime ] ) does not apply , and the second line is valid in the whole range @xmath388 since @xmath389 the dotted lines in fig . 3 correspond to the limiting case of large @xmath2 , and the squares indicate the points where this limiting case is reached . for minimum - error discrimination , where @xmath4 , the maximum probability of correct results does not depend on the relation between @xmath282 and @xmath274 . ( [ solq1 ] ) and ( [ pure3prime ] ) yield @xmath390 \ ; & \mbox{if $ \ ; \eta\leq \eta_{cr } $ } \quad\\ \!\!\eta\left(1 + 2\sqrt{b(1-b}\right ) \ ; & \mbox{if $ \ ; \eta \geq \eta_{cr } $ } , \quad\\ \end{array } \right.\ ] ] which coincides with the result obtained already in ref . when @xmath391 all three states are guessed in the measurement performing minimum - error discrimination , while otherwise only the states @xmath380 and @xmath381 are guessed to occur . before concluding the paper , we briefly discuss the relation of our method to previous investigations of the minimum - error discrimination of arbitrary qubit states , where @xmath4 . from eq . ( [ perp ] ) we obtain the representation @xmath392 @xmath393 , where due to eq . ( [ compl1 ] ) for minimum - error discrimination the condition @xmath394 has to be satisfied with non - negative values of @xmath175 . upon eliminating the operator @xmath24 , we obtain from the first equality in eq . ( [ mx ] ) for any pair of states the equation @xmath395 ( @xmath396 ) . ( [ my ] ) has been recently derived in an alternative way @xcite , and has been applied to study the minimum - error discrimination of qubit states using a geometric formulation @xcite . in contrast to this , in our method , which refers to the general case @xmath397 , the operator @xmath24 is not eliminated . rather , our approach essentially rests on determining @xmath24 and is therefore for @xmath4 related to the treatments of minimum - error discrimination in refs . @xcite,@xcite and @xcite . in this paper we investigated the discrimination of @xmath0 mixed quantum states by an optimal measurement that yields the maximum probability of correct results , @xmath32 , while the probability of inconclusive results is fixed at a given value @xmath397 . for the discrimination of qudit states in a @xmath1-dimensional hilbert space , we discussed the general properties of the optimal measurement . moreover , we derived the analytical solution for optimally discriminating with fixed @xmath2 between a uniformly mixed and a pure qudit state . in the main part of the paper we specialized on the optimal discrimination of qubit states in a two - dimensional hilbert space and developed a general method to obtain the solution . we studied the special case where the prior probabilities of the qubit states are equal , and we also treated the discrimination between four or less arbitrary qubit states with fixed @xmath2 . as an illustrative application of our method , we derived explicit analytical results for discriminating qubit states which posses a partial symmetry . we emphasize that apart from determining @xmath32 , our method also allows to consider the various possible realizations of the optimal measurement for a given discrimination problem . in particular , we found that for discriminating @xmath0 symmetric qubit states the maximum probability of correct results with fixed @xmath2 can be for instance also achieved by a measurement where only three of the states are guessed to occur when @xmath0 is odd , and only two of the states when @xmath0 is even , instead of guessing all @xmath0 states . note added : after submitting this work a related paper @xcite appeared . in this appendix we provide the detailed derivations for the results presented in sec . we start by treating the case where only states from the first group are guessed , that is @xmath398 using eq . ( [ first ] ) and the first line of eq . ( [ two1a ] ) , we obtain @xmath399 with @xmath400 and @xmath401 . from eq . ( [ z0 ] ) it follows that the condition @xmath402 is equivalent to @xmath403 where @xmath404^{2}\ ] ] with @xmath405 for @xmath234 . making use of eq . ( [ f ] ) we therefore arrive at @xmath406 and due to eq . ( [ first ] ) we obtain the explicit result @xmath407 in the upper line we took into account that @xmath408 corresponds to @xmath409 which implies that @xmath410 , and we used a similar relation for the lower line . since @xmath397 , eq . ( [ solq2 ] ) can only be fulfilled and hence eq . ( [ z0 ] ) can only determine the optimal solution if @xmath411 where @xmath50 corresponds to @xmath412 from eqs . ( [ z0 ] ) and ( [ sol ] ) we obtain the maximum probability of correct results @xmath413 with @xmath414 since the expressions in the upper and lower lines of eqs . ( [ solq2 ] ) - ( [ a-2 ] ) are identical for @xmath415 and @xmath416 we can replace the conditions @xmath417 by @xmath418 , and similarly @xmath419 by @xmath420 , if we extend the restriction @xmath45 to @xmath145 . ( [ a-2 ] ) describes the optimal solution provided that @xmath205 and @xmath421 determined by eq . ( [ z0 ] ) , satisfy the positivity constraints given by eq . ( [ constr1 ] ) . analogous results can be obtained for the case where only states from the second group are guessed to occur , that is where @xmath422 . we still need to study the case where states from both groups are guessed in the optimal measurement , that is where @xmath300 and @xmath301 . we shall do this in the following , where we derive the complete solutions for the problems discussed in secs . iii c and iii d. _ equiprobable states with equal purity . _ in accordance with eq . ( [ r1 ] ) we assume that @xmath334 and @xmath332 , which for @xmath423 and @xmath329 corresponds to @xmath335 . supposing that states from both groups are guessed , that is @xmath326 , we obtain from eq . ( [ two1a ] ) the solution @xmath424 which clearly satisfies the positivity constraints given by eq . ( [ constr1 ] ) . taking into account that @xmath425 , eqs . ( [ compl2a ] ) - ( [ f ] ) yield the constants @xmath426 and @xmath427 . since @xmath333 , cf . ( [ r1 ] ) , @xmath318 can not be negative . on the other hand , the condition @xmath320 requires that @xmath2 does not exceed the critical value @xmath428 . with the help of eq . ( [ sol ] ) we thus arrive at the first line of eq . ( [ solq ] ) . the second line of eq . ( [ solq ] ) refers to the case where the solution determined by eqs . ( [ z0 ] ) and ( [ a-2 ] ) is optimal , as will be shown below by verifying that eq . ( [ constr1 ] ) is satisfied . the third line corresponds to eq . ( [ lim2 ] ) . with the help of eq . ( [ solq2a ] ) we find after a little algebra that @xmath429 if @xmath334 and @xmath332 . hence for @xmath430 it follows that @xmath431 which means that for @xmath432 always the first line of eq . ( [ a-2 ] ) applies . it remains to be shown that for @xmath347 the positivity constraints given by eq . ( [ constr1 ] ) are fulfilled when @xmath205 and @xmath200 are determined by eq . ( [ z0 ] ) . from eqs . ( [ constr2 ] ) and ( [ constr2a ] ) with @xmath433 and @xmath329 we find after minor algebra that eq . ( [ constr1 ] ) is satisfied if @xmath434 , which because of eq . ( [ solq2 ] ) yields the two conditions @xmath435 if @xmath418 and @xmath436 if @xmath437 the first condition requires that @xmath438 . the second condition is fulfilled for @xmath439 , as becomes obvious from the second line of eq . ( [ solq2 ] ) and from the validitiy of the relation @xmath440 hence eq . ( [ constr1 ] ) is indeed satisfied and we have derived eq . ( [ solq ] ) . _ pure states with different prior probabilities . _ for @xmath358 we obtain with the help of eq . ( [ maxconf1 ] ) the relation @xmath441 where @xmath442 and @xmath443 , while @xmath444 with @xmath359 . supposing that states from both groups are guessed , we get from eq . ( [ two1a ] ) the solutions @xmath445 and @xmath446 which have to positive when eq . ( [ constr2 ] ) holds . to be specific , we again assume that @xmath332 . the condition @xmath447 then implies that @xmath448 , and a corresponding relation is valid when @xmath449 . from eq . ( [ sol ] ) we obtain the solution @xmath450 which holds true for certain regions of @xmath2 where the positivity constraint given by eq . ( [ constr1 ] ) is satisfied and where eqs . ( [ compl2a ] ) - ( [ f ] ) , resulting from the completeness relation , can be fulfilled with positive values of @xmath318 and @xmath316 . outside these regions only states from one of the groups will be guessed in the optimal measurement . now we specialize on the discrimination of three mirror - symmetric pure states , given by eq . ( [ mirror ] ) , where @xmath368 and @xmath451 . ( [ maxconf1 ] ) yields the respective maximum confidences @xmath452 for @xmath380 and @xmath381 , and @xmath453 for the state @xmath454 . provided that all three states are guessed , that is @xmath455 , we obtain from eq . ( [ two1a ] ) the solution @xmath456 and @xmath457 . for @xmath360 , that is for @xmath458 , it follows that the relation @xmath459 holds true when all three states are guessed . ( [ compl2a ] ) - ( [ f ] ) then yield the solutions @xmath460 and @xmath461 . here we introduced a critical value @xmath236 , given by @xmath462 ^ 2 } , \quad{\rm where } \;\ ; q_{cr}\geq0\;\ ; { \rm if}\;\;\eta \leq \eta_{cr}\ ] ] with @xmath463^{-1}$ ] . while @xmath318 is always positive , the condition @xmath320 requires that @xmath464 due to eq . ( [ sol ] ) we find that @xmath465 for @xmath466 where @xmath467.\ ] ] clearly , this result only applies when @xmath468 or @xmath469 , respectively , which implies that @xmath470 and that eq . ( [ constr1 ] ) is therefore satisfied . as will be shown below , for @xmath471 a measurement is optimal where @xmath323 , that is where eq . ( [ z0 ] ) determines the solution . extending eq . ( [ a-2 ] ) to the case @xmath415 , we obtain @xmath472 for @xmath473 , where @xmath474 hence we have obtained the first two lines of eq . ( [ solq1 ] ) , and the third line follows again from eq . ( [ lim2 ] ) . in eq . ( [ mirror1a ] ) we took into account that with @xmath475 the condition @xmath476 is equivalent to the condition @xmath477 . we note that for @xmath478 the condition @xmath479 , or @xmath480 , respectively , can be only fulfilled if @xmath481 , which requires that @xmath482 for @xmath483 we therefore get @xmath484 , or @xmath485 , respectively , which means that the solution is given by the lower line of eq . ( [ mirror1a ] ) in the whole range @xmath486 next we treat the case where @xmath375 , or @xmath487 , respectively , which means that @xmath488 if all three states are guessed . from eqs . ( [ compl2a ] ) - ( [ f ] ) we obtain @xmath489 and @xmath490 . it is useful to introduce @xmath378 with @xmath491 , that is @xmath492 ^ 2}{\eta^2b(1- b ) } , \quad{\rm where } \;\ ; q_{cr}^{\prime}\geq 0\;\ ; { \rm if}\;\;\eta \geq \eta_{cr}.\ ] ] provided that @xmath493 the conditions @xmath319 and @xmath494 hold true . ( [ sol ] ) we therefore get @xmath495 for @xmath496 , where @xmath497.\ ] ] thus we have obtained the second line of eq . ( [ pure3prime ] ) . for @xmath498 , where @xmath422 , only the state @xmath454 is guessed . from eq . ( [ mirror3 ] ) it follows that @xmath499 when @xmath500 this means that in eq . ( [ lim2 ] ) @xmath47 corresponds to @xmath501 , and we arrive at the third line of eq . ( [ pure3prime ] ) . on the other hand , for @xmath502 we get @xmath398 as will be shown below , when @xmath503 a measurement is optimal where @xmath323 and where the solution is therefore determined by @xmath339 , given by eq . ( [ mirror1a ] ) . this yields the first line of eq . ( [ pure3prime ] ) . for @xmath504 the condition @xmath505 , or @xmath506 , respectively , can be only fulfilled if @xmath507 , which requires that @xmath508 for @xmath509 we therefore get @xmath510 , which means that the solution is given by eq . ( [ mirror3 ] ) in the whole range @xmath511 . let us finally consider the positivity constraints given by eq . ( [ constr1 ] ) for the case where the solution is described by @xmath339 , that is where @xmath205 and @xmath200 are determined by eq . ( [ z0 ] ) . making use of eqs . ( [ constr2 ] ) and ( [ constr2a ] ) with @xmath234 and @xmath364 we find after minor algebra that eq . ( [ constr1 ] ) is satisfied for @xmath512 ^ 2 } .$ ] taking the definition of @xmath236 into account , we therefore obtain from eq . ( [ solq2 ] ) the two conditions @xmath513 if @xmath418 and @xmath514 if @xmath515 for @xmath360 , or @xmath516 , respectively , the first condition requires that @xmath438 , while the second condition is always fulfilled for @xmath517 , as follows from the second line of eq . ( [ solq2 ] ) and from the relation @xmath518 which holds for @xmath516 . for @xmath375 the second condition requires that @xmath519 since @xmath520 , which means that it only applies if @xmath521 , or @xmath522 , respectively , where the first condition is always satisfied . ( [ constr1 ] ) indeed restricts the range of validity of eq . ( [ z0 ] ) to the regions @xmath523 if @xmath375 and @xmath473 if @xmath524 the author acknowledges partial financial support by deutsche forschungsgemeinschaft dfg ( sfb 787 ) .

the propagation of dissipationless , hydromagnetic , purely toroidal alfvn waves in a realistic background three - fluid solar wind with axial symmetry and differential proton - alpha flow is investigated . the short wavelength wkb approximation is not invoked . instead , the equations that govern the wave transport are derived from standard multi - fluid equations in the five - moment approximation . the alfvnic point , where the combined poloidal alfvn mach number , is found to be a singular point for the wave equation , which is then numerically solved for three representative angular frequencies , and rad s with a fixed wave amplitude of 10 km s imposed at the coronal base ( 1 ) . the wave energy and energy flux densities as well as wave - induced ion acceleration are computed and compared with those derived in the wkb limit . between 1 and 1 au , the numerical solutions show substantial deviation from the wkb expectations . even for the relatively high frequency , a wkb - like behavior can be seen only in regions . in the low - frequency case , the computed profiles of wave - related parameters show a spatial dependence distinct from the wkb one , the deviation being particularly pronounced in interplanetary space . in the inner corona , the computed ion velocity fluctuations are considerably smaller than the wkb expectations in all cases , as is the computed wave - induced acceleration exerted on protons or alpha particles . as for the wave energy and energy flux densities , they can be enhanced or depleted compared with the wkb results , depending on . with the chosen base wave amplitude , the wave acceleration has negligible effect on the ion force balance in the corona . hence processes other than the non - wkb wave acceleration are needed to accelerate the ions out of the gravitational potential well of the sun . however , at large distances beyond the alfvnic point , the low - frequency waves can play an important role in the ion dynamics , with the net effect being to equalize the speeds of the two ion species considered .

astro-ph0702383

ever since their identification by @xcite , alfvn waves have been extensively studied using _ in situ _ measurements , such as by helios and ulysses , covering the heliocentric distance from 0.29 to 4.3 au @xcite . on the other hand , the non - thermal broadening of a number of ultraviolet lines , such as those measured with the sumer ( solar ultraviolet measurements of emitted radiation ) and uvcs ( ultraviolet coronagraph spectrometer ) instruments on soho ( the solar and heliospheric observatory ) , is usually attributed to the transverse velocity fluctuations , thereby enabling one to infer the amplitudes of these fluctuations in the inner corona below @xmath11 @xmath5 @xcite . moreover , the faraday rotation measurements , which yield information regarding the line - of - sight magnetic field fluctuations , have been shown to support indirectly the presence of alfvn waves inside 10 @xmath5 @xcite . the solar wind in intermediate regions , for the time being , can be explored only by radio scintillation measurements which allow one to derive the velocity fluctuations ( e.g. , * ? ? ? * ; * ? ? ? it is noteworthy that although the hourly - scale fluctuations seem to be more frequently studied , the fluctuation spectrum measured by helios nevertheless spans a broad frequency range from @xmath3 to @xmath12 @xmath13 @xcite . most of the theoretical investigations into the interaction between alfvn waves and the solar wind have been performed in the short wavelength wkb limit which makes the problem more tractable mathematically . for instance , by employing the wkb approximation , @xcite derived an expression for the ponderomotive force through which the alfvn waves may provide further acceleration to the solar wind . the wave acceleration was later incorporated in detailed numerical models by , e.g. , @xcite . it was soon realized that alfvn waves may also heat the solar wind via dissipative processes such as the cyclotron resonance interaction between ions and high frequency , parallel propagating waves generated by a turbulent cascade ( cf . the extensive review by @xcite ) . such a parallel cascade scenario has been successful in explaining a number of observations , to name but one , the significant thermal anisotropy of ions as established by uvcs measurements @xcite . as pointed out by @xcite , the applicability of the wkb approximation in which the processes are formulated is questionable in the near - sun region in view of the large alfvn speeds . furthermore , a turbulent cascade requires a non - vanishing magnetic reynolds stress tensor , which , however , is zero in the wkb limit since the particle and field components of the tensor cancel each other exactly . a non - wkb analysis is therefore required to account for the wave reflection and the consequent driving of any turbulence cascade . as a matter of fact , non - wkb analysis of alfvn waves in the solar wind has been carried out for decades ( e.g. , * ? ? ? * ; * ? ? ? this , however , is almost exclusively done in the framework of ideal mhd , which allows waves propagating in opposite directions to be explicitly separated when the elssser variables are used . the adoption of the elssser variables has also enabled a new turbulence phenomenology concerning the nonlinear coupling between counter - propagating waves @xcite . this coupling term , if interpreted as the energy cascaded towards fluctuations with increasingly large perpendicular wavenumbers , is also more consistent with theoretical expectations . despite substantial advances achieved in ideal mhd , which is appropriate for the description of the gross properties of the solar wind , the non - wkb analysis of alfvn waves has rarely been done using multi - fluid transport equations . in contrast , multi - fluid , alfvn wave driven solar wind modeling formulated in the wkb limit has reached considerable sophistication ( see * ? ? ? * ) . such a multi - fluid approach is particularly necessary for the solar wind since the alpha particles must be included given their non - negligible abundance and the fact that the proton - alpha differential speed can be a substantial fraction of the proton speed in the fast stream @xcite . there is therefore an obvious need to extend available non - wkb analyses of alfvn waves from the ideal mhd to the multi - fluid case . the intent of this paper is to present an analysis of alfvn waves in a 3-fluid solar wind assuming axial symmetry , without assuming that the wavelength is small compared with the spatial scales at which the background parameters vary . the perturbed velocity and magnetic field are assumed to be in the azimuthal direction , i.e , only purely toroidal waves are investigated . to further simplify the treatment , the wave dissipation is neglected . this simplification is necessary here since if one wants to gain some quantitative insights into the wave dissipation , and to maintain a reasonable self - consistency at the same time , one should formulate the dissipation in terms of the amplitudes of waves propagating outwards and inwards . to this end , a full multi - fluid elssser analysis is required but is unfortunately unavailable at the present time . however , if the wave dissipation is neglected and therefore the nonlinear interaction between waves and the multicomponent wind is entirely through the agent of ponderomotive forces , then the problem can be formulated without distinguishing explicitly between the directions of wave propagation ( cf . the paper is organized as follows . in section [ sec_mathform ] , we show how to reduce the general multi - fluid transport equations to the desired form . the resultant equations governing the fourier amplitudes of toroidal alfvn waves at a given frequency are then solved analytically in two limiting cases , namely the wkb and zero - frequency limits , in section [ sec_ana_sol ] . apart from these analytically tractable cases , the equations have to be solved numerically . in section [ sec_num_model ] , we re - formulate the model equations for numerical convenience , describe the background flow parameters and detail the solution procedure as well . the numerical solutions for three different frequencies are presented in section [ sec_num_res ] . finally , section [ sec_conc ] summarizes the results , ending with some concluding remarks . presented in this section is the mathematical development of the equations that govern the toroidal fluctuations in a solar wind which consists of electrons ( @xmath14 ) , protons ( @xmath15 ) and alpha particles ( @xmath16 ) . each species @xmath17 ( @xmath18 ) is characterized by its mass @xmath19 , electric charge @xmath20 , number density @xmath21 , mass density @xmath22 , velocity @xmath23 , and partial pressure @xmath24 . if measured in units of the electron charge @xmath14 , @xmath20 may be expressed by @xmath25 with @xmath26 by definition . to simplify the mathematical treatment , a number of assumptions have been made and are collected as follows : 1 . it is assumed that the solar wind can be described by the standard transport equations in the five - moment approximation . 2 . quasi neutrality and quasi - zero current are assumed , i.e. , @xmath27 and @xmath28 where @xmath29 . symmetry about the magnetic axis is assumed , i.e. , @xmath30 in a heliocentric spherical coordinate system ( @xmath31 ) . the time - independent solar wind interacts with the waves only through the wave - induced ponderomotive forces . the wave frequency considered is in the hydromagnetic regime , i.e. , well below the ion gyro - frequencies . the perturbed velocity and magnetic field are assumed to be in the @xmath32 direction only . the effects of the solar rotation on the background solar wind are neglected such that there is no need to consider the coupling of alfvn waves to the compressional modes in the presence of a spiral magnetic field . the effects of the coulomb friction on the waves are neglected , and so is the wave - induced modification of the coulomb friction between background ion flows . the equations appropriate for a multi - component solar wind plasma in the standard five - moment approximation are as follows ( for the derivation see appendix a.1 in @xcite ) @xmath33 \nonumber \\ & - & \frac{z_k}{4\pi n_e m_k}\left(\nabla\times\vec{b}\right)\times\vec{b } \nonumber \\ & + & \frac{z_k e}{m_k c } \frac{n_j z_j}{n_e } \left(\vec{v}_j-\vec{v}_k\right)\times\vec{b } = 0 , \label{gen_vec_vk } \\ \frac{\partial } { \partial t } \frac{p_s}{\gamma -1 } & + & \vec{v}_s\cdot\nabla\frac{p_s}{\gamma -1 } + \frac{\gamma}{\gamma-1}p_s(\nabla\cdot\vec{v}_s ) \nonumber \\ & + & \nabla\cdot\vec{q}_s -\frac{\delta e_s}{\delta t } -q_s = 0 , \label{gen_ps } \\ \frac{\partial \vec{b}}{\partial t } & -&\nabla\times\left(\vec{v}_e \times \vec{b}\right ) = 0 , \label{gen_vec_magind}\end{aligned}\ ] ] where the subscript @xmath17 refers to all species ( @xmath34 ) , while @xmath35 stands for ion species only ( @xmath36 ) . the gravitational constant is denoted by @xmath37 , @xmath38 is the mass of the sun , @xmath39 the magnetic field and @xmath40 the speed of light . the momentum and energy exchange rates due to the coulomb collisions of species @xmath17 with the remaining ones are denoted by @xmath41 and @xmath42 , respectively . moreover , @xmath43 is the heat flux carried by species @xmath17 , and @xmath44 stands for the heating rate applied to species @xmath17 from non - thermal processes . in equation ( [ gen_vec_vk ] ) , the subscript @xmath45 stands for ion species other than @xmath35 , namely , @xmath46 for @xmath47 and vice versa . as can be seen , in addition to the term @xmath48 , the lorentz force possesses a new term in the form of the cross product of the ion velocity difference and magnetic field . physically , this new term represents the mutual gyration of one ion species about the other , the axis of gyration being in the direction of the instantaneous magnetic field . equations ( [ gen_nk ] ) to ( [ gen_vec_magind ] ) form a complete set if supplemented with the description of species heat fluxes @xmath43 and heating rates @xmath44 . as such , they can be invoked to depict self - consistently the interaction between alfvn waves and the solar wind species by explicitly introducing these waves via boundary conditions . on the one hand solving this set of equations presents a computationally formidable task ; on the other hand , one can extract the necessary information concerning the dynamical feedback of the waves to the plasma by going beyond the wkb limit . in the non - wkb approach to be adopted here , one assumes that the governing equations can still be separated into those governing the background time - independent flow and those governing the transport of waves . as noted by @xcite ( also see the discussion ) , this separation does not necessarily require the waves be linear as long as sufficiently small wave amplitudes are imposed at the sun . further simplification also results from the choice of a flux tube coordinate system , in which the base vectors are @xmath49 , where @xmath50 with the subscript @xmath51 denoting the poloidal component . moreover , the independent variable @xmath52 is the arclength along the poloidal magnetic field line . this choice permits the decomposition of the magnetic field and species velocities into background ones and fluctuations , @xmath53 where @xmath54 . from the assumption of azimuthal symmetry , and the assumption that @xmath55 is time - independent , one can see from the poloidal component of equation ( [ gen_vec_magind ] ) that @xmath56 should be strictly in the direction of @xmath55 . in other words , @xmath57 to a good approximation . now let us consider the @xmath32 component of the momentum equation ( [ gen_vec_vk ] ) . since the wave frequencies in question as well as other frequencies associated with the spatial dependence are well below the ion gyro - frequency @xmath58 ( @xmath29 ) , from an order - of - magnitude estimate one can see that @xmath59 . combined with the fact that @xmath57 , this leads to that both @xmath60 and @xmath61 should be very small and can be safely neglected unless they appear alongside the ion gyro - frequency . with this in mind , one can find from the @xmath62 component of equation ( [ gen_vec_vk ] ) that @xmath63 that is , the ion velocity difference is aligned with the instantaneous magnetic field . this alignment condition further couples one ion species to the other . note that due to the assumption of quasi - zero current , equation ( [ ion_vdiff ] ) leads to @xmath64 where @xmath29 . given the aforementioned assumptions , the time - independent multicomponent solar wind in which the toroidal alfvn waves propagate is governed by @xmath65 where the prime denotes the derivative with respect to the arclength @xmath52 which becomes the only independent spatial variable . in addition , @xmath66 denotes the acceleration exerted on ion species @xmath35 ( @xmath29 ) by the toroidal fluctuations , @xmath67 where @xmath68 is the distance from a point along the poloidal magnetic field line to the magnetic axis . besides , the angular brackets stand for the time - average over one wave period . the variable @xmath69 distinguishes the present study from those using the ideal mhd in which case @xmath70 ( e.g. , * ? ? ? apart from this , the wave - induced acceleration @xmath66 includes the inertial centrifugal acceleration ( the first term on the right hand side ) , and the usual @xmath48 term ( the second term ) . the transport of toroidal alfvn waves is governed by the azimuthal component of the momentum equation ( [ gen_vec_vk ] ) for ion species @xmath35 ( @xmath29 ) together with that of the magnetic induction law ( [ gen_vec_magind ] ) . to be more specific , these equations read @xmath71 = 0 , \label{bphi } \\ \frac{\partial u_k}{\partial t } & + & u_k\left[\frac{\partial u_{k}}{\partial l } + u_{k } ( \ln r)'\right ] \nonumber\\ & -&\frac{z_k}{4\pi n_e m_k}b_l\left [ \frac{\partial b}{\partial l } + b ( \ln r)'\right ] = x_k . \label{vkphi1 } \end{aligned}\ ] ] the wave propagation is characterized by several key parameters , namely , the wave energy and energy flux densities as well as the wave - induced acceleration . these parameters can be found by considering the energy conservation for the alfvn waves , @xmath72 \right\ } \nonumber \\ & = & -\sum_k \rho_k u_k u_k^2(\ln r ) ' + \frac{u_e}{4\pi } \left[b\frac{\partial}{\partial l}b + b^2(\ln r ) ' \right ] \nonumber \\ & & + \sum_k \rho_k u_k x_k . \label{enercons}\end{aligned}\ ] ] on the left hand side ( lhs ) , the first term is the time derivative of the perturbation energy density , while the second is the divergence of the perturbation flux density . the physical meaning of the right hand side ( rhs ) can be revealed as follows . by using relation ( [ ukandue ] ) one finds @xmath73 since @xmath74 , one can identify the time - average of the rhs of equation ( [ enercons ] ) as the negative of the work done by the wave - induced forces on the solar wind , i.e. , @xmath75 ( cf . eq.([def_wave_acce ] ) ) . in other words , taking the time - average of equation ( [ enercons ] ) yields @xmath76 where @xmath77 is the time - average of the perturbation flux density and will be termed wave flux density for simplicity . ( actually , the wave properties to be discussed always refer to time - averaged values . ) the total energy is therefore conserved for the system comprised of a multi - fluid solar wind and toroidal alfvn waves : the gain in the solar wind kinetic energies is at the expense of the wave energy . the appearance of @xmath78 makes equation ( [ vkphi1 ] ) inconvenient to work with . instead one may consider the azimuthal component of the total momentum to eliminate @xmath78 , the resultant equation being @xmath79 \nonumber \\ & -&\frac{b_l}{4\pi r}\frac{\partial}{\partial l}\left(rb\right ) = 0 . \label{totvphi}\end{aligned}\ ] ] now one may proceed by introducing the fourier amplitudes at a given angular frequency @xmath10 , @xmath80 = [ \tilde{b}(l ) , \tilde{u}_s(l ) ] \exp(-i\omega t ) , \label{def_fourier}\end{aligned}\ ] ] in which @xmath54 . as a result , the faraday s law ( [ bphi ] ) and equation ( [ totvphi ] ) now take the form @xmath81\tilde{b } -b_l(\ln r ) ' \tilde{u}_e , \label{induc_w } \\ & & \left(\sum_k \rho_k u_k \frac{u_k - u_e}{b_l}-\frac{b_l}{4\pi}\right)\tilde{b } ' + \sum_k \rho_k u_k \tilde{u}_e ' \nonumber \\ & = & -\left[\sum_k \frac{\rho_k u_k}{r}\left(r\frac{u_k - u_e}{b_l}\right ) ' -\frac{b_l}{4\pi}\left(\ln r\right)'\right]\tilde{b } \nonumber\\ & - & \sum_k \rho_k u_k\left(\ln r\right)'\tilde{u}_e \nonumber\\ & + & i\omega\sum_k\rho_k\left(\frac{u_k - u_e}{b_l}\tilde{b}+\tilde{u}_e\right ) , \label{totmomen_w}\end{aligned}\ ] ] in which we have used relation ( [ ukandue ] ) to express the ion velocity fluctuation @xmath82 in terms of the electron one @xmath83 . once the background flow parameters and proper boundary conditions are given , equations ( [ induc_w ] ) and ( [ totmomen_w ] ) can be solved for the fourier amplitudes of magnetic fluctuation @xmath84 and the electron velocity fluctuation @xmath83 for a given angular frequency @xmath10 . the ion velocity fluctuations @xmath82 can then be found in virtue of relation ( [ ukandue ] ) . the three wave - related parameters can be evaluated by forming a time - average over one wave period . specifically , the wave energy density @xmath85 and wave energy flux density @xmath77 are given by [ def_efa_wave ] @xmath86 while the wave - induced acceleration @xmath66 has already been given by equation ( [ def_wave_acce ] ) . it can be seen that @xmath85 consists of the ion kinetic as well as the magnetic energies , while @xmath77 is comprised of the kinetic energy flux convected by ion fluids and the poynting flux . to evaluate the time - average @xmath87 for two wave - related fluctuations @xmath88 \sim [ \tilde{f},\tilde{g } ] \exp(-i\omega t)$ ] , one may use the expression ( cf . * ? ? ? * ) @xmath89 where @xmath90 denotes the real part of a complex variable @xmath91 , and the superscript asterisk denotes the complex conjugate . although in general one has to rely on a numerical integrator to solve equations ( [ induc_w ] ) and ( [ totmomen_w ] ) , as shown below , they are analytically tractable for two limiting cases . in this section , we will examine the analytical solutions to equations ( [ induc_w ] ) and ( [ totmomen_w ] ) in the wkb and zero - frequency ( @xmath92 ) limits . these analytical treatments not only help to validate numerical solutions , but also allow one to gain insights into the mathematical properties of the governing equations . extensive studies have been made on the alfvn waves in the wkb limit . in particular , the alfvn wave force exerted on differentially streaming ions was first derived by @xcite and later by @xcite for a spherically expanding solar wind . based on the idea of the wave average lagrangian , the work by @xcite not only obtained the expression for the wave force but further showed that there exists an adiabatic invariant , namely the wave action flux , in the absence of wave dissipation . the derivation is rather general and does not require a specific flux tube geometry . alternatively , the action flux conservation has been independently obtained by @xcite who used the more familiar wkb analysis . although @xcite assumed that the solar wind is again spherically symmetric , his results can be readily extended to an axisymmetric configuration . in what follows , such an extension is presented . the formal development of the wkb analysis starts with the introduction of the expansion [ def_wkb_expan ] @xmath93 in which @xmath94 . it is assumed that @xmath95 and @xmath96 ( @xmath97 ) vary at the same spatial scale @xmath98 as the background flow parameters , and @xmath99 is small compared with the wave number @xmath100 , i.e. , @xmath101 . in addition , it is assumed that @xmath102 . substituting the expansion ( [ def_wkb_expan ] ) into equations ( [ induc_w ] ) and ( [ totmomen_w ] ) , one can find by sorting different terms according to orders of @xmath103 that [ wkb_eqs ] @xmath104 ' , \label{wkb_ind } \\ & & \sum_k \rho_k \omega_k \left(u_{e , n } + \frac{u_k - u_e}{b_l}b_n\right ) + k\frac{b_l}{4\pi } b_n \nonumber\\ & = & -i\sum_k \frac{\rho_k u_k}{r}\left [ r\left(u_{e , n-1}+\frac{u_k - u_e}{b_l } b_{n-1}\right)\right ] ' \nonumber \\ & + & i \frac{b_l}{4\pi r}\left(r b_{n-1}\right ) ' , \label{wkb_mom}\end{aligned}\ ] ] where the definition @xmath105 ( @xmath54 ) has been used . at the lowest order @xmath106 , the rhs of equation ( [ wkb_eqs ] ) is zero . for @xmath107 and @xmath108 not to be identically zero , the determinant of the coefficient matrix has to be zero . as a result , one finds the local dispersion relation @xmath109 where @xmath110 is the local phase speed of alfvn waves . furthermore , one can find the local eigen - relations at order @xmath106 , @xmath111 with @xmath94 . at an arbitrary order @xmath112 , it follows from equation ( [ wkb_ind ] ) that @xmath113 in which @xmath114'\end{aligned}\ ] ] arises from quantities at order @xmath115 . at order @xmath116 , one can eliminate @xmath117 and @xmath118 on the lhs of equation ( [ wkb_eqs ] ) by using the dispersion relation ( [ wkb_disprel ] ) , the resultant equations taking the form @xmath119 where [ def_sntn ] @xmath120 , \label{def_sn } \\ t_n & = & \sum_k \rho_k \omega_k r b_l \left(\frac{u_{ph } b_n}{r b_l}\right ) ' \nonumber \\ & + & k b_l \sum_k \frac{\rho_k u_k}{r } \left(r\frac{u_{ph}-u_k}{b_l } b_n\right ) ' \nonumber \\ & + & k \frac{b_l^2}{4\pi r}\left(r b_n\right ) ' . \label{def_tn}\end{aligned}\ ] ] the barred and unbarred parts are again distinguished . on expanding @xmath121 , by repeatedly using the dispersion relation ( [ wkb_disprel ] ) one finds that @xmath122 ' , \label{wkb_tn}\end{aligned}\ ] ] in which the following definitions have been used ( cf . * ? ? ? * ) [ def_rhovmvj2 ] @xmath123 with the last one defined for future use . if putting @xmath106 , from @xmath124 one finds @xmath125 , i.e. , @xmath126 note that equation ( [ wkb_action_txt ] ) can be interpreted in terms of the conservation of wave action flux , which was first found by @xcite . on the other hand , expanding equation ( [ vkphi1 ] ) to first order in @xmath103 and using the resultant @xmath78 to evaluate @xmath127 in equation ( [ def_wave_acce ] ) , with the aid of the eigen - relation ( [ wkb_eigen1 ] ) , one can find a compact expression for the wave - induced acceleration on ion species @xmath35 ( @xmath29 ) , @xmath128 where the time - average @xmath129 . it is noteworthy that the shape of the line of force , which enters into the discussion through @xmath68 , does not show up via @xmath130 unless one examines the evolution of higher order fluctuations @xmath95 and @xmath96 ( @xmath112 ) . some remarks on the wave acceleration @xmath66 given by equation ( [ wkb_acce ] ) . one may find that @xmath131 if the flow speeds are entirely determined by the wave forces . usually @xmath132 increases with increasing distance . the net effect of @xmath66 is therefore to limit the proton - alpha differential speed @xmath133 . in this sense @xmath66 was likened to an additional frictional force ( e.g. , * ? ? ? as noted by @xcite , the differential nature of @xmath66 derives eventually from a wave - induced transverse drift velocity @xmath134 ( @xmath29 ) , which is different for different species because they see different wave electric fields when they flow differentially . the contribution to the wave force from @xmath134 is twofold . first , it induces a centrifugal force that is always non - negative . second , it contributes to the species electric current flow @xmath135 which in turn exerts on ion species @xmath35 a @xmath136 force , where @xmath137 is the wave magnetic field . this force may become negative . as a matter of fact , the two terms may even add up to a negative value : the wave force may decelerate rather than accelerate the ion species . consider a simplified situation similar to that considered by @xcite . one assumes that the spatial variation of both @xmath138 and @xmath139 can be neglected , and @xmath140 where @xmath141 is the bulk alfvn speed . one may further assume that the alpha particles are test particles , the background magnetic field is purely radial , and that @xmath142 with @xmath143 being a positive constant . in this case , from equation ( [ wkb_acce ] ) one finds that @xmath144 -\left(4-\epsilon\right)\left(u_{\alpha p}+u_p\right)^2 \right\ } .\end{aligned}\ ] ] it follows that @xmath145 when @xmath146 where the critical value @xmath147 can be roughly approximated by @xmath148 which may be smaller than @xmath149 by a factor of @xmath150 . noting that now the wave phase speed is @xmath151 , one finds that the waves may decelerate alpha particles even though @xmath152 . it turns out that this much simplified picture roughly represents the behavior of @xmath153 in the region between 20 and 30 @xmath5 where @xmath143 can be taken to be @xmath154 for the adopted background flow parameters ( see figure [ fig_aw]b which indicates that @xmath153 in the wkb case becomes negative beyond 22.7 @xmath5 where @xmath133 exceeds @xmath147 ) . the other extreme will be @xmath92 . in this case , equations ( [ bphi ] ) and ( [ totvphi ] ) can be integrated to yield [ zero_intg ] @xmath155 = a_l , \label{zero_totang}\end{aligned}\ ] ] where @xmath156 and @xmath157 are two integration constants . when deriving equation ( [ zero_totang ] ) , we have used the fact that @xmath158 is a constant . equation ( [ zero_intg ] ) demonstrates a clear connection to the problem of angular momentum transport in a multicomponent solar wind , as has been discussed by @xcite . with the aid of equation ( [ ukandue ] ) , equation ( [ zero_intg ] ) can be solved to yield @xmath159 where @xmath160 is the square of the combined alfvn mach number . for typical solar winds , there exists one point where @xmath0 between 1 @xmath5 and 1 au . call this point the alfvnic point , and let it be denoted by subscript @xmath161 . it then follows that for @xmath162 not to be singular at this point , one must require @xmath163 . as a result , one may obtain [ zero_exprs ] @xmath164 \label{zero_us } , \end{aligned}\ ] ] where @xmath34 . apart from the wkb and zero - frequency limits , equations ( [ induc_w ] ) and ( [ totmomen_w ] ) can be integrated only numerically . nevertheless , the treatment in the zero - frequency limit in the previous section reveals that the alfvnic point where @xmath165 is a singular point for the system of equations . one may expect that this is also the case for an arbitrary finite @xmath10 . in their present form , however , equations ( [ induc_w ] ) and ( [ totmomen_w ] ) do not show explicitly the existence of such a singular point . hence they are not convenient to work with numerically and need to be further developed in what follows . moreover , we will also describe how to specify the background flow parameters , and the solution procedure as well . for the convenience of presentation , one may define , in addition to @xmath166 , the following dimensionless parameters @xmath167 in which @xmath168 is the bulk alfvn speed . now equation ( [ totmomen_w ] ) can be expressed as @xmath169 taking into account equation ( [ induc_w ] ) , one finally arrives at [ num_eqs ] @xmath170 where two dimensionless variables have been introduced for convenience , @xmath171 moreover , the coefficients are [ def_num_coefes ] @xmath172 \\ f_{22 } & = & \frac{i \omega}{u_e}x_{ee}+\left(z - x_{me}\right)\left(\ln r\right ) ' \nonumber\\ & - & ( m_t^2 - 1)(\ln u_a)'.\end{aligned}\ ] ] it is now clear that the alfvnic point where @xmath165 is a singular point for equation ( [ num_eqs ] ) . the ensuing task is therefore to find solutions that pass smoothly through the alfvnic point . in principle , one needs to solve equations ( [ bkg_nk ] ) to ( [ bkg_ps ] ) to find a realistic background solar wind by neglecting the wave - induced acceleration in equation ( [ bkg_vk ] ) and using a suitable heat input @xmath44 ( @xmath54 ) . however , it is observationally established that in the heliocentric range @xmath173 au , the proton - alpha differential speed @xmath174 closely tracks the local alfvn speed in the fast solar wind with @xmath175 @xmath176@xcite . so far this fact still poses a theoretical challenge : adjusting the heating parameters proves difficult to yield such a behavior for @xmath133 . in what follows , we shall adopt a two - step approach to find the background flow parameters . first , equations ( [ bkg_nk ] ) to ( [ bkg_ps ] ) are solved by using a suitable set of heating parameters along a prescribed meridional magnetic field line . as a result , we have @xmath177 , @xmath139 , @xmath178 and @xmath179 for the whole heliocentric range from 1 @xmath5 out to 1 au . note that the subscript @xmath180 denotes the results obtained from this step . second , the alpha parameters are altered as follows . we first impose an _ ad hoc _ profile for @xmath133 which is identical to @xmath179 in the range @xmath181 au but undergoes a smooth transition to a profile that is roughly 0.8 @xmath149 everywhere for @xmath182 au . the alpha speed and mass density are then given by @xmath183 and @xmath184 , respectively . for the meridional magnetic field , we will adopt an analytical model given by @xcite . in the present implementation , the model magnetic field consists of dipole and current - sheet components only . a set of parameters @xmath185 , @xmath186 , @xmath187 and @xmath188 are chosen such that the last open magnetic field line is anchored at heliocentric colatitude @xmath189 on the sun . the background magnetic field configuration and flow parameters are depicted in figure [ fig_bkgrnd ] . in fig.[fig_bkgrnd]a , the field line along which we integrate the solar wind equations ( [ bkg_nk ] ) to ( [ bkg_ps ] ) and the wave equation ( [ num_eqs ] ) is delineated by the thick contour . this line of force is rooted at colatitude @xmath190 on the sun where the poloidal magnetic field strength @xmath191 is @xmath192 g. it reaches @xmath193 at 1 au where @xmath191 is 3.3@xmath194 , compatible with the ulysses measurements @xcite . in fig.[fig_bkgrnd]b , the solid lines give the distribution with heliocentric distance @xmath195 of the proton ( @xmath139 ) and alpha ( @xmath138 ) speeds , while the dashed curve is for the phase speed ( @xmath196 ) of alfvn waves in the wkb limit as determined by equation ( [ wkb_disprel ] ) . furthermore , fig.[fig_bkgrnd]c shows the distribution with @xmath195 of the proton - alpha differential speed @xmath133 ( the solid line ) and alfvn speed @xmath149 ( the dashed curve ) . the asterisks in figs.[fig_bkgrnd]b and [ fig_bkgrnd]c denote the alfvnic point where @xmath165 , which is located at @xmath197 @xmath5 in the chosen background model . it can be seen in fig.[fig_bkgrnd]b that @xmath196 is much larger than the ion flow speeds below @xmath198 @xmath5 but is close to @xmath138 beyond @xmath199 au . this latter behavior is understandable since one can find from equation ( [ wkb_disprel ] ) that @xmath200 where @xmath201 ( @xmath29 ) . the second expression is very accurate for @xmath202 which is the case for the assumed background flow ( cf . fig.[fig_bkgrnd]c ) . hence @xmath203 in the region @xmath182 au where @xmath204 and @xmath205 . moreover , it can be seen from fig.[fig_bkgrnd]c that @xmath133 drops from nearly zero at @xmath5 to @xmath206 @xmath176 at 1.42 @xmath5 , beyond which @xmath133 increases gradually . in interplanetary space @xmath182 au , @xmath133 can be seen to follow @xmath149 closely , as is required . at the alfvnic point @xmath207 , the proton speed @xmath139 is 602 @xmath176 , the alpha one @xmath138 is 721 @xmath176 . eventually @xmath139 reaches 648 @xmath176 at 1 au where @xmath138 is 676 @xmath176 . as for the density parameters @xmath208 and @xmath209 , the model yields @xmath210 @xmath211 and @xmath212 at 1 au . as a result , the proton ( alpha ) flux @xmath213 ( @xmath214 ) is 2.32 ( 0.11)@xmath215 @xmath216 s@xmath4 when scaled to 1 au . all these values are consistent with typical measurements for the low - latitude fast solar wind streams , e.g. , those made by ulysses @xcite . given the background flow parameters , the coefficients in equation ( [ def_num_coefes ] ) can be readily evaluated for a given angular frequency @xmath10 . equation ( [ num_eqs ] ) is ready to solve once proper boundary conditions are supplemented . as is well known , one boundary condition has to be imposed at the alfvnic point to ensure the solution is regular . to establish this , let us consider the region adjacent to the alfvnic point @xmath217 , such that any function @xmath218 can be taylor - expanded , @xmath219 in which @xmath220 . when substituting this expansion into equation ( [ num_eqs ] ) , one finds [ num_alfvpnt ] @xmath221 it is easy to recognize that equations ( [ num_alf_zero1 ] ) and ( [ num_alf_zero2 ] ) are not independent from each other , since at the alfvnic point @xmath217 , @xmath222 . hence at @xmath217 , the unknowns @xmath223 , @xmath224 and @xmath225 can all be expressed in terms of @xmath226 , which can be arbitrarily chosen . once these four quantities are known , the wave quantities @xmath227 and @xmath228 can be obtained by the expansion ( [ def_alf_expan ] ) at the two grid points immediately astride the alfvnic point . equation ( [ num_eqs ] ) is then integrated by using a fourth order runge - kutta method both inwards to the coronal base @xmath5 and outwards to 1 au ( see * ? ? ? we then rescale the obtained solution such that @xmath229 is fixed at @xmath5 for all @xmath10 to be considered . in this section , the solutions to equation ( [ num_eqs ] ) with a number of different angular frequencies @xmath10 are presented . all these solutions have the same amplitude for the electron velocity fluctuation @xmath230 @xmath176 , or equivalently the time average @xmath231 @xmath176 . note that such a value is only about @xmath232 of the upper limit of the nonthermal velocity amplitude derived from line width measurements @xcite . with this choice at the coronal base , we can avoid awkwardly large wave amplitudes in interplanetary space , a natural consequence of the assumption that no wave dissipation is present . nevertheless , the base amplitude can be seen as a free scaling parameter for equation ( [ num_eqs ] ) . the relative deviation of the non - wkb from wkb results is independent from this choice , even though the absolute magnitudes are affected . figure [ fig_reim ] presents the radial profiles of the real ( solid lines ) and imaginary ( dashed lines ) parts of the fourier amplitudes with three different angular frequencies , @xmath1 ( left column ) , @xmath2 ( middle ) and @xmath3 @xmath6(right ) . since the modulus of the magnetic fluctuation @xmath233 spans several orders of magnitude , @xmath234 is plotted instead of @xmath84 in panels ( a ) , ( d ) and ( g ) . note that panel ( g ) uses a scale different from ( a ) and ( d ) . panels ( b ) , ( e ) and ( h ) depict the proton velocity fluctuation @xmath235 , while panels ( c ) , ( f ) and ( i ) give the alpha one @xmath236 . the most prominent feature of figs.[fig_reim]a , [ fig_reim]b and [ fig_reim]c for @xmath1 @xmath6 is that the radial dependence of the real and imaginary parts of fluctuations exhibits a clear oscillatory behavior . this is particularly true beyond @xmath237 @xmath5 . further inspection of the region @xmath238 @xmath5 indicates that the magnetic and fluid velocity fluctuations are well correlated , as would be expected from the eigen - relation ( [ wkb_eigen1 ] ) obtained in the wkb limit . the wkb nature is further revealed by examining the phase relation , i.e. , the displacement between the real and imaginary parts , for any of the three fourier amplitudes . for instance , for the @xmath235 profile the nodes in the real part correspond well to the troughs or crests in the imaginary part . now if examining the envelopes of the fourier amplitude profiles , one can see that @xmath239 ( fig.[fig_reim]a ) tends to increase with @xmath195 while @xmath240 tends to decrease beyond say 10 @xmath5 . on the other hand , @xmath241 exhibits a non - monotonic behavior and possesses a local minimum at @xmath242 @xmath5 . moreover , the magnitude of @xmath236 is considerably smaller than that of @xmath235 beyond 0.3 au . this is understandable in light of equation ( [ wkb_eigen1 ] ) since the local phase speed of alfvn waves @xmath196 is close to the alpha flow speed @xmath138 ( see fig.[fig_bkgrnd]b ) . the fourier amplitudes for @xmath243 @xmath6 ( the middle column ) also possess an oscillatory behavior , however the exact correlation between the magnetic ( fig.[fig_reim]d ) and ion velocity ( figs.[fig_reim]e and [ fig_reim]f ) fluctuations is gone , as is the exact phase relation between the real and imaginary parts . furthermore , one can see that the magnitude of the alpha velocity fluctuations ( fig.[fig_reim]f ) in interplanetary space is substantially larger than that for @xmath1 @xmath6 . now let us move on to the right column for which @xmath8 @xmath6 . distinct from the preceding two columns , the oscillatory feature disappears altogether . instead , the real and imaginary parts of the fourier amplitudes evolve slowly with radial distance @xmath195 . we note that such a transition from wave - like to quasi - steady dependence on @xmath195 with decreasing @xmath10 was explored in detail by @xcite ( see also * ? ? ? * ; * ? ? ? * ) . that there exists a critical frequency below which the transition occurs was interpreted in terms of the coupling between inwardly and outwardly propagating waves . although in the present study , the propagation in opposite directions has not been explicitly separated , one can see that for a realistic 3-fluid solar wind , there also exists a similar critical frequency which may be approximated by @xmath244 ( cf . eq.(55 ) of * ? ? ? * ) . for the chosen background flow , we find @xmath245 @xmath6 . the numerical experiments have confirmed the validity of this approximation . figure [ fig_ampltd ] shows the radial evolution of the time - averages of the magnetic @xmath246 as well as the fluid velocity fluctuations @xmath247 ( @xmath29 ) for the three angular frequencies @xmath1 ( solid lines ) , @xmath2 ( dashed lines ) and @xmath3 @xmath6 ( dash - dotted lines ) , and for the wkb case ( dotted lines ) as well . the wkb results are evaluated from equations ( [ wkb_eigen1 ] ) and ( [ wkb_action_txt ] ) by using the same value for @xmath248 at @xmath5 as in the numerical solutions . it is obvious that for all the frequencies considered , the profiles demonstrate substantial deviations from the wkb one , not only in the absolute values but also in spatial dependence . a wkb - like spatial dependence is recovered only in profiles with @xmath1 @xmath6 and @xmath7 @xmath5 or those with @xmath243 @xmath6 in regions @xmath249 @xmath5 . it is interesting to see that , the @xmath246 profiles are strikingly similar in the region @xmath250 @xmath5 for all three frequencies . this behavior can be understood as follows . first of all , it can be readily shown from equation ( [ num_eqs ] ) that @xmath251 for the solutions with all three frequencies , it turns out that the first term on the rhs always dominates the second one in the inner corona . in addition , in the first several solar radii , the flow is very sub - alfvnic , i.e. , @xmath252 , and @xmath253 . it follows that @xmath254 is dominated by @xmath255 . therefore @xmath256 is approximately proportional to @xmath257 . from the definition of @xmath227 ( eq.([def_xi_eta ] ) ) , one can see that @xmath258 . that is , @xmath246 shows little frequency dependence . on the other hand , from equation ( [ wkb_action_txt ] ) it follows that in the wkb limit @xmath259 for @xmath260 since @xmath261 . hence the difference between the profiles for the three frequencies and the wkb one in the region @xmath262 @xmath5 reflects the fact that the mass density has a scale height less than @xmath263 . now consider the portion @xmath264 @xmath5 where @xmath265 . one may see that the slope of the @xmath246 profile for @xmath243 @xmath6 becomes similar to that for @xmath266 @xmath6 ( or equivalently the wkb one ) asymptotically . on the other hand , the profile with @xmath8 @xmath6 has a flatter slope than the wkb one . this is understandable by noting that the @xmath246 profile for @xmath8 @xmath6 can be rather accurately represented by the zero - frequency solution ( [ zero_b ] ) . by noting that @xmath158 is a constant and @xmath267 for @xmath264 @xmath5 , one can see that @xmath233 and therefore @xmath268 . on the other hand , for the wkb profile one can see from equation ( [ wkb_action_txt ] ) that asymptotically @xmath269 since @xmath270 . moving on to figs.[fig_ampltd]b and [ fig_ampltd]c , one may notice that the magnitude of the ion velocity fluctuations @xmath247 ( @xmath29 ) have nearly the same values at @xmath5 as assumed for @xmath248 , which can be expected from equation ( [ ukandue ] ) together with the fact that @xmath271 for @xmath272 . let us examine fig.[fig_ampltd]b in some detail . one can see that between 1 @xmath5 and 1 au , the wkb estimate always exceeds the computed @xmath273 value . furthermore , for @xmath274 @xmath5 , the @xmath273 profile for @xmath1 @xmath6 differs little from that for @xmath243 @xmath6 , however they are significantly larger than that for @xmath8 @xmath6 . take values at 5 @xmath5 for instance . one finds that @xmath273 is 83.9 @xmath176 for the wkb estimate , but 44.6 and 41.9 @xmath176 for @xmath1 and @xmath2 @xmath6 , respectively . as to the case @xmath8 @xmath6 , @xmath273 is 17.2 @xmath176 . in other words , the wkb estimate yields a value that is 1.88 , 2 and 4.88 times the computed values for @xmath1 , @xmath2 , and @xmath3 @xmath6 , respectively . at 1 au , the difference between the wkb estimate and computed results is also substantial . one finds for @xmath1 , @xmath2 , and @xmath3 @xmath6 , the wkb estimate is 1.83 , 1.18 and 1.32 times the values obtained numerically . an interesting feature of the @xmath273 profile for @xmath182 au is that it approaches a constant asymptotically . this can be explained by noting that the magnitudes of ion velocity fluctuations for @xmath8 @xmath6 can also be roughly represented by the zero - frequency solution ( [ zero_us ] ) . ( the zero frequency solutions ( [ zero_exprs ] ) can be accurately reproduced if @xmath10 is further reduced . ) at distances @xmath275 where @xmath276 , equation ( [ zero_us ] ) can be written as @xmath277,\end{aligned}\ ] ] where the constant @xmath278 is the ion mass flux ratio , and is 0.19 for the supposed background solution . in the region considered , the first term is found to dominate the second in the square parentheses . as a result , @xmath279 . since @xmath280 , one finds that @xmath240 and therefore @xmath273 approach a constant asymptotically . let us proceed to fig.[fig_ampltd]c , from which one can see that the @xmath281 profiles also deviate considerably from the wkb estimate . in particular , below @xmath11 @xmath5 , the @xmath281 profiles behave in a fashion similar to @xmath273 , however the difference between the case for @xmath8 @xmath6 and the wkb estimate is even larger . for instance , at 5 @xmath5 , the wkb estimate is about 10 times the value with @xmath8 @xmath6 . on the other hand , for @xmath1 ( @xmath2 ) @xmath6 , one finds the ratio of the wkb value to that numerically derived is 1.89 ( 2.1 ) . at larger distances , one may notice the local minimum in the profile with @xmath1 @xmath6 at @xmath282 au , consistent with fig.[fig_reim]c . furthermore , one can see that the values for @xmath8 @xmath6 become an order of magnitude larger than the wkb one . at 1 au , for @xmath1 , @xmath2 , and @xmath3 @xmath6 , the values obtained numerically are 0.55 , 1.44 and 11.9 times the wkb one , respectively . in addition , similar to @xmath273 , @xmath281 also approaches a constant value asymptotically . this is understandable since in the region @xmath283 where @xmath265 , one finds from equation ( [ zero_us ] ) that @xmath284,\end{aligned}\ ] ] in which the first term in the square parentheses , as in the case for @xmath273 , is found to dominate the second . therefore @xmath285 asymptotically . as a result , @xmath281 should show little spatial dependence . figure [ fig_ewfw ] presents the radial distribution of ( a ) : the wave energy density @xmath85 and ( b ) : the energy flux density @xmath77 for the three angular frequencies and wkb expectations . in both figs.[fig_ewfw]a and [ fig_ewfw]b , the profiles for @xmath1 @xmath6 approach a wkb - like behavior when @xmath7 @xmath5 . when @xmath243 @xmath6 , a wkb - like spatial dependence of @xmath85 and @xmath77 can be seen for @xmath286 @xmath5 . now let us consider fig.[fig_ewfw]a first . at 1 @xmath5 , for @xmath8 @xmath6 , @xmath85 is larger than the wkb result , but for @xmath1 or @xmath2 @xmath6 , it is smaller . this behavior is determined by the magnetic component in @xmath85 ( eq.([def_efa_wave ] ) ) since the particle part is nearly frequency independent . note that in the wkb limit , the wave energy is equally distributed between the kinetic and magnetic energies , @xmath287 . hence at the coronal base , for @xmath1 and @xmath2 @xmath6 , the kinetic energy is larger than the magnetic one , whereas for @xmath8 @xmath6 , the tendency is reversed . it is interesting to note that among the profiles for the three frequencies the highest @xmath10 corresponds to a profile that deviates the most from the wkb profile in the region @xmath288 @xmath5 . in the region @xmath289 @xmath5 , one can see that for @xmath243 @xmath6 the @xmath85 profile becomes wkb - like asymptotically . however , for @xmath8 @xmath6 the slope of the @xmath85 profile is flatter than the wkb one . this can be understood since for @xmath283 , the wkb limit yields @xmath290 . on the other hand , for @xmath283 , the zero - frequency result yields @xmath85 that is largely determined by the magnetic energy , which leads to @xmath291 . inspection of fig.[fig_ewfw]b reveals that the the spatial dependence of the profiles of the energy flux density @xmath77 is remarkably similar in the region @xmath288 @xmath5 . in contrast , in the same region , fig.[fig_ewfw]a shows that the @xmath85 profiles with different @xmath10 have rather different @xmath195 dependence . this behavior of @xmath77 stems from the fact that the loss of the wave energy flux in the form of the work done on the plasma is negligible . in other words , @xmath77 is diluted only by the flux tube expansion below 2 @xmath5 ( cf . eq.([enercons_aver ] ) ) . for instance , for @xmath243 @xmath6 at 2 @xmath5 , @xmath292 is @xmath293 erg @xmath216 s@xmath294 , amounting to only @xmath295 of @xmath296 . as a result , @xmath297 , or @xmath298 . that is , in this region the spatial profile of @xmath77 should show little frequency dependence . this is found to be true for all the frequencies considered , and for the wkb result as well . on the other hand , in the region @xmath283 , one can see that for @xmath243 @xmath6 the @xmath77 profile becomes wkb - like asymptotically , whereas the profile for @xmath8 @xmath6 decreases more slowly than the wkb one . it turns out that asymptotically for the low frequency waves @xmath77 is dominated by the poynting vector which evolves like @xmath299 . in the wkb limit , however , the ratio of the contributions of particles to the poynting vector is roughly @xmath300 . as a result , @xmath301 . from the perspective of solar wind modeling , one may be more curious about the feedback from the waves to the background flow . to this end , figure [ fig_aw ] presents the radial distribution of the acceleration exerted on ( a ) the protons @xmath302 and ( b ) alpha particles @xmath153 . one can see that for @xmath1 and @xmath2 @xmath6 , the wave acceleration is less effective than that in the wkb limit throughout the computational domain . however , in the low frequency case @xmath8 @xmath6 , @xmath302 exceeds the wkb expectation considerably beyond 22.8 @xmath5 . nevertheless , in all cases the wave force tends to accelerate the protons , i.e. , @xmath303 everywhere between the coronal base and 1 au . however this is not the case when the alpha particles are concerned . it can be seen from fig.[fig_aw]b that the wave acceleration @xmath153 is negative in the interval between 22.7 and 66.1 @xmath5 in the wkb limit as well in the case @xmath1 @xmath6 . this coincidence of the positions where @xmath153 changes sign stems from the fact that beyond @xmath237 @xmath5 , the wave is wkb - like . on the other hand , the @xmath304 profile for @xmath243 @xmath6 changes sign at 21.8 and 61.8 @xmath5 . these locations are slightly different from their counterparts in the wkb limit . when it comes to @xmath8 @xmath6 , the @xmath153 profile does not show any resemblance to the wkb expectation . in particular , the waves start to decelerate the alpha particles even in the inner corona : @xmath304 is negative everywhere beyond @xmath305 @xmath5 . an interesting aspect of the wave - induced acceleration is that asymptotically both @xmath302 and @xmath153 approach constant values . this is understandable since in the zero - frequency limit , in the region where @xmath275 , both @xmath306 and @xmath307 show little radial dependence . it then follows from the expressions ( [ zero_up_mtll1 ] ) and ( [ zero_ui_mtll1 ] ) that for ion species @xmath35 ( @xmath29 ) , @xmath308 . since in the region considered , the line of force is nearly perfectly radial , one can see that @xmath309 is a constant . as a result , @xmath66 should approach a constant asymptotically . the extent to which the wave forces may alter the ion flows can be obtained only through a self - consistent modeling by using , e.g. , the iterative approach adopted by @xcite : the wave equation ( [ num_eqs ] ) and the solar wind equations ( [ bkg_nk ] ) to ( [ bkg_ps ] ) incorporating the wave contribution are solved alternately until a convergence is met . as a first step , however , we may simply evaluate the ion speeds @xmath310 ( @xmath29 ) corrected for the wave force , i.e. , @xmath311 figure [ fig_ucorr ] presents the radial distribution of both @xmath312 and @xmath313 for all the frequencies considered . the background flow speed profiles are also plotted for comparison . one may see that , with the present choice of the wave amplitude , the waves have negligible effects on the ion acceleration below the alfvn point . beyond the alfvn point , the effects introduced by the waves on the speed profiles become more important , especially in the low frequency case . as a matter of fact , for @xmath8 @xmath6 the corrected proton speed @xmath312 reaches 785 @xmath176 at 1 au where the background value is 648 @xmath176 . as for the alpha speed , @xmath313 becomes negative beyond 100 @xmath5 due to the significant deceleration exerted on the alphas by the waves . of course , such a situation will not appear in reality . what will happen is that the protons are accelerated whereas alphas are decelerated by the low frequency waves until the ions move at nearly identical speeds . the net effect of low - frequency waves is thus to limit the speed difference between the protons and alpha particles at large distances . however , it should be pointed out that in these regions the net work done by the low - frequency wave on the solar wind as a whole is negligible : @xmath314 and @xmath77 is nearly divergence free as discussed in reference to fig.[fig_ewfw ] . this study has been motivated by the apparent lack of a non - wkb analysis of alfvn waves in a multi - fluid solar wind with differentially flowing ions . to be more specific , this study is concerned with the propagation of dissipationless , hydromagnetic ( angular frequency @xmath10 well below ion gyro - frequencies ) , purely toroidal alfvn waves that propagate in a background 3-fluid solar wind comprised of electrons , protons and alpha particles . azimuthal symmetry is assumed throughout . no assumption has been made that the wavelength is small compared with the spatial scales at which the background flow parameters vary . the wave behavior at a given @xmath10 is governed by equation ( [ num_eqs ] ) , which is derived from the general transport equations in the five - moment approximation . the alfvnic point , where the combined alfvn mach number @xmath0 ( cf . eq.([def_mt2 ] ) ) , is a singular point of equation ( [ num_eqs ] ) and a regularity condition has to be imposed . for the other boundary condition , we impose a velocity amplitude of 10 @xmath176 at the coronal base ( 1 @xmath5 ) . for the given background model of a realistic low - latitude fast solar wind , equation ( [ num_eqs ] ) is integrated numerically for three representative angular frequencies @xmath1 , @xmath2 and @xmath3 @xmath6 to yield the radial distribution of the wave energy and energy flux densities as well as the wave - induced acceleration exerted on ion species . the first conclusion concerns the applicability of the wkb approximation . between 1 @xmath5 and 1 au , the numerical solutions show substantial deviation from the wkb expectations . even for the relatively high frequency @xmath1 @xmath6 , a wkb - like behavior can be seen only in regions where @xmath7 @xmath5 . in the low - frequency case @xmath8 @xmath6 , the computed profiles of wave - related parameters show a spatial dependence that is distinct from the wkb one , the deviation being particularly pronounced in interplanetary space . in the inner corona @xmath9 @xmath5 , the computed ion velocity fluctuations are considerably smaller than the wkb expectations in all cases , as is the computed wave - induced acceleration exerted on protons or alpha particles . as for the wave energy and energy flux densities , they can be enhanced or depleted compared with the wkb results , depending on @xmath10 . the second conclusion is concerned with how the wave acceleration may alter the background flow parameters . in reference to fig.[fig_ucorr ] , it is found that with the current choice of base wave amplitude , the wave acceleration has little effect on the force balance for protons or alpha particles in the corona . that is , one has to invoke processes other than the non - wkb wave acceleration to accelerate the ions out of the gravitational potential well of the sun . however , at large distances beyond the alfvnic point , low - frequency waves may play an important role in the ion dynamics , with the net effect being to equalize the speeds of the two ion species considered . strictly speaking , the separation of the flow into fluctuations and a time - independent background implies that the waves are linear . however , one may have noticed that the wave amplitude @xmath246 at 1 au for @xmath8 s@xmath4 @xmath6 is substantially larger than the background poloidal magnetic field strength ( cf . fig.[fig_ampltd]a ) . that the transverse magnetic field dominates the poloidal one demands a careful examination of the nonlinear effects other than the wave - induced acceleration . in particular , one needs to look at the generation of secondary waves and structures by the primary alfvn waves through the source terms in the momentum equation . as discussed by @xcite , in the case of ideal mhd these source terms decrease sufficiently fast with radial distance asymptotically . consequently , the first - order wave amplitudes are valid provided that the amplitude imposed at the coronal base is sufficiently small . the basic picture is expected to be the same even if a second ion species is included , although a similar discussion in the 3-fluid framework will be complicated by the richness of wave modes due to the differential proton - alpha streaming ( e.g. , * ? ? ? as has been mentioned in the introduction , alfvn waves are dissipated in some way , and this dissipation of the primary waves should be described self - consistently . a possibility to do this is to perform a full elssser analysis extended to the multi - fluid case , and to express the dissipation in terms of the amplitudes of counter - propagating waves . we note that this already complicated issue will become even trickier considering the necessity to apportion the dissipated wave energy among different species . in closing , we note that the low - frequency waves may also be important for outflows from stars other than the sun . for instance , in the radiatively driven stellar winds , these waves will provide a further channel of momentum exchange between passive ions and line - absorbing ions in addition to the coulomb friction . this possibility was first pointed out by @xcite in connection with the effects of stellar rotation . due to the clear resemblance between the low - frequency alfvn waves and stellar rotation ( cf . section [ sec_zero_freq ] ) , their discussion also applies to the case where the star persistently emits alfvn waves with frequencies lower than the critical one defined by equation ( [ def_omegac ] ) . consequently , the mass loss rate may be significantly altered . a quantitative study of this effect is beyond the scope of the present paper though . this research is supported by a pparc rolling grant to the university of wales aberystwyth . we thank shadia rifai habbal for her careful reading of the first draft , and hui - nan zheng for helpful discussions .

we calculate the reionization history for different models of the stellar population and explore the effects of primordial magnetic fields , dark matter decay and dark matter annihilation on reionization . we find that stellar populations based on a scalo - type initial mass function for population ii stars can be ruled out as sole sources for reionization , unless star formation efficiencies of more than or very high photon escape fractions from the parental halo are adopted . when considering primordial magnetic fields , we find that the additional heat injection from ambipolar diffusion and decaying mhd turbulence has significant impact on the thermal evolution and the ionization history of the post - recombination universe and on structure formation . the magnetic jeans mass changes the typical mass scale of the star forming halos , and depending on the adopted stellar model we derive upper limits to the magnetic field strength between andng ( comoving ) . for dark matter annihilation , we find an upper limit to the thermally averaged mass - weighted cross section of . for dark matter decay , our calculations yield a lower limit to the lifetime of dark matter particles ofs . these limits are in agreement with constraints from recombination and provide an independent confirmation at a much later epoch .

0807.3802

over the last decades , our understanding of cosmological reionization has improved considerably . observations of high - redshift quasars clearly indicate that reionization must end around @xmath5 @xcite , and the wmap 5-year - measurement finds a reionization optical depth of @xmath6 @xcite , yielding clear evidence that reionization started significantly earlier and is thus a continuous process . reionization through miniquasars would produce a soft - x - ray background that is significantly higher than the observed background , and can thus be ruled out @xcite . simulations of stellar reionization further show that the process is highly inhomogeneous and based on the growth and merging of ionized bubbles around the first stellar sources @xcite . understanding the essential physical processes , it is possible to build semi - analytic models that describe stellar reionization , which can be tested over a wide parameter space @xcite . additional constraints on the cosmic star formation rate are now available from gamma - ray burst observations @xcite . the reionization framework thus provides an increasingly reliable test for the stellar population during reionization and can be used to constrain global physical conditions in the early universe . + + regarding the stellar population , the works @xcite and @xcite suggest that the first stars were top - heavy with a peak in the imf at around @xmath7 . on the contrary , @xcite and @xcite indicate that gas especially in more massive systems can fragment because of dips in the equation of state , which may lead to the formation of a stellar cluster . it was shown that cooling in previously ionized gas is enhanced and leads to typical stellar masses of @xmath8 @xcite . it was further suggested that the presence of weak magnetic fields is sufficient to lead to a more present - day like mode of star formation , resulting in considerably lower masses @xcite . recently , a new phase of stellar evolution was suggested in which the stars would be powered by dark matter annihilation instead of nuclear fusion @xcite . various follow - up works have explored such a scenarios in more detail . studies by @xcite explored the main - sequence and pre - main - sequence phase of dark stars , and other works calculated the effect of dark matter capture by off - scattering from baryons @xcite . while many predictions are still model - dependent , it has often been suggested that such stars may have typical masses of @xmath9 , giving rise to a very bright main - sequence phase , or may have much longer lifetimes due to modifications in the stellar evolution . in the end , all these suggestions must face the constraint that the stellar population must be able to provide the correct reionization optical depth . + + complications may arise through the presence of additional physics that are often not considered in standard reionization calculations and simulations on the first stars . such possibilities include the presence of primordial magnetic fields , dark matter decay and dark matter annihilation . magnetic fields have been observed on all scales in the universe , and recently , it has been demonstrated that they were present already in high - redshift galaxies @xcite . they are found in the interstellar gas as well as in the intergalactic medium @xcite , but their origin is still unclear . there is a viable possibility that these fields have a primordial origin @xcite . so far , the most stringent constraints on the strength of these putative fields come from the measurements of the cosmic microwave background radiation ( cmbr ) and from big - bang nucleosynthesis ( bbn ) calculations . a homogeneous magnetic field would produce temperature anisotropies in the cmbr @xcite whose maximal amplitudes are limited by the cobe satellite measurements which in turn limit the field strength , as measured today , to @xmath10 @xcite . the presence of any primordial field would also alter the cmbr power spectrum by changing the characteristic velocities . with the sensitivity of the planck satellite one should be able to detect fields with present day strength of @xmath11 @xcite . so far , measurements by the wmap satellite are compatible with the absence of primordial magnetic fields . strong magnetic fields in the early universe can also change the abundance of relic @xmath12he and other light elements during the big bang nucleosynthesis @xcite . to comply with observational limits on light element abundances these primordial fields must not exceed @xmath13 at the time when the universe was @xmath14 which corresponds to a present day field @xmath15@xcite . effects of primordial magnetic fields have already been considered in @xcite , finding that density perturbations can be enhanced by the lorentz force from tangled magnetic fields . both the evolution of perturbations in the presence of magnetic fields as well as their effect on the thermodynamics via ambipolar diffusion heating and decaying mhd turbulence was considered in @xcite . recent calculations of @xcite show that the enhancement of structure due to magnetic fields is pronounced at about @xmath16 , but becomes less effective on larger mass scales and appears as a subdominant contribution on the scale of the magnetic jeans mass . consequences for @xmath17 cm observations have been explored as well @xcite . in this work , we examine the consequences for reionization in more detail and calculate the backreaction on structure formation according to the work of @xcite . the wmap 5 year data @xcite have measured the thomson scattering optical depth from reionization and allow to constrain different reionization scenarios in the early universe . the nature of dark matter is still unclear , and consequences of various particle physics like massive neutrinos or axion decay have been explored early ( e. g. * ? ? ? * ; * ? ? ? * ; * ? ? ? observational progress allowed to refine these studies and to explore such scenarios in the framework of @xmath18cdm cosmology @xcite . recently , the possibility was discussed to detect such effects using future @xmath17 cm telescopes@xcite , and it was shown that the fraction of the energy absorbed into the igm can be calculated in detail for specific models of dark matter decay and annihilation @xcite . indeed , secondary ionization through the decay / annihilation products can provide a way to ionize the igm which is independent of the stellar contribution @xcite , whereas the additional heat input increases the jeans mass and delays the formation of the first structures . in this work , we show how reionization constrains the properties of the stellar population as well as some additional heat sources like primordial magnetic fields , dark matter annihilation and decay . in [ reionization ] , we present our model for stellar reionization , which considers the igm as a two - phase - medium consisting of ionized bubbles and overall neutral gas . based on the thermal evolution , we self - consistently determine the minimal mass scale of halos which can collapse . in [ bfields ] , we explain our treatment of primordial magnetic fields and show how they modify the thermal evolution . the treatment of dark matter annihilation and decay , as well as some implications for dark stars , are discussed in [ darkness ] . the optical depth for different models is given in [ cmbfast ] . further discussion and outlook is given in [ outlook ] . the thermal and ionization history of the igm between recombination and the end of reionization is determined by a number of different processes . at redshifts @xmath19 , compton scattering of cmb photons couples the gas temperature @xmath20 to the cmb temperature @xmath21 . at lower redshifts , this coupling is less efficient and the gas decouples from the cmb due to adiabatic expansion . this standard scenario can be altered if additional energy injection mechanisms are present . additional heat input can lead to an earlier redshift of decoupling , whereas an increase in the ionized fraction , for instance due to secondary ionization from the decay or annihilation products of dark matter , tends to make compton cooling more efficient . once structure formation sets in , the thermal and ionization history is further influenced from x - rays produced in star forming regions and uv photons that escape from the first galaxies . we modified the recfast code @xcite and included all these feedback processes as described below , modeling the igm as a two - phase medium of ionized and partially - ionized gas . in this picture , the fully - ionized gas refers to the gas in the hii regions of the first luminous sources , while the partially ionized gas describes gas that is not yet affected by uv feedback . in the absence of additional energy injection mechanisms , recombination in the early universe and freeze - out of electrons was calculated with unprecedent accuracy , following the detailed level populations of hundreds of energy levels for @xmath22 , @xmath23 and @xmath24 and self - consistently calculating the radiation field@xcite . they developed the recfast code , a simplified version of the multi - level calculations , based on an effective three - level model for the hydrogen atom . recfast is capable of fully reproducing the results of the more detailed calculation @xcite . both the detailed calculation and the recfast code were recently updated and improved in @xcite . for the partially - ionized gas , we modify the equations describing the thermal and ionization history in the following way : the equation for the temperature evolution is given as @xmath25 where @xmath26 is the new heating term ( see [ ambi ] , [ decay ] and [ stellar ] ) , @xmath27 the new cooling term including lyman @xmath28 cooling , bremsstrahlung cooling and recombination cooling using cooling functions of @xcite , @xmath29 is the thomson scattering cross section , @xmath30 the stefan - boltzmann radiation constant , @xmath31 the electron mass , @xmath32 the speed of light , @xmath33 boltzmann s constant , @xmath34 the total number density , @xmath35 the electron fraction per hydrogen atom , @xmath36 is the hubble factor and @xmath37 is the number ratio of @xmath23 and @xmath22 nuclei , which can be obtained as @xmath38 from the mass fraction @xmath39 of @xmath23 with respect to the total baryonic mass . the evolution of the ionized fraction of hydrogen , @xmath40 , is given as @xmath41 } { h(z)(1+z)[1+k_{\mathrm{h}}(\lambda_{\mathrm{h}}+\beta_{\mathrm{h}})n_{\mathrm{h}}(1-x_{\mathrm{p}})]}\\ & \times&[1+k_{\mathrm{h}}\lambda_{\mathrm{h}}n_{\mathrm{h}}(1-x_{\mathrm{p}})]-\frac{k_{{\mbox{{\scriptsize ion}}}n_h x_{\mathrm{p}}}}{h(z)(1+z)}-f_{{\mbox{{\scriptsize ion}}}}\nonumber.\label{ion}\end{aligned}\ ] ] in this equation , @xmath42 is the number density of hydrogen atoms and ions , @xmath43 planck s constant , @xmath44 is the collisional - ionization coefficient @xcite , @xmath45 describes ionization from x - rays ( see [ stellar ] ) , and the parametrized case b recombination coefficient for atomic hydrogen @xmath46 is given by @xmath47 with @xmath48 , @xmath49 , @xmath50 , @xmath51 and @xmath52 , which is a fit given in @xcite to the coefficient of @xcite . this coefficient takes into account that direct recombination into the ground state does not lead to a net increase of neutral hydrogen atoms , since the photon emitted in the recombination process can ionize other hydrogen atoms in the neighbourhood . the fudge factor @xmath53 serves to speed up recombination and is determined from comparison with the multilevel - code . we further introduce the clumping factor @xmath54 to take into account the increase in the recombination rate in structures of increased density at low redshifts . we use the fit formula @xmath55 obtained from simulations of @xcite at redshifts @xmath56 and set @xmath57 at higher redshifts . the photoionization coefficient @xmath58 is calculated from detailed balance at high redshifts as described in @xcite . at lower redshifts , however , radiative recombination is no longer balanced by photoionization in the presence of additional energy injection mechanisms like ambipolar diffusion heating . once the ionized fraction drops below @xmath59 , we thus calculate the photoionization coefficient directly from the photoionization cross section given in @xcite . the frequency @xmath60 corresponds to the lyman-@xmath28 transition from the @xmath61 state to the @xmath62 state of the hydrogen atom . the two - photon transition between the states @xmath63 and @xmath62 is close to lyman-@xmath28 . consequently we use the same frequency for both processes . finally , @xmath64 is the two - photon rate for the transition @xmath63-@xmath62 according to @xcite and @xmath65 $ ] the cosmological redshifting of lyman @xmath28 photons . the additional terms for eq . ( [ temp ] ) and ( [ ion ] ) that are needed to describe the effects of dark matter annihilation and decay will be discussed in [ darkness ] . the fully - ionized component in the hii regions is described with the volume filling factor @xmath66 that denotes the volume fraction of ionized hydrogen bubbles . it is needed to compute the reionization optical depth and takes uv feedback into account . it evolves as @xmath67 @xcite . the uv photon production rate @xmath68 will be described in [ stellar ] in more detail , @xmath69 denotes the mean electron number density in fully - ionized regions , @xmath42 the neutral hydrogen density in regions that are still unaffected from uv photons and @xmath70^{-0.7}\ \mathrm{cm^3/s}$ ] is the case a recombination coefficient @xcite which we have chosen here , as recombinations will preferably occur in dense regions where the recombination photons are unlikely to escape into the igm @xcite . it is evaluated at the temperature with @xmath71 to account for the fact that the ionized regions should be heated at least to @xmath72 k. if heating via ambipolar diffusion and decaying mhd turbulence increased the gas temperature above this threshold , it is evaluated at the gas temperature instead . to compare the models with the observational constraints from wmap , we calculate the thomson scattering optical depth of free electrons , given as @xmath73 where @xmath74 is the comoving number density of ionized and neutral hydrogen , @xmath75 and @xmath76 the usual cosmological density parameters , @xmath77 the hubble constant and @xmath78 is the redshift where reionization starts , which we define as the point where the effective ionized fraction @xmath79 becomes larger than @xmath80 . this point has been chosen after a comparison of the actual te cross calibration spectra with those assuming sudden reionization , but the results are not very sensitive to this choice . in summary , we adopt the following picture : * the partially - ionized gas unaffected by uv feedback is modeled with eqs ( [ temp ] ) , ( [ ion ] ) that describe the evolution of its temperature and ionization degree . * the fully - ionized gas is described by the volume - filling factor @xmath66 calculated from eq . ( [ volumefilling ] ) . its temperature is given as the maximum of @xmath72 k and the temperature of the partially - ionized gas . * recombination both in the partially- and fully - ionized gas is enhanced by the clumping factor given in eq . ( [ clump ] ) . * for the partially - ionized component , recombination is described using the case b recombination coefficient , appropriate for low - density gas far from the ionizing sources . * for the fully - ionized gas , we use the case a recombination coefficient , assuming that most recombinations take place in high - density regions near the ionizing sources . the implementation of different feedback mechanisms is described below in more detail . the universe becomes reionized due to stellar feedback . we assume here that the star formation rate ( sfr ) is proportional to the change in the fraction of collapsed halos @xmath81 . as shown in @xcite , @xmath81 is given as @xmath82,\label{fcoll}\ ] ] where @xmath83 ^ 2\;,\label{sigma}\ ] ] and where @xmath84 is the minimum mass of haloes that are able to collapse at a given redshift . here , @xmath85 is the spherical bessel function , @xmath86 is related to the halo mass @xmath87 via @xmath88 , @xmath89 is the mean density , @xmath90 the linearized density threshold for collapse in the spherical top - hat model and @xmath91 the linear growth factor . in the absence of magnetic field , @xmath84 is determined by the filtering mass @xcite . as discussed in the introduction , tangled magnetic fields can potentially create more small scale structure via the lorentz force . the calculations by @xcite show that this effect is pronounced in the minihalo regime . while they adopted a constant minimal collapse mass of @xmath92 which is independent of the magnetic field , we use the framework of @xcite to take into account the change in the mass scale of halos that can form stars . indeed , we find that the mass scale is changed significantly ( see fig . [ field ] ) , such that the additional small - scale structure from tangled magnetic fields does not contribute to star formation . we introduce the magnetic jeans mass , @xcite @xmath93 and the thermal jeans mass , @xmath94 the filtering mass in the presence of magnetic fields is then given as @xmath95,\ ] ] where @xmath96 is the scale factor and @xmath97 . the minimum halo mass to consider is given as @xmath98 , where @xmath99 denotes the minimal halo mass for which the baryons can efficiently cool after collapse . we adopt here the fiducial value @xmath100 @xcite . similar estimates have been given in @xcite , @xcite and @xcite . for the subsequent analysis , we furthermore assume that only a certain fraction @xmath101 of the collapsing halo mass turns into stars . as x - ray photons have long mean free paths , they can play an important role in the ionization and heating of the gas . assuming that the local correlation between the sfr and the x - ray luminosity ( from @xmath102 kev ) holds up to a renormalization factor @xmath103 ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the x - ray heating function @xmath104 is given as @xmath105 x - ray emission has two major sources , inverse - compton scattering of cmb photons with relativistic electrons accelerated in supernovae , and high - mass x - ray binaries . the former may play an increasingly important role at high redshifts , as the cmb photons are more energetic at high redshifts @xcite . in the early universe the factor @xmath106 if @xmath107 erg are released per @xmath7 , corresponding to an overall efficiency of 5% @xcite . the abundance of x - ray binaries depends on metallicity and the stellar initial mass function , and could be especially large if very massive pop . iii stars dominate @xcite . in the model presented here , we adopt @xmath108 as a generic value and assume that the uncertainty can be ascribed to the star formation efficiency @xmath101 . in general , x - rays lose their energy through three channels . a fraction @xmath109 goes into heating , a fraction @xmath110 into ionization , and a fraction @xmath111 into excitation . these parameters should not be confused with @xmath45 introduced in eq . ( [ ion ] ) and @xmath81 introduced in eq . ( [ fcoll ] ) . we calculate them using the fit formulae in @xcite . we can thus write @xmath112 the contribution from the x - rays to the term @xmath45 in eq . ( [ ion ] ) is then given as @xmath113 uv photons from stellar sources are in general absorbed locally , only a fraction @xmath114 that manages to escape from the first galaxies can contribute to the ionization of the igm . again , the production of ionizing photons is associated with the star formation rate @xcite , yielding a contribution @xmath115 where @xmath116 with @xmath117 and @xmath118 is the number of ionizing photons per stellar baryon . the nature of the first stellar sources is still under discussion , and even the question whether the primordial imf is top - heavy or closer to the locally measured imf is not solved . while for instance @xcite and @xcite found a top - heavy imf using adaptive - mesh refinement ( amr ) or smoothed - particle hydrodynamics ( sph ) simulations , it was shown in @xcite that primordial and low - metallicity gas can fragment if the evolution of the gas is followed further after the formation of the first clump , due to a dip in the equation of state . based on similar arguments , @xcite argued that even a small metallicity fraction can lead to fragmentation in the first protogalaxies . it was further suggested that magnetic fields can have a crucial influence on the primordial imf @xcite . given these uncertainties , we discuss different models for the stellar population . dark stars have also been suggested as some of the first luminous sources . we will discuss this possibility in more detail in [ dark star ] . .summary of adopted stellar models . the first column gives the model name . the second column indicates the stellar populations that contribute to reionization . the third column gives the number of ionizing photons per baryon used in eq ( [ xi ] ) . the fourth column lists the adopted escape fractions . model a is a highly extreme case in which we assume all ionizing photons come from massive pop iii stars and can escape the star - forming halo . in model b photons can escape efficiently only from halos less massive than a corresponding virial temperature of @xmath119k , while for higher - mass halos , this fraction is reduced to @xmath0 . in model c we assume that pop iii as well as pop ii stars contribute to reionization , and we adopt the same escape probabilities as in b. model d is another extreme case which assumes that all ionizing photons come from low - mass pop ii stars , with escape fraction @xmath120 . [ cols="^,^,^,<",options="header " , ] in models a and b , we assume that reionization is solely due to metal - free massive pop iii stars . this situation can only be examined from a theoretical point of view . most investigations of uv feedback from high - mass zero - metallicity stars indicate that ionizing photons can easily escape the star - forming halo and drive large hii regions into the igm . this suggests that the escape fraction in high - redshift galaxies could be very high , of order unity , being much higher than in the present - day universe ( for instance * ? ? ? * ; * ? ? ? * ; * ? ? ? detailed numerical simulations of @xcite show that indeed the shock bounding hii regions of massive pop . iii stars can easily photo - evaporate the minihalo and lead to an escape fraction of one . however , the situation is not fully clear . there are other studies that suggest that the escape fraction could remain small @xcite , depending on the mass of the collapsing halo . this is why we adopt two different models . model a is an extreme case , in which we assume that the escape fraction is @xmath121 independent of the halo mass . the more realistic case is probably model b , where we distinguish whether the virial temperature corresponding to the generalized filtering mass is smaller or larger than @xmath72 k. if it is smaller , we still assume that the halo is easily photoevaporated . if it is larger , we assume that most of the mass is kept within the halo and adopt an escape fraction @xmath122 . the virial mass corresponding to @xmath72 k is given as @xmath123 @xcite . for model a and b , we adopt a total number of ionizing photons per baryon of @xmath124 , following @xcite . the heavy elements produced by the very first stars will gradually enrich the igm . it is very likely that some contribution to reionization comes from low - metallicity pop ii stars as well . as these stars are expected to have lower masses than pop iii stars , we introduce model c , which corresponds to a stellar population of intermediate mass . it has @xmath125 and an escape fraction according to model b , suggesting that such stars may photo - evaporate only low - mass halos . finally , we consider in model d the extreme case of very rapid chemical enrichment and assume cosmic reionization is entirely driven by low - mass pop ii stars . we assume a stellar mass distribution similar to the local imf in the galactic halo @xcite with a metallicity of @xmath126 of the solar value . this corresponds to @xmath127 . we furthermore adopt escape fractions that are typical for local star forming galaxies . many upper limits and a few detections have been found observationally @xcite , suggesting @xmath128 . a detection from lyman - break galaxies at @xmath129 implied a higher fraction of @xmath0 @xcite , while more recent observations place upper limits of @xmath130 or claim detections at even lower levels @xcite . as the relevant rates scale only linear in @xmath114 , it seems reasonable to adopt @xmath131 for our pop ii model . as it is unlikely that cosmic reionization occurs instantaneously with the onset of pop iii star formation nor that chemical enrichment is so rapid that all ionizing photons come from metal - enriched pop ii stars , we adopt the intermediate scenario c with a mixed population as our fiducial model . all global cosmological parameters are chosen according to the wmap 5 year data @xcite . as discussed in @xcite , magnetic fields can significantly alter the thermal evolution of the igm via ambipolar diffusion heating and decaying mhd turbulence , which can have a significant influence on the filtering mass scale . we explain our treatment of these heating terms in this section . the presence of magnetic fields introduces two different contributions to the heating rate , one coming from ambipolar diffusion and one resulting from the decay of mhd turbulence . + in the first case , the contribution can be calculated as @xcite : @xmath132 here , @xmath133 , @xmath134 and @xmath135 are the mass densities of neutral hydrogen , ionized hydrogen and all baryons . the ion - neutral coupling coefficient for primordial gas is given as @xcite @xmath136},\end{aligned}\ ] ] where @xmath137 is the number density of @xmath23 and @xmath138 the mass of the hydrogen atom . collisions with electrons are neglected here , as their contribution is suppressed by a factor @xmath139 . we adopt the zero drift velocity momentum transfer coefficients of @xcite for collisions of @xmath140 with @xmath22 and @xmath23 , which is a good approximation in the absence of shocks . they are given by @xmath141 as the power spectrum for the magnetic field is unknown and eq . ( [ ambiheat ] ) can not be solved exactly , we adopt a simple and intuitive approach to estimate the integral for a given average magnetic field @xmath142 with coherence length @xmath143 . the heating rate can then be evaluated as @xmath144 the coherence length @xmath143 is in principle a free parameter that depends on the generation mechanism of the magnetic field . it is constrained through the fact that tangled magnetic fields are strongly damped by radiative viscosity in the pre - recombination universe on scales smaller than the alfvn damping scale @xmath145 given by @xcite @xmath146 where @xmath147 denotes the comoving magnetic field . in fact , we expect fluctuations to be present on all scales . as the heating term goes as @xmath148 , we thus adopt the minimal value @xmath149 . for decaying mhd turbulence , we adopt the prescription of @xcite , @xmath150^{\tilde{\alpha } } h(t)}{[\ln(1+t_d / t_i)+\ln(t / t_i ) ] ^{\tilde{\alpha}+1}},\label{decayheat}\ ] ] where @xmath151 is the cosmological time at redshift @xmath152 , @xmath153 is the dynamical timescale , @xmath154 the time where decay starts , i. e. after the recombination epoch when velocity perturbations are no longer damped by the large radiative viscosity , @xmath155 is the corresponding redshift . for a power spectrum of the magnetic field strength with power - law index @xmath28 , the parameter @xmath156 is given as @xmath157 @xcite . in the generic case , we expect the power spectrum of the magnetic field to have a maximum at the scale of the coherence length , and the heat input by mhd decay should be determined from the positive slope corresponding to larger scales @xcite . we thus adopt @xmath158 for the calculation . we estimate the dynamical timescale as @xmath159 , where @xmath160 is the alvn velocity and @xmath135 the baryon mass density . the evolution of the magnetic field as a function of redshift can be determined from the magnetic field energy @xmath161 , which evolves as @xcite @xmath162 while the redshift of reionization of course depends on the model for the stellar population , the mechanism which delays reionization in the presence of magnetic fields is always the same . we illustrate this for our fiducal model c adopting a star formation efficiency @xmath101 of 1% , but point out that the general discussion is valid also for the other stellar population models . we use the cosmological parameters of wmap 5 @xcite . fig . [ temperature ] shows the thermal evolution of the igm for different magnetic field strengths . in the absence of magnetic fields , the gas temperature follows the temperature of radiation until @xmath163 , where compton scattering becomes inefficient . afterwards , the gas cools adiabatically until it is reheated during reionization . in the presence of magnetic fields , gas and radiation decouple earlier due to the ambipolar diffusion heating and decaying mhd turbulence and stays at higher temperatures , which prevents collapse in low - mass halos and thus delays reionization . the additional heat increases the ionized fraction at early times ( see fig . [ ionization ] ) , while the redshift where the igm becomes fully ionized is delayed , due to the increased generalized filtering mass ( see fig . [ filt ] ) . collisional ionization introduces a natural temperature plateau of the order @xmath164 k , as any further temperature increase will have a backreaction on the ionized fraction of the gas , and thus make ambipolar diffusion less efficient . a large variety of different particle physics models has been suggested to explain the dark matter content of the universe @xcite . many of these models predict some interactions in the dark sector and include some form of dark matter decay and annihilation . the consequences of such scenarios for the thermal evolution have been discussed in detail for instance in @xcite . in this work , we discuss the implications of such scenarios on reionization . as we have seen for the case of primordial magnetic fields , the additional heat input increases the filtering mass and thus increases the minimal halo mass in which the first luminous objects can form , making stellar reionization less effective . on the other hand , once heating through dark matter decay or annihilation is effective , secondary ionization will likely be effective as well and thus increase the thomson scattering optical depth . as a consequence of dark matter annihilation models , a new phase of stellar evolution was suggested in @xcite , in which stars are not powered by nuclear fusion , but by annihilating dark matter within them . these stars could reach masses of up to @xmath165 @xcite . once dark matter annihilation becomes ineffective and such stars enter the main sequence phase , they would thus be extremely bright sources of uv photons . the consequences of such a phase are discussed below . further models taking into account a dark matter capturing phase have been suggested as well @xcite . as they are highly parameter - dependent , we study them in a separate work @xcite . the energy of the annihilation products can be deposited into heating , ionization and collisional excitation . the latter is quickly radiated away , but may contribute to the build - up of a lyman @xmath28 background that helps to couple the spin temperature of hydrogen to the gas temperature , which may be relevant for @xmath17 cm observations @xcite . for given models of dark matter , it is possible to work out the detailed absorbed fractions as a function of redshift @xcite . in this work , however , we adopt the more generic approach of @xcite . for definiteness , they assume that the dark matter particles annihilate to high - energy photons , which allows to determine the energy fractions going into heat and ionization by the fitting formulae of @xcite . as they show , heating by dark matter annihilation is accompanied also by a significant increase in the electron fraction due to secondary ionization . for other models of dark matter annihilation , the results can be rescaled by appropriate absorption efficiencies . to eq . ( [ temp ] ) describing the temperature evolution , we thus add a term @xmath166 according to @xcite , where @xmath167 , @xmath168 , @xmath37 is the helium fraction by number , @xmath169 is the fraction of energy going into heating ( see * ? ? ? * ) and @xmath170 the effective baryon - normalized `` lifetime '' , given as @xmath171 here , @xmath172 and @xmath173 denote the cosmological parameters for dark and baryonic matter , @xmath174 is the critical density at redshift zero , @xmath175 is the dark matter particle mass and @xmath176 the velocity - averaged cross section . in the same way , we add a term to eq . ( [ ion ] ) that describes the evolution of the ionized fraction : @xmath177 where @xmath178 ev is the hydrogen ionization threshold and @xmath179 the fraction of the energy going into ionization , for which we use the fitting formulae of @xcite . as the decay rate scales with the dark matter density squared , it is most efficient at early times and may thus modify the recombination history , which allows to place upper limits on the dark matter annihilation . as shown in @xcite , the wmap 1-year - data yield an upper limit of @xmath180 where @xmath181 corresponds to the energy fraction actually absorbed into the igm and is of order 1 for particle masses in the mev range @xcite . in a similar way , dark matter decay can alter the thermal and ionization history of the universe . the lifetimes of dark matter particles considered here will be considerably larger than the age of the universe to ensure that the abundance of dark matter does not change significantly . still , the conversion of particle mass to thermal energy can have an important impact on the thermal evolution of the universe @xcite . in this limit , the decay rate is constant over time and the effect at lower redshifts is more pronounced compared to the case of dark matter annihilation . to calculate the heating and ionization rate , the same formalism can be employed as for dark matter annihilation , but we adopt the baryon - normalized decay rate @xmath182 where @xmath183 is the lifetime of the dark matter particle . based on the modified recombination and reionization histories in the presence of dark matter decay , it was shown that @xcite @xmath184 where @xmath185 is the fraction of the particle mass - energy released through decay and @xmath181 the fraction of the released energy deposited into the igm . if data of large - scale surveys are included , the constraint can be slightly improved , yielding @xmath186 dark stars have been suggested as a possible consequence of dark matter annihilation in the early universe @xcite . if indeed the first stars form in the central peaks of the dark matter distribution in the early minihaloes , they might be powered by dark matter annihilation instead of nuclear fusion . such stars could have up to @xmath165 and surface temperatures in the range of @xmath187 k , i. e. significantly colder than conventional pop . . however , this dark phase will gradually come to an end as the dark matter in the cusp annihilates , so the star may contract and finally reach a main sequence phase . a linear stability analysis and 1d simulations including hydrodynamics and radiation indicates the stability of such objects @xcite . in this case , one can expect a similar number of photons per stellar baryon as for conventional pop . iii stars , perhaps even higher by a factor of 2 @xcite . as it is crucial for these objects to form on the peak of the dark matter density , we expect them to form only in low - mass halos with a virial temperature below @xmath72 k , as the so - called atomic cooling halos are very turbulent @xcite and are more likely to form a stellar cluster instead of a single massive star @xcite . in addition , such massive halos are more likely to have accreted material from previous metal enrichment @xcite , which may cause fragmentation as well @xcite . in previously ionized regions , the thermodynamics of collapse are significantly altered and simulations generally find lower - mass stars @xcite , which might be an important limitation for dark stars as well . in this section , we show the results for the different stellar populations and in the context of different additional physics . we first concentrate on the effect of primordial magnetic fields and compare the results of different models for the stellar population . we run our reionization model for a range of different star formation efficiencies and primordial magnetic fields and obtain the corresponding optical depth . to constrain the parameter space , some assumptions need to be made . the first is obvious : the calculated optical depth must agree with the optical depth measured by wmap 5 at least within @xmath188 . in addition , we require that the star formation efficiency may not be unreasonably high . for pop . iii stars with @xmath189 forming in halos of @xmath190 ( dark matter ) , star formation efficiencies of the order @xmath191 are expected . it seems thus reasonable to reject star formation efficiencies higher by one order of magnitude , i. e. of the order @xmath192 . further requirements are that reionization must be complete by redshift 6 , and that there should be a stellar contribution to reionization , so that the universe becomes metal - enriched . now we discuss the constraints resulting from these criteria . + + for all stellar models , we can exclude magnetic fields larger than @xmath193 ng , based on the optical depth alone ( see fig . [ taua ] to [ taud ] ) . a transition occurs at about @xmath194 ng : for lower magnetic fields , reionization is still mostly due to stellar radiation , whereas for higher magnetic fields , the igm is heated to such high temperatures that collisional ionization becomes very efficient and increases the optical depth . in this regime , structure formation is strongly suppressed , as the magnetic jeans mass scales with @xmath195 and dominates over the thermal jeans mass . independently of the stellar model , such a regime is always found between @xmath194 and @xmath193 ng . therefore , structure formation is considerably impeded such that metals do not form and the universe is not fully ionized by redshift @xmath196 . we can thus exclude magnetic field strength above @xmath194 ng independent of the stellar model . + + if we further require that the star formation efficiency must be lower than @xmath192 , the results become model - dependent . in model a , which is an extreme case where very high escape fractions are assumed even for very massive halos , no further constraint on the magnetic field is possible on the @xmath188 level . still , even in this extreme case , one can reject fields larger than @xmath197 ng on the @xmath198 level . the other extreme case , model d with low escape fractions and a low number of ionizing photons per stellar baryon , can be rejected completely , as it always requires unphysically high star formation efficiencies . + + for more realistic cases , essentially model b and c , we find that a critical magnetic field strength exists above which a very high star formation efficiency is needed to get into the @xmath188 interval around the measured optical depth . for model b , the critical value is @xmath199 ng , for model c , it is at @xmath1 ng . as it seems likely that a transition to less massive pop . ii stars occurs during reionization , model c might indeed be the most realistic case and provide an upper limit of @xmath1 ng . for the case of dark matter annihilation and decay , we will show only the results for the pop . iii star model b. regarding the competition between stellar reionization and secondary ionization from high - energetic annihilation / decay products , this corresponds to a rather conservative choice if one assumes that stellar reionization should do the main contribution . such a stellar population could be interpreted either as conventional pop . iii stars or as so - called dark stars in a main sequence phase . we find that dark matter annihilation starts to become important for @xmath200 and completely dominates the stellar contribution for higher values ( see fig . [ img : annihilation ] ) . values higher than @xmath201 are inconsistent with the wmap 5-year data . the results for dark matter decay are given in fig . [ img : decay ] . decay starts to become important for lifetimes below @xmath202 s. however , lifetimes smaller than @xmath203 s are incompatible with wmap 5 . as pointed out recently , constraints on dark matter annihilation can be significantly improved taking into account the dark matter clumping factor @xmath204 @xcite . however , this quantity is highly uncertain and may vary by six orders of magnitude @xcite . in the framework of light dark matter , we recently showed that is is constrained to be smaller than @xmath205 at redshift zero @xcite , yielding an enhancement of the annihilation rate by a factor of @xmath206 or more at redshift @xmath193 . other dark matter models @xcite even yield a clumping factor that drops down to @xmath207 at these redshifts , such that our calculation altogether provides a conservative and firm upper limit . from fig . [ img : annihilation ] , we can further see that very massive dark stars in a main sequence phase , which should have higher star formation efficienes that are an order of magnitude higher than conventional pop . iii stars , are ruled out by wmap 5 , as the models require rather low star formation efficiencies to reproduce the thomson scattering optical depth . of course , this conclusion holds only if one assumes that reionization is completely due to dark stars , which seems rather unlikely ( see discussion in [ dark star ] ) , and we will discuss some alternatives in [ outlook ] . we have calculated the reionization optical depth for different stellar models in the presence of primordial magnetic fields as well as dark matter annihilation and decay . the results indicate which star formation efficiencies are required in the presence of some additional heating mechanism for a given stellar model . considering different stellar models and primordial magnetic fields , we find the following results : 1 . independent of the model for the stellar population , we can securely exclude primordial magnetic fields larger than @xmath194 ng . 2 . for the most realistic case with a mixed stellar population ( model c ) , we even find an upper limit of @xmath1 ng , as higher magnetic fields would require star formation efficiencies larger than @xmath192 , which is unrealistic . similar results are found for model b , assuming reionization completely due to pop . iii stars . 3 . reionization only due to population ii stars ( model d ) is ruled out completely . 1 . dark matter annihilation provides noticeable contributions to the reionization optical depth only for thermally averaged mass - weighted cross sections @xmath208 . parameters @xmath209 can be ruled out by @xmath188 on the basis of wmap 5 year data . dark matter decay becomes important for the reionization optical depth for lifetimes below @xmath202 s. 4 . dark matter lifetimes below @xmath203 s are ruled out by @xmath188 . these results are in agreement with conservative constraints obtained from the gamma - ray background @xcite . we further showed that reionization can not be due to @xmath210 dark stars alone , as the corresponding optical depth would be significantly too high . one might wonder whether heating from dark matter annihilation might help to significantly delay stellar reionization , in order to reconcile this model with observations . however , as can be seen in fig . [ img : annihilation ] , this is more than compensated by the effects of secondary ionization , once that dark matter annihilation starts to have a significant influence on the igm . in case collider experiments like the lhc or other dark matter detection experiments find evidence for a self - annihilating dark matter candidate , this can be seen as evidence for a rapid transition towards a different mode of star formation , or a problem in our understanding of dark stars . to reconcile dark stars with observations , the following scenarios seem feasible : 1 . dark stars with @xmath211 may be extremely rare objects , and their actual mass scale is closer to the mass scale for typical pop . iii stars . in such a case , reionization could not distinguish between dark stars and conventional pop . iii stars . 2 . the transition to lower - mass stars occurs very rapidly , such that dark stars only contribute to the very early phase of reionization . reasons for that might be chemical , radiative as well as mechanical feedback ( see also the discussion in [ dark star ] ) . in fact , even a double - reionization scenario might be conceivable , in which very massive dark stars ionize the universe at high redshifts . due to chemical and radiative feedback , formation of such stars might be suppressed and the universe might become neutral again , until the formation of less massive stars becomes efficient enough to reionize the universe . 3 . dark stars do not reach a main - sequence phase , but are disrupted earlier by some non - linear instability . such an instability would be constrained by the fact that it neither appears in a linear stability analysis nore in 1d simulations . violent explosions of dark stars might even be considered as a source for gamma - ray bursts . while too definite conclusions on the existence of dark stars are not yet possible , it is at least indicated that the possibilities mentioned above should be explored in more detailed , and a better understanding of the properties of the dark stars . a better understanding of their evolution after the dark phase will certainly help to better understand their possible role during reionization . the constraints derived here are independent of other works that essentially rely on the physics of recombination to derive upper limits on additional physics ( e. g. * ? ? ? * ; * ? ? ? * ; * ? ? ? in particular , magnetic fields can evolve dynamically and their field strength may thus change between these epochs . it has recently been suggested that the biermann battery effect creates magnetic fields in the presence of an electron pressure gradient@xcite . we thus need to probe magnetic fields at different epochs . as we will show in a separate work @xcite , upcoming @xmath17 cm measurements will allow to probe the thermal history before and during reionization in great detail , and may allow to detect primordial magnetic fields of the order of @xmath212 ng . additional ways of probing the reionization history and the dark ages exist as well : scattering of cmb - photons in fine - structure lines of heavy elements may lead to a frequency - dependent cmb power spectrum and may allow to measure metal - abundances as a function of redshift @xcite . before reionization , molecules may form in the igm and introduce further frequency - dependent features in the cmb @xcite . such features are likely enhanced in the presence of either primordial magnetic fields or dark matter decay / annihilation , as the increased electron fraction catalyses the formation of molecules , and the additional heat input leads to a departure of the level populations from the radiation temperature . further improvements are expected from the upcoming measurement of planck , that will measure the reionization optical depth with unprecedent accuracy and thus allow to strengthen the constraints obtained here on the stellar populations and additional physics . in addition , a more accurate determination of the cosmological parameters will remove further uncertainties in the present models of reionization . we thank paul clark , andrea ferrara , katherine freese , daniele galli , simon glover , thomas greif and francesco palla for many exciting discussions on this topic . drgs thanks the heidelberg graduate school of fundamental physics ( hgsfp ) , the lgfg for financial support and the inaf - osservatorio astrofisico di arcetri for financial support . the hgsfp is funded by the excellence initiative of the german government ( grant number gsc 129/1 ) . rb is funded by the emmy - noether grant ( dfg ) ba 3607/1 . rsk thanks for support from the emmy noether grant kl 1358/1 . all authors also acknowledge subsidies from the dfg sfb 439 _ galaxies in the early universe_. we thank the anonymous referee for help improving the manuscript . becker , r. h. et al . , 2001 , @jnlaj , 122 , 2850 komatsu , e. , et al . , 2008 , @jnlapjs , submit ted nolta , m. , et al . , 2008 , @jnlapjs , submit ted dijkstra , m. , haiman , z. , loeb , a. , 2004 , , 613 , 646 salvaterra , r. , haardt , f. , ferrara , a. , 2005 , monthly notices of the royal astronomical society : lett ers , 362 , l50 gnedin , n. y. , 2005 , , 542 , 535 ciardi , b. , ferrara , a. , white , s. d. m. , 2003 , @jnlmnras , 344 , l7 kohler , k. , gnedin , n. y. , ha milton , a. j. s. , 2005 , , 657 , 15 shapiro , p. r. , giroux , m. l. , 1987 , , 321 , l107 haiman , z. , loeb , a. , 1997 , , 483 , 21 barkana , r. , loeb . , a. , 2001 , phys . , 349 , 125 loeb , a , barkana , r. , 2001 , @jnlara&a , 39 , 19 l choudhury , t. r. , ferrara , a. , 2 005 , @jnlmnras , 361 , 577 schneider , r. , salvaterra , r. , ferra ra , a. , ciardi , b. , 2006 , @jnlmnras , 369 , 825 yuksel , h. , kistler , m. d. , beacom , j. f. , 2008 , , 683l , 5 abel , t. , bryan , g. l. , norman , m. l . , 2002 , science , 295 , 93 bromm , v. , larson , r. , 2004 , @jnlara&a , 42 , 7 9 clark , p. c. , glover , s. c. o. , klessen , r. s. , 2008 , , 672 , 757 omukai , k. , schneider , r. , h aiman , z. , 2008 , submitted to yoshida , n. , oh , s. p. , kitayama , t. , h ernquist , l. , 2007 , , 663 , 687 yoshida , n. , omukai , k. , hernquist , l. , 2007 , , 667 , l117 barrow , j. d. , ferreira , p. g. , & silk , j. 1997 , physical review letters , 78 , 3610 silk , j. , langer , m. , 2006 , @jnlmnras , 371 , 44 4 spolyar , d. , freese , k. , gondolo , p. , 2008 , , 100 , 051101 iocco , f. , 2008 , , 677 , l1 freese , k. , bodenheimer , p. , spolyar , d. , iocco , f. , bressan , a. , ripamonti , e. , schneider , r. , ferrara , a. , marigo , p. , 2008 , mnras , in press yoon , s .- c . , iocco , f. , akiyama , s. , 2008 , eprint arxiv:0806.4016 taoso , m. , bertone , g. , meynet , g. , ekstrom , s. , 2008 , eprint arxiv 0806.2681 freese , k. , spolyar , d. , aguirre , a. , 2008 , eprint arxiv:0802.1724 gondolo , p. , 2008 , eprint arxiv:0806.0617 bernet , m. l. , miniati , f. , lilly , s. j. , kronberg , p. p. , dessauges - zavadsky , m. , 2008 , nature , 454 , 302 , r. , brandenburg , a. , moss , d. , shukurov , a. , & sokoloff , d. , r. 2001 , space science reviews , 99 , 243 , c. l. , & taylor , g. b. 2002 , @jnlara&a , 40 , 319 , d. , & rubinstei n , h. r. 2001 , @jnlphys . , 348 , 163 , l. m. 2002 , reviews of modern phy sics , 74 , 775 , i. b. , & n ovikov , i. d. 1983 , p. translation . ) , j. , danielsson , u. h. , grasso , d. , & rubinstein , h. 1996 , physics letters b , 388 , 253 , j. j. , & oconnell , r. f. 1969 , physical review , 180 , 1289 , g. 1969 , , 223 , 938 kim , e .- j . , olinto , a. v. , rosner , r. , 1996 , , 468 , 28 sethi , s. k. , subramanian , k. , 2005 , @jnlmnras , 356 , 778 tashiro , h. , sugiyama , n. , 2006 , @jnlmnras , 368 , 965 tashiro , h. , sugiyama , n. , 2006 , @jnlmnras , 372 , 1060 schleicher , d. r. g. , banerjee , r. , klessen , r. s. , , accepted ( eprint arxiv:0808.1461 ) doroshkevich , a. g. , khlopov , m.yu . , klypin , a.a . , 1989 , @jnlmnras , 239 , 923 berezhiani , z.g . , khlopov , m.yu . , 1991 , z.phys.c- particles and fields , 49 , 73 berezhiani , z.g . , khlopov , m.yu . , khomeriki , r.r . , 1990 , yadernaya fizika , 52 , 104 [ english translation : sov.j.nucl.phys . ( 1990 ) v. 52 , pp . 65 - 68 ] chen , x. , kamionkowski , m. , 2004 , , 70 , 043502 hansen , s. h. , haiman , z. , , 600 , 26 avelino , p. p , barbosa , d. , 2004 , , 70 , 067302 kasuya , s. , kawasaki , m. , 2004 , , 70 , 103519 pierpaoli , e. , 2004 , , 92 , 031301 bean , r. , melchiorri , a. , silk , j. , 2003 , , 68 , 083501 padmanabhan , n. , finkbeiner , d. p. , 2005 , , 72 , 023508 furlanetto , s. , oh , s. p. , pierpaoli , e. , 2006 , , 74 , 103502 ripamonti , e. , mapelli , m. , ferrara , a. , 2007 , @jnlmnras , 374 , 1067 mapelli , m. , ripamonti , e. , 2007 , proceedings of the 11th marcel grossmann meeting held in berlin , germany , july 23 - 29 2006 seager , s. , sasselov , d. , scott , d. , 1999 , , 523 , l1 seager , s. , sasselov , d. , scott , d. , 2000 , , 128 , 407 wong , w. y. , moss , a. , scott , d , 2007 , @jnlmnras , 386 , 1023 anninos , p. , zhang , y. , abel , t. , norman , m. l. , 1997 , new a , 2,209 t. abel , p. anninos , y. zhang , m. l. norman . new astronomy , vol . august 1997 . pquignot , d. , petitjean , p. , boisson , c. , 1991 , @jnla&a , 251 , 680 hummer , d. g. , 1994 , @jnlmnras , 268 , 109 mellema , g. , iliev , i. t. , pen , u .- l . , shapiro , p. r. , 2006 , @jnlmnras , 372 , 679 sasaki , s. , takahara , f. , 1993 , pasj , 45 , 655 goldman , s. p.,1989 , phys . a. , 40 , 1185 osterbrock , d. e. , astrophysics of gaseous nebulae and active galactic nuclei , university science books miralda - escud , j. , 2003 , , 597 , 66 press , w. h. , schechter , p. , 1974 , , 187 , 425 gnedin , n. y. , hui , l. , 1998 , @jnlmnras , 296 , 44 greif , t. h. , johnson , j. l. , klessen , r. s. , bromm , v. , 2008 , @jnlmnras , in press yoshida , n. , abel , t. , hernquist , l. , sugiyama , n. , 2003 , , 592 , 645 mackey , j. , bromm , v. , hernquist , l. , 2003 , , 586 , 1 furlanetto , s. r. , oh , s. p. , briggs , f. h. , 2006 , physics reports , 433 , 181 grimm , h .- j . , gilfanov , m. , sunyaev , r. , 2003 , @jnlmnras , 339 , 793 ranalli , p. , comastri , a. , setti , g. , @jnla&a , , 399 , 39 gilfanov , m. , grimm , h .- j . , sunyaev , r. , 2004 , @jnlmnras , 347 , l57 glover , s. c. o. , brand , p. w. j. l. , 2003 , @jnlmnras , 340 , 210 koyama , k. , petre , r. , gotthelf , e. v. , hwang , u. matsuura , m. , ozaki , m. , holt , s. s. , 1995 , , 378 , 255 oh , s. p. , 2001 , , 553 , 499 dove , j. b. , shull , j. m. , ferrara , a. , 2000 , , 531 , 846 ciardi , b. , bianchi , s. , ferrara , a. , 2002 , @jnlmnras , 331 , 463 fujita , a. , martin , c. l. , mac low , m .- m . , abel , t. , 2003 , , 599 , 50 whalen , d. , abel , t. , norman , m. l. , 2004 , , 610 , 14 oh , s. p. , haiman , z. , 2002 , , 569 , 558 wood , k. , loeb , a. , 2000 , , 545 , 86 scalo , j. , 1998 , in asp conf . ser . 142 , the stellar initial mass function , ed . g. gilmore & d. howell ( san francisco : asp ) , 201 kroupa , p. , 2002 , science , 295 , 82 chabrier , g. , 2003 , pasp , 115 , 763 steidel , c. c. , pettini , m. , adelberger , k. l. , 2001 , , 546 , 665 hurwitz , m. , jelinsky , p. , dixon , w. v. d. , 1997 , , 481 , l31 heckman , t. m. , sembach , k. r. , meurer , g. r. , leitherer , c. , calzetti , d. , martin , c. l. , 2001 , , 558 , 56 deharveng , j .- m . , buat , v. , le brun , v. , millard , b. , kunth , d. , shull , j. m. , gry , c. @jnla&a , 375 , 805 bland - hawthorn , j. , maloney , p. r. , 1999 , , 510 , l33 bland - hawthorn , j. , maloney , p. r. , 2001 , , 550 , l231 sethi , s. k. , nath , b. b. , subramanian , k. , 2008 , @jnlmnras , accepted cowling , t. g. , 1956 , @jnlmnras , 116 , 114 draine , b. t. , 1980 , , 241 , 1021 shang , h. , glassgold , a. e. , shu , f. h. , lizano , s. , 2002 , , 564 , 853 giallongo , e. , cristiani , s. , dodorico , s. , fontana , a. , 2002 , , 568 , l9 malkan , m. , webb , w. , konopacky , q. , 2003 , , 598 , 878 , a. , lanzetta , k. m. , chen , h .- w . , 2003 , @jnlmnras , 342 , 1215 inoue , a. k. , iwata , i. , deharveng , j .- m . , buat , v. , burgarella , d. , 2005 , @jnla&a , 435 , 471 pinto , c. , galli , d. , 2008 , @jnla&a , submitted jedamzik , k. , katalini , v. , olinto , a. v. , 1998 , , 57 , 3264 subramanian , k. , barrow , j. d. , 1998 , , 58 , 83502 seshadri , t. r. , subramanian , k. , 2001 , , 87 , 101301 olesen , p. , 1997 , phys . lett . b , 398 , 321 shiromizu , t. , 1998 , phys . b , 443 , 127 christensson , m. , hindmarsh , m. , brandenburg , a. , 2001 , , 64 , 56405 banerjee , r. , jedamzik , k. , 2003 , , 91 , 251301 - 1 mller , w. , biskamp , d. , 2000 , , 84 , 475 banerjee , r. , jedamzik , k. , 2004 , , 70 , 123003 olive , k. a. , 2008 , advances , in space research , 42 , 581 schleicher , d. r. g. , banerjee , r. , klessen , r. s. , 2008 , , submitted ( eprint arxiv:0809.1519 ) shull , j. m. , van steenberg , m. e. , 1985 , , 298 , 268 zhang , l. , chen , x. , lei , y .- a . , si , z .- g . , 2006 , , 74 , 103519 zhang , l. , chen , x. , kamionkowski , m. , si , z .- g . , 2007 , , 76 , 061301(r ) gamgami , f. , 2007 , phd thesis , university of heidelberg bromm , v. , kudritzki , r. p. , loeb , a. , 2001 , , 552 , 464 tornatore , l. , ferrara , a. , schneider , r. , 2007 , @jnlmnras , 382 , 945 chuzhoy , l. , 2008 , apjl , 679 , 65 cumberbatch , d. t. , lattanzi , m. , silk , j. , 2008 , @jnlmnras , submitted schleicher , d. r. g. , glover , s. c. o. , banerjee , r. , klessen , r. s. , 2008 , , submitted ( eprint arxiv:0809.1523 ) mack , g. d. , jacques , t. d. , beacom , j. f. , bell , n. f. , yuksel , h. , 2008 , , accepted xu , h. , oshea , b. w. o. , collins , d. c. , norman , m. l. , li , h. , li , s. , 2008 , @jnlapj , submitted basu , k. ; hernndez - monteagudo , c. ; sunyaev , r. a. , 2004 , a&a , 416 , 447 schleicher , d. r. g. , galli , d. , palla , f. , camenzind , m. , klessen , r. , bartelmann , m. , glover , s. c. o. , 2008 , @jnla&a , 490 , 521

a pair of counter - propagating majorana edge modes appears in chiral _ p_-wave superconductors and in other superconducting systems belonging to the same universality class . these modes can be described by an ising conformal field theory . we show how a superconducting flux qubit attached to such a system couples to the two chiral edge modes via the disorder field of the ising model . due to this coupling , measuring the back - action of the edge states on the qubit allows to probe the properties of majorana edge modes .

1101.0604

chiral majorana fermion edge states were originally predicted to exist in the @xmath0 fractional quantum hall plateau . @xcite these edge states support not only neutral fermionic excitations but also more exotic edge vortices . a single edge vortex corresponds to a @xmath1 phase shift to all fermions situated to one side of it . @xcite two edge vortices may either fuse into an edge fermion or annihilate each other , with the outcome depending on the preceding evolution of the system . in other words , the edge theory ( together with the corresponding bulk theory ) possesses non - abelian statistics . @xcite this unusual physics and its potential applications to topological quantum computation are the reasons why the majorana edge states have attracted much attention recently . @xcite similar non - abelian anyons and their corresponding edge states appear in superconducting systems as well . initially it was discovered that _ p_-wave superconductors support non - abelian anyons in the bulk and chiral majorana edge states . @xcite later it was shown that depositing a conventional @xmath2-wave superconductor on the surface of a topological insulator while breaking time - reversal symmetry provides an alternative route to realize these non - abelian states . @xcite alternative proposals include substituting the topological insulator by a two - dimensional electron gas with spin - orbit coupling @xcite or by a half - metal . @xcite the realizations of majorana edge states using _ s_-wave superconductors have the following advantages : first , they rely on combining simple , well - studied ingredients . second , the materials do not have to be extremely pure unlike samples needed to support the fractional quantum hall edge states . finally , the superconducting implementations of majorana fermions may feature a larger bulk excitation gap and may therefore be operated at higher temperatures . the downside of the superconducting implementations of majorana edge states is the lack of means to manipulate edge vortices . @xcite different from the 5/2 fractional quantum hall state , the edge vortices are not coupled to charge and thus can not be controlled by applying voltages . @xcite therefore , the standard proposal to probe the edge vortices in superconducting systems is to inject fermion excitations into the edge , to let them split into edge vortices , and finally to conclude about the behavior of the edge vortices from the detection of the fermion excitations after the subsequent fusion of edge vortices.@xcite in this paper , we propose a more direct way to manipulate and measure edge vortices using a flux qubit consisting of a superconducting ring interrupted by a josephson junction . @xcite our main idea is based on the following observations : first , an edge vortex is created when a superconducting vortex crosses the edge . second , the motion of the superconducting vortices can be fully controlled by a flux qubit , since by applying a flux bias to the qubit one can tune the energy cost for a vortex being present in the superconducting ring . @xcite in this way , attaching a flux qubit to a system supporting majorana edge states allows to directly create , control , and measure edge vortices without relying on splitting and fusing fermionic excitations . we note that our proposal is not necessarily advantageous for the purposes of topological quantum computing since quantum computing with majorana fermions may even be realized without ever using edge states.@xcite instead the aim of our investigation is to develop a better tool for probing the fractional excitations of the edge theory . the paper is organized as follows . in sec . [ sec : setup ] , we discuss a schematic setup of a system where a pair of chiral majorana fermion edge modes couple to a flux qubit as a probe of the edge states and briefly list our main findings . in sec . [ sec : ising ] , we review the connection between the one - dimensional critical transverse - field ising model and majorana fermion modes . we identify the vortex tunneling operators between two edge states as the disorder fields of the ising model , and subsequently derive an effective hamiltonian for the flux qubit coupled to majorana modes . in sec . [ sec : formalism ] , we provide the necessary formalism for evaluating the expectation values for the flux qubit state and qubit susceptibilities . in sec . [ sec : expectation - value ] and sec . [ sec : corr - qubit ] , we compute the qubit expectation values and the two - point qubit correlation functions in the presence of the edge state coupling , and use these results to derive the qubit susceptibility . in sec . [ sec : higher ] , we analyze higher order corrections to correlation functions of the qubit state . we summarize our results in sec . [ sec : conclusion ] . additionally , we provide a brief overview of the flux qubit hamiltonian in app . [ app : flux - qubit ] . in app . [ app : eff - two - level ] we reduce the flux qubit hamiltonian to that of a two - level system and derive the coupling between the flux qubit and the majorana modes . in app . [ app : correlation - function ] , we give the form of the four point correlation function for the disorder field of the ising model . finally in app . [ app:2nd - order - xx ] , we provide the detailed derivation of the higher order corrections to the correlation functions of the qubit state . in this work , we consider the following setup : a strip of _ s_-wave superconductor is deposited on the surface of either a three - dimensional topological insulator or a semiconductor with strong spin - orbit coupling and broken time - reversal symmetry ( or any other superconducting setup supporting majorana edge states ) . as depicted in fig . [ fig : setup ] , a pair of counter - propagating majorana fermion edge modes appears at the two opposite edges of the superconductor . @xcite to avoid mixing between counter - propagating edge states , the width of the superconductor should be much larger than the superconducting coherence length @xmath3 . here and in the following , @xmath4 denotes the fermi velocity of the topological insulator ( semiconductor ) and @xmath5 the proximity - induced superconducting pair - potential . in order to avoid mixing of the two counter - propagating edge modes at the ends of the sample , we require the length of the superconducting strip to be longer than the dephasing length . a flux qubit , consisting of a superconducting ring with a small inductance interrupted by a josephson junction , is attached to the heterostructure supporting the majorana edge modes , as shown in fig . [ fig : setup ] . by applying an external flux @xmath6 , the two classical states of the superconducting ring corresponding to the phase difference of @xmath7 and @xmath8 across the junction can be tuned to be almost degenerate . @xcite in this regime , the flux qubit can be viewed as a quantum two - level system with an energy difference @xmath9 ( which we choose to be positive ) between the states @xmath10 and @xmath11 and a tunneling amplitude @xmath12 between them . as described in app . [ app : flux - qubit ] , the energy difference @xmath9 can be easily tuned by the external flux @xmath6 threaded through the ring . the transition between the two qubit states is equivalent to the process of a vortex tunneling through the josephson junction in or out of the superconducting ring . for convenience , we will refer to the hilbert space spanned by the qubit states @xmath10 and @xmath13 as a spin-@xmath14 system . for example , we are going to call the pauli matrices @xmath15 acting on the qubit states the qubit spin . a vortex tunneling through the weak link in the superconductor from one edge to the other is a phase slip of @xmath16 of the superconducting phase difference at the tunneling point . due to this event , all fermions to one side of the weak link gain a phase of @xmath1 . as will be shown below , the vortex tunneling operator can be identified with the operator of the disorder field of a one - dimensional critical ising model onto which the majorana edge modes can be mapped . since vortex tunneling events couple the qubit spin to the majorana edge modes , we expect various observables of the qubit to carry signatures of this coupling . the main theory parameter that we are after is the scaling dimension @xmath17 of the edge vortex operator ( disorder field ) . our main results apply to the regime when vortex tunneling is weak @xmath18 . we find that the reduction of the spin expectation value in the @xmath19-direction due to the vortex tunneling acquires a nontrivial scaling exponent @xmath20 similarly , the spin expectation value along the @xmath21-direction is proportional to @xmath22 thereby probing the scaling dimension of the disorder field . the finite frequency susceptibilities that characterize the response of the polarization of the qubit spin to a perturbation with frequency @xmath23 provide additional information about the majorana edge states . the susceptibility @xmath24 , which characterizes the change of @xmath25 due to a modulation of @xmath26 with frequency @xmath23 , is measurable with current experimental techniques . it can be measured by modulating the external flux @xmath6 and reading out the current from a dc - squid coupled to the qubit . @xcite the frequency dependence of the susceptibilities exhibits a non - lorentzian resonant response around the frequency @xmath27 ( here and in the following , we set @xmath28 ) . it is modified by the coupling to the majorana edge states and shows the scaling behavior @xmath29 , and the distance @xmath30 from the resonance is larger than the width of the resonance . the phase change of susceptibility at the resonance @xmath31 is different from the @xmath1 phase change for a usual oscillator . the origin of the extra @xmath32 phase shift is the abelian part of the statistical angle of the vortex excitations . the flux qubit has two low energy states , corresponding to a phase difference @xmath33 or @xmath34 across the josephson junction at @xmath35 . the hamiltonian of the qubit is given by @xmath36 the energy difference @xmath9 can be tuned by applying an external flux to the qubit while the tunneling amplitude @xmath37 can be manipulated by changing the josephson coupling of the junction . @xcite as discussed in app . [ app : flux - qubit ] , the two levels described in represent the two lowest energy states localized at the two energy minima of a double well potential . in order for the two - level approximation to be accurate , the energies @xmath38 as well as the driving frequency @xmath23 have to be much smaller than the level spacing at each well . the tunneling phase @xmath39 is proportional to the charge induced on the sides of the junction and its fluctuations are the main source of qubit decoherence . for simplicity we neglect the charge noise so that we can assume that @xmath39 is static and set it to zero without loss of generality . the qubit hamiltonian now reads @xmath40 when there is no phase difference across the josephson junction ( @xmath33 ) , the hamiltonian of the chiral majorana modes appearing at the edges of the superconductor , as shown in fig . [ fig : setup ] , reads @xmath41,\ ] ] where @xmath42 is the velocity of the majorana modes , and @xmath43 and @xmath44 are the majorana fermion fields at the upper and lower edges of the superconductor in fig . [ fig : setup ] . the sign difference between the terms containing @xmath45 and @xmath46 is due to the fact that the modes are counter - propagating . the majorana fermion fields obey the anti - commutation relations @xmath47 a vortex tunneling through the weak link at @xmath35 advances the phase of each cooper pair in the region @xmath48 by @xmath8 . for majorana fermions , just like any other fermions , this results in phase shift of @xmath1 . the effect of this phase shift is a gauge transformation @xmath49 where the parity operator @xmath50 is given by @xmath51,\ ] ] with the fermion density @xmath52 and @xmath53 . we refer to app . [ app : eff - two - level ] for a derivation of the qubit hamiltonian and the gauge transformation eq . . when the phase difference between two sides of the josephson junction is exactly @xmath1 , the majorana modes approaching the junction are fully reflected . @xcite since this phenomenon occurs only very close to the phase difference of @xmath1 , where the system only spends a short amount of time during the process of a phase slip , we will neglect the effect of this backscattering . the relation between the phase slip and the parity operator was discussed and used in previous work focusing on the @xmath0 fractional quantum hall state . @xcite combining the hamiltonian of the majorana edge states ( [ eq : h - mf ] , [ eq : h-2pi ] ) with the qubit hamiltonian , we get the full hamiltonian of the coupled system in the basis of @xmath10 and @xmath11 : @xmath54 the first part of hamiltonian represents the chiral majorana edge states coupled to the phase slip of the superconductor while the second part is the bare flux qubit hamiltonian . because the parity operator is highly nonlocal if expressed in terms of majorana fermions , it is desirable to map the majorana modes on a system where the vortex tunneling event becomes a local operator . to this end , we establish the equivalence of the chiral majorana edge modes with the long wavelength limit of the one - dimensional transverse - field ising model at its critical point . @xcite the lattice hamiltonian of the ising model at the critical point is given by @xcite @xmath55 where @xmath56 are the spin-@xmath14 operators at site @xmath57 . with the jordan - wigner transformation , @xmath58 the ising model can be cast in terms of fermions as @xmath59.\ ] ] here @xmath60 obey the usual onsite spin commutation relations while the fermions operators @xmath61 and @xmath62 obey canonical anti - commutation relations . for each fermion , we introduce a pair of majorana operators @xmath63 and @xmath64 such that @xmath65 the majorana fermions satisfy the clifford algebra @xmath66 in terms of the majorana operators , the hamiltonian assumes the form @xmath67 in the long wavelength limit , the hamiltonian reduces to with the identification of the continuum majorana operators @xmath68 and the velocity @xmath69 . to complete the mapping , the bandwidth of the ising model should be related to the cutoff energy @xmath70 of the linear dispersion of the majorana edge states , @xmath71 . thereby , a pair of counter - propagating majorana edge states , @xmath43 and @xmath44 , can be mapped on the low energy sector of the one - dimensional transverse - field ising model at its critical point . for the parity operator , we obtain a representation in terms of the ising model with the following procedure : we first discretize @xmath72 using the mapping and identify @xmath73 as a lattice point on the ising model . thereafter , we obtain an expression for the vortex tunneling operator @xmath50 in terms of the ising model @xmath74 by using eq . and the jordan - wigner transformation . here , @xmath75 is the disorder field of the ising model , i.e. , the dual field of the spin field.@xcite the ising hamiltonian has a form identical to eq . when expressed through @xmath76 operators , @xmath77 with @xmath78 . @xcite we see that the parity operator is indeed a local operator in the dual description of the ising model . after mapping on the ising model eq . becomes ( here and in the following , we use the shortcut notation @xmath79 ) @xmath80 and the full hamiltonian of majorana edge states and the flux qubit maps onto @xmath81 finally , an additional unitary transformation @xmath82 yields @xmath83 here , @xmath84 are the pauli matrices acting in the hilbert space spanned by @xmath10 and @xmath85 . the operators of the qubit spin can be expressed through @xmath86 as @xmath87 we use the hamiltonian in the form of eq . and the qubit spin operators in the rest of the paper . the way of identifying _ two _ edge majorana states with a _ complete _ transverse field ising model presented above is different from the one commonly used in preceding research . usually , the _ chiral part _ of the ising model is identified with a _ majorana edge.@xcite the advantages of our method are the possibility to write a complete hamiltonian of the problem and simplified book - keeping , while its drawback is the need for the right - moving edge and the left - moving edge to have the same geometries . overall the differences are not important and both methods can be used interchangeably . to probe the universal properties of majorana edge states , the energy scales of the qubit should be much smaller than the cutoff scale of the ising model , @xmath88 . in the weak coupling limit @xmath18 , we construct a perturbation theory in @xmath89 by separating the hamiltonian @xmath90 into an unperturbed part and a perturbation @xmath91 without loss of generality we set @xmath92 , so that the ground state of the unperturbed qubit is @xmath10 . for brevity we omit the spatial coordinate of the @xmath76 operator in the following since it is always the same in the setup that we consider . we use the interaction picture with time - dependent operators @xmath93 the perturbation @xmath94 in this picture is given by @xmath95,\ ] ] where @xmath96 are the time - dependent raising and lowering operators . the structure of the raising and lowering operators leads to physics similar to the kondo and luttinger liquid resonant tunneling problems . @xcite in the calculation we need the real - time two - point and four - point correlation functions of @xmath76 in the long - time limit @xmath97 . the two - point correlation function is @xmath98 where sgn@xmath99 denotes the sign of @xmath21 , and @xmath100 the scaling dimension of the @xmath76 field . @xcite the phase shift @xmath101 of the two - point correlator is the abelian part of the statistical angle for the ising anyons braiding rules . @xcite correlation functions involving combination of multiple fields can be obtained via the underlying ising conformal field theory or via a bosonization scheme . @xcite the expression for the four - point correlation function is given in app . [ app : correlation - function ] due to its length . for brevity we will measure energies in units of @xmath70 and times in units of @xmath102 in the following calculation and restore the dimensionality in the final result . we are interested in observables of the flux qubit : the spin expectation values and the spin susceptibilities . we use time - dependent perturbation theory to calculate these quantities . @xcite this method is straightforward because of the simple form of the perturbing hamiltonian in terms of raising and lowering operators . assuming that the system is in the unperturbed ground state at time @xmath103 , the expectation value of a qubit spin operator @xmath104 is expressed through the s - matrix @xmath105 , @xmath106 here , @xmath107 is the time - ordering operator and @xmath108 is the expectation value with respect to the unperturbed ground state . similarly , the two - point correlation functions of the qubit spin are given by @xmath109 the perturbative calculation for both the expectation values and correlation functions is done by expanding the @xmath110-matrices in @xmath111 order by order . this procedure is equivalent to the schwinger - keldysh formalism with the expansion of @xmath110 and @xmath112 corresponding to insertions on the forward and backward keldysh contour . according to linear response theory , the susceptibility is given by the fourier transform of the retarded correlation function of the qubit@xcite @xmath113 \rangle_c \\ = - 2 \int_{0}^{\infty } \!dt\ , e^{i \omega t } { \rm i m } \langle \sigma^{\alpha}(t ) \sigma^{\beta}(0 ) \rangle_c,\end{gathered}\ ] ] where @xmath114 denotes the cumulant , @xmath115 and we have used @xmath116 . we see that in order to calculate the susceptibilities only the imaginary part of the correlation functions for @xmath117 is required . in this section , we calculate the expectation values of the qubit spin due to coupling with the majorana edge states to the lowest non - vanishing order . using the identity @xmath118 we obtain @xmath119 since @xmath120 . the first non - vanishing correction in the perturbative calculation of @xmath121 is of second order in @xmath122 . by expanding @xmath110 and @xmath123 in eq . , we obtain @xmath124 the integrand @xmath125 originates from the first order expansion of both @xmath110 and @xmath123 . the second order contributions from the same @xmath110- or @xmath123-matrix vanish due to the structure of @xmath122 in the qubit spin space . substituting and into the integrand @xmath126 yields @xmath127 by evaluating the integral in eq . ( [ eq : integral - sz ] ) , we find @xmath128 where @xmath129 denotes the gamma function . the expectation value of @xmath130 in the unperturbed ground state vanishes . the first non - vanishing contribution to @xmath131 arises to first order in @xmath89 . expanding @xmath110 and @xmath123 in eq . to the first order yields @xmath132 \rangle_0 = \frac{\sin(-\varepsilon t_1 + \tfrac{\pi}{8 } ) \delta}{|t_1|^{2 \delta_\mu } } \nonumber\end{gathered}\ ] ] after substituting @xmath130 from eq . and employing the two point correlator , eq . . evaluating , we find @xmath133 finally , @xmath134 to all orders in perturbation theory since the hamiltonian is invariant under @xmath135 . since we are interested in the behavior of susceptibilities at frequencies close to the resonance @xmath27 , we only need to obtain the long - time asymptotic of the correlation functions of the qubit spin . using and , we immediately obtain that @xmath136 is non - vanishing to zeroth order . this is due to the fact that flipping the qubit spin automatically involves creation of an edge vortex , and @xmath137 is exactly the spin flip operator . in the same manner , one obtains that @xmath138 to zeroth order . concentrating next on the mixed correlator , the relations and yield @xmath139 the leading non - vanishing term in this correlation function is of first order in @xmath12 and given by @xmath140 in the long - time limit . the leading order contribution to @xmath141 can be evaluated using with expansions of @xmath110 and @xmath123 to second order in @xmath12 . in the long - time limit , the leading contribution of the correlation function is given by @xmath142 correlators containing a single @xmath143 vanish because of the invariance under @xmath135 . we see that all the non - vanishing two - point correlation functions are the same up to overall prefactors . therefore , we will focus on @xmath144 in the following . the coupling of the flux qubit to the continuum majorana edge states can be thought of as a two - level system coupled to an environment via the interaction . this coupling leads to self - energy corrections @xmath145 for the qubit hamiltonian @xmath146 that effectively shifts the energy spectrum and can also induce damping . @xcite since we are only interested in qubit observables , we focus on the structure of @xmath145 for the two - level system and do not discuss the self - energy correction of the majorana edge states . to second order , the self - energy correction for two spin states can be written in terms of the perturbed hamiltonian as @xcite @xmath147 where @xmath148 is the energy for the spin-@xmath149 qubit states and @xmath150 indicates that the ising model is in its ground state with spin-@xmath39 for the qubit state . due to the structure of the hamiltonian , the first order correction to the self - energy vanishes additionally , the off - diagonal self - energy corrections vanish also to second order . by inserting a complete set @xmath151 of the hilbert space of @xmath152 with @xmath153 denoting the complete set of eigenstates with energy @xmath153 for the ising sector , the diagonal elements of the self - energy become @xmath154 because @xmath155 , only terms with @xmath156 give non - vanishing contributions such that @xmath157 where @xmath158 corresponds to @xmath159 , and @xmath160 to @xmath161 . the diagonal elements of the self - energy in eq . can be cast to the form @xmath162 to see that is equal to , we first insert a complete set of states of the ising model , then write the time evolution of @xmath76 in the heisenberg picture , and finally evaluate the integral . evaluating eq . with eq . yields @xmath163 where we have used @xmath164 . the absence of the imaginary part for @xmath165 indicates that the spin - up state is stable . the self - energy thus gives an energy shift to the spin - up state while it gives an energy shift with a damping to the spin - down state , @xmath166 the energy renormalization and damping alter the time evolution of the ground state correlation function @xmath167 where the energy renormalization and damping @xmath168 are given by @xmath169 at zero temperature , this correlator is the only non - vanishing qubit correlator that enters in the perturbative calculation . therefore , the effect of the self energy can be captured by replacing @xmath170 in the qubit correlation functions computed in the long - time limit excluding the self - energy correction . using the replacement rule , one obtains the zero temperature correlator @xmath171 the energy renormalization and the induced damping do not arise explicitly in the lowest - order perturbation and require the resummation of the most divergent contributions to all orders in perturbation theory . in a system where wick s theorem applies , the resummation for the self - energy can be derived explicitly from a diagrammatic perturbation scheme . @xcite because the correlation functions of multiple @xmath76 s do not obey the wick s theorem ( see app . [ app : correlation - function ] ) , the resummation procedure for our system becomes more complicated . in the long time limit , however , the most divergent contributions in all orders can be collected by using the operator product expansion for two @xmath76 fields that resembles the structure of the wick s theorem . @xcite besides @xmath172 , finite temperature is an alternative source of decoherence . the finite temperature correlators of disorder fields are readily obtained from the zero temperature correlators using a conformal transformation @xcite @xmath173^{2 \delta_\mu } } , \ ] ] where @xmath174 denotes temperature and @xmath175 the boltzmann constant . the finite temperature correlator @xmath176 in the long - time limit can be obtained by substituting eq . into with the proviso @xmath177 such that the temperature has no direct effect on the qubit dynamics . with the correlation functions derived above , we are now in the position to evaluate susceptibilities of the qubit . we should keep in mind that these correlators are valid only in the long - time limit and can only be used to study the behavior of the susceptibilities close to the resonant frequency @xmath27 . evaluating eq . with eq . yields the susceptibility at zero temperature around the resonance , @xmath178^{1- 2 \delta_\mu}},\ ] ] where @xmath179 and @xmath172 are given in . here , we note that the susceptibility shows non - lorentzian response . this is in contrast to the conventional lorentzian response of a two - level system weakly coupled to the environment . @xcite if we neglect @xmath179 and @xmath172 , which are of higher order in @xmath89 , this susceptibility reduces to @xmath180 so it diverges and changes the phase by @xmath181 at the resonant frequency . we can attribute this phase change to the phase shift of the correlator of two disorder fields in eq . . the presence of damping @xmath172 in eq . provides a cutoff for the divergence of the response on resonance . the maximal susceptibility is reached at @xmath182 , and its value is given by @xmath183 using the proportionality of the correlation functions and , one gets that @xmath184 and @xmath185 . it is interesting to note that when @xmath186 both @xmath187 and @xmath188 are divergent while @xmath189 vanishes at the resonance . in fig . [ fig : sus - xx ] , the absolute value of the susceptibility @xmath190 close to the resonance is plotted as a function of frequency . the dotted line shows the modulus of eq . for @xmath191 while the dashed line shows that of eq . . a renormalization of the resonant frequency @xmath179 becomes clearly visible when comparing the peak position of the dashed line to that of the dotted line . the conformal dimension of the vortex excitation can be measured in the region with @xmath192 where latexmath:[\[\label{eq : abs - chi - xx - scaling - frequency } @xmath188 and @xmath189 exhibit the same scaling behavior . as a function of frequency @xmath23 close to resonance @xmath9 . the dotted line shows the zero temperature susceptibility in the absence of the damping and energy renormalization while the dashed line shows the result in the presence of the energy shift and the damping in eq . . the parameters used for the plot are @xmath194 and @xmath195 . the solid line shows a plot of the finite temperature susceptibility with @xmath196 . ] the finite temperature susceptibility of @xmath197 can be evaluated from the correlation function subjected to the transformation . the result is plotted as the solid line in fig . [ fig : sus - xx ] . an immediate effect of the temperature is that it also introduces a cutoff for the divergence on resonance . for instance , the resonance peak of the susceptibility yields a different scaling behavior with respect to the temperature @xmath198 as long as @xmath199 . the zero temperature scaling behavior of the resonance peak will be masked by a finite temperature with a crossover at @xmath200 . these scaling and crossover behaviors of the resonance strength are features of the coupling of the majorana edge states and the flux qubit . @xcite the finite temperature susceptibility shows a resonance at @xmath201 , as shown in fig . [ fig : sus - xx ] . around the resonance , the frequency dependence at finite temperature will be given by the power law but with the region constrained by @xmath202 instead of @xmath172 if @xmath203 . so far , we have computed the qubit susceptibilities to their first non - vanishing orders and the lowest order self - energy correction @xmath204 . as a consequence , we only used the two - point correlation functions @xmath205 in our evaluations . the next nontrivial corrections to the qubit correlators involve the equal position four - point correlator of the disorder fields @xmath206 . as discussed in appendix [ app : correlation - function ] , the four - point correlator , in principle , contains information about the non - abelian statistics of the particles because changing the order of the fields in the correlation function not only alters the phase but can also change the functional form of the correlator . @xcite it is thus interesting to go beyond the lowest non - vanishing order . additionally , doing so allows to check the consistency of the calculation of the self - energy correction done in sec . [ subsec : energy - shift ] . as an example we focus on the second order correction to the @xmath207 correlator in the long - time limit . the details of the calculation are given in app . [ app:2nd - order - xx ] and the result in eq . . the dominant correction is a power law divergence @xmath208,\ ] ] which is just the second order in @xmath12 expansion of the modified correlation function @xmath209 hence , we confirm that the second order perturbative correction is consistent with the the self - energy correction calculation . the leading correction to the susceptibility @xmath187 in second order is due to the logarithmic term @xmath210 in the correlator and has the form @xmath211^{1 - 2 \delta_\mu } } \\ \times \ln \left(\frac { ( \gamma/2)^2 + ( \omega - \varepsilon-\nu)^2}{\varepsilon^2 } \right ) \end{split}\ ] ] where we have included the self - energy correction , and omitted terms without logarithmic divergence . unfortunately the effects of nontrivial exchange statistics of disorder fields are not apparent in this correction . we have proposed a novel scheme to probe the edge vortex excitations of chiral majorana fermion edge states realized in superconducting systems utilizing a flux qubit . to analyze the coupling we mapped the hamiltonian of the majorana edge states on the transverse - field ising model , so that the coupling between the qubit and the majorana edge modes becomes a local operator . in the weak coupling regime @xmath212 we have found that the ground state expectation values of the qubit spin are given by @xmath213 additionally , the susceptibility tensor of the qubit spin in the basis @xmath214 is given by @xmath215^{1- 2 \delta_\mu } \lambda^{2\delta_\mu}},\end{gathered}\ ] ] with the real part @xmath179 and the imaginary part @xmath216 of the self - energy given by @xmath217 we see that all of these quantities acquire additional anomalous scaling @xmath218 due to the fact that each spin flip of the qubit spin couples to a disorder field @xmath76 . similar scaling with temperature appears in interferometric setups,@xcite but using a flux qubit allows to attribute its origin to the dynamics of vortices much more easily and also gives additional tunability of the strength of the coupling . another effect of the vortex tunneling being present is the phase change @xmath219 of the susceptibility around the resonance.@xcite this phase shift occurs due to the anomalous scaling and the presence of the abelian statistical angle of the disorder field , in view of the fact that @xmath187 is just a correlator of two disorder fields in the frequency domain . the long wavelength theory which we used is only applicable when all of the energy scales are much smaller than the cutoff energy of the majorana modes this is an important constraint for the flux qubit coupled to the majorana edge states . in systems where the time - reversal symmetry is broken in the bulk ( unlike for topological insulator - based proposals @xcite ) , the velocity of the majorana edge states can be estimated to be @xmath220 and the dispersion stays approximately linear all the way up to @xmath5 . the cutoff of the majorana modes is related to the energy scale of the ising model @xmath221 . equating @xmath222 and @xmath223 , we obtain the lattice constant of the ising model @xmath224 , with @xmath225 the fermi wavelength . the fermi wavelength is typically smaller than any other length scale , and so the long wavelength approximation we have used is well - justified . for a typical flux qubit the tunneling strength @xmath12 is indeed much smaller than the superconducting gap , the level splitting @xmath9 may vary from zero to quantities much larger than the superconducting gap . our proposal provides a way to measure properties of the non - abelian edge vortex excitations different from the conventional detection scheme that requires fusing vortices into fermion excitations . however , none of our results for the single flux qubit can be directly connected to the non - abelian statistics of the quasiparticles , even after including higher - order corrections . thus , it is of interest for future research to investigate a system where the edge vortex excitations are coupled to two qubits such that braiding of vortex excitations can be probed . @xcite another feature of systems with several qubits worth to investigate is the ability of the majorana edge modes to mediate entanglement between different flux qubits . we thank c. w. j. beenakker for useful discussions . this research was supported by the dutch science foundation nwo / fom ( c .- y . h. , a. a. , and f. h. ) and the swedish research council ( vetenskapsrdet ) ( j. n. ) . the flux qubit which we consider consists of a superconducting ring interrupted by a josephson junction which is parameterized by its critical current @xmath226 , its capacitance @xmath227 , and the self - inductance @xmath228 of the ring threaded by a magnetic flux @xmath6 . the hamiltonian in the phase basis reads @xcite @xmath229 where @xmath230 is the phase difference across the josephson junction and @xmath231 is the superconducting flux quantum . we have introduced the charging energy @xmath232 , the josephson energy @xmath233 , and the inductive energy @xmath234 . the potential energy is given by the last two terms of the hamiltonian . neglecting for a moment the inductive energy , the cosine potential favors states with @xmath235 . the transition between these state involves a change of the phase difference by @xmath8 which corresponds to driving a vortex in or out of the superconducting loop . the inductive energy breaks the degeneracy of the states with a different number of vortices @xmath57 in the loop by favoring states with @xmath236 . when the flux @xmath6 is tuned close to @xmath237 , the system becomes frustrated since the states @xmath33 and @xmath238 are then nearly degenerate in energy . when the inductive energy is smaller than the josephson energy but still large enough such that states with more vortices in the superconducting loop are not accessible , the potential takes the form of a double - well with the minima close to @xmath7 and @xmath8 . these requirements are met when @xmath239 . the charging energy @xmath240 describes the influence of quantum dynamics . if the level spacing @xmath241 in each well is large enough and additionally the two wells are well separated , only the lowest energy states @xmath10 and @xmath11 , which are localized near the classical minima @xmath242 , are relevant . hence , the low energy hamiltonian of the system reduces to eq . . for @xmath243 , the energy detuning of the two minima is given by @xmath244 which can be tuned via the flux @xmath6 in the superconducting loop . the tunneling amplitude is given by @xmath245 . @xcite let us now discuss the experimental parameters for the flux qubit . assuming that the superconductor order parameter @xmath5 is about @xmath246 k , the corresponding coherence length is of the order of @xmath247 @xmath76 m . to avoid the mixing of the majorana edge states , the width of josephson junction needs to be larger than the coherence length which is in the range of micrometers this is consistent with most experiments . @xcite although the design of the flux qubit in ref . is more complicated than the simplest design discussed here , the idea of a @xmath8 phase shift for a full vortex tunneling through the josephson junction is the same . thus , as a concrete example , we quote the experimentally achieved parameters from ref . : @xmath248 ghz , @xmath249 ghz and @xmath250 ghz . the tunneling amplitude is measured and estimated to be @xmath251 mhz . @xcite moreover , the level spacing is estimated to be @xmath252 ghz . the euclidean action of the superconductor phase corresponding to the hamiltonian reads @xmath253.\ ] ] as discussed in app . [ app : flux - qubit ] , the double - well potential @xmath254 has two energy minima located at @xmath33 and @xmath16 such that @xmath255 and @xmath256 . for later convenience , we have shifted the potential energy by @xmath257 . in our discussion , we will assume that the level spacing @xmath258 is the same at both wells and that the potential profile connecting two minima can be approximated by @xmath259 . the concrete form of the potential does not affect the qualitative feature of our discussion . @xcite the action of the majorana fermions can be inferred from the hamiltonian as @xmath260,\ ] ] where @xmath261 and @xmath262 . the action describing the coupling between the phase field and the majorana fermions is given by @xmath263 where @xmath264 with @xmath265 is the fermion density of the majorana fermions . the origin of this coupling is the electrostatic energy @xmath266 where @xmath267 is the voltage from the josephson relation and @xmath268 is the charge of the superconductor island at one side of the josephson junction . @xcite here we have chosen a gauge such that the superconductor phase at the other side of the josephson junction is fixed . observe that the equation of motion of the phase field is not affected by the coupling term when the integration of the fermion density yields no explicit time dependence . the total action of the system thus becomes @xmath269 . from the euclidean ( imaginary time ) version of feynman s path integral , the transition rate reads @xmath270 \int [ d \phi ] e^{- s } , \ ] ] where @xmath271 is the corresponding hamiltonian , @xmath272 represent the initial and final phase eigenstates , and @xmath273 is the normalization constant . because the leading contribution to eq . ( [ eq : euclidean - path ] ) at large times @xmath274 comes from the lowest - lying energy eigenstates , the hamiltonian at the left hand side can be approximated by an effective hamiltonian that contains only a few low energy states . @xcite for the double well potential @xmath254 , there exist two low energy states that are localized at the two classical minima at @xmath33 and @xmath8 . by considering the transition rates within and between two minima @xmath275 for @xmath276 , we would like to show that the effective hamiltonian is a two - level system coupled to the majorana fermions . to compute transition rates for @xmath277 , we first observe that the phase field is mostly localized at one of the wells and behaves as a simple harmonic oscillator . therefore , the main contributions to the transition rates come from the phase field in the localized states and are given by @xmath278 for @xmath279 and @xmath8 states , respectively . @xcite notice that the phase field and majorana fermions are effectively decoupled when the phase field is localized . the other contributions to the transition rates come from trajectories of the phase field that contain tunneling events between two wells . these tunneling events are so called instantons and anti - instantons that occur in a very short time interval @xmath280 . in the dilute gas approximation , each instanton or anti - instanton event centered at time @xmath281 contributes to the transition rates with a factor @xmath282 here , @xmath283 is the action from the tunneling of the phase field through the barrier , and @xmath284 is due to the coupling with the approximation that the time interval of instanton @xmath285 is small such that the density field can be replaced by @xmath286 . since the integration of fermion density is an integer , we define @xmath287 . we also note that @xmath288 . moreover , we are not free to distribute the instantons and anti - instantons . they have to be alternated in time and the first tunneling event is determined by the initial state . @xcite let us consider the transition rate for @xmath289 . because the phase field needs to tunnel an even number of times in order to be back to the initial well , the number of ( anti-)instantons has to be even for a non - vanishing contribution . a trajectory of the phase field that contains @xmath290 ( anti-)instantons ordered in time , @xmath291 , gives the contribution to @xmath289 as @xmath292 which is integrated over the centers of ( anti-)instantons @xmath281 . here @xmath293 is the path integral summation over fermion fields such that @xmath294 \ ; o(\psi_u,\psi_d ) \ ; e^{-s_{\psi } } , \ ] ] for an arbitrary fermion field combination @xmath295 . the total transition rate can be written as @xmath296 where @xmath297 is defined in eq . . the transition rate of @xmath298 can be derived in the same manner and takes the same form as @xmath299 in eq . by summing over @xmath300 $ ] . the transition rates between two wells , @xmath301 and @xmath302 , can also be computed by properly counting the ( anti-)instanton events . the crucial difference is now that an odd number of tunneling events are needed for the final state to be in a different well than the initial state . the total transition rate of @xmath301 then reads @xmath303 by summing over contributions from trajectories with odd tunneling events @xmath304 finally , @xmath305 also takes the same form as @xmath301 with a substitution of @xmath306 $ ] in eq . . by using the interaction picture , we can explicitly show that the effective hamiltonian @xmath307 reproduces the transition rates within and between two wells in eqs . ( [ eq : r00-summed ] , [ eq : r-2pi-0-summed ] ) up to an overall constant , see eq . . here , @xmath308 is defined in eq . ( [ eq : h - mf ] ) and @xmath309 are pauli matrix acting on the two - level basis @xmath310 , @xmath311 , of the superconducting phase difference @xmath230 . @xcite note that @xmath312 enters in the hamiltonian together with @xmath50 . this is a consequence of gauge invariance whenever the superconducting phase difference changes by @xmath8 the phase of the majorana to the left of the junction has to be changed by @xmath1 . every physical observable has to be gauge invariant that is why @xmath312 ( and @xmath313 for that matter ) always have to occur together with @xmath50 . in this spirit , we define the ( observable ) qubit degrees of freedom as @xmath314 see . we thus conclude that the two - level hamiltonian together with the identification gives the effective low energy description of the system . the one - dimensional critical transverse - field ising model is a conformal field theory ( cft ) with central charge @xmath315 . this cft contains the following primary fields : @xmath316 , @xmath317 , @xmath2 , and @xmath76 . here @xmath316 is the identity operator , @xmath318 is the energy field ( a product of the right and left moving majorana fermion fields @xmath319 and @xmath320 ) , and @xmath2 is the ising spin field with its dual field @xmath76 . @xcite the dual field @xmath76 is also called the disorder field and has the same scaling behavior as the ising spin field @xmath2 at the critical point . on the lattice , the disorder fields @xmath76 are non - linear combinations of ising spin fields @xmath2 and reside on the bonds of lattice ising model . they are hence not independent of the ising spin field @xmath2 . in the continuum and in imaginary time , the two - point correlation function of disorder fields @xmath76 can be obtained from cft @xcite @xmath321^{\delta_\mu}},\ ] ] with @xmath322 and @xmath323 . following ref . , the real - time correlators can be obtained by analytical continuation @xmath324 . here @xmath325 is introduced to ensure the correct phase counting and is important for the abelian part of the statistics . the equal position two - point correlation function is given by @xmath326 by using the identity @xmath327 one obtains the two - point correlation function in the form of eq . . the four - point correlation function of @xmath76 s can be obtained in a similar manner . in imaginary time , the correlation function is given by @xcite @xmath328 where @xmath329 is the conformally invariant cross ratio , and the absolute values should be understood as @xmath330 . because we are interested in tunneling at a single point , we can set @xmath331 . in this limit the four - point correlation function can be evaluated to be @xmath332 the real - time correlation function can be obtained by first taking a square root of eq . followed by the analytical continuation , @xmath333 , @xcite @xmath334 \\ + & f_{13}(t_1,t_2,t_3,t_4 ) \left [ \theta(1234 ) + \theta(1432 ) + \theta(2143 ) + \theta(2341)+ \theta(3214)+\theta(3412)+ \theta(4123)+\theta(4321)\right ] \\ + & f_{14}(t_1,t_2,t_3,t_4 ) \left [ \theta(1243 ) + \theta(1342 ) + \theta(2134 ) + \theta(2431)+ \theta(3124)+\theta(3421)+ \theta(4213)+\theta(4312)\right ] , \end{split}\ ] ] where @xmath335 for @xmath336 and is otherwise zero . the corresponding functions @xmath337 are given by @xmath338^{1/4 } [ \xi + i ( t_3-t_4 ) ] ^{1/4 } } { [ \xi + i ( t_1-t_3 ) ] ^{1/4 } [ \xi + i ( t_1-t_4 ) ] ^{1/4 } [ \xi + i ( t_2-t_3 ) ] ^{1/4 } [ \xi + i ( t_2-t_4 ) ] ^{1/4 } } , \\ f_{13}(t_1,t_2,t_3,t_4 ) = \frac{[\xi + i ( t_1-t_3 ) ] ^{1/4 } [ \xi + i ( t_2-t_4 ) ] ^{1/4 } } { [ \xi + i ( t_1-t_2 ) ] ^{1/4 } [ \xi + i ( t_1-t_4 ) ] ^{1/4 } [ \xi + i ( t_2-t_3 ) ] ^{1/4 } [ \xi + i ( t_3-t_4 ) ] ^{1/4 } } , \\ f_{14}(t_1,t_2,t_3,t_4 ) = \frac{[\xi + i ( t_1-t_4 ) ] ^{1/4 } [ v + i ( t_2-t_3 ) ] ^{1/4 } } { [ \xi + i ( t_1-t_2 ) ] ^{1/4 } [ \xi + i ( t_1-t_3 ) ] ^{1/4 } [ \xi + i ( t_2-t_4 ) ] ^{1/4 } [ \xi + i ( t_3-t_4 ) ] ^{1/4 } } . \end{split}\ ] ] here @xmath339 , @xmath340 , and @xmath341 are the three characteristic functions appearing in the fourth - order correlation functions . for an abelian state , they usually appear in quasi - symmetric combinations and exchanging two of the times alters various phase factors , which is a characteristic of fractional statistics . for the current non - abelian case , however , exchanging two of the times not only alters phase factors but can also change the form of the correlation function from one of the characteristic functions to another . this is a special feature of non - abelian statistics . and @xmath343 coordinates used in appendix [ app:2nd - order - xx ] . ] because our ultimate goal is to compute the qubit susceptibility , we are interested in the correlator with @xmath344 in the long - time limit @xmath345 . let us first recall the perturbative part of hamiltonian in the interaction picture : @xmath346.\ ] ] since the vortex tunneling in or out of the superconducting ring directly couples to the disorder field of the ising model @xmath347 in the transformed basis , the evaluation of the second order correction for the correlator @xmath348 requires the knowledge of the four - point correlation function derived in appendix [ app : correlation - function ] . we expand the @xmath110 and @xmath123-matrices in to second order with insertions at times @xmath349 and @xmath343 . nonzero contributions to the correlator come from three regions : ( a ) @xmath350 , ( b ) @xmath351 and ( c ) @xmath352 . these three regions are shown in fig . [ fig : integral - domain ] . in what follows , we will evaluate the second order contributions from each region in the long - time limit . the contribution from region a is given by @xmath353 with the integrand @xmath354 where @xmath355 is a shorthand notation . the plus and minus signs come from the location of the insertions . the plus sign corresponds to having both insertions located on the same branch ( either forward @xmath110 or backward @xmath123 ) while the minus sign corresponds to the situation where the two insertions are located on different branches . because only certain orderings of insertions of raising and lowering operators @xmath356 or @xmath357 , coming both from the interaction term and the @xmath358 , give non - vanishing contributions , the integrand is given by @xmath359 here , the four - point correlation function can be read off from eq . and simplified using the identity . remarkably , these correlators have the same time dependence function and differ only by phase factors . this feature is characteristic also to regions b and c. after some algebra , the integrand simplifies to to evaluate the integral , we first simplify it by introducing new variables such that @xmath361 and @xmath362 with the new integrating domain @xmath363 and @xmath364 . the second order correction from region a becomes @xmath365 where we have introduced a regularization factor @xmath366 , with @xmath367 . the integral in eq . will not generate any oscillatory dependence but is divergent when both @xmath174 and @xmath368 are large . it is thus convenient to separate the algebraic part of the integrand into three parts @xmath369 combined with the exponential prefactor , the integration of @xmath370 is regular , the integral of @xmath371 diverges linearly while that of @xmath372 diverges logarithmically . integrating @xmath371 with all the exponential prefactors gives @xmath373 , \quad \eta \to 0^+\ ] ] since the the linear long time divergence is purely imaginary , it does not contribute to the correlation function . we now add the real parts of the three integrals , , and and then multiply them with the prefactors in . the result is the leading long - time contribution from region a to the qubit spin correlator : @xmath376 in the long - time limit , the leading contribution is given by the term @xmath377 . since the non - oscillatory contribution from is dominated by @xmath392 , we can expand the integrand around this point to get the leading contribution . because we are interested in the correlator in the long - time limit , we then deform the integration contour in the complex plane such that both @xmath384 and @xmath385 change from @xmath7 to @xmath393 . the leading non - oscillatory contribution is given by @xmath394 the oscillatory contribution @xmath390 is dominated by @xmath395 and @xmath396 , we can thus expand the integrand around this point to get the leading contribution . again we are interested in the correlator in the long - time limit and thus deform the integration contour such that @xmath384 varies from @xmath397 to @xmath246 and @xmath385 varies from @xmath7 to @xmath393 . after these transformations @xmath398 evaluates to @xmath399 where @xmath400 . summing up , the leading contributions to @xmath389 are @xmath401 the leading non - oscillatory contribution of @xmath389 is a constant while the leading oscillatory contribution has a power law decay @xmath402 . to integrate @xmath403 , we again use the variables @xmath386 and @xmath387 such that @xmath404 once again , the non - oscillatory contribution is dominated by @xmath392 . we expand the algebraic part of the integrand around @xmath405 , deform the integration contour such that @xmath384 runs from @xmath7 to @xmath393 and @xmath385 from @xmath7 to @xmath406 , and get @xmath407 to evaluate the oscillatory part @xmath408 of @xmath409 , we expand the integrand around @xmath410 and @xmath411 for the leading contribution . the necessary deformation of the integration contour is now given by @xmath384 changing from @xmath397 to @xmath246 and @xmath385 from @xmath7 to @xmath406 . the leading oscillatory contribution from eq . is now given by @xmath412 with @xmath413 . the final expression for @xmath409 is @xmath414 similarly to @xmath389 , the leading non - oscillatory contribution of @xmath409 is a constant , while the leading oscillatory contribution has a power law decay @xmath415 . we deform the integration contour in eq . such that @xmath368 goes from @xmath7 to @xmath428 and then back from @xmath429 to @xmath246 . the leading contribution in the long - time limit is dominated by the region near the real axis . the expansion around @xmath430 leads to an oscillatory contribution while the expansion around @xmath431 leads to a non - oscillatory contribution . to the lowest few orders , the asymptotic behavior in the long - time limit is given by @xmath432 the oscillatory contribution contains a power - law divergent @xmath433 term . as we discuss later , this term contributes to the shift of the resonant frequency and to the damping for the @xmath434 correlation function . to integrate @xmath435 , we first change the integration variables to @xmath368 and @xmath174 defined by @xmath436 and @xmath437 such that the integral separates into two parts @xmath438 after changing @xmath439 in the second integral and then introducing @xmath440 , this equation simplifies to @xmath441 again , we deform the integration contour in the integral over @xmath442 with @xmath442 changing from @xmath7 to @xmath443 and then from @xmath444 to @xmath246 . now the oscillatory contribution comes from @xmath445 while the non - oscillatory one from @xmath446 . by expanding the integrand around these two points , we get the leading contributions : @xmath447 the gaussian hypergeometric function evaluates to @xmath448 . the second order correction to the correlation function @xmath450 can be obtained by adding up the contributions from all the three regions , given by eqs . , and and then subtracting @xmath451 as calculated in eq . . the full expression for the correlator in the long - time limit is @xmath452 e^{-i \pi/8 } } { 8 \gamma(\tfrac{3}{4 } ) } + \frac{3 \cos ( \frac{\pi } { 8 } ) \gamma ( \tfrac{5}{8 } ) } { 4 \sqrt{2 } \gamma(\tfrac{11}{8 } ) } \right . \\ & \left . \qquad \qquad \qquad \qquad - \frac{2\gamma(\tfrac{7}{4})e^ { i \pi/8 } } { \sqrt{\pi } } + \frac{2 \cos ( \frac{\pi } { 8 } ) \ , _ 2f_1\left(-\frac{1}{4},\frac{3}{4};\frac{3}{2};-1\right ) \gamma ( \tfrac{7}{4})}{\sqrt{\pi } } \right\ } . \end{split}\ ] ] this result agrees well with numerical evaluation of the integral . a power law divergence @xmath453 and a logarithmic contribution @xmath454 dominate the long - time behavior of the correlator . however , this logarithmic contribution will be cut off either by the induced damping or by a finite temperature . a heuristic way to see that the term diverging as @xmath433 corresponds to self - energy correction is to add it to the zeroth order correlator of @xmath455 given by . the sum of these two terms equals to @xmath456 with @xmath179 and @xmath172 the same as in eq . . it then becomes apparent that is exactly the expansion of the renormalized correlator to the second order in @xmath12 @xmath457 we thus conclude that the explicit evaluation of the higher order correction gives a result consistent with the self - energy calculation . the leading contribution to the second order corrections comes from region @xmath227 with integration of @xmath458 when the integration variable @xmath368 is around @xmath459 , cf . eq . . since @xmath460 , this expansion to the zeroth order is equivalent to making an operator product expansion of @xmath461 for @xmath462 in the four - point correlation function of @xmath76 operators . @xcite in the n@xmath463 order of perturbation theory with insertion times @xmath464 , we expect that the most divergent contribution arises when all the insertion times belong to the interval @xmath465 $ ] . by ordering the times @xmath466 and using the operator product expansion for the pairs @xmath467 for @xmath468 , we get a perturbative structure resembling wick s theorem . the resummation of these terms would give the contributions for the self energy which we calculated in sec . [ subsec : energy - shift ] . in the present work , the jordan - wigner transformation is introduced for the ising spin fields @xmath469 . if the transformation is introduced for the disorder field , as in refs . and , one should interchange the ising spin field and the disorder field in our discussion . for a lorentzian response , the susceptibility is given by @xmath470 $ ] that leads to a phase change of @xmath1 , compared to @xmath471 in eq . ( [ eq : chi - xx-0nd - order ] ) , across the resonance . here , the deacy rate can generally be expressed as @xmath472 with @xmath473 for high temperature @xmath474 . @xcite in contrast to eq . ( [ eq : chi - xx - resonant - scaling - delta ] ) and eq . ( [ eq : chi - xx - resonant - scaling - t ] ) , the resonance peak for a lorentzian response thus scales as @xmath475 at zero temperature while it scales as @xmath476 at finite temperature . for topological insulator - based proposal with time - reversal symmetry in the bulk , cf . ref . , the velocity of majorana edge modes is further suppressed and is given by @xmath477 when @xmath478 . the cutoff energy for the linear dispersion is constrained to the region @xmath479 . in this case , we still get @xmath480 the fermi wavelength . the basis for the two - level system , @xmath481 and @xmath482 , used in the app . [ app : eff - two - level ] directly denotes the superconductor phase difference across the junction . this basis is in turn different from the basis @xmath10 and @xmath11 defined in the main text , right above eq . ( [ eq : h - mf - qubit - coupled ] ) . they are however related by @xmath483 and @xmath484 .

we discuss recent results in the study of the evolution of strongly coupled field theories in the presence of time dependent couplings using the holographic correspondence . the aim is to understand ( i ) thermalization and ( ii ) universal behavior when the coupling crosses a critical point . our emphasis is on situations where a subset of bulk fields can be treated in a probe approximation . we consider two different setups . in the first , defect conformal field theories are described by probe branes in ads space - times , and an initial vacuum state evolves due to a time dependent coupling in the probe sector . while a black hole formation is invisible in this approximation , we show that thermalization can nevertheless happen - this is signalled by formation of apparent horizons on the brane worldvolume . in the second setup , we consider a probe bulk scalar field in the background of a ads black brane . in equilibrium , this system undergoes a critical phase transition at some temperature when the source for the dual operator vanishes . for a time dependent source which goes across this critical point , we show that a zero mode of the bulk field dominates the dynamics and leads to scaling behavior of the order parameter as a function of the rate of change . uk/11 - 08 * holographic quantum quench * + sumit r. das + department of physics and astronomy , + university of kentucky , lexington , ky 40506 , usa

1111.7275

the problem of quantum quench , i.e. the response of a system to a time dependent coupling , has recently attracted a lot of attention in several areas of many - body physics , particularly because of progress in cold atom experiments @xcite,@xcite,@xcite . this problem is interesting for at least two reasons . the first relates to the question of thermalization . suppose we start with the ground state and then turn on a time dependent coupling which again approaches a constant at late times . does the system evolve into some kind of steady state ? if so , is the state `` thermal '' in any sense ? the second question deals with the situation where the quench takes place across a value of the parameter where there is an equilibriium critical point . in this case there is some evidence that the time evolution carries some universal features of the critical point . assuming that there is only one scale which governs the behavior of the system in the critical region , one can , e.g. derive scaling properties of one point functions following adaptations of the early work of kibble and zurek @xcite . suppose the coupling approaches the critical coupling linearly , i.e. ( g - g_c ) ~vt these arguments then show that the one point function of an operator with conformal dimension @xmath0 at the critical point has a scaling behavior @xcite , ~(v)^ f(tv^ ) where @xmath1 is the dynamical critical exponent and @xmath2 is the correlation length exponent . another manifestation of such universal behavior appears in @xmath3 dimensional field theories which are quenched _ suddenly _ to a critical point . in this case , powerful methods of boundary conformal field theory can be used to obtain the time dependence of correlation functions @xcite . for example the one point function of a generic operator with conformal dimension @xmath0 behaves as ~e^- where @xmath4 denotes a length scale which characterizes the initial theory away from criticality . @xmath4 is of course not universal , so neither is the relaxation time @xmath5 . however , the ratio of the relaxation times for two _ different _ operators @xmath6 and @xmath7 with conformal dimensions @xmath8 and @xmath9 is universal , = unlike equilibrium critical phenomena there is no general theoretical framework to understand such scaling relations . in particular , there are very few theoretical tools available to study such systems when they are strongly coupled . it is , therefore , natural to explore if the ads / cft correspondence @xcite - @xcite is useful in this problem . in this contribution we will summarize results obtained in this approach . the work on thermalization is in collaboration with tatsuma nishioka and tadashi takayanagi @xcite . the work on quench across critical points is in in collaboration with pallab basu @xcite . in the ads / cft correspondence , couplings of the boundary field theory are boundary values of a bulk field . in the regime where supergravity is valid , the problem of quantum quench then reduces to a classical problem with given initial and boundary conditions . this problem has been studied when the time dependent coupling is the boundary metric or the gauge theory coupling ( i.e. the boundary value of the dilaton ) . suppose this coupling is a constant in the far past and future , and has a smooth time dependent profile at intermediate times . in the bulk description this corresponds to a disturbance created on the boundary which propagates in the bulk . under suitable conditions , this leads to black hole formation in the bulk @xcite . the correlators at future time would then be thermal with a temperature characterized by the hawking temperature . the time scale after which this happens depends on the nature of the correlators , but turns out to be always smaller than what one would expect from a conformally invariant system evolving to a thermal state . thus , in this case thermalization of the field theory is signalled by black hole formation . for homogeneous planar collapse ( i.e. a space independent coupling in poincare patch ) , a black hole is always formed . in the case of homogeneous collapse in global ads , a black hole is formed when the rate of change is fast enough compared to the scale set by the radius of the sphere on which the boundary theory lives . in other situations , e.g. a slow variation of the coupling for global ads , a black hole is not formed so long as the supergravity approximation is valid . rather , if the coupling becomes _ weak _ at some time , the bulk string frame curvature grows large , leading to a breakdown of the supergravity approximation - thus mimicking a space - like singularity @xcite . for the case of a slow variation of the coupling it turns out that the gauge theory remains well defined and may be used to show that a smooth passage through this region of small coupling is possible without formation of a large black hole . related scenarios appear in @xcite and @xcite . one of our main aims is to study quantum quench across critical points . many such critical points can be studied in setups where a subset of the bulk fields can be treated in a probe approximation . in this approximation , these probe fields provide the essential physics and their backreaction to the background gravity can be ignored , typically suppressed by @xmath10 . this motivates us to study the problem of quantum quench in situations where such a probe approximation is valid . we will first consider defect field theories which arise as dual descriptions of a set of probe branes in the @xmath11 bulk @xcite . this approach has been used extensively to study flavor physics , as well as models with possible applications to condensed matter systems . the nice feature of this approach is that the boundary field theory is known , though they typically have supersymmetry . in this case a quantum quench of couplings in this subsector becomes a classical motion of these probe branes , with specified time dependent boundary conditions at the @xmath12 boundary @xcite . we investigate the question of thermalization in this context . since the background geometry is unchanged in this approximation any black hole which is formed due to the quench is not visible . we will find that thermalization is nevetheless visible - this manifests itself as the formation of an apparent horizon on the brane worldvolume . in the second setup we consider a `` bottom - up '' bulk theory of gravity with a neutral scalar field @xcite , where one writes down a bulk theory and _ assume _ that there is some dual field theory on the boundary . in this specific instance , the background is a @xmath13 charged black brane and the coupling of the scalar is large , so that its backreaction to the geometry is small . when the mass of the scalar lies in the range @xmath14 there is always a critical temperature below which the trivial solution is unstable . equivalently , for a given temperature , whenever the mass is below a certain value , the trivial solution is unstable . in this regime there is a new nontrivial static , stable solution whose `` non - normalizable '' part vanishes . this means that in the dual theory there is a new phase where the expectation value of the operator dual to this scalar is non - zero , even in the absence of any source . for any nonzero temperature there is a continuous phase transition with mean field exponents at the critical mass . at zero temperature ( i.e. when the background is an extremal brane ) the transition persists , but is of the berezinskii - kosterlitz - thouless type . this setup is similar to that of holographic superconductors @xcite and has been proposed as models for antiferromagentic transitions . we consider quench across this critical point by working at the critical mass , but turning on a time dependent source which crosses zero ( i.e. the critical point ) at some time @xcite . we show that the dynamics of the bulk scalar is dominated by a zero mode of the radial operator in the critical region when the rate of change of the source is small , this leads to a landau - ginsburg type dynamics with dynamical critical exponent @xmath15 , and a resulting scaling behavior of the order parameter . probe branes in the bulk of @xmath12 have been used to introduce flavor in the standard ads / cft correspondence . consider for concreteness @xmath16 whose dual is @xmath17 super - yang - mills in @xmath18 dimensions with gauge group @xmath19 . let us introduce @xmath20 dp branes which wrap a @xmath21 . possible supersymmetric wrappings are summarized in table ( [ tabl1 ] ) . .[tabl1]probe branes in @xmath16 . [ cols="<,^,>",options="header " , ] the dp branes give rise to new hypermultiplet fields . these live on the intersection of the dp branes with the d3 branes which gave rise to the @xmath16 geometry . from the point of view of the @xmath22 theory the hypermultiplets live on a lower dimensional defect . in the strong coupling regime , the bulk theory is the original supergravity together with the action of branes coupled to it . in the limit of @xmath23 the backreaction of the probe branes on the background @xmath16 geometry can be ignored and the entire bulk theory is given by the action of these branes moving in the fixed background geometry . we will take the brane action to be of the dbi type . in the dual theory this means we can consider the defect field theory by itself , and ignore the effect of hypermultiplet loops . as is standard in the ads / cft correspondence , the boundary values of the dbi fields are identified with sources for the dual operators in the dual field theory . consider for example the case of a d5 brane . let us write the @xmath16 metric in the form ds^2 = ( y^2+r^2)[-dt^2+dx_1 ^ 2+dx_2 ^ 2+dx_3 ^ 2]+ [ dr^2+r^2d_2 ^ 2+dy^2+y^2d(^_2)^2 ] [ one]the d5 brane is wrapped along @xmath24 . the fields in the dbi action are @xmath25 , which are the transverse coordinates to the brane . the value of @xmath26 is then the mass of the hypermultiplet fields coming from @xmath27 open strings joining the d5 brane with the stack of @xmath28 three branes which produce the background geometry . thus a time dependent boundary value of @xmath29 is a time dependent mass for the hypermultiplets . therefore a quantum quench in this dual theory may be implemented simply by providing a time dependent boundary condition for the dbi field . however the dbi fields are the transverse coordinates of the branes - so this corresponds to a motion of the edge of the brane . this disturbance sets up a wave along the brane and therefore correponds to an excited state of the defect field theory . our aim is to figure out the nature of this state at late times . in the full theory , such a disturbance would lead to a deformation of the background geometry and possibly lead to black hole formation , which would appear as thermalization in the boundary field theory . we want to explore if any signature of thermalization remains in the probe approximation . the following sections summarize some salient points of work with tatsuma nishioka and tadashi takayanagi @xcite . the essential physics is in fact apparent in the simplest example - d1 branes in @xmath16 . for this purpose it is convenient to write the @xmath16 metric as ds^2 = 2drdv - f(r)dv^2+r^2 d^2+(d^2+^2d^2+^2d^2_3 ) if we are using the poincare patch , @xmath30 and @xmath31 is the flat metric on @xmath32 , while in the global patch @xmath33 and @xmath31 is the round metric on @xmath34 . we have used eddington - finkelstein coordinates in the @xmath35 part . the d1 brane is along @xmath36 and its action is given by the standard dbi action obtained from the induced metric . the dynamical fields on the brane are @xmath37 and the coordinates contained in @xmath31 . it is clear from the symmetries that one can have a class of solutions of the form ( r , v ) , = with all the other coordinates held constant . the equations of motion which follow from the dbi action is best written by using the advanced ef coordinate @xmath38 as well as @xmath39 _ u_v + _ v _ u ( ) + _ u _ v ( ) = 0 , [ uveqnmotion ] where l = 1-_u _ v . it is easy to see that any @xmath40 which satisfies either @xmath41 or @xmath42 is a solution of ( [ uveqnmotion ] ) . in particular , a solution which is a function of @xmath39 alone represents the retarded effect of a boundary value of @xmath40 . the induced metric produced by such a retarded solution is given by [ indf ] ds_ind^2 = -f(r ) dudv + ( _ v ) ^2 dv^2 = 2drdv - [ f(r)-(_v ) ^2]dv^2 . this is a two - dimensional ads vaidya metric which has an apparent horizon at @xmath43 , provided this equation has a solution for real @xmath44 . in the poincare metric there is always a solution for real @xmath44 , while in global coordinates , this is not guaranteed . depending on the profile of @xmath45 , the apparent horizon may or may not develop into an event horizon . an example where it does is given by the profile ( v ) = _ 0 ( v + ( kv ) ) [ prof1 ] which leads to the following equation for the location of the apparent horizon for the poincare patch r = _ 0 ( 1 + ( kv ) ) [ ahprof1 ] this asymptotes to an event horizon at @xmath46 . this profile represents a d1 brane which starts from rest and spins with an increasing spin , asymptoting to a constant rotation rate . the function ( [ prof1 ] ) and the location of the apparent horizon is shown in figure ( [ fig : prof1 ] ) and ( [ fig : ahprof1 ] ) location of the apparent horizon . @xmath39 as a function of @xmath44 from equation ( [ ahprof1 ] ) . at late time this becomes an event horizon ] location of the apparent horizon . @xmath39 as a function of @xmath44 from equation ( [ ahprof1 ] ) . at late time this becomes an event horizon ] for strings which eventually stop spinning , the apparent horizon does not develop into an event horizon , but recedes back to @xmath47 . for example if ( v ) = _ 0(1 + ( kv ) ) [ prof2 ] the apparent horizon is located at r = [ ahprof2 ] the function ( [ prof2 ] ) and the location of the apparent horizon is shown in figure ( [ fig : prof2 ] ) and ( [ fig : ahprof2 ] ) location of the apparent horizon . @xmath39 as a function of @xmath44 from equation ( [ ahprof2 ] ) . the apparent horizon now recedes back to @xmath47 ] location of the apparent horizon . @xmath39 as a function of @xmath44 from equation ( [ ahprof2 ] ) . the apparent horizon now recedes back to @xmath47 ] in global @xmath12 the equation which determines the apparent horizon is given by @xmath49 which does not always have a solution for real @xmath44 . for example , a brane whose end point is uniformly rotating has @xmath50 - this would lead to an event horizon only when @xmath51 . the poincare patch solutions represent injection of energy from the boundary , which flows into the poincare horizon . in the global solutions , the energy flows from a point of the boundary to the antipodal point of the @xmath34 . fluctuations of the brane around this classical solution will feel the effect of an apparent horizon on the worldsheet . let us choose a static gauge where the worldsheet coordinates are identified with two of the space - time coordinates @xmath52 . the transverse coordinates are @xmath53 . the @xmath11 metric can be then written as ds^2 = g_ab ( x^a , x^i ) dx^adx^b + g_ij(x^a , x^k ) dx^i dx^j . [ fluc1 ] expanding around a classical solution @xmath54 , x^i(x^a ) = x_0^i ( x^a ) + y^i(x^a ) , [ fluc3 ] this leads to the following action for quadratic fluctuations s_2 = d^2 _ 0^ab g_ij ( ^a , x_0^i ) _ a y^i _ b y^j . [ fluc8 ] where @xmath55 denotes the induced metric due to the background solution @xmath56 . in particular , the fluctuations of @xmath57 i.e. all fluctuations in @xmath58 directions are minimally coupled massless scalars on the worldsheet , while the fluctuations of the boundary gauge theory spatial directions @xmath59 have an additional factor of @xmath60 coming from the fact that @xmath61 along these directions . it is well known that fields which live on a space - time with an apparent horizon behave approximately thermally if the apparent horizon lasts long enough @xcite . while the standard derivation of hawking radiation assumes the presence of an event horizon , the essential physics is the large redshift , which is present near an apparent horizon as well . in our case , profiles like ( [ prof1 ] ) lead to exact thermality at late times since the apparent horizon evolves into an event horizon . on the other hand profiles like ( [ prof2 ] ) lead to an effective `` time dependent temperature '' in the dual theory , which of course makes sense when the time variation is slow enough . the thermal nature of the state produced by time dependence becomes clear from a calculation of the fluctuation of the end - point of the string . in @xcite it has been shown that the fluctuations of a string suspended from the horizon of a ads black brane ended at a flavor d - brane near the boundary of ads are dual to brownian motion of the corresponding quark in the hot @xmath62 gauge theory . in this case the bulk black brane metric induces a worldsheet metric which has a horizon . the fluctuations then reflect hawking radiation from the worldsheet horizon . in the d - brane solutions considered above , the bulk metric has no horizon . however due to the motion of the d - brane , the induced metric on the worldvolume can develop a horizon . since the fluctuations of @xcite comes purely from properties of the induced metric it is natural to expect that a similar phenomenon appears in our case . the result of this calculation for fluctuations in the @xmath40 direction is ( y^(t - t^))^2 & ~ & , ( t - t^ ) , ( y^(t - t^))^2 & ~ & - , ( t - t^ ) , while for fluctuations in the @xmath63 direction are ^2 & ~ & , t m^2 , & ~ & |t - t^| , t m^2 . the rotating @xmath64 brane solution corresponds to a time dependent coupling in the @xmath22 theory coupled to hypermultiplets living on the zero dimensional defect . the d1-d3 system is 1/4-bps and the d1-d3 open strings lead to the two complex scalars @xmath65 of hypermultiplets which belong to the fundamental and anti - fundamental representations of the color @xmath66 gauge group . let us express the three complex adjoint scalar fields in the @xmath67 super yang - mills by @xmath68 . these correspond to cartesian coordinates in the transverse @xmath69 composed of @xmath70 where @xmath71 represents the 5-sphere . we choose @xmath72 such that its phase rotation describes the one in the @xmath40 direction and that @xmath73 is equivalent to @xmath74 . the time dependent coupling term which corresponds to a uniformly rotation d1-brane is given by dt . [ intti ] the justification for this is given in @xcite . for non - uniform rotation with a profile @xmath75 the exponential factors are simply replaced by @xmath76 . thus , from the boundary theory point of view we have a time dependent coupling - this leads to thermalization . note that this is thermalization of only the hypermultiplet sector - the vector multiplet sector is unchanged in the lowest order of this approximation . thermalization in higher dimensional field theories can be also investigated by considering higher dimensional branes in the bulk of @xmath11 . it is possible to construct uniformly rotating d7 and d5 branes using a combination of analytic and numerical methods . while these d5 and d7 solutions have been obtained earlier in @xcite , the implications to thermalization was not realized . quench - like solutions ( i.e. solutions where the rotation vanishes in the asymptotic past and the asymptotic future ) can be obtained numerically as well @xcite . rotating d5 branes are dual to a time dependent mass of the hypermultiplets in the dual 2 + 1 dimensional defect field theory . rotating d7 branes are dual to a time dependent phase of the mass of fermions in the hypermultiplet , as well as a time dependent bosonic potential . the essential physics is similar to the d1 brane descibed in the previous subsection , viz . an apparent horizon is formed and the fluctuations respond in a thermal fashion . an additional signature of dissipation appears when we consider probe d3 branes obtained by performing t - duality on the d1 brane solution described above along @xmath8 and @xmath9 directions . in this case the worldvolume is a @xmath18 dimensional theory and the dual defect field theory is @xmath77 dimensions . it turns out that in this case one can turn on a background electric field on the worldvolume of a uniformly rotating d3 brane . this means we are turning on a chemical potential and a charge density for the corresponding global charge in the dual field theory , in addition to a time dependent coupling . the fluctuations of the gauge field around this background now react to the apparent horizon of the induced metric . this leads to an electrical conductivity @xmath78 , whose behavior as a function of the frequency @xmath2 is quite similar to drude theory at low frequencies . however the real part of @xmath79 approaches a constant at large frequency , as is typical in a 2 + 1 dimensional critical theories . this is shown in figures ( [ fig : one ] ) and ( [ fig : two ] ) imaginary part of @xmath78 . ] imaginary part of @xmath78 . ] an interesting example of the process of thermalization due to formation of an apparent horizon on the worldvolume concerns the thermalization of the meson sector of @xmath22 yang - mills theory due to a time dependent chemical potential @xcite . as usual , quarks are introduced by placing a set of d7 branes and a chemical potential corresponds to a worldvolume electric field . this is made time dependent by coupling to a time dependent external current , i.e. by injecting quarks from outside . as in the previous examples this results in the formation of an apparent horizon with a characteristic temperature . for other effects of this phenomenon see @xcite . finally the formation of apparent horizons in these examples is similar to acceleration horizons on worldvolumes discussed in @xcite and related phenomena have been studied in @xcite . as we discussed above , quantum quench is particularly interesting when the time dependent coupling crosses a critical point . the probe approximation has turned out to be quite useful for studying holographic critical points . such phase transitions are known to occur for various probe branes @xcite . another class of probe fields appear in discussions of holographic superconductors @xcite . this setup consists of a charged scalar field in the presence of a charged black brane . when the gauge coupling is large , the backreaction of the scalar and the gauge field to the background geometry can be ignored . in this case , for a given mass of the scalar field there is a critical temperature below which the scalar condenses - this is interpreted as superfluidity in the boundary theory . the phase transition between the ordered and disordered phases is a critical point . an even simpler setting consists of a neutral scalar field with a quartic coupling which is large enough to ensure that a probe approximation is reliable . as shown in @xcite for suitable values of the parameters , this model displays a critical point of the type encountered in antiferromagentic phase transitions . in this section i will describe a recent study of quantum quench in this model in collaboration with pallab basu @xcite . the model of @xcite has a neutral scalar field @xmath80 in the background of a charged @xmath13 black brane . the lagrangian is given by = [ -()^2 -(^2+m^2)^2- ] [ 1 - 1 ] the background metric is given by ( in @xmath81 units ) ds^2 = [ -r^2 f(r)dt^2+r^2 d^2]+ [ 1 - 2 ] where f(r ) = [ 1 + - ] 0 1 [ 1 - 3 ] the associated hawking temperature is then given by t = ( 1- ) [ 1 - 4 ] in the following we will replace @xmath82 . this means all dimensional quantities are expressed in units of @xmath83 . in the limit of large @xmath84 the field @xmath85 can be regarded as a probe field . in @xcite it was shown that when the mass lies in the range - < m^2 < - [ 1 - 5 ] there is a critical phase transition at some value of @xmath86 when the source to the dual operator vanishes . conversely , for a given @xmath87 there is a value of @xmath88 where the theory is critical . the upper limit in ( [ 1 - 5 ] ) is the bf bound for the near - horizon @xmath89 geometry which appears in the extremal ( @xmath90 ) metric . ( note that the ads scale for this infrared @xmath89 is given by @xmath91 in our units ) . the lower bound is the bf bound for the asymptotic @xmath13 . field configurations which are translationally invariant in the @xmath63 directions satisfy the equations of motion [ -_t^2+_r(r^4 f(r)_r)]-m^2 -^3 = 0 [ 1 - 6 ] near the @xmath13 boundary the asymptotic behavior of the solution to the linearized equation is of the form , it is a valid concern whether we will be able to neglect the non - linear term near the boundary . this can be done as long as @xmath92 or @xmath93 . ] ( r ) = j(t)r^-_-[1+o(1/r^2)]+<o>(t ) r^-_+[1+o(1/ r^2 ) ] [ 1 - 7 ] where @xmath94 is given by _ = [ 1 - 8 ] in the range of masses of interest , both the solutions are normalizable , so that there is a choice of quantization . the standard quantization considers the coefficient @xmath95 as the source in the dual field theory and @xmath96 then gives the expectation value of the dual operator . in the alternative quantization the expectation and source change the role . consider first the linearized problem , ignoring the cubic term . by a standard change of coordinates to tortoise coordinates @xmath97 and a field redefinition to @xmath98 , d=- ( r , t ) = [ 1 - 9 ] the horizon is then at @xmath99 and the boundary is at @xmath100 . at the linearized level , the equation ( [ 1 - 6 ] ) becomes -_t^2 = -_^2 + v_0 ( ) _ [ 1 - 10 ] with v_0 ( ) = r^2 f(r)[(m^2 + 2 ) - + ] [ 1 - 11 ] where in @xmath101 we need to express @xmath44 in terms of @xmath97 using ( [ 1 - 9 ] ) . for solutions of the type @xmath102 , equation ( [ 1 - 11 ] ) is a schrodinger problem in a potential @xmath101 . the potential goes to zero at the horizon @xmath99 and behaves as @xmath103 near the boundary @xmath100 . note that for a brane background at any finite temperature , @xmath104 near the horizon , while @xmath105 so that @xmath106 as we approach the horizon . a typical @xmath101 for @xmath107 is shown in figure([fig : potential1 ] ) . the potential @xmath101 ] in contrast , for the extremal background @xmath109 while @xmath110 so that @xmath111 . this makes the analysis for the extremal background rather subtle . we will work with the non - extremal case . in @xcite it was shown that when @xmath112 is below a critical value , @xmath113 there are bound states of this schrodinger problem , showing that the trivial @xmath114 is unstable . at @xmath88 a zero energy bound state appears , which vanishes in an appropriate fashion at the boundary and is in addition regular at the horizon . in the complex frequency plane some quasinormal mode(s ) hit the origin at @xmath88 . this critical mass is @xmath115 when @xmath116 and decreases with decreasing @xmath117 or increasing temperature . for @xmath118 there is a stable nontrivial static solution @xmath119 of the full nonlinear equation of motion with the condition that @xmath120 . this means that in the conventional quantization , the expectation value of the dual operator is nonzero even in the absence of a source , i.e. the field condenses . the the critical point is at @xmath121 and @xmath120 . similarly there is a nontrivial solution with @xmath122 which means that there is a condensate in the alternative quantization as well . the critical point has mean field exponents at any finite @xmath87 . if the operator dual to the field @xmath85 is @xmath123 and the source is @xmath124 then an analysis identical to that presented in @xcite leads to ( for @xmath125 ) _ j=0~(m^2-m_c^2)^1/2 |_j=0~(m^2-m_c^2)^-1 < > _ m = m_c~j^1/3 [ 1 - 12 ] exactly at zero temperature the phase transition is of bkt type and the order parameter depends exponentially _ j=0 ~[1 - 13 ] in the following , we will study quench across this critical point by considering a time dependent source @xmath95 which asymptotes to constant values at early and late times and crosses zero at some time , e.g. j(t ) = j_0 ( vt ) in the conventional quantization this means that in the dual boundary field theory , we have a source coupling to the operator dual to @xmath85 . for any static @xmath126 we of course have a nontrivial @xmath127 and hence a nonzero @xmath128 . our first aim is to get some insight into the time dependence of @xmath129 when we have a nontrivial @xmath95 . if the time dependence is slow enough one would expect that far away from the critical point the dynamics is adiabatic , while near the critical point adiabaticity should break down . it is instructive to examine the way this happens . it is well known that to study low frequency modes in the background of a black brane it is convenient to use ingoing edddington - finkelstein coordinates , u = t- , where @xmath97 is defined in ( [ 1 - 9 ] ) . in terms of these coordinates the equation of motion [ 1 - 6 ] ) becomes -2_u_= -_^2 + v ( , ) . [ 4 - 1 ] where v(,)= v_0()+ f(r ) ^3 . this equation has to be solved with the boundary condition that the field is _ regular at the horizon _ , which at the linearized level is equivalent to requiring that the waves are purely ingoing at the horizon @xcite . we need to solve ( [ 4 - 1 ] ) with the condition ( u , ) ^-1+_- j(u ) as 0 [ 4 - 1 - 1 ] where @xmath130 are defined in ( [ 1 - 8 ] ) . to perform the adiabatic expansion , let us decompose the field @xmath131 as ( , u)=_l(,u)+_s(,u ) [ 4 - 1 - 2 ] where @xmath132 and @xmath133 as @xmath134 . for a constant @xmath124 , @xmath135 is time independent . in this case there is a static solution @xmath136 , which is the equlibrium configuration . in the presence of a source which is _ slowly varying _ in units of the horizon radius , one can therefore expand the field @xmath137 in an adiabatic expansion of the form _ s(,u ) = _ 0 ( , j(u ) ) + _ 1(,u ) + . [ expan ] here @xmath138 ( recall that we are using @xmath139 units ) is an adiabaticity parameter which keeps track of the adiabatic expansion . if we scale @xmath140 , each @xmath141 derivative is of order @xmath142 . the idea then is to insert ( [ expan ] ) into the equations of motion and obtain equations for @xmath143 order by order in @xmath144 . to the lowest order one ^(1 ) _ 1 = \ { [ -_^2 + v_0 ( ) ] + f(r ) ( 3 ^ 2_0 + 6 _ l_0 + 3_l^2 ) } _ 1= - 2 _ u__l - 2 _ u__0 [ 1 - 16 ] the solution to this equation is _ 1=d g(,)_u^_^ ( _ 0+_l)(^ ) . [ 1 - 17a ] where @xmath145 is the green s function of the operator @xmath146 with the boundary conditions @xmath147 and @xmath148 is regular : g^(1)(, ) & = & _ 1( ) _ 2 ( ) , , where @xmath149 and @xmath150 are solutions of homogeneous part of eqn ( [ 1 - 17a ] ) satisfying appropriate boundary condition at the horizon @xmath151 and the boundary @xmath100 respectively , and @xmath152 is the wronskian which is independent of @xmath97 in this case . we have normalized @xmath153 and @xmath154 in such a fashion that @xmath155 at the horizon and @xmath156 near the boundary . regularity of the functions @xmath157 and @xmath158 mean that @xmath159 are finite at the horizon . since @xmath160 , this implies that @xmath161 . this ensures that the integral in ( [ 1 - 17a ] ) is finite even though the green s function approaches a constant in the region near the horizon @xmath162 . furthermore , near the horizon @xmath150 can be expressed as a linear combination of a regular and irregular solution , i.e @xmath163 . this implies that @xmath164 . thus @xmath165 is finite so long as @xmath166 is finite . at the critical point , @xmath124 becomes small . then @xmath158 and @xmath157 in the left hand side of ( [ 1 - 16 ] ) vanish , and the operator is identical to the operator acting on the linearized small fluctuations at @xmath88 around the trivial solution @xmath167 , i.e. the operator @xmath168 which appears on the right hand side of ( [ 1 - 10 ] ) . we know that this operator has a zero mode which is regular at the horizon _ and _ vanishes as @xmath169 at the boundary @xmath100 . this means that at this point @xmath170 . therefore , the first adiabatic correction diverges . for small @xmath171 , the leading departure from the critical operator comes from the term which is proportional to @xmath172 . thus we can use perturbation theory in @xmath124 to estimate @xmath173 . as argued before , @xmath174 , while @xmath175 . hence the leading divergence in @xmath176 can be estimated as @xmath177 . adiabaticity breaks down when _ 1(,u ) ~_0 ~j^5/3 in particular for profiles of @xmath171 where @xmath178 near the critical point at @xmath179 , adiabaticity breakdown occurs at u ~v^-2/5 the scaling behavior found above is identical to that in a landau - ginsburg dynamics with dynamical critical exponent @xmath180 . the dynamics of an spatially homogeneous order parameter @xmath40 is given by + m^2 + ^3+j(t ) = 0 [ three ] the equilibrium critical point is at @xmath181 . for @xmath182 the equilibrium value of the order parameter is _ 0(j ) = [ -j]^1/3 as usual an adiabatic expansion is of the form ( t ) = _ 0 ( j(t ) ) + _ 1 ( t ) + and to lowest order _ 1 = and adiabaticity breaks down when @xmath183 which becomes the condition ~j^1/3 ~j^5/3 exactly as in our system . when adiabaticity breaks down our system enters a scaling region . suppose the function @xmath95 behaves linearly with time in the critical region . then it is straightforward to see from ( [ three ] ) that in this region the solution is of the form ( t , v ) = v^1/5 ( tv^2/5,1 ) this means , in particular , that the time at which the order parameter hits zero scales as @xmath184 while the value of the order parameter at @xmath185 scales as @xmath186 . a numerical solution of the equation ( [ three ] ) with adiabatic initial conditions is consistent with this scaling , as shown in figure ( [ fig : lg ] ) the scaled order parameter as a function of scaled time for a @xmath187 at @xmath182 with @xmath188 ( from the bottom on the left ) . the adiabatic solution ( dashed ) is also shown as a comparison . ] note that the order parameter hits zero _ later _ than the location of the equilibrium critical point . this is a manifestation of the phenomenon of raising the critical temperature when the temperature is time dependent @xcite which has been holographically realized in @xcite . we will now argue that the behavior of our holographic system in the critical region is fairly well described by such a lg dynamics consider the dynamics of the bulk field in the critical region in the presence of a linear quench @xmath189 for small @xmath39 . to do this , first substitute ( [ 4 - 1 - 2 ] ) in the equation ( [ 4 - 1 ] ) and rescale _ s v^ _ s , u v^-u [ rescale ] the equation ( [ 4 - 1 ] ) then becomes _ s+v^ [ f(r ) ( _ s)^3 + u [ -_^2 + v_0 ( ) ] _ l+2 _ u _ _ s ] + = 0 [ veq ] the ellipsis denote terms which contains higher powers of @xmath39 . let us expand the sub - leading part of the scalar field in terms of eigenfunctions of the operator @xmath168 ( defined in equation ( [ 1 - 10 ] ) at the critical point , _ s(,u ) = a_k(u ) _ k ( ) dk the @xmath190 satisfy _ ^c _ k = [ -_^2 + v_0^c()]_k = k^2 _ k [ eigeneqn ] where @xmath191 denotes the potential in ( [ 1 - 11 ] ) at @xmath192 . the eigenfunctions @xmath193 are delta function normalized and obey the condition _ 0 [ ^1-_- _ k ( ) ] = 0 in terms of the eigen - coefficients @xmath194 the equation ( [ veq ] ) becomes , k^2 a_k+ v^ ( u j_k + b_kk _ u a_k dk + a_k a_k a_k c_k , k,k,k d k dk dk ) + = 0 [ modeeqn ] where _ k & = & _ k ( ) [ -_^2 + v_0 ( ) ] _ l d + b_kk & = & d _ k _ _ k + c_k , k,k,k & = & d _ k _ k _ k _ k f(r ) . [ moments ] the equation ( [ modeeqn ] ) suggests that there is a solution in a perturbation expansion of powers of @xmath195 , a_k(u ) = ( k ) _ 0(u ) + v^ _ k(u)+ , [ vexpn ] if this expansion makes sense , the dominant behavior of the solution for @xmath196 is given by the zero mode @xmath197 . substituting ( [ vexpn ] ) in the equation ( [ modeeqn ] ) we get to the lowest order in the small @xmath39 expansion , & & u j_0 + b_00 _ 0(u ) + c_0000 _ 0(u)^3 = 0 + & & _ k(u ) = -(u j_k- b_k0 _ 0(u)-c_k000 _ 0(u)^3 ) [ scalingeqn ] the first equation in ( [ scalingeqn ] ) determines the time dependence of the zero mode @xmath197 , while the second equation determines the leading correction from nonzero modes in terms of the solution for @xmath197 . it is useful to rewrite the second equation by subtracting the first from it , _ k(u ) = - ( u ( j_k - j_0 ) - ( b_k0-b_00 ) _ 0 ( u)-(c_k000-c_0000 ) _ 0(u)^3 ) [ scalingeqn2 ] the expansion in powers of @xmath198 would be valid if @xmath199 remains finite . however , @xmath200 is a continuous parameter starting from zero . this means that there is a potential divergence in the @xmath201 limit . indeed , as will be argued in the next subsection , for generic potential @xmath202 the numerator on the right hand side of ( [ scalingeqn2 ] ) behaves as @xmath200 for small @xmath200 , so that @xmath203 indeed diverges at @xmath204 . however exactly at the critical point , the small @xmath200 behavior changes to @xmath205 so that @xmath203 remains finite and the expansion in @xmath198 remains valid . to examine this issue we need to consider the eigenvalue problem _ k = k^2 _ k [ egen ] as discussed above the potential @xmath206 as @xmath207 . this potential is shown in figure ( [ fig : potential1 ] ) . the basic features of the eigenfunctions can in fact be gleaned from a simpler problem in which we replace the potential by the following potential which has the same qualitative features . & = & = 0 + u ( ) & = & -u_0 0 1 + & = & 0 1 [ squarewell ] this problem is of course solvable . the eigenfunctions of the schrodinger operator with eigenvalue @xmath208 are _ k ( ) & = & ( ) 0 1 + _ k ( ) & = & ( k+ ( ) ) [ 12 - 1 ] where the constants @xmath209 and @xmath210 are determined by matching at @xmath211 , a(k ) & = & + ( k)&= & ^-1()-k . [ 12 - 2 ] the solution for @xmath204 is _ 0 ( ) & = & ( ) 0 1 + _ 0 ( ) & = & a+ b [ 12 - 3 ] the matching conditions at @xmath211 now yield ( ) & = & a+b + ( ) & = & a [ 12 - 4 ] for any @xmath212 the solution blows up at @xmath99 . thus regular solutions require @xmath170 . however the second equation in ( [ 12 - 4 ] ) then imply that = ( n + ) [ 12 - 5 ] these are the zero modes . as we increase the depth of the potential , the first zero mode appears at @xmath213 . in the context of our model this is the potential where we have a critical point . the small @xmath200 behavior of @xmath209 and @xmath214 can be read off from the expressions ( [ 12 - 2 ] ) . for a generic @xmath215 these are a(k ) & ~ & + o(k^2 ) + ( k ) & ~ & k [ - 1 ] + o(k^3 ) [ 12 - 6 ] whereas for critical potentials we have a(k ) & ~ & 1 - + o(k^4 ) + ( k ) & ~ & - - [ 12 - 7 ] thus the small-@xmath200 behavior of the eigenfunctions are drastically different for the critical potentials . this has important implications for the coefficients like @xmath216 and @xmath217 in ( [ scalingeqn2 ] ) . consider for example the quantity @xmath218 . this is an integral of the form _ 0^d j ( ) _ k ( ) where @xmath219 is a smooth function ( which is @xmath220 \ch_l $ ] ) . if we replace the true eigenfunctions by those of our simplified problem , we get _ k = a(k ) _ 0 ^ 1 d ( ) j ( ) + _ 1^d(k+ ( k ) ) using ( [ 12 - 6 ] ) and ( [ 12 - 7 ] ) we therefore see that _ k -j_0 ~k k 0 for generic potentials , whereas _ k -j_0 ~k^2 k 0 for critical potentials . it is straightforward to see that the behavior of the other coefficients @xmath221 and @xmath217 are similar . the small-@xmath200 behavior of the eigenfunctions for the potential which is relevant for us . @xmath222 is quite similar . this has been discussed in the appendix c of @xcite . going back to ( [ scalingeqn2 ] ) we therefore see that the small @xmath39 expansion is generically _ not valid _ since the corrections diverge at small @xmath200 . however for the critical potential , @xmath223 remain finite as @xmath201 and the expansion in powers of @xmath198 makes sense . the validity of this expansion means that the solution for @xmath196 is of the form _ s(,u ) v^_0 ( v^u ) _ 0()+v^ _ k(v^u ) _ k ( ) dk + thus the dynamics for small @xmath39 is dominated by the zero mode in the critical region . the equation for @xmath224 is , however , exactly the same as the equation for the order parameter @xmath40 in the lg model in the previous subsection , after going back to the original variables prior to the rescaling in ( [ rescale ] ) . thus in this region the system behaves as one with dynamical critical exponent @xmath180 . therefore we conclude that to leading order in small @xmath39 , we have ( u , v ) ~lim_0 [ ^1-_+ _ s ( , u ) ] ~v^1/5 < > ( tv^2/5,1 ) and the time scale behaves as @xmath184 . the fact that the dynamics in the critical region is governed by an equation with a first order time derivative is made manifest in our treatment using eddington - finkelstein coordinates . the whole analysis can be of course performed in principle in the @xmath225 coordinates . we suspect that in this case one has to exercise extreme care , just as one had to extract out a leading horizon behavior in the linearized problem of fields in a black brane background @xcite . note that beyond the critical region , @xmath171 departs from the form @xmath178 , and the expansion in powers of @xmath198 is no longer valid . now all the modes are important , and integrating out higher modes can give rise to higher time derivatives in the effective equation for @xmath128 . an analysis similar to the above can be carried out when the quenching is performed by making the mass parameter of the bulk theory a function of the retarded time @xmath141 , keeping @xmath226 . this does not have a direct interpretation in the boundary field theory . however , as explained in @xcite , the field @xmath85 can acquire a mass because of a coupling to some other field @xmath227 . a time dependent boundary value of the field @xmath227 can then lead to a time dependent mass . when @xmath228 near the critical point , we now get a behavior @xmath229 . we have performed some preliminary numerical work for the case of a mass quench . our results clearly display the propagation of disturbances towards the horizon along a light cone , and a decay of the order parameter in a fashion similar to the @xmath180 lg dynamics , which is shown in figure([fig : ord2 ] ) . however our results are not accurate enough to verify the scaling behavior found above . the late time decay should be governed by the quasinormal modes @xcite . the order parameter @xmath230 in boundary theory . ] we have shown that holographic methods are useful in providing insight into various questions related to quantum quench in strongly coupled field theories which have a gravity dual . interestingly , this is possible in the probe approximation , which is typically much easier to study . so far we have been able to study in some detail quench dynamics near holographic critical points which are described by mean field exponents . not surprisingly , we found scaling behavior characteristic of landau - ginsburg models with @xmath180 . this value of the dynamical critical exponent is consistent with the results of @xcite . it is important to perform our quench analysis for zero temperature , where the equilibrium transition is of the bkt type @xcite , similar to brane models of zero temperature chiral symmetry breaking transition in @xcite . this case is rather subtle , but can be studied using similar methods . we expect that in this case we will have a @xmath231 theory dominating the critical region . this would provide results for quench dynamics which are not easily obtainable by other methods . finally , a more extensive numerical investigation should throw light on the question of thermalization at late times , after the system has crossed the critical region . i would like to thank my collaborators pallab basu , tatsuma nishioka and tadashi takayanagi for very enjoyable collaborations and many insightful discussions . i would also like to thank karl landsteiner , satya majumdar , gautam mandal , shiraz minwalla , takeshi morita , ganpathy murthy , omid saremi , alfred shapere , sandip trivedi and especially kristan jensen and krishnendu sengupta for disucssions . s.r.d . would like to thank institut de fisica teorica at madrid , tata institute of fundamental research at mumbai and indian association for the cultivation of science at kolkata for hospitality during the final stages of this work . finally i thank the the organizers of 11th workshop on non - perturbative qcd in paris , sixth crete regional meeting on string theory in milos and quantum theory and symmetries 7 in prague for organizing stimulating conferences . this work is partially supported by national science foundation grants phy-0970069 and phy-0855614 . 99 p. calabrese and j. l. cardy , j. stat . mech . * 0504 * ( 2005 ) p010 [ arxiv : cond - mat/0503393 ] ; p. calabrese and j. l. cardy , phys . * 96 * ( 2006 ) 136801 [ arxiv : cond - mat/0601225 ] p. calabrese and j. cardy , [ arxiv:0704.1880 [ cond-mat.stat-mech ] ] ; s. sotiriadis and j. cardy , j. stat . ( 2008 ) p11003 , [ arxiv:0808.0116 [ cond-mat.stat-mech ] ] ; s. sotiriadis , p. calabrese and j. cardy , epl * 87 * ( 2009 ) 20002 , [ arxiv:0903.0895 [ cond-mat.stat-mech ] ] ; s. sotiriadis and j. cardy , arxiv:1002.0167 [ quant - ph ] . t. w. b. kibble , j. phys . aa * 9 * , 1387 ( 1976 ) . w. h. zurek , nature * 317 * , 505 ( 1985 ) . j. m. maldacena , adv . theor . math . * 2 * ( 1998 ) 231 [ int . j. theor . * 38 * ( 1999 ) 1113 ] [ arxiv : hep - th/9711200 ] ; o. aharony , s. s. gubser , j. m. maldacena , h. ooguri and y. oz , phys . rept . * 323 * , 183 ( 2000 ) [ arxiv : hep - th/9905111 ] . s. r. das , t. nishioka , t. takayanagi , jhep * 1007 * , 071 ( 2010 ) . [ arxiv:1005.3348 [ hep - th ] ] . r. a. janik and r. b. peschanski , phys . d * 74 * , 046007 ( 2006 ) [ arxiv : hep - th/0606149 ] ; r. a. janik , phys . lett . * 98 * , 022302 ( 2007 ) [ arxiv : hep - th/0610144 ] ; p. m. chesler and l. g. yaffe , phys . * 102 * , 211601 ( 2009 ) [ arxiv:0812.2053 [ hep - th ] ] ; p. m. chesler , l. g. yaffe , phys . rev . * d82 * , 026006 ( 2010 ) . [ arxiv:0906.4426 [ hep - th ] ] ; s. bhattacharyya and s. minwalla , jhep * 0909 * ( 2009 ) 034 [ arxiv:0904.0464 [ hep - th ] ] . j. abajo - arrastia , j. aparicio , e. lopez , jhep * 1011 * , 149 ( 2010 ) . [ arxiv:1006.4090 [ hep - th]];h . ebrahim , m. headrick , [ arxiv:1010.5443 [ hep - th ] ] ; v. balasubramanian , a. bernamonti , j. de boer , n. copland , b. craps , e. keski - vakkuri , b. muller , a. schafer _ et al . _ , phys . lett . * 106 * , 191601 ( 2011 ) . [ arxiv:1012.4753 [ hep - th ] ] ; v. balasubramanian , a. bernamonti , j. de boer , n. copland , b. craps , e. keski - vakkuri , b. muller , a. schafer _ et al . _ , phys . rev . * d84 * , 026010 ( 2011 ) . [ arxiv:1103.2683 [ hep - th ] ] ; d. garfinkle , l. a. pando zayas , phys . rev . * d84 * , 066006 ( 2011 ) . [ arxiv:1106.2339 [ hep - th ] ] ; d. garfinkle , l. a. p. zayas , d. reichmann , [ arxiv:1110.5823 [ hep - th ] ] . t. albash , c. v. johnson , new j. phys . * 13 * , 045017 ( 2011 ) . [ arxiv:1008.3027 [ hep - th ] ] ; t. takayanagi , t. ugajin , jhep * 1011 * , 054 ( 2010 ) . [ arxiv:1008.3439 [ hep - th ] ] ; c. t. asplund , s. g. avery , [ arxiv:1108.2510 [ hep - th ] ] . s. r. das , j. michelson , k. narayan and s. p. trivedi , phys . d * 74 * , 026002 ( 2006 ) [ arxiv : hep - th/0602107 ] ; a. awad , s. r. das , a. ghosh , j. h. oh and s. p. trivedi , phys . d * 80 * ( 2009 ) 126011 [ arxiv:0906.3275 [ hep - th ] ] . t. hertog and g. t. horowitz , jhep * 0407 * , 073 ( 2004 ) [ arxiv : hep - th/0406134 ] ; t. hertog and g. t. horowitz , jhep * 0504 * , 005 ( 2005 ) [ arxiv : hep - th/0503071 ] ; n. turok , b. craps and t. hertog , arxiv:0711.1824 [ hep - th ] ; b. craps , t. hertog and n. turok , arxiv:0712.4180 [ hep - th ] . j. mcgreevy and e. silverstein , jhep * 0508 * , 090 ( 2005 ) [ arxiv : hep - th/0506130 ] ; e. silverstein , phys . d * 73 * , 086004 ( 2006 ) [ arxiv : hep - th/0510044 ] ; g. horowitz , a. lawrence and e. silverstein , jhep * 0907 * , 057 ( 2009 ) [ arxiv:0904.3922 [ hep - th ] ] . for a review and references to the original literature see e.g. j. erdmenger , n. evans , i. kirsch , e. threlfall , eur . j. * a35 * , 81 - 133 ( 2008 ) . [ arxiv:0711.4467 [ hep - th ] ] ; s. s. gubser , phys . rev . * d78 * , 065034 ( 2008 ) . [ arxiv:0801.2977 [ hep - th ] ] ; s. a. hartnoll , c. p. herzog , g. t. horowitz , jhep * 0812 * , 015 ( 2008 ) . [ arxiv:0810.1563 [ hep - th ] ] ; s. a. hartnoll , c. p. herzog , g. t. horowitz , phys . . lett . * 101 * , 031601 ( 2008 ) . [ arxiv:0803.3295 [ hep - th ] ] . p. basu , a. mukherjee , h. -h . shieh , phys . rev . * d79 * , 045010 ( 2009 ) . [ arxiv:0809.4494 [ hep - th ] ] ; c. p. herzog , p. k. kovtun , d. t. son , phys . rev . * d79 * , 066002 ( 2009 ) . [ arxiv:0809.4870 [ hep - th ] ] . d. arean , p. basu , c. krishnan , jhep * 1010 * , 006 ( 2010 ) . [ arxiv:1006.5165 [ hep - th ] ] ; s. a. hartnoll , class . * 26 * ( 2009 ) 224002 [ arxiv:0903.3246 [ hep - th ] ] ; c. p. herzog , j. phys . a * 42 * , 343001 ( 2009 ) [ arxiv:0904.1975 [ hep - th ] ] ; see e.g. m. visser , `` essential and inessential features of hawking radiation , '' int . j. mod . d * 12 * , 649 ( 2003 ) [ arxiv : hep - th/0106111 ] . j. de boer , v. e. hubeny , m. rangamani and m. shigemori , `` brownian motion in ads / cft , '' jhep * 0907 * , 094 ( 2009 ) [ arxiv:0812.5112 [ hep - th ] ] ; a. n. atmaja , j. de boer and m. shigemori , `` holographic brownian motion and time scales in strongly coupled plasmas , '' arxiv:1002.2429 [ hep - th ] ; d. t. son and d. teaney , `` thermal noise and stochastic strings in ads / cft , '' jhep * 0907 * , 021 ( 2009 ) [ arxiv:0901.2338 [ hep - th ] ] . a. obannon , jhep * 0901 * , 074 ( 2009 ) . [ arxiv:0811.0198 [ hep - th ] ] ; n. evans , e. threlfall , [ arxiv:0807.3679 [ hep - th ] ] ; n. evans , e. threlfall , phys . rev . * d79 * , 066008 ( 2009 ) . [ arxiv:0812.3273 [ hep - th ] ] . k. -y . kim , j. p. shock , j. tarrio , jhep * 1106 * , 017 ( 2011 ) . [ arxiv:1103.4581 [ hep - th ] ] ; s. prem kumar , phys . rev . * d84 * , 026003 ( 2011 ) . [ arxiv:1104.1405 [ hep - th ] ] . s. janiszewski , a. karch , [ arxiv:1106.4010 [ hep - th ] ] ; c. hoyos , t. nishioka , a. obannon , [ arxiv:1106.4030 [ hep - th ] ] . u. gursoy , e. kiritsis , l. mazzanti , f. nitti , nucl . * b820 * , 148 - 177 ( 2009 ) . [ arxiv:0903.2859 [ hep - th ] ] . u. gursoy , e. kiritsis , l. mazzanti , f. nitti , jhep * 1012 * , 088 ( 2010 ) . [ arxiv:1006.3261 [ hep - th ] ] . j. g. russo , p. k. townsend , class . grav . * 25 * , 175017 ( 2008 ) . [ arxiv:0805.3488 [ hep - th ] ] ; m. chernicoff and a. guijosa , jhep * 0806 * ( 2008 ) 005 [ arxiv:0803.3070 [ hep - th ] ] ; a. paredes , k. peeters and m. zamaklar , jhep * 0904 * , 015 ( 2009 ) [ arxiv:0812.0981 [ hep - th ] ] ; c. athanasiou , p. m. chesler , h. liu , d. nickel and k. rajagopal , arxiv:1001.3880 [ hep - th ] ; e. caceres , m. chernicoff , a. guijosa and j. f. pedraza , arxiv:1003.5332 [ hep - th ] ; t. hirata , s. mukohyama and t. takayanagi , jhep * 0805 * ( 2008 ) 089 [ arxiv:0804.1176 [ hep - th ] ] ; t. hirayama , p. w. kao , s. kawamoto and f. l. lin , arxiv:1001.1289 [ hep - th ] ; d. mateos , r. c. myers and r. m. thomson , `` holographic phase transitions with fundamental matter , '' phys . lett . * 97 * , 091601 ( 2006 ) [ arxiv : hep - th/0605046 ] ; s. kobayashi , d. mateos , s. matsuura , r. c. myers and r. m. thomson , `` holographic phase transitions at finite baryon density , '' jhep * 0702 * , 016 ( 2007 ) [ arxiv : hep - th/0611099 ] ; d. mateos , s. matsuura , r. c. myers and r. m. thomson , `` holographic phase transitions at finite chemical potential , '' jhep * 0711 * , 085 ( 2007 ) [ arxiv:0709.1225 [ hep - th ] ] . k. jensen , a. karch , e. g. thompson , jhep * 1005 * , 015 ( 2010 ) . [ arxiv:1002.2447 [ hep - th ] ] . k. jensen , a. karch , d. t. son , e. g. thompson , phys . lett . * 105 * , 041601 ( 2010 ) ; [ arxiv:1002.3159 [ hep - th]];k . jensen , phys . rev . * d82 * , 046005 ( 2010 ) . [ arxiv:1006.3066 [ hep - th]];n . evans , k. jensen , k. -y . kim , phys . rev . * d82 * , 105012 ( 2010 ) . [ arxiv:1008.1889 [ hep - th ] ] .

we study an effective field theory of interacting nucleons at distances much greater than the pion s compton wavelength . in this regime the potential is conjectured to be the sum of a delta function and its derivatives . the question we address is whether this sum can be consistently truncated at a given order in the derivative expansion , and systematically improved by going to higher orders . regularizing the lippmann - schwinger equation using a cutoff we find that the cutoff can be taken to infinity only if the effective range is negative . a positive effective range which occurs in nature requires that the cutoff be kept finite and below the scale of the physics which has been integrated out , i.e. . comparison of cutoff schemes and dimensional regularization reveals that the physical scattering amplitude is sensitive to the choice of regulator . moreover , we show that the presence of some regulator scale , a feature absent in dimensional regularization , is essential if the effective field theory of scattering is to be useful . we also show that one can define a procedure where finite cutoff dependence in the scattering amplitude is removed order by order in the effective potential . however , the characteristic momentum in the problem is given by the cutoff , and not by the external momentum . it follows that in the presence of a finite cutoff there is no small parameter in the effective potential , and consequently no systematic truncation of the derivative expansion can be made . we conclude that there is no effective field theory of scattering with nucleons alone . = 0.0 in = 0.0 in = 6.5 in = 8.9 in = 0.0 in = 0.0 in = 0.0 in = 0.25 in = -1 cm u.md . pp # 98 - 024 + + department of physics , university of maryland , college park , md , 20742 - 4111

nucl-th9709062

there exist many nucleon - nucleon potentials which reproduce phase shifts and nuclear properties with remarkable accuracy @xcite . three fundamental features are shared by these potential models : ( i ) pions dominate at long distances , ( ii ) there is some source of intermediate - range attraction , and ( iii ) there is some source of short - distance repulsion . however , in general , distinct physical mechanisms in these models account for the same feature of the nuclear force . agreement with experiment is maintained in spite of these differences because of the large number of fit parameters . it would be a considerable advance if a systematic approach to the nucleon - nucleon interaction could be developed based solely on symmetries and general physical principles . systematic approaches to the scattering of strongly interacting particles , such as chiral perturbation theory , are based on the ideas of effective field theory ( eft ) . effective field theory says that for probes of a system at momentum @xmath2 , details of the dynamics at scale @xmath3 are unimportant . what is important at low energies is the physics that can be captured in operators of increasing dimensionality which take the form of a power - series in the quantity @xmath4 it is important to realize that even if eft ideas can be applied to the @xmath0 system they will probably not lead to any startlingly new predictions for @xmath0 scattering . indeed , it is entirely possible that the resulting fits to phase shifts will not be as good as those produced by conventional @xmath0 potentials with the same number of parameters . however , the real motivation for constructing an eft is to relate one process to another . for instance , one would like to relate @xmath0 scattering systematically to scattering processes with more nucleons , such as @xmath5 and @xmath6 scattering , and to say something predictive about processes involving pionic and photonic probes of few - nucleon systems . one reason to hope that this can be achieved in nuclear physics is provided by the pattern of chiral symmetry breaking in qcd . the fact that chiral symmetry is spontaneously broken implies that the pion is light and interacts weakly at low energies of course the lightness of the pion in itself guarantees that it should play a fundamental role in nuclear physics . the weakness of pion interactions at low energies allows pion interactions with nucleons to be systematized using chiral perturbation theory . this procedure has proved remarkably successful in describing the interactions of pions with a single nucleon @xcite . these ideas have been extended to processes involving more than one nucleon , leading to arguments concerning the relative importance of various pion - exchange mechanisms in the nuclear force @xcite . however , these eft arguments assume the existence of a systematic power - counting scheme for multi - nucleon processes . in our view there is no convincing argument in the literature that such a power counting exists . the purpose of this paper is to investigate this issue in the simplest possible context . naive transposition of eft ideas to nuclear physics immediately suggests a puzzle . in nuclear physics there are bound states whose energy is unnaturally small on the scale of hadronic physics . in order to generate such bound states within a `` natural '' theory it is clear that one must sum a series to all orders . therefore , weinberg proposed @xcite implementing the eft program in nuclear physics by applying the power - counting arguments of chiral perturbation theory to an @xmath7-nucleon effective potential rather than directly to the s - matrix . only @xmath7-nucleon irreducible graphs should be included in the @xmath7-nucleon effective potential . the potential obtained in this way is then to be inserted into a lippmann - schwinger or schrdinger equation and iterated to all orders . there will of course be unknown coefficients in the effective potential , but these can be fit to experimental data as in ordinary chiral perturbation theory @xcite . thus , an eft treatment of the @xmath0 interaction differs in a fundamental way from conventional eft applications like @xmath8 scattering in chiral perturbation theory . in both cases operators are ordered in an effective lagrangian in the same way . however , in @xmath9 scattering the operator expansion in the effective lagrangian maps to a power series in @xmath4 in the scattering amplitude . it is straightforward to see that eft treatments where there is a direct mapping from the lagrangian to the s - matrix are systematic @xcite . on the other hand , when the mapping is from the lagrangian to an effective potential which is subsequently iterated to all orders , many issues arise which lead one to question the existence of a systematic power counting in the potential . in order to raise some of these issues , consider @xmath0 scattering in the @xmath10 channel at momentum scales @xmath11 . the eft at these scales involves only nucleons since the pion is heavy and may therefore be `` integrated out '' . the effective lagrangian thus consists of contact operators of increasing dimensionality constrained by spin and isospin . throughout the paper this is the eft which we consider . we do not intend that this eft should provide a quantitative description of the @xmath0 phase shifts . instead , we study it because scattering amplitudes can be calculated analytically . it therefore allows us to elucidate issues of principle in eft for @xmath0 scattering . one might naively expect to be able to calculate the @xmath0 scattering amplitude directly from the effective lagrangian as a power series in @xmath12 , where @xmath13 is the heavy scale in this problem . it is instructive to show in some detail why this fails and one is led to consider an expansion in the potential . consider an expansion of the amplitude in the @xmath14-wave channels : @xmath15 where @xmath16 , and , with the prevailing prejudice of eft , we anticipate that the dimensionless coefficients @xmath17 , @xmath18 , etc . will be natural ; i.e. of order unity . we know that in the @xmath10 and @xmath19 channels there are , respectively , a quasi - bound state and a bound state at low energies . the power series expansion ( [ eq : texp ] ) with natural coefficients can only be correct if these bound states are at energies @xmath20 . however , in these channels the bound states occur at unnaturally low energies , i.e. at energies @xmath21 . therefore , the coefficients in the expansion must be unnatural they are fixed by the pole positions of the low - lying bound states rather than by the scale of the physics that has been integrated out . this limits the usefulness of an expansion in the amplitude to an extremely restricted domain of validity . on the other hand , if one makes the following expansion of the potential in @xmath14-wave channels : @xmath22 and iterates it via the lippmann - schwinger equation ( see fig . [ fig1 ] ) @xmath23 one may hope to generate ( quasi-)bound states at the appropriate energies while maintaining natural coefficients in the potential . at face value this procedure appears promising . the expansion ( [ eq : vexp ] ) may be truncated at some finite order in the quantities @xmath24 and @xmath25 and , provided @xmath26 the neglected terms will be small . however , in making the expansion in the potential of eq . ( [ eq : vexp ] ) and then iterating , a number of issues arise which are absent in standard eft treatments , and which render this procedure suspect . in what follows we will identify and clarify some of these issues . first , the physical scattering amplitude that is generated when the potential ( [ eq : vexp ] ) is iterated is exactly unitary and therefore necessarily contains arbitrarily high powers in energy or momenta . this occurs regardless of the order to which one is working in the momentum expansion of the potential @xmath27 . this suggests that as long as one is interested in physics near the bound state pole where exact unitarity is important the scattering amplitude may be sensitive to physics at arbitrarily short - distance scales . while this observation does not necessarily invalidate the eft approach , it does mean that one should not simply translate intuition gained about how short - distance physics decouples in perturbative eft calculations ( e.g. in chiral perturbation theory ) to the non - perturbative problem at hand . thus , the first question we face is whether a systematic expansion in the potential translates to a systematic determination of the scattering amplitude . a necessary condition for this to occur is that the integrals in the ls equation , which formally extend up to infinity , be dominated by momenta of order @xmath28 , where @xmath29 is the on - shell momentum , since otherwise there is no small parameter in the expansion ( [ eq : vexp ] ) for @xmath27 . second , non - perturbative regularization and renormalization is required when iterating to all orders using the ls equation . this is an issue because of the presence of ultraviolet divergences , which generally can arise in two ways . when explicit pions are included in the eft , the potential itself may contain loop graphs which require regularization and renormalization . these divergences are easily handled using standard perturbative methods . a second type of ultraviolet divergence arises because in solving the ls equation one integrates the potential over all momenta . given the hard asymptotic behavior inherent to the momentum expansion this necessarily introduces new divergences . the divergences which arise from iterating the potential become more severe as one goes to higher order in the eft expansion . these divergences apparently violate the assumption that @xmath30 , so the existence of a procedure to regularize these divergences and renormalize in such a way that the momentum scales probed inside loops are ultimately well below @xmath13 provides a non - trivial condition on the existence of an eft . in ordinary perturbative eft , all regularization schemes lead to the same renormalized amplitude . this insensitivity to the short - distance physics implied by the regulator makes perturbative eft methods like chiral perturbation theory possible . similarly , there is no hope of defining a sensible eft for @xmath0 scattering unless there is some degree of regulator independence . an interesting feature that arises when regularizing and renormalizing the lippmann - schwinger equation is that not all regulators lead to the same physical results . for instance , we will show that dimensional regularization ( dr ) and cutoff schemes lead to different physical scattering amplitudes . this result should give practitioners of eft pause , since it suggests a sensitivity to short - distance physics which violates the basic tenets of eft . we further demonstrate that in order to generate low - energy ( quasi-)bound states in the @xmath0 system within this eft , one must use a scale - dependent regulator . cutoff schemes provide the most physical means of regularizing the effective theory . we therefore focus on whether it is possible to implement the regularization and renormalization program in a consistent fashion using cutoff schemes . within this restricted class of regulators physical results are insensitive to the specific choice of regulator . however , we find that taking the cutoff to infinity requires that the effective range parameter in the scattering amplitude be negative . in physical processes of interest the effective range parameter is positive . the impossibility of maintaining a positive effective range when the cutoff is removed follows from an old theorem of wigner , which depends only on general physical principles @xcite . therefore , either the cutoff must be kept finite or else all orders in the effective potential must be retained . using an extension of wigner s theorem to the eft where pions are explicitly included , we further argue that the inclusion of pions as explicit degrees of freedom in the eft does not resolve these difficulties . having found that an eft with only nucleons as explicit degrees of freedom can work only if there is a finite cutoff or regulator scale , we investigate the possibility of a cutoff effective field theory . in cutoff eft , cutoff dependence in physical observables is removed systematically by adding higher - dimensional operators to the effective action . however , this procedure makes sense only if there is a small parameter which allows one to conclude that higher - order operators in the action are negligible . we show that there is no such small parameter in @xmath0 scattering when there are low - energy ( quasi-)bound states . we stress that we _ are not _ questioning the existence of an effective field theory of @xmath0 scattering . since the principles underlying eft are causality and locality , an eft of @xmath0 scattering must exist unless a sacred principle is violated . the question we address is : `` what are the relevant low - energy degrees of freedom which must be included explicitly in the eft ? '' the problem of @xmath0 scattering is subtle in this respect because generally there are singularities at unnaturally low energies ( e.g. , the deuteron ) in scattering amplitudes . these singularities are not present as fields in the effective action . what our analysis here shows is that if low - lying bound states are present then there is no systematic eft for @xmath0 scattering in which there are only nucleon fields in the effective lagrangian . this would suggest the necessity of introducing the physics of ( quasi-)bound states in terms of explicit degrees of freedom in the eft description . a step in this direction has been taken in ref . @xcite , but in this paper we will not discuss such ideas further . therefore , we use the term eft to mean an effective field theory in which low - energy bound states are not included as fields in the effective action , i.e. an eft in which only nucleons ( and pions ) are explicit degrees of freedom . in sec . 2 we discuss the @xmath10 channel in @xmath0 scattering and introduce a simple potential model that reproduces data remarkably well . this model demonstrates how an unnaturally small scale can emerge from a theory with only large scales . in sec . 3 we review the effective field theory power - counting argument for @xmath0 scattering , originally proposed by weinberg . in particular , we examine arguments for power counting in the potential rather than in the amplitude in the @xmath0 problem . in sec . 4 we perform a `` leading - order '' eft calculation for @xmath0 scattering in the @xmath10 channel and show that all regularization schemes give the same scattering amplitude . in sec . 5 we perform a `` second - order '' eft calculation using cutoff regularization . we show that if a cutoff is introduced such that the regulated theory respects all physical principles , then the cutoff can not be taken to infinity . we discuss how this result relates to an old theorem of wigner . in sec . 6 we compare dimensional regularization ( dr ) and cutoff regularization schemes . we show that dr and cutoff regularization give different physical scattering amplitudes in the second - order eft calculation , indicating that there is a sensitivity to the choice of regulator , and therefore to short - distance physics . we find that the presence of a regulator scale is necessary to reproduce unnaturally low - lying ( quasi- ) bound states . in sec . 7 we consider the possibility of an effective field theory with a finite cutoff . we find that although a procedure whereby cutoff dependence in the amplitude is removed by adding higher orders in the effective potential can be defined , the small parameter which is necessary to conclude that this is systematic is simply not present . we summarize and conclude in sec . the @xmath31 phase shift in @xmath0 scattering is remarkably well reproduced ( see fig . [ fig2 ] ) up to center of mass momenta of order @xmath13 by the first two terms in the effective - range expansion : @xmath32 , \label{eq : reamp}\ ] ] where @xmath33 is the scattering length , @xmath34 is the effective range , and @xmath29 is the on - shell momentum , @xmath35 . experimentally these parameters are determined to be @xmath36 the extremely large ( negative ) value of the scattering length implies that there is a virtual bound state in this channel very near zero energy . while the value of @xmath34 is consistent with what one might expect for a `` natural '' theory where pions dominate the low - energy physics ( @xmath37 ) , the value of @xmath33 is far from being natural ( @xmath38 ) . = 8 cm in order to understand the emergence of an unnaturally large scattering length it is instructive to consider a simple separable potential model whose scattering amplitude exactly reproduces the effective - range expansion , eq . ( [ eq : reamp ] ) , with no terms of higher order in momenta . consider the potential @xmath39 where @xmath40 is a dimensionless coupling constant and @xmath41 is a scale that characterizes the range of the potential . clearly this model has nothing do with effective field theory . it is a simple model for the @xmath10 channel which captures some of the salient features of the scattering . inserting this potential into the ls equation ( [ eq : lse ] ) gives the on - shell amplitude : @xmath42 \nonumber \\ & = & -\frac{m}{4\pi}\left [ \frac{\lambda}{g}(1+\frac{k^2}{\lambda^2 } ) -(\lambda + ik)\right ] . \label{eq : beanemodel}\end{aligned}\ ] ] this is simply the effective - range expansion to order @xmath43 . one can easily read off the scattering length and effective range , by comparing eqs . ( [ eq : beanemodel ] ) and ( [ eq : reamp ] ) : @xmath44 the choices @xmath45 mev and @xmath46 then reproduce the values ( [ eq : cexpo2 ] ) . note that the effective range is directly proportional to the range , @xmath47 , of the potential . the expression for @xmath33 makes clear that an unnaturally large scattering length occurs if @xmath40 is `` accidentally '' close to one . in this model the large scattering length is the result of a cancellation between two distinct contributions to the scattering amplitude . the first of these contributions comes from the `` strength '' of the potential ( the blob in fig . [ fig1 ] ) since it originates in the @xmath48 term in eq . ( [ eq : beanemodel ] ) . the second term comes from virtual excitations in the loops in fig . this contribution is sensitive to the range of the potential , i.e. @xmath47 . note that both of these contributions to @xmath49 are natural , i.e. given by @xmath50 and @xmath41 respectively . nevertheless , if @xmath51 there is a cancellation between the `` strength '' and the `` range '' , which results in an unnaturally large scattering length . we will see that this qualitative feature persists in the eft treatment . in its standard form chiral perturbation theory is not useful in @xmath0 scattering because of the presence of low - lying bound states which imply a breakdown of perturbation theory @xcite . weinberg argued that this breakdown manifests itself through the presence of infrared divergences in reducible loop graphs in the limit of static heavy nucleons . this argument is not entirely correct it _ is _ possible to have a `` perturbative '' effective field theory for the interaction of heavy particles , provided that there are no low - lying bound or quasi - bound states . ( see , for instance ref . however , since there _ are _ low - lying bound states in nuclear systems , weinberg s main point remains true : if power - counting arguments are relevant in nuclear physics , they must be applied to an effective potential , rather than to the scattering amplitude . in general , the effective @xmath0 potential is the sum of all two - particle - irreducible ( 2pi ) @xmath52 diagrams . when there are pions and nucleons in the eft , the effective potential contains pion - exchange diagrams , short - range contact interactions and a mixture of the two . however , in this paper we are interested in energy scales where @xmath13 is large compared to the initial and final momenta @xmath53 and @xmath54 . in this case eft arguments imply that the pion can be integrated out and replaced by contact interactions . the starting point of any eft calculation is an effective lagrangian , where operators are ordered according to the number of derivatives . the most general effective lagrangian consistent with spin and isospin , including only operators relevant to @xmath10 scattering is @xmath55 there are only tree graphs in the @xmath0 potential obtained from this lagrangian . hence there is a one - to - one correspondence between the operators in the effective lagrangian and the terms in the potential . when iterated using the lippmann - schwinger equation , this potential gives the full @xmath0 amplitude of the theory . this procedure is entirely equivalent to solving the quantum partition function with an effective action to a given order in the derivative expansion , and thereby generating the four - point correlation function . there is therefore , at least for this eft , actually no need to formulate the discussion in terms of potentials . nevertheless , here we use the familiar language of non - relativistic quantum mechanics . the effective potential to order @xmath56 in the derivative expansion can be expressed in the form @xcite @xmath57^{\bar\nu } c_{\bar\nu } \label{eq : chiralexp}\ ] ] where the sum here is over all possible terms extracted from ( [ eq : lag ] ) and @xmath41 is the scale of the physics integrated out , taken to be @xmath13 in the present context . equation ( [ eq : chiralexp ] ) is intended to be symbolic ; @xmath58 indicates that either of these quantities may appear in the expansion , in any combination consistent with symmetry , with only their total power constrained . for instance , at @xmath59 we have the structures @xmath60 and @xmath61 with coefficients @xmath62 and @xmath63 , respectively . it is assumed that the coefficients @xmath64 are natural ; that is , of order unity . the fundamental assumption underlying effective field theory for the @xmath0 interaction is that this expansion in the potential , or equivalently that in the lagrangian , may be sensibly truncated at some finite order , @xmath56 . it is the purpose of this paper to test this assumption . the physical scattering amplitude is obtained by iterating the potential ( [ eq : chiralexp ] ) using the lippmann - schwinger equation @xmath65 which generates the @xmath66-matrix . this procedure is illustrated in fig . [ fig1 ] . by assumption , truncating the expansion ( [ eq : chiralexp ] ) and retaining only its first few terms will be valid only for nucleon momenta well below @xmath41 . it is clear that a method of regularizing the otherwise - divergent integrals which occur when potentials such as ( [ eq : chiralexp ] ) are inserted into the ls equation must be specified . we first investigate these questions by considering the @xmath10 @xmath0 amplitude generated by the `` zeroth - order '' eft potential . this potential is @xmath68 the solution of the ls equation with this potential is easily found to be @xmath69 where @xmath35 is the on - shell momentum , and @xmath70 this integral is linearly divergent . regularizing with a sharp cutoff , @xmath71 , gives @xmath72 where we have neglected terms that vanish as powers of @xmath73 . we choose the renormalization condition @xmath74 where @xmath75 is the `` renormalized potential '' and @xmath33 is the physical scattering length . note that just as in the example of sec . [ sec - sm ] there is a cancellation in eq . ( [ eq : t0cutoff ] ) between the contributions to @xmath49 from the strength of the potential , and the integration over loop momenta . in the limit @xmath76 we then find the physical renormalized scattering amplitude , @xmath77 , \label{eq : t0}\ ] ] derived by weinberg @xcite and kaplan _ et al . _ @xcite . note that this result is completely insensitive to choice of regularization scheme . after renormalization the only piece of the integration over internal loop momenta which survives is the imaginary part , which gives the unitarity branch cut and depends only on the on - shell momentum , @xmath29 . the real piece of @xmath78 , which comes from integration over virtual nucleon momenta , is absorbed into the constant @xmath79 . this is exactly how one renormalizes in standard eft , where power - law divergent pieces of loop integrals are absorbed by bare coefficients in the amplitude . one efficient way to ensure that the power - law divergent parts of loop graphs never enter the calculation is to use dimensional regularization . however , note that if dr is used , as in ref . @xcite , then the cancellation seen in eq . ( [ eq : t0cutoff ] ) is destroyed , since the linear divergence is zero in dr . although of no physical consequence here , this effect will prove important at second order in the eft . the result ( [ eq : t0 ] ) appears promising : we have reproduced the scattering length term of the effective - range expansion . moreover , we see that at `` leading order '' in the effective field theory calculation the scattering amplitude is regularization - scheme independent ; the real part of the integration over loop momenta does not have any effect on the final physical result . in order to test the robustness of this result we must calculate corrections . at the next order the potential has momentum dependence and therefore is more sensitive to short - distance physics . from the effective lagrangian ( [ eq : lag ] ) we extract the following `` second - order '' eft potential : @xmath80 an easy way to solve the ls equation with this potential is to observe that @xmath81 may be written as a two - term separable potential . the on - shell t - matrix is easily found to be ( see ref . @xcite for details ) : @xmath82 where @xmath83 the integrals @xmath84 , @xmath85 and @xmath86 are all divergent , and so this scattering amplitude must be regularized and renormalized . it is instructive to carry the renormalization a certain distance without specifying a regularization scheme . we choose as renormalized parameters the experimental values of the scattering length , @xmath33 , and the effective range , @xmath34 . in other words , we fix @xmath87 and @xmath88 by demanding that @xmath89.\ ] ] unitarity guarantees that the imaginary parts agree . equating the real parts at @xmath90 yields @xmath91 equating the @xmath43 coefficients gives @xmath92 . \label{eq : recondn}\ ] ] note that the physical parameters are complicated non - linear functions of divergent integrals and bare parameters . the scattering amplitude can now be written as @xmath93 with @xmath94 notice that the scattering amplitude is now expressed in terms of the physical parameters @xmath33 and @xmath34 and the linearly divergent integral @xmath95 . it is not surprising that we have a divergence left over after imposing our renormalization conditions since we had three distinct divergences@xmath95 , @xmath84 and @xmath85and only two free parameters@xmath87 and @xmath88 . if we now regularize @xmath95 using a sharp cutoff , @xmath71 , we find in the @xmath76 limit : @xmath32 . \label{eq : cutoff2}\ ] ] which is the effective - range expansion to order @xmath43 . given the success of eq . ( [ eq : cutoff2 ] ) in describing the @xmath10 @xmath0 phase shifts this result seems quite promising . however , there is a subtlety hidden in this result . the second term in eq . ( [ eq : recondn ] ) disappears when the cutoff is removed . as @xmath96 @xmath97 from which we see that @xmath98 , ( recalling that @xmath99 defined in eq . ( [ in ] ) is negative definite ) , _ no matter what real value we choose for the constant @xmath88 in the bare lagrangian_. this result is at first sight rather peculiar since it implies that the sign of @xmath34 is predicted in the effective theory with the cutoff removed . it is the behavior of the power - law divergences arising from integrals over internal loop momenta which leads to the result @xmath100 . this is in contrast to the `` zeroth - order '' amplitude , where these integrals over internal loop momenta did not affect the final on - shell amplitude . moreover , the result @xmath98 as @xmath101 clearly means that in order to describe a system with @xmath102 using this potential one can not remove the cutoff but must instead maintain the hierarchy @xmath103 . ( the necessity of keeping the cutoff small and below @xmath104 is supported by direct calculation @xcite . ) the fact that one can not take the cutoff @xmath71 to infinity and still obtain a positive effective range is , in fact , not surprising . long ago , assuming only causality and unitarity , wigner proved that if a potential vanishes beyond range @xmath105 then the rate at which phase shifts can change with energy is bounded by @xcite @xmath106 it is straightforward to show that this translates to @xcite @xmath107 . \label{eq : wb}\ ] ] in light of this general result it is no longer surprising that the effective range was found to be negative in the eft calculation when the cutoff @xmath71 was removed . wigner s bound implies that no theory obeying the general physical principles of causality and unitarity can support a positive effective range with a zero - range force . when using a sharp cutoff the regulated hamiltonian satisfies all physical principles and is therefore a legitimate physical theory , with the cutoff identified with the range ; i.e. @xmath108 . consequently in the limit @xmath96 one is trying to use zero - range potentials to produce a positive effective range . equation ( [ eq : wb ] ) shows that this is impossible . therefore , wigner s bound indicates that there is no eft treatment of @xmath0 scattering with nucleons as the only explicit degrees of freedom in which it is possible to remove the cutoff from the problem . one might think that including explicit pions in the effective theory description will ameliorate the situation . this is not so ; it has been shown that the wigner bound is a general feature of contact interactions and continues to apply in the presence of pions @xcite . as we have seen , _ a priori _ there are two sources of short - distance physics in the effective field theory calculation : operators in the effective lagrangian of increasing dimensionality whose coefficients encode information about short - distance physics , and the effects of virtual particle excitations within loop graphs . in ordinary perturbative eft calculations the existence of a consistent power - counting scheme in the s - matrix relies on removing the short - distance physics that arises in loop graphs via the renormalization procedure @xcite . since the short - distance physics is removed in a manner insensitive to choice of regularization scheme it is economical to use dr to regularize and renormalize , because dr respects chiral and gauge symmetries . when considering the relevance of eft methods in nuclear physics one might choose to extrapolate intuition gained from perturbation theory and regulate using dr . however , it is important to realize that the use of dr implicitly _ assumes _ that eft can work only if the short - distance physics buried in loop graphs _ does not _ contribute to low - energy physics . of course there is no need to make this assumption , cutoff regularization is the most physically - transparent method of regularization and keeps track of all sources of short - distance physics in the calculation . in this section we compare dr and cutoff schemes . there are two fundamental points we wish to make : ( i ) dr and cutoff regularization do not lead to the same physical scattering amplitude . therefore in the nonperturbative context of @xmath0 scattering , short - distance physics from loops is important , in contrast to the situation in perturbative calculations ; ( ii ) when low - lying ( quasi-)bound states are present , a necessary ( but not sufficient ) condition for a workable eft treatment of @xmath0 scattering is that short - distance effects from loops contribute to the physical scattering amplitude . we now use dimensional regularization ( dr ) to regulate the integrals @xmath95 , @xmath84 , and @xmath85 which appear in the expression for @xmath109 above . as discussed above , dr is a convenient way to implement the idea that is central to the success of perturbative eft : the power - law divergent pieces of integrals over internal loop momenta should not affect the final physical scattering amplitude . if this holds true then dr should lead to the amplitude found using cutoff regularization . in dr all power - law divergences vanish , therefore @xmath110 . consequently , eqs . ( [ eq : tsimple ] ) and ( [ eq : adef ] ) reduce to @xmath111 , \label{eq : tdr}\ ] ] in agreement with kaplan _ et al . _ the renormalized scattering amplitude is not the same as that obtained using cutoff regularization . therefore the amplitude calculated using the second - order eft potential is not independent of regularization scheme , and so the scattering amplitude must be sensitive to the short - distance physics from loops . this raises the question of whether the eft treatment based on potentials has really managed to separate the long - distance physics from the short - distance physics . this separation is a precondition for the successful application of eft ideas . clearly the dr amplitude maps to the effective - range expansion only for momenta @xmath112 . this is consistent with eq . ( [ eq : cutoff2 ] ) if both @xmath33 and @xmath34 are natural , as then eqs . ( [ eq : tdr ] ) and ( [ eq : cutoff2 ] ) will agree within the domain of validity of the eft , @xmath113 . the effective theory is then perturbative . however , if the scattering length is unnaturally large the momentum domain within which the two forms are equivalent becomes small . in fact , eq . ( [ eq : tdr ] ) reproduces the data in the @xmath10 extremely poorly , since it only agrees with the phenomenologically efficacious amplitude ( [ eq : reamp ] ) for very small @xmath29 . it was argued elsewhere that this failure implies that the effective theory calculation is not valid in the presence of a large scattering length @xcite . however , this failure of dimensional regularization is no surprise when one considers the simple model of the effective - range expansion discussed in sec . [ sec - sm ] . there it was clear that an unnaturally large scattering length occurs through the cancellation between two terms in the scattering amplitude , one of which arises from integration over loop momenta . this cancellation between `` range '' and `` strength '' is a general feature whenever one solves the schrdinger equation with some finite - range potential . by its very nature , dr _ can not _ give a scattering amplitude that feels this cancellation , since it discards all short - distance physics that comes from loops . in fact , since there are no logarithmic divergences in this problem and the loop graphs have no finite real part , it is straightforward to show that dr gives an amplitude in which only the absorptive parts of the loop graphs are retained . ( see fig . [ fig3 ] . ) hence when there are low - lying bound states the bound - state energy sets the scale of the coefficients in the potential of the dimensional regularization calculation . but recall that this was precisely the problem that iterating the potential via the ls equation was supposed to avoid . thus the implementation of the general cancellation between range and strength must be an element of any eft description of @xmath0 scattering . in general , information about the range of the potential enters through these power - law divergences , and dr s neglect of all power - law divergences means it discards this information on the range of the interaction . this is no accident , dr is designed to be a scale - independent regularization scheme . this aspect of the dr prescription allows it to violate the wigner bound and give an amplitude which has a positive effective range from an eft potential which is apparently zero range . therefore we argue that a regulator that introduces a new scale in @xmath0 scattering must be used . however , in the next section we will show that effects due to this new scale actually destroy power counting . therefore , we believe that the intuition that eft can work only if scattering amplitudes are insensitive to the short - distance physics arising from loop graphs , which is the primary motivation for using dr , is actually correct . in other words , in all circumstances in which eft is systematic dr can , and for simplicity should , be used . in the calculation of @xmath0 scattering in the presence of low - energy bound states dr fails only because the sensitivity to short - distance physics inherent in this problem means that no sensible eft can be formulated for this problem . the arguments of the previous section show that an eft approach to @xmath0 scattering will only be efficacious in the presence of low - lying ( quasi-)bound states if the short - distance physics in loop graphs contributes to the physical scattering amplitude . if a cutoff is introduced , so that we can keep track of this physics , the bound ( [ eq : wb ] ) indicates that it can not be taken to infinity . therefore , in this section we investigate the possibility of an eft for @xmath0 scattering in the presence of a finite cutoff . this physically intuitive approach has been advocated by lepage @xcite . the idea is that one takes an underlying theory of @xmath0 interactions and introduces a ( sharp or smooth ) momentum cutoff @xmath71 representing the scale at which the first new physics becomes important . all loops now only include momenta @xmath114 . of course , one must compensate for the effects of these neglected modes . however , lepage argues that since these modes are highly virtual , one may approximate their effects by a sequence of local contact interactions . furthermore , if the cutoff @xmath71 is placed well below the mass @xmath41 of some exchanged quantum , then , for momenta @xmath115 and @xmath116 below the cutoff , the exchange of this quantum : @xmath117 may be replaced by a contact interaction , since @xmath118 . therefore the effects coming from exchanges of quanta with masses well above the cutoff scale @xmath71 may also be approximated by contact interactions . for the numerical application of these ideas to the @xmath0 problem see @xcite . now , all that has been said in the previous paragraph still applies if we set our cutoff @xmath71 below the scale @xmath13 . then the only explicit degrees of freedom in the problem are nucleon modes with momentum @xmath119 . all higher - momentum nucleon modes _ and _ all exchanged mesons are integrated out . this cutoff effective field theory of the @xmath0 interaction is of little practical use , but can be investigated analytically in a way that raises issues of principle . the effective lagrangian is @xmath120 where the dots signify operators with more than two derivatives . as discussed in sec . [ sec - eft ] , in the limit that the nucleon is very heavy we can formally calculate the four - point function exactly . to do this we merely write down the effective potential which corresponds to this lagrangian : @xmath121 \theta(\beta - p ) \theta(\beta - p ' ) , \label{eq : veffexp}\ ] ] and iterate via the lippmann - schwinger equation ( [ eq : lse2 ] ) . here the theta function imposes a sharp cutoff , and so all integrals ( not just the divergent ones ) will be cut off sharply at momentum @xmath71 . after renormalization the coefficients @xmath87 , @xmath88 , etc . , will , of course , depend on the cutoff scale @xmath71 , as well as physical scales in the problem . of course , the expression ( [ eq : veffexp ] ) is an infinite series , and for practical computation some method of truncating it must be found . the fundamental philosophy of cutoff eft provides a rationale for this as follows . if we work to any finite order in the effective potential , cutoff - dependent terms in the scattering amplitude will appear . these must be in correspondence with neglected higher - order operators in @xmath27 . if such terms are progressively added to @xmath27 , one may remove the cutoff dependence order - by - order . below we will see that one can indeed define such a `` systematic '' procedure in this problem . however , _ in order that it really make sense to truncate the effective potential ( [ eq : veffexp ] ) at some finite order it must be that the operators that are neglected are in some sense small . _ naively one would expect that a sufficient condition for the operator with @xmath7 derivatives to be a small correction to the overall potential is : @xmath122 where @xmath123 is the momentum operator . this operator yields some characteristic momentum in the problem . now , in perturbation theory the short - distance physics can be removed from loops via renormalization , thereby achieving a clean separation of scales , and so the characteristic momentum is simply the external momentum . thus , in perturbative calculations the operator hierarchy ( [ eq : pcondn ] ) can be maintained , as long as one restricts oneself to small enough external momenta . however , in a nonperturbative context we have already seen that physics from loops must enter into the calculation if low - lying ( quasi-)bound states are to be generated . consequently , we expect that momenta at scales @xmath124 are important . below we show that @xmath125 more generally , condition ( [ eq : cc2 ] ) becomes : @xmath126 combining eqs . ( [ eq : cmcnbeh ] ) and ( [ eq : pcondn ] ) we obtain @xmath127 equation ( [ eq : success ] ) reflects the fact that the crucial momenta in the problem must be well below the scale where the cutoff effective field theory breaks down . the external momentum can certainly be kept in this energy regime . however , below we also show that , even if the external momentum scales are small , the quantum mechanical averages of operators @xmath128 involve @xmath71 . in this circumstance there is little justification for truncating the effective lagrangian and keeping only a finite number of operators . consider once again the @xmath10 @xmath0 amplitude generated by the `` zeroth - order '' eft potential . this potential in cutoff eft is given by @xmath129 the solution of the lippmann - schwinger equation is then @xmath130 where @xmath131 matching the solution to the effective - range expansion gives @xmath132,\ ] ] where the renormalization condition for @xmath133 is still eq . ( [ eq : t0cutoff ] ) . note that @xmath33 is cutoff independent as before , but now we have a non - zero effective range in this zeroth - order calculation : @xmath134 this @xmath135 arises from the energy dependence introduced by the cutoff . the appearance of a cutoff - dependent quantity like @xmath135 in the renormalized amplitude only reflects the limited accuracy of this calculation . below we will see that @xmath34 can be made cutoff independent by including higher - order terms in the effective potential . we now turn our attention to the amplitude and renormalization conditions that arise when one takes the `` second - order '' effective potential : @xmath136 \theta(\beta - p ) \theta(\beta - p'),\ ] ] and iterates it via the lippmann - schwinger equation . repeating the analysis of sec . [ sec - nu2 ] , but now with all integrals cutoff sharply at momentum @xmath71 , we again obtain the amplitude : @xmath137 while @xmath84 , and @xmath85 are as before , provided that it is understood that a sharp cutoff is to be used to render the divergences finite . > from these equations we obtain the following equations for @xmath87 and @xmath88 : @xmath138 where , as above , the last term in eq . ( [ eq : c2beta ] ) arises because the presence of the cutoff generates additional energy dependence when the integral @xmath139 is evaluated . of course , as @xmath71 is varied the @xmath87 and @xmath88 that satisfy eqs . ( [ eq : cbeta ] ) and ( [ eq : c2beta ] ) will change significantly . however , because one is fitting to low - energy scattering data different values of @xmath71 will not lead to any fundamental differences in the low - energy t - matrix . since @xmath87 and @xmath88 are fit to the first two terms in the effective - range expansion sensitivity to the cutoff enters first through the fourth - order term in the effective - range expansion . therefore the effects of choosing different cutoffs can only appear in the on - shell amplitude at order @xmath140 , where @xmath29 is the on - shell momentum . this is precisely the sense in which working to higher order in the potential will improve the fit `` systematically '' . if we constructed the `` next - order '' effective potential , @xmath141 and refitted the ( four ) coefficients appearing in it to the low - energy data then sensitivity to the cutoff will be pushed back to an @xmath142 correction . thus , the introduction of higher - dimension operators allows for the `` systematic '' removal of cutoff dependence in the amplitude . in general , the two - body scattering amplitude will be insensitive to the choice of regulator , provided one restricts oneself to the domain of validity of the cutoff effective field theory , @xmath143 . this implies that so long as we renormalize the coefficients to the low - energy scattering data we `` systematically '' improve our fits to the scattering data . this is the systematicity seen in the work of lepage @xcite , and that of scaldeferri _ et al . _ @xcite . nevertheless , this does not yield a systematic eft potential in the sense we have defined it here . after all , _ any _ parameterization of a potential which is rich enough to fit the lowest terms in the effective - range expansion will similarly be improved as one adds parameters to it . the question is whether one can use the power counting to argue _ a priori _ that certain contributions to the potential of eq . ( [ eq : veffexp ] ) will be systematically small and hence can be neglected at some specified level of accuracy . to answer this question we must examine the values of @xmath87 and @xmath88 which are required to solve the equations above . assuming that we are working in the regime where @xmath144 it follows that the second of these two equations becomes @xmath145 where @xmath146 . this leads to a quadratic equation for @xmath147 , which for values of @xmath71 up to some @xmath148 has real solutions . for @xmath149 the renormalization condition for @xmath88 has no real solution if @xmath150 , as discussed in sec . [ sec - nu2 ] and ref . however , if @xmath151 , then we see that @xmath152 . consequently , we infer @xmath153 this behavior for @xmath87 is the same as that obtained in the zeroth - order calculation in the case @xmath144 . since we demand that the physical observable @xmath33 be the same in both the zeroth and second - order calculations , the renormalization conditions ( [ eq : t0cutoff ] ) for @xmath133 and ( [ eq : cbeta ] ) for @xmath87 give @xmath154 where @xmath155 ( @xmath156 ) is @xmath95 times the original dimensionful coefficient . now @xmath157 , provided that @xmath151 . thus , eq . ( [ eq : cc0comp ] ) shows that if @xmath158 , @xmath155 differs significantly from the lowest - order result . hence , if both the zeroth and second - order effective potentials are to reproduce the scattering length then there must be a large difference between them . this already suggests that a systematic truncation of the cutoff eft may not be possible . note that such problems do not arise if @xmath33 is natural , i.e. of order @xmath159 , and @xmath71 is chosen to be much less than @xmath13 ; then the leading order behavior of the coefficients @xmath87 and @xmath88 is very different . in fact , @xmath160 in this case all loop effects coming from @xmath88 are suppressed by a factor of at least @xmath161 . indeed , all loop effects are suppressed by a factor of at least @xmath162 . therefore , if @xmath163 a non - perturbative calculation is not necessary . in other words , if the experimental parameters are natural then cutoff field theory with @xmath163 gives a perturbative eft in which loop graphs are consistently suppressed @xcite . however , if a perturbative calculation is performed then the regularization scheme chosen becomes immaterial , as the short - distance physics may be renormalized away . dimensional regularization would be an equally valid , and considerably simpler , way of implementing a systematic perturbative eft . therefore we do not consider the case where @xmath49 is of order @xmath13 any further here , but instead return to the case where @xmath49 is unnaturally small , as is true in the @xmath10 and @xmath19 channels in nature . for unnaturally long scattering lengths ( [ eq : cc2beh ] ) shows that : @xmath164 consequently the condition ( [ eq : pcondn ] ) becomes , for @xmath165 , @xmath166 . it is easy to give a heuristic justification of why this behavior of the coefficients arises in a non - perturbative cutoff eft calculation , and why we expect similar behavior to all orders in the effective potential . after all , the choice of a theta function to regulate the momentum - space integrals as in eq . ( [ eq : veffexp ] ) is entirely arbitrary . all that has been said above could be reformulated with a smooth cutoff . this would result in an effective potential of the form @xmath167 g(p^2/\beta^2,p'^2/\beta^2 ) \label{eq : veffexp2}\ ] ] where @xmath168 obeys @xmath169 , @xmath170 and @xmath171 faster than any power of @xmath172 as @xmath173 with @xmath174 held fixed . in a non - perturbative calculation the effective potential should be essentially unaltered by this change in the form of the cutoff . however , this necessarily means that the ratios @xmath175 differ from those @xmath176 by terms of order @xmath177 . therefore for a generic cutoff function @xmath40 the ratio @xmath176 must be of order @xmath177 . now , if the ratio @xmath176 goes like @xmath177 , then the condition ( [ eq : pcondn ] ) becomes eq . ( [ eq : success ] ) . however , the effective potential is to be used in a momentum regime which extends up to @xmath71 , and at the upper end of this momentum regime it is clear that all terms in the expansion for @xmath27 are equally important . of course , if internal loops were dominated by the external momentum , @xmath29 , and so @xmath178 , then this behavior of the coefficients would not be cause for concern , since @xmath179 could be maintained . however , virtual momenta up to @xmath71 flow through all internal loops , and we have already argued that the final amplitude is sensitive to these virtual effects which come from the range of the underlying interaction . therefore we believe it makes sense to consider a quantum average in evaluating the condition ( [ eq : pcondn ] ) , since such an average is sensitive to these virtual effects . although there is no bound state in this channel , all arguments about the size of operators in the effective action would apply equally well if there was a low - energy bound state in the channel under consideration . so , let us evaluate quantum averages of the operator @xmath180 using the bound - state wave function obtained from the zeroth - order potential given by eq . ( [ eq : v0coeft ] ) . we use a bound - state wave function because scattering wave functions are not easily normalizable , and so quantum averages become difficult to calculate . the zeroth - order potential yields a wave function for the bound state of energy @xmath181 , @xmath182 where @xmath183 is some normalization constant that is yet to be determined . in the regime @xmath184 this wave function gives @xmath185 thus , condition ( [ eq : success ] ) is apparently satisfied . however , if this condition is to hold as an operator equation it must be that , for any @xmath186 : @xmath187 for @xmath188 this does not hold . that is to say , if we calculate @xmath189 with the wave function @xmath190 there is no reason to truncate the expansion at any finite order , since all terms beyond zeroth order contribute with equal strength to the quantum average . if there was systematic power counting for the @xmath0 potential in cutoff field theory then the contribution of these `` higher - order '' terms in the potential should get systematically smaller as the `` order '' is increased . however , it is clear that this does not happen rather , all terms beyond zeroth order contribute to the potential at the same order . therefore one can not justify a truncation of eqs . ( [ eq : veffexp ] ) and ( [ eq : veffexp2 ] ) at some finite order in @xmath53 and @xmath54 . for the reasons explained above , such a truncation may result in a good fit to the experimental data for on - shell momenta @xmath179 , but it is not based on a systematic expansion of the @xmath0 potential in powers of momentum . our results are discouraging for any effective field theory description of @xmath0 scattering in which low - lying bound states are not included as explicit degrees of freedom . these low - energy bound and quasi - bound states exist in nuclear physics , and so , if efts in which they are not explicitly included are to be useful , power - counting arguments can only apply to the @xmath0 potential , rather than to the @xmath0 amplitude . solving such an eft with nucleons would appear to be a simple problem in non - relativistic quantum mechanics . however , the problem is complicated by the appearance of power - law divergences which arise from the internal loop integrations in the lippmann - schwinger equation . in perturbative efts such divergences are renormalized away and have no effect on the physical amplitude . in fact , if this were not the case there would be no consistent power counting in these theories . by contrast , when the second - order eft potential is iterated to all orders , we find that the resulting scattering amplitude is inevitably regularization scheme dependent : dimensional regularization and cutoff regularization give different renormalized scattering amplitudes . we are therefore forced to conclude that the physical scattering amplitude is sensitive to the power - law divergences arising from loop integrations . one way to understand this sensitivity is to realize that when cutoff regularization is used the potential acquires a range , which is naively @xmath73 . when the schrdinger equation is solved all physical observables depend on the upper limit in internal loop integrations , @xmath71 . in any perturbative calculation this dependence can be removed by the introduction of a finite number of higher - dimensional operators in the lagrangian . however , for some observables , e.g. the effective range , this can not be achieved for arbitrary cutoff in our non - perturbative calculation : these observables still depend on the cutoff ( equivalently , on the `` range '' of the potential ) , even after renormalization , as evidenced by the bound ( [ eq : wb ] ) . when dimensional regularization discards all power - law divergences in order to maintain its scale independence it removes , by fiat , the power - law divergences which represent an important part of the physics of the range of the @xmath0 interaction . this suggests dr should not be used to regularize this potential . it is straightforward to show that if dr is used to regularize a potential which generates low - energy bound or quasi - bound states , the resulting coefficients in the dimensionally regularized potential will be governed by the bound - state energy , and not by the underlying scale of the eft @xcite . thus , if dr is used to regularize the eft @xmath0 potential the radius of convergence of the resulting expansion will be small as seen in the work of kaplan _ et al . _ @xcite . again , this may be thought of as a consequence of dr s discarding power - law divergences : such power - law divergences implement the cancellation between `` range '' and `` strength '' which was seen in the simple model of sec . [ sec - sm ] ( see also ref . @xcite ) to be the key to allowing a theory with natural scales to produce `` unnaturally '' low - lying bound states . _ therefore the only hope for a systematic and successful eft description of @xmath0 scattering with nucleons alone lies with methods of regularization which preserve information on these power - law divergences . _ cutoff regularization is such a regularization scheme . however , the effective range is positive in nature , and so we discover that if we wish to use an effective field theory with a cutoff to describe the real world we must keep the cutoff finite and below the scale @xmath41 of the short - distance physics . while this may provide a procedure for fitting low - energy scattering data , such an approach can only be regarded as systematic if the important momenta inside internal loops are well below the cutoff scale . if this is not true physical observables will be sensitive to details of the potential in precisely the region where the artificially - introduced cutoff plays an important role . as we saw in sec . [ sec - coeft ] there will then be no justification for keeping some terms in the eft potential and neglecting others . now , while the external momentum @xmath29 can clearly be kept small compared to the cutoff , we have just argued that in order to have a successful eft in the presence of low - lying bound states the virtual momenta of order the cutoff must play a significant role ! this becomes apparent mathematically through the strong cutoff dependence of the quantum averages of momentum operators . _ we therefore conclude that there is no small parameter which allows one to consistently truncate the regularized @xmath0 potential at some finite order in the derivative expansion . _ the preceding argument that power counting fails might be regarded as incomplete since we have demonstrated a violation of power counting at the level of the bare parameters in the regularized potential . in general one does not expect power counting in effective field theories to be respected at the level of the bare parameters in the lagrangian but rather at the level of the renormalized parameters . for example , in ordinary chiral perturbation theory , the bare parameters which enter the perturbative calculations are strictly infinite , while the renormalized amplitudes have natural power counting . here we have shown that sensible regularization schemes inevitably involve a finite cutoff and therefore the only object in which systematic power counting can be defined is the potential with bare parameters ; _ there simply is no renormalized potential_. one might attempt to _ define _ the renormalized potential as the @xmath191 matrix , @xmath192)^{-1}$ ] . this quantity is certainly well defined . however , as discussed elsewhere in this paper , the entire reason for doing eft at the level of the potential is that we are interested in developing an eft in the vicinity of the low - energy pole , and power counting in @xmath66 or @xmath191 fails in the presence of this pole . it might also be suggested that the inclusion of pions as explicit degrees of freedom in the effective field theory will ameliorate the situation . however , we do not believe this to be the case . after all , provided the energy under consideration is low enough , nothing in the general eft arguments applied to the @xmath0 potential in refs . @xcite requires the inclusion of explicit pions . when pions are added as fields in the lagrangian of the eft the basic object of the calculation remains an eft potential which is to be iterated via the lippmann - schwinger equation . this potential still requires regularization . if cutoff regularization is used a generalization of the wigner bound implies that even in this eft with explicit pions the cutoff can not be taken to infinity @xcite . dr could also be used to render finite the divergent integrals which arise upon iteration , but again , doing this implicitly assumes that details of the short - distance physics do not affect the physical scattering amplitude . ( in fact , additional problems , beyond those discussed here , arise when dr is applied to the eft with pions , as noted in ref . therefore it seems that cutoff eft , with a finite cutoff @xmath71 , is the safest way to regularize the eft with explicit pions . whether the inclusion of explicit pions modifies our conclusions about cutoff eft is a matter for numerical investigation . previous numerical calculations indicate that the inclusion of explicit pions does not change the scales of the coefficients @xmath87 and @xmath88 @xcite . hence , it does not appear that the explicit inclusion of pions in the eft will modify our conclusion that there is no systematic eft for @xmath0 scattering . richardson _ et al . _ have independently argued that the eft approach to @xmath0 scattering is not systematic @xcite . however , their definition of systematic is more restrictive than ours . they define systematicity to mean that the scattering length @xmath33 should receive no contribution from the term with two derivatives in the effective lagrangian ( the coefficient @xmath88 ) . since @xmath33 does receive a contribution from this term ( see eq . ( [ eq : cbeta ] ) ) they claim that the eft ( [ eq : lag ] ) is not systematic . this definition of systematicity is patterned on perturbative eft , and may be unduly rigid . we would argue that the appearance of @xmath88 in @xmath33 does not of itself imply non - systematicity . after all , the contribution of @xmath88 to @xmath33 might be systematically small . this happens , for instance , if a finite cutoff @xmath163 is used to regularize the divergences and the physical observables are all natural . on the other hand , in the case of interest here , where @xmath49 is unnaturally large , we saw in the previous section that the contribution of the appropriately - rescaled @xmath88 to @xmath33 _ is _ of order one . the coefficients @xmath87 that are obtained in the zeroth and second - order calculations then differ significantly . perhaps this is only symptomatic of the more general malaise already diagnosed : the higher - order terms in the effective action can not be safely ignored . in summary , the appearance of higher - order coefficients in the expression for low - momentum observables does not necessarily imply non - systematicity . but here we expect the contribution of these higher - order operators to low - momentum observables to be large , because the effective potential can not be systematically truncated . the potential of eft for the @xmath0 interaction lies partly in embedding the @xmath0 scattering amplitude derived in the eft in scattering processes involving three or more nucleons . one would hope to apply power - counting arguments to determine those contributions that are important in the nuclear force and those that are systematically suppressed . in principle , this approach has predictive power . for example , as noted by weinberg , in a nucleus three - body forces are characteristically down by two powers of @xmath193 compared to two - body forces , and so might be expected to yield small contributions to nuclear observables . furthermore , four - body forces are down by four powers of @xmath193 and thus could be expected to be negligible @xcite . this suggests that at some reasonably crude level one might simply neglect @xmath194-body force effects , @xmath195 in nuclei . an analysis of realistic phenomenological nuclear force calculations @xcite and the relative contributions of @xmath194-body forces to nuclear binding energies shows qualitative agreement with this prediction . unfortunately , in light of the results of this paper one must question in what sense it is valid to use power counting to deduce a hierarchy of @xmath194-body forces . if eft power counting is a valid way to deduce that certain terms in the hamiltonian are suppressed then we would have seen that higher - derivative terms in the potential were negligible . in fact , in none of the regularization schemes discussed here was this the case . nevertheless , we stress that effective field theory continues to be the most promising method of systematizing nuclear physics . what we have found here is that an eft for @xmath0 scattering with only nucleons must contain _ all _ operators in the effective action , and therefore is not useful . this has a simple physical interpretation . the @xmath0 scattering amplitude is sensitive to all operators in the effective action because it has a singularity corresponding to the ( quasi-)bound state pole . by definition , this pole `` feels '' all distance scales and therefore naturally requires that operators to all orders in momentum in the effective action be present in order to describe it . the only way around this dilemma is to include the ( quasi-)bound state pole as a `` fundamental '' degree of freedom in the effective theory , while retaining the four - point @xmath0 interactions of the lagrangian ( [ eq : lag ] ) @xcite . in the resulting eft , these @xmath0 contact interactions can be treated perturbatively with consistent power counting . application of these ideas in the three - nucleon scattering problem is being pursued in ref . we thank m. k. banerjee , m. c. birse , w. d. linch , m. a. luty , j. a. mcgovern , e. f. redish , u. van kolck , s. j. wallace , and s. weinberg for useful and enjoyable discussions on the topics discussed in this paper . this research was supported in part by the u. s. department of energy , nuclear physics division ( grant de - fg02 - 93er-40762 ) .

melnikov s method is an analytical way to show the existence of classical chaos generated by a smale horseshoe . it is a powerful technique , though its applicability is somewhat limited . in this paper , we present a solution of type iib supergravity to which melnikov s method is applicable . this is a brane - wave type deformation of the adss background . by employing two reduction anstze , we study two types of coupled pendulum - oscillator systems . then the melnikov function is computed for each of the systems by following the standard way of holmes and marsden and the existence of chaos is shown analytically . * melnikov s method in string theory * + yuhma asano , hideki kyono and kentaroh yoshida _school of theoretical physics , dublin institute for advanced studies , + 10 burlington road , dublin 4 , ireland _ +_department of physics , kyoto university , + kitashirakawa oiwake - cho , kyoto 606 - 8502 , japan _

1607.07302

the gauge / gravity correspondence is a fascinating topic in the study of string theory . a certain class of it preserving a conformal symmetry , in which a string theory is defined on an anti - de sitter ( ads ) space and its dual is a conformal field theory ( cft ) , is called the ads / cft correspondence @xcite . a prototypical example is a duality between type iib string theory on the ads@xmath0s@xmath1 background and the four - dimensional @xmath4 @xmath5 super yang - mills theory in the large @xmath6 limit . remarkably , the integrability structure exists behind it and hence one can check the conjectured relations in a rigorous way by employing various integrability techniques ( for a big review , see @xcite ) . in particular , the ads@xmath0s@xmath1 superstring is classically integrable in the sense of kinematical integrability @xcite . the integrability exposed in this case is , however , rarely exceptional and such a good property is not equipped with general examples of the ads / cft ( or gauge / gravity ) correspondence . holographic interpretations in the gauge / gravity correspondence are usually concerned with curved string backgrounds and hence classical motions of a string are described by non - linear equations . but most of the non - linear equations are not integrable and therefore the behavior of classical string solutions should become chaotic . more intriguingly , the holographic counterpart of chaotic strings should exist on the gauge - theory side as well , but it has not been clarified yet . if it has been done , then one could open up a new frontier in the study of the gauge / gravity correspondence . a well - studied example of non - integrable backgrounds is ads@xmath7 @xcite , where @xmath8 is a five - dimensional sasaki - einstein space @xcite has been completed with a supertrace operation @xcite . ] . chaotic string solutions were found in @xcite by computing poincar sections . the chaotic behavior remains even in a near - penrose limit @xcite . similar studies have been done for many backgrounds in the preceding works @xcite . similarly , chaotic motions of d0-branes can also be studied as well as classical strings . the d0-brane dynamics is described by a matrix model proposed by banks , fischler , shenker and susskind ( bfss ) @xcite . chaotic d0-branes in the bfss matrix model were studied in @xcite by following a seminal paper on chaos in a classical yang - mills theory @xcite . in comparison to the bfss case , a matrix model on a pp - wave background , which was proposed by berenstein , maldacena and nastase ( bmn ) @xcite , has a strong advantage that there is no flat direction and all of the trajectories are definitely bounded . classical chaos in the bmn matrix model was shown in @xcite by following @xcite . the chaos at finite temperature was also studied in @xcite in relation to the fast scrambler scenario @xcite . a very recent work @xcite investigated chaotic motions of chiral condensates in a holographic qcd setup @xcite and displayed the dependence of lyapunov exponent on the rank of gauge group @xmath6 and t hooft coupling @xmath9 . a lot of achievements have been obtained for the chaotic behavior of strings and d - branes as introduced above . motivated by this progress , we are concerned here with an application of melnikov s method @xcite in the context of the gauge / gravity correspondence . this method can show analytically the existence of chaos generated by a smale horseshoe , though its applicability is somewhat restricted . in this paper , we will present a string background to which melnikov s method is applicable . this is a brane - wave type deformation of ads@xmath0s@xmath1 presented in @xcite . by employing two reduction anstze , we study two types of coupled pendulum - oscillator systems . then the melnikov function is computed for each of the systems in the standard way of holmes and marsden @xcite . this paper is organized as follows . section 2 gives a concise review of melnikov s method and provides simple examples . in section 3 , we introduce a brane - wave type deformation of ads@xmath0s@xmath1 , and study two anstze . the associated melnikov function is computed for each of the cases . section 4 is devoted to conclusion and discussion . in appendix [ elliptic functions ] , the definitions and useful identities of elliptic functions are summarized . appendix [ apx : calc ] explains a detailed computation of melnikov function for a spinning string ansatz . in this section , we shall give a brief introduction of melnikov s method . this is an analytical method to argue the existence of classical chaos . the key ingredient in this method is the so - called melnikov function and its simple zero points are related to a discrete dynamical system , smale s horseshoe , which generates a certain class of chaos . let us concentrate on a continuous dynamical system composed of two particles which has the hamiltonian represented by a direct sum like @xmath10 . the four - dimensional phase space is spanned by the coordinates @xmath11 and the conjugate momenta @xmath12 . we are concerned with evolution of @xmath13 and @xmath12 in time @xmath14 . it is also supposed that @xmath15 has a homoclinic orbit ( fig . [ fig : separatrix ] ) with a hyperbolic saddle point @xmath16 . the existence of the homoclinic orbit is surely crucial so as to apply melnikov s method . in other words , this condition somewhat restricts the applicability of this method . phase space with a hyperbolic saddle point @xmath16 and the other two types of orbits . the concentric rings inside the homoclinic orbit describe oscillation while the open trajectories outside it come from the infinity , surround the homoclinic orbit and go away to the infinity again . ] this system is integrable because the hamiltonian is separable and one can solve the equation of motion as a one dimensional hamiltonian system . then , once a certain small perturbation is added to the integrable system , the integrability is broken and orbits near the homoclinic one are likely to be chaotic ( c.f . , kam theorem @xcite ) . when the perturbation is turned on , the homoclinic orbit opens up in general as in fig . [ fig : homoclinic - orbit ] , while the saddle point survives up to a slight deviation of the location ( i.e. , the saddle is not resolved but just moves a little bit ) . then there still remain a stable manifold @xmath17 and an unstable one @xmath18 which meats at the saddle point . the stable manifold is defined as the set of points that get infinitely close to the saddle point by time evolution . similarly , the unstable manifold is the set of points getting closer to the saddle point by the inverse evolution . in general , the two manifolds are separated after the system is perturbed , while each of them is closed at the saddle point and forms a closed loop called a homoclinic orbit ( i.e. , the degeneracy of the stable and unstable manifolds ) in the non - perturbed system . [ cols="^,^ " , ] the idea of melnikov s method is to argue whether the stable manifold and the unstable manifold have a point of transverse intersection , i.e. , a transverse homoclinic point . when the system is perturbed by a small interaction term @xmath19 where @xmath20 is a small real parameter , the orbits on these manifolds are expressed as perturbative expansions from the original ones by an infinitesimal parameter @xmath20 . then , one can define the separation between those manifolds , which is proportional to the melnikov function . in particular , in the case that at least one transverse homoclinic point exists , a certain class of chaos is generated by smale s horseshoe ( see fig . [ fig : horseshoe phase ] ) . the saddle point expands small phase - space volume @xmath21 around the saddle point @xmath16 into long thin volume @xmath22 along the unstable manifold to @xmath16 , @xmath18 , and its end tip reaches the stable manifold @xmath17 at a transverse homoclinic point @xmath23 . then the area @xmath24 along the stable manifold is compressed into a vicinity around the saddle point , @xmath21 . this one cycle of process transforms @xmath21 to a u - shaped form , which looks like a horseshoe . to be more precise , two horizontal strips @xmath25 and @xmath26 in @xmath21 which move to the intersections of @xmath22 and @xmath24 by successive poincar maps , are mapped into two vertical strips @xmath27 and @xmath28 in @xmath21 , as shown in fig . [ fig : horseshoemap ] . this is smale s horseshoe map . the smale - birkhoff homoclinic theorem states that this map has a hyperbolic invariant set , which is a direct product of cantor sets . by the repetition of this map , a chaotic behavior appears near the saddle point and the transverse homoclinic points . if the melnikov function has a simple zero point , then the stable manifold and the unstable manifold have a transverse intersection and the system exhibits chaos due to the above argument . thus the criterion to determine whether chaos of smale s horseshoe type appears or not is boiled down to a simple calculation of the melnikov function . is mapped to the red area @xmath22 , and thereafter the blue area @xmath24 is mapped to @xmath21 . note that these areas overlap with one another and the colors in the overlapping regions are different from the original ones . the entire map is regarded as a horseshoe map . a similar argument is applicable to the case with a heteroclinic orbit . ] and @xmath26 are mapped onto the vertical strips @xmath27 and @xmath28 . in the end , the initial square is transformed to another square . these squares correspond to the region @xmath21 in the phase space . ] in order to apply melnikov s method , it is convenient to transform a canonical pair , say @xmath29 and @xmath30 , into the action - angle variables @xmath31 and @xmath32 , while the others are kept so as to preserve the homoclinic structure . section [ sec : stand - meln - meth ] is devoted to the standard review of melnikov s method . some simple examples are presented in sec . [ 2.2 ] . if the resulting variables do not form a canonical pair , the hamiltonian generally can not be separated into a direct sum , and hence melnikov s method has to be modified . such a case often arises in some studies of a string sigma model with an ads - like geometry . this modification of melnikov s method will be discussed in sec . [ sec : spec - meln - meth ] . first of all , we will briefly introduce the reduction method of a four - dimensional system with @xmath11 and the conjugate momenta @xmath12 to a two - dimensional system , as presented in @xcite . the non - perturbed hamiltonian @xmath33 is supposed to be a direct sum of two classically integrable systems like @xmath34 and the interaction between them is turned on as a small perturbation @xmath35 , where @xmath20 is an infinitesimal real constant . that is , the total hamiltonian @xmath36 is given by @xmath37 in the following , we assume that @xmath38 is periodic in @xmath29 and @xmath15 has a homoclinic orbit . the existence of the homoclinic orbit is crucial in melnikov s method . it is convenient to transform @xmath29 and @xmath30 into the angle - action variables @xmath32 and @xmath31 . then the hamiltonian @xmath36 can be rewritten as @xmath39 the hamilton equations are given by @xmath40 where the symbol `` @xmath41 '' stands for the derivative with respect to @xmath14 . in addition , @xmath42 is also supposed to be periodic in @xmath32 . this dynamical system is essentially three - dimensional because the energy @xmath43 is conserved . just for convenience , the unperturbed frequency is supposed to be positive , namely @xmath44 when @xmath20 is sufficiently small , @xmath32 becomes a monotonically increasing function from the first equation in ( [ eom ] ) . then , thanks to the @xmath14-independence in the dynamical system , one can set @xmath32 as a new time for the system by deleting @xmath14 . the next task is to reduce the remaining degree of freedom @xmath31 . by solving the energy conservation @xmath45 , @xmath31 is represented by @xmath46 here @xmath47 and @xmath48 depend on @xmath43 and hence these are also expressed as @xmath49 then one can obtain a two - dimensional system which is composed of @xmath50 and @xmath51 with `` time '' @xmath32 : @xmath52 here the prime stands for the derivative with respect to @xmath32 , and @xmath53 , @xmath54 , @xmath55 and @xmath56 are given by , respectively , @xmath57 \frac{\partial_{p_2}h_2(q_2,p_2)}{\omega(l^{(0)}(q_2,p_2))^2}\ , , \nonumber \\ f^{(0 ) } & = -\frac{1}{\omega(l^{(0)}(q_2,p_2))}\,\partial_{q_2}h_2(q_2,p_2)\ , , \nonumber \\ f^{(1 ) } & = -\frac{\partial_{q_2}h_{\rm int}(\theta , l^{(0)},q_2,p_2)}{\omega(l^{(0)}(q_2,p_2 ) ) } + \biggl [ \partial_{i}h_{\rm int}(\theta , l^{(0)},q_2,p_2)\ , , \nonumber \\ & \hspace{55 mm } + \omega'(l^{(0)}(q_2,p_2))\ , l^{(1)}(q_2,p_2,\theta ) \biggr ] \frac{\partial_{q_2}h_2(q_2,p_2)}{\omega(l^{(0)}(q_2,p_2))^2}\,.\end{aligned}\ ] ] note that in the above computation we have used the expressions in and the following relations : @xmath58 we are now ready to introduce melnikov s method . to explain the role of the melnikov function in this method , we employ the poincar map at @xmath59 with a perturbation parameter @xmath20 by @xmath60 , where @xmath61 is a value modulo the period . by this poincar map , a point on a orbit at @xmath61 is mapped to another on the orbit at @xmath61 , modulo the period . let us consider a small vicinity @xmath62 around the transverse homoclinic point on the orbit . this can be mapped to a neighborhood of the saddle point @xmath21 by finite numbers of forward or backward actions of @xmath60 . since these forward and backward actions can be connected at @xmath21 and @xmath62 , there is an integer @xmath63 where @xmath64 maps a point in @xmath21 to another in @xmath21 itself . this is smale s horseshoe . therefore , what we have to do here is to show the existence of the transverse homoclinic point . to verify the condition to ensure its existence , let us calculate a separation between the stable and unstable manifolds passing through the saddle point , at @xmath59 . one can express a solution of on the stable manifold as @xmath65 for @xmath66 , and on the unstable manifold as @xmath67 for @xmath68 . here the variables @xmath69 and @xmath70 describe a separatrix solution to at @xmath71 . also , let @xmath72 and @xmath73 lie on the normal to the unperturbed homoclinic orbit at @xmath61 for clarity . then the separation at @xmath61 can be defined as @xmath74 where @xmath53 and @xmath55 are functionals of @xmath75 and @xmath76 . this is the projection to a vector @xmath77 which is normal to the original homoclinic orbit . i.e. , this vector is orthogonal to @xmath78 . is a distance between the stable and unstable manifolds . ] for convenience , let us introduce @xmath79 defined as @xmath80 where @xmath81 and @xmath82 . then the derivative of @xmath79 is represented by @xmath83 here @xmath54 and @xmath56 are functionals of @xmath75 , @xmath76 , and explicit @xmath32 . note that we have utilized the following relations : @xmath84 by employing the numerator in ( [ distance - d ] ) , let us define the melnikov function as @xmath85 hence , as far as the leading order is concerned , the separation @xmath86 is proportional to the melnikov function like @xmath87 by using @xmath88 , the melnikov function ( [ melnikov fn 1d ] ) can be rewritten as @xmath89 here we have used the following relations : @xmath90 note that near the saddle point , the derivatives of @xmath50 and @xmath51 are equal to zero . hence @xmath91 and the above equations follow ) and a non - perturbed homoclinic solution . ] . now the argument of the melnikov function ( [ mel - t ] ) is given by @xmath61 . it is more useful to rewrite the melnikov function as a function of @xmath14 , rather than @xmath32 . let us define @xmath92 as the initial value of @xmath31 . then the time @xmath14 can be expressed as @xmath93 finally , the melnikov function is represented by @xmath94 where @xmath69 and @xmath70 have been redefined as functions of @xmath14 , and @xmath95 is the poisson bracket defined with @xmath50 and @xmath51 like @xmath96 according to melnikov s method , the existence of simple zeros in @xmath97 means that this system has a horseshoe for sufficiently small @xmath20 lying near the homoclinic orbit of @xmath98 on the energy surface @xmath99 . it would be helpful to see examples of computing the melnikov functions . let us show here two examples : 1 ) a dynamical system composed of a coordinate variable @xmath23 and its conjugate momentum @xmath16 with a time - dependent perturbation , 2 ) a four - dimensional dynamical system composed of two coordinate variables @xmath29 , @xmath50 and their conjugate momenta @xmath30 , @xmath51 with a time - independent perturbation . these examples are discussed in detail in @xcite and @xcite , respectively . more precisely speaking , the first example is a three - dimensional dynamical system . this case has not been covered in the previous subsection , but the computation scheme is almost the same . a central difference is that the reduction process is not necessary . it would really be instructive to study this case and to capture the essential idea of the melnikov function . we consider a pendulum system with a periodic external force described by the hamiltonian , @xmath100 and it depends on time @xmath14 explicitly . then the hamilton equations are given by @xmath101 here @xmath20 is an infinitesimal constant . note that these equations can be regarded as those of reduced systems like ( [ eq : q , p ] ) by replacing @xmath14 with @xmath32 . this point is also concerned with the footnote 5 . when the perturbation is turned off ( i.e. , @xmath71 ) and the energy @xmath43 is set to 1 , this system exhibits a separatrix solution given by @xmath102 the pendulum simply oscillates around the stable fixed point @xmath103 below @xmath104 , while the pendulum rotates all the way above @xmath104 . then by substituting the reduced equations ( [ example1 ] ) and the homoclinic solution ( [ example1-sol ] ) into the formula ( [ melnikov fn reduced ] ) , the melnikov function can be computed as @xmath105 it has nontrivial simple zeros at @xmath106 . hence , when @xmath43 is close to @xmath107 , the pendulum behaves in a weird manner it switches oscillation and rotation randomly . this is nothing but the origin of chaotic motion . melnikov s method tells us that this chaotic motion is generated by smale s horseshoe . as the second example , let us look at a simple pendulum - oscillator system with a small perturbation . this is a four - dimensional dynamical system composed of @xmath29 , @xmath50 and the conjugate momenta @xmath30 and @xmath51 . by taking @xmath108 as a perturbation term , the perturbed hamiltonian is given by @xmath109 here @xmath20 is an infinitesimal parameter again and @xmath110 is a frequency of the oscillator . when the perturbation is turned off ( i.e. , @xmath71 ) , the system is completely separable : the one is a simple oscillator and the other is a pendulum . the variables of the oscillator @xmath29 and @xmath30 can be transformed into the action and angle variables , @xmath31 and @xmath32 . then the hamiltonian can be rewritten as @xmath111 for the unperturbed system , the oscillation can be written as @xmath112 , and the separatrix solution of the unperturbed subsystem with @xmath98 is the same as in the previous . when the perturbation is turned on , the pendulum and oscillator systems begin to interact each other weakly , and the pendulum feels the motion of the oscillator through the interaction term . this situation is quite similar to the dynamical system . thus , as in sec . 3.1 , the reduction method leads to a two - dimensional dynamical system to which melnikov s method can be applied . in the present case , one can straightforwardly employ the formula . by putting @xmath42 and @xmath113 into , the melnikov function can be evaluated as @xmath114 \nonumber \\ & = \frac{1}{\omega}\int_{-\infty}^\infty\!\ ! d\tau \ ; \sqrt{\frac{2i^{(0)}}{\omega}}\ , \frac{2\sin\omega \tau}{\cosh(\tau-\tau_0 ) } = \sqrt{\frac{2i^{(0)}}{\omega}}\ , \frac{2\pi\sin\omega \tau_0}{\cosh(\frac{\pi\omega}{2})}\ , . \label{ex2}\end{aligned}\ ] ] it is easy to see that the melnikov function ( [ ex2 ] ) has simple zeros at @xmath115 . hence this system also exhibits chaotic motions generated by smale s horseshoe . so far , we have considered melnikov s method in the standard case . for later purpose , however , we need to generalize it slightly . the method we discuss here is a generalization of the reduction method in @xcite . in this subsection , we are interested in the melnikov function in more intricate dynamical systems . in general , one of @xmath116 s may depend on the other one . namely , we have in our mind the hamiltonian of the following type @xmath117 where the entire @xmath42 is supposed to be periodic in @xmath14 hereafter . then a system composed of @xmath29 and @xmath30 continues to feel a potential due to the energy of @xmath15 , even though the perturbation is turned off , i.e. , @xmath71 . in such a system , one can formally change @xmath29 and @xmath30 to @xmath31 and @xmath32 ( keeping @xmath50 and @xmath51 unchanged ) . but @xmath31 and @xmath32 fail to be a canonical pair . the hamilton equations are given by @xmath118 now we have to take account of an additional equation coming from the explicit time - dependence of the hamiltonian : @xmath119 for a moment , let us consider the case with @xmath71 . then the variables of the system become separable . when we concentrate on @xmath38 by setting @xmath15 as a constant @xmath120 , the action - angle variables can be written as @xmath121 here @xmath31 and @xmath32 satisfy @xmath122 and inversely , @xmath123 let us return to the interacting case ( @xmath124 ) . by using the relations ( [ 2.29 ] ) and ( [ 2.31 ] ) , @xmath31 and @xmath32 can be used as new variables , instead of @xmath29 and @xmath30 , even if @xmath20 is turned on . although the new variables are not canonical , one can still track their time evolution through @xmath125 in general , @xmath126 depends on @xmath15 and is assumed to be a positive constant . furthermore , suppose that a ratio of @xmath126 and the period of @xmath42 is rational for a reason explained later . then , @xmath14 can be expressed as @xmath127 and @xmath42 is periodic in @xmath32 to the leading order in @xmath20 . since the @xmath14-dependence appears only in @xmath42 , @xmath14 can be replaced by @xmath128 as far as the quantities are computed up to the first order of @xmath20 . when @xmath20 is sufficiently small , @xmath129 becomes positive ; hence @xmath32 may be regarded as a new time coordinate for the system again . now it is a turn to reduce the degrees of freedom from @xmath130 to @xmath131 . let us first remove @xmath14 in @xmath42 by @xmath132 and the derivative with respect to @xmath14 by @xmath133 . the next step is to rewrite functions of @xmath29 and @xmath30 in terms of @xmath31 , @xmath32 and @xmath15 . namely , the concerned quantities are rewritten as follows : @xmath134 finally , one can eliminate @xmath31 through the relation @xmath135 with the help of the following implicit function : @xmath136 the resulting expressions are given by @xmath137 where we have introduced the following quantities : @xmath138 \partial_{p_2}h_2(q_2,p_2)\ , , \nonumber \\ f^{(0 ) } & = -\frac{g_{1}(h_2,l^{(0)},\theta)}{\omega}\,\partial_{q_2}h_2(q_2,p_2)\ , , \nonumber \\ f^{(1 ) } & = -\frac{g_{\mathrm{int},q}(q_2,p_2,l^{(0)},\theta)}{\omega } + \biggl [ \frac{g_{1}(h_2,l^{(0)},\theta)}{\omega^2}\ , \tilde\omega(q_2,p_2,l^{(0)},\theta ) \nonumber \\ & \hspace{55 mm } -\frac{\partial_ig_{1}(h_2,l^{(0)},\theta)}{\omega}\,l^{(1)}(q_2,p_2,h,\theta ) \biggr ] \partial_{q_2}h_2(q_2,p_2)\ , . \label{reduced}\end{aligned}\ ] ] now three variables : @xmath50 , @xmath51 and @xmath36 are concerned with our analysis . here there are some comments . the first is that @xmath36 is a slow variable compared to the other two . the second is that @xmath54 and @xmath56 are periodic in @xmath32 because the ratio of @xmath126 and the period of @xmath42 has been taken to be rational . this periodicity is required in melnikov s method ; hence the rational condition has been supposed just below ( [ rational ] ) . finally when @xmath36 satisfies appropriate conditions , the reduced system ( [ reduced ] ) exhibits a horseshoe for sufficiently small @xmath20 , as we will show below . because the energy @xmath36 is not conserved , one needs to check whether the poincar map would map a volume element on a surface with constant energy to the same surface or another surface but close to the original . let us assume that there is @xmath63 where @xmath139 has a horseshoe . @xmath63 is related to @xmath20 , and the limit @xmath140 makes @xmath141 go zero . the average change of energy by @xmath139 is @xmath142 where the bar stands for the average over explicit @xmath32 with the other variables fixed . the integral over @xmath32 is performed by all dependence of @xmath32 including implicit one . it is sufficient to evaluate this on the homoclinic orbit because the region where a horseshoe structure appears is close to the homoclinic orbit for small @xmath20 . now , we assume that , for a certain value of @xmath36 , @xmath143 then , @xmath139 maps the energy surface to itself to leading order in @xmath20 . this means that there exists a map from a vicinity near the saddle point to itself under this condition . before defining the melnikov function in this case , one needs to remove the explicit @xmath32-dependence in @xmath53 and @xmath55 . since @xmath36 and @xmath15 are nearly constant as far as the poincar section is concerned , one can expand those in @xmath20 as @xmath144 and @xmath145 here , explicit @xmath32 dependence in @xmath146 comes from that in the separatrix solution . with an assumption that @xmath147 is positive - definite , we now define a new time @xmath148 by @xmath149 then , functions periodic in @xmath32 are periodic in @xmath148 as well if @xmath150 is finite since @xmath147 is periodic in @xmath32 . one finds the equations of motion of @xmath50 and @xmath51 finally become @xmath151 where we have introduced the following quantities : @xmath152 \,.\end{aligned}\ ] ] because this is a one - dimensional system , to which one can apply melnikov s method , the melnikov function is straightforwardly defined as @xmath153 here @xmath69 and @xmath70 are the separatrix solution to at @xmath71 , and @xmath33 is the initial total energy . let us rewrite ( [ mel ] ) as a function of @xmath14 , rather than @xmath148 . we define @xmath154 as @xmath155 and denote the value of @xmath15 with the separatrix solution by @xmath156 . then , by using the following notations : @xmath157 the melnikov function can be rewritten as @xmath158 this function is proportional to the separation of the stable and unstable manifolds at @xmath159 near the surface of @xmath160 because @xmath161 is assumed . hence , melnikov s method says that the existence of simple zeros in @xmath97 indicates that this system has a horseshoe for sufficiently small @xmath20 lying near the homoclinic orbit of @xmath98 and near the energy surface @xmath160 . in this section , we shall consider an application of melnikov s method in the context of string theory . in other words , we are concerned with an appropriate string background to which melnikov s method is applicable . an example of such backgrounds is a brane - wave type deformation of the ads@xmath0s@xmath1 background constructed in @xcite . the background is composed of the metric and the five - form field strength @xmath162\ , , \label{metric } \\ { \rm ds}^2_{\rm s^5 } & = & { \rm ds}^2_{{\mathbb c}\rm p^2}+({\rm d}\chi+\omega)^2\ , , \nonumber \\ f_5 & = & 4 [ \omega_{\rm ads_5 } + \omega_{\rm s^5}]\ , , \nonumber \end{aligned}\ ] ] where @xmath163 satisfies @xmath164 with @xmath165 an integer greater than 0 . the dilaton is constant and the other fluxes are zero . here the metric of round s@xmath1 is expressed as a @xmath166 fibration over @xmath167p@xmath168 , where @xmath169 is the local coordinate on the hopf fibre and @xmath110 is the one - form potential for the khler form on @xmath167p@xmath168 . the metric of @xmath167p@xmath168 and @xmath110 are given by @xmath170 where @xmath171 are defined as @xmath172 the remaining task is to fix an arbitrary function @xmath173 . a remarkable point is that the metric ( [ metric ] ) is a pp - wave type deformation ; hence the only non - trivial equation of motion is the @xmath174-component of the equations of motion for the metric . thus all we have to do is to solve the following constraint : @xmath175 note here that this constraint ( [ cons ] ) is obviously satisfied when @xmath176 and @xmath173 is a constant . this case corresponds to the undeformed ads@xmath0s@xmath1 . it is easy to derive the general solution of ( [ cons ] ) , but some simple ones are enough for our purpose here . hereafter , we will concentrate on two solutions given by @xmath177 here the overall constant can be absorbed by rescaling @xmath178 ; hence we have simply set it to 1 . we will study the motion of a string propagating on the background presented above . to show classical chaos in the bosonic sector , we will not touch on the fermionic sector . in addition , to make our analysis simpler , we will adopt two anstze . in the following , we will compute melnikov s function under each of the anstze . in this subsection , we will consider an oscillating string ansatz . let us perform a coordinate transformation @xcite : @xmath179 \tan \left(m \tilde x^+\right)\ , , \nonumber \\ & x^i=\frac{\tilde x^i}{\cos \left(m \tilde x^+\right ) } \qquad ( i=1,2)\ , , \qquad z=\frac{\tilde z}{\cos \left(m \tilde x^+\right)}\ , . \label{singular}\end{aligned}\ ] ] note here that the points with @xmath180 are singular . after this transformation , the system gets restricted and a certain part of the deformed ads background is magnified . the resulting metric is given by @xmath181\,.\end{aligned}\ ] ] note here that the undeformed ads@xmath182 is also deformed to a pp - wave type metric . in the following , we are concerned with a string theory defined on this background . we will concentrate on the bosonic part and study the classical action of polyakov type with the light - cone gauge . for simplicity , let us take the following ansatz : @xmath183 here @xmath184 is a winding number along the @xmath185-direction . note that this ansatz is consistent with the original equations of motion . from now on , let us set @xmath186 for simplicity and the function @xmath173 is given in ( [ f(mu ) ] ) . this is just because the case with @xmath187 is integrable under the ansatz ( [ ansatz1 ] ) and hence the case with @xmath186 is the simplest and non - integrable , though the case with @xmath188 would be non - integrable as well . then the light - cone hamiltonian is simplified as @xmath189\ , . \label{light cone hamiltonian g_--=0}\end{aligned}\ ] ] this is nothing but a coupled pendulum - oscillator system . it should be remarked that the hamiltonian ( [ light cone hamiltonian g_=0 ] ) has the explicit @xmath14-dependence due to the singular coordinate transformation ( [ singular ] ) . the unperturbed hamiltonian is @xmath190 where @xmath15 is the hamiltonian of a pendulum system , @xmath191 the perturbation term is given by @xmath192 where @xmath178 is assumed to be infinitesimal . for this system , one can compute the melnikov function by following the generalized method developed in sec . [ sec : spec - meln - meth ] . in the non - perturbed case ( i.e. , @xmath193 ) , @xmath15 can be regarded as a constant @xmath120 because @xmath194 . then the hamiltonian can be rewritten in terms of the action - angle variables by performing a canonical transformation to @xmath195 : @xmath196 the resulting hamiltonian is expressed in terms of @xmath195 as follows : ) is conceivable because the third term in ( [ lc1 ] ) can be regarded as a centrifugal force with a angular momentum @xmath197 under the identification @xmath198 . ] @xmath199 here we have introduced the energy @xmath43 . from the equation of motion of @xmath195 , @xmath200 now @xmath201 is written as a function of @xmath14 through the transformation ( [ canonical - trf ] ) , @xmath202}\,.\end{aligned}\ ] ] in order to calculate the melnikov function , the separatrix solution of the pendulum system @xmath15 is necessary . if one takes @xmath203 then @xmath204 and @xmath205 in , in this system , are given by @xmath206 and the equations of motion for the pendulum system with time @xmath148 by @xmath207 which can easily be solved by @xmath208\,.\ ] ] by integrating ( [ tr ] ) , the following solution is obtained , @xmath209\right]\,.\ ] ] the new period becomes @xmath210 . now , with @xmath211 , the separatrix solution can be rewritten as @xmath212 \,.\end{aligned}\ ] ] note that its energy @xmath120 is @xmath213 . one must check the energy condition . the equation of the energy is @xmath214 its average on the homoclinic orbit is given by @xmath215 here @xmath216 is an approximation in the region sufficiently near @xmath217 we also use the following expression , @xmath218 which is sufficiently small . then , the change of the energy by a poincar map with @xmath63 shifts of time by the period @xmath219 is estimated as @xmath220 at @xmath221 . in the above calculation , we have omitted the subleading order in @xmath22 and used an approximate relation @xmath222 the equation suggests that the total energy generally changes monotonically . eventually , we can expect this change from the beginning because the frequency of the factor oscillating explicitly in the perturbed hamiltonian coincides with the frequency of unperturbed @xmath201 . hence , it possibly provides a `` resonance '' source . the condition is however satisfied when @xmath223 if a string with this condition starts its trajectory from the vicinity near the homoclinic point , then the energy will not change drastically when the string comes back to the vicinity near the homoclinic point . now that we can apply the melnikov s method explained in sec . [ sec : spec - meln - meth ] , the melnikov function is computed as @xmath224 here @xmath156 is defined as @xmath225 . this poisson bracket is reduced to @xmath226 ( 1 - 5\cos[2\mu^{(0)}(\tau-\tau_0 ) ] ) ( z^{(0)}(\tau))^{-4 } \cos^{2}(m\tau)\,.\end{aligned}\ ] ] then the melnikov function is calculated as @xmath227 } ( 1 - 5\cos[2\mu^{(0)}(\tau ) ] ) \nonumber \\ & \hspace{35 mm } \times ( z^{(0)}(\tau))^2(z^{(0)}(\tau+\tau_0))^{-4 } \cos^{2}[m(\tau+\tau_0 ) ] \nonumber \\ & = -nl^2\int_{-\infty}^\infty dt\ , \left ( 3\frac{\sinh[\frac{nl^2}{2}t]}{\cosh^{3}[\frac{nl^2}{2}t ] } -5\frac{\sinh[\frac{nl^2}{2}t]}{\cosh^{5}[\frac{nl^2}{2}t ] } \right ) \nonumber \\ & \hspace{35 mm } \times ( z^{(0)}(\tau(t)))^2(z^{(0)}(\tau(t)+\tau_0))^{-4 } \cos^{2}[m(\tau(t)+\tau_0 ) ] \nonumber \\ & = -\frac{4\pi m}{3nl^2\sinh\pi } \biggl [ \sin[2m(\tau_0+\tau_1 ) ] \nonumber \\ & \qquad -4a \biggl ( \sin(2m\tau_0 ) + \frac{16}{\cosh\pi}(\sin[2m(2\tau_0+\tau_1)]-\sin[2m(\tau_0+\tau_1 ) ] ) \biggr)\biggr ] \nonumber \\ & \qquad + o(\eta^2,a^2)\,.\end{aligned}\ ] ] to derive the above expression , we have used the following approximation of @xmath148 and @xmath14 @xmath228 this melnikov function has simple zeros at least when @xmath22 is sufficiently small for any @xmath229 . thus the result indicates the existence of chaos generated by a smale s horseshoe for trajectories satisfying @xmath230 , that is , @xmath231 although such trajectories are quite rare , trajectories with @xmath232 remain in a small range of the energy after performing each of the poincar maps if the observation time is taken to be sufficiently short . in this subsection , we consider a spinning string ansatz . in order to obtain a desired ansatz , we convert the poincar coordinate of the ads part to the global coordinate by performing the following transformation : @xmath233 as in the previous subsection , we will study the bosonic part of the polyakov action . as a remark , we will take here the spatial direction of the string - world - sheet to be infinite ( i.e. , an infinitely long string ) , rather than the usual closed string . with the gauge condition @xmath234 , let us consider the following ansatz : @xmath235 because we are considering an infinite string , there is no need to impose a boundary condition on @xmath236 . the relevant part of the metric is rewritten as @xmath237\,.\end{aligned}\ ] ] for simplicity , we concentrate on the case with @xmath187 hereafter . the function @xmath173 is given in ( [ f(mu ) ] ) . note here that this case is not integrable under the ansatz ( [ ansatz2 ] ) in comparison to the previous ansatz ( [ ansatz1 ] ) . then the lagrangian density is given by @xmath238 \ , , \label{lagrangian}\end{aligned}\ ] ] where the prime here is defined as the derivative with respect to @xmath239 . @xmath240 has been absorbed into @xmath241 , @xmath239 , @xmath242 , and @xmath178 so that @xmath243 and @xmath244 are canonically normalized . also , we impose a condition on @xmath241 and @xmath245 like @xmath246 so that the system should be bounded . if one regards @xmath239 as time , `` hamiltonian , '' which governs the @xmath239-dependence of @xmath243 and @xmath244 , can be defined as @xmath247 thanks to the condition ( [ kappa ] ) , the second term is positively definite . the classical solution should satisfy the virasoro constraints , @xmath248 note that this condition ( [ vir ] ) is exactly the same as the equation for a noether charge for the translation along the @xmath239-direction in where the charge is @xmath249 . thus there is not any contradiction as far as the total `` energy '' is @xmath249 . moreover , as far as the virasoro constraints are satisfied , the ansatz is consistent with the original equations of motion . in the following , we treat @xmath239 like time and use the symbol `` @xmath14 '' , instead of @xmath239 . this system is regarded as a coupled pendulum - oscillator system . the hamiltonian of an oscillator system is given by @xmath250 and that of a pendulum system is @xmath251 while the latter system is a usual pendulum one , the former is oscillated in a squared hyperbolic sine potential . the interaction term is written as @xmath252 where @xmath178 is assumed to be infinitesimal . when @xmath193 , the hamiltonian system is a direct sum of two systems where one is an oscillator and the other has a homoclinic orbit . hence to compute the melnikov function , we can follow the standard method described in sec . [ sec : stand - meln - meth ] . the solution of the oscillator system with its energy @xmath43 is @xmath253\ , , \label{solution_osc}\end{aligned}\ ] ] where the function am@xmath254 is the jacobi amplitude function , defined as the inverse of the elliptic integral of the first kind @xmath255 as @xmath256 the last expression in is obtained as an approximation when @xmath43 is sufficiently small . let us compute the melnikov function by using the separatrix solution of the pendulum system , @xmath257\ , . \end{aligned}\ ] ] the energy of this solution is @xmath258 . as the total energy is @xmath249 , the value of @xmath43 is estimated as @xmath259 , if the interaction term is negligible . the present case does not require the energy condition because the system is not disturbed by an external force . hence the melnikov function defined in is given by @xmath260 where @xmath261 is defined as @xmath262 . the poisson bracket is now reduced to @xmath263 where we have used the following identities : . ] @xmath264 = { \operatorname{cn}}(ix , ik ) = { \operatorname{nd}}\left ( \sqrt{1+k^2}\ , x,\frac{1}{\sqrt{1+k^2 } } \right)\,.\end{aligned}\ ] ] note here that due to the condition ( [ kappa ] ) , @xmath245 is real and @xmath265 . the final step is to perform the integration with respect to @xmath14 . it is really complicated and messy ; hence the detailed calculation is explained in appendix [ apx : calc ] . after a lengthy calculation , the melnikov function is given by @xmath266 \bigg\}\ , , \label{melnikov_spinning}\end{aligned}\ ] ] where we have introduced the following quantities @xmath267 and the functions @xmath268 and @xmath269 are defined as , respectively , @xmath270 ^ 4 } \biggl [ { \operatorname{nd}}^2\left ( \omega'_1\tau_0,s \right ) { \operatorname{cd}}^2\left ( \omega'_1\tau , s ' \right ) \nonumber \\ & \qquad\hspace{20 mm } -\left(\frac{2e}{\omega'^2_1}\right)^2 { \operatorname{sd}}^2\left ( \omega'_1\tau_0,s \right ) { \operatorname{cd}}^2\left ( \omega'_1\tau_0,s \right ) { \operatorname{sd}}^2\left ( \omega'_1\tau , s ' \right ) { \operatorname{nd}}^2\left ( \omega'_1\tau , s ' \right ) \biggr]\ , , \nonumber \\ g_2(\tau ) & \equiv \frac{-\omega_2 ^ 2 } { 24\left ( 1-\frac{2e}{\omega'^2_1}\right)^2\omega'^4_1\cosh^6(\frac{\omega_2\tau}{2 } ) } \biggl [ 16(\omega'^2_1-e+\omega_2 ^ 2 ) + ( 8(\omega'^2_1-e)-13\omega_2 ^ 2)\cosh(\omega_2\tau ) \nonumber \\ & \qquad\hspace{50 mm } + ( -8(\omega'^2_1-e)+\omega_2 ^ 2)\cosh^2(\omega_2\tau ) \biggr]\ , . \notag\end{aligned}\ ] ] the factor @xmath271 in the first term of has non - trivial zeros periodically at @xmath272 , where @xmath165 is integer . then @xmath273 also depends on @xmath154 and has a non - zero value at general @xmath154 . although im@xmath274 $ ] in the second term is not periodic in @xmath154 , its absolute value decreases exponentially as @xmath154 goes to infinity . then it gets smaller than the maximum of the absolute value of the first term when @xmath154 is sufficiently large . furthermore , when @xmath43 is small enough , the second term is negligible , and it is easier to see simple zeros as explained in appendix [ apx : calc ] . thus the melnikov function has simple zeros at sufficiently large @xmath154 , and hence this system exhibits classical chaos associated with smale s horseshoe . in this paper , we have considered an application of melnikov s method in the context of the gauge / gravity correspondence . in particular , we have presented a string background to which melnikov s method is applicable . this is a brane - wave type deformation of the ads@xmath0s@xmath1 background . by employing two reduction anstze , we have studied two types of coupled pendulum - oscillator systems . then the melnikov function has been computed for each of the systems and the existence of chaos has been shown analytically . a strong advantage of melnikov s method is that the existence of classical chaos can be shown analytically and the mechanism is explained as a smale s horseshoe . the melnikov functions here have been computed on the string - theory side . one of the most interesting question is what is the gauge - theory interpretation of the melnikov function . this function knows all about the classical chaotic string solutions , including the associated fractal structure . according to the standard ads / cft dictionary , classical string solutions should correspond to composite operators on the gauge - theory side . the information on the fractal structure would be crucial in determining the alignment of component fields in the associated composite operators . hence the melnikov function may play an important role in determining the associated composite operators in some manner . from more technical aspects , it is significant to consider a generalization to include the friction . recall that an external force was induced with the oscillating ansatz . if the friction could be introduced as well , one can realize the system with the energy injection and dissipation . such a system is not a conserved system but dissipative . the chaos in the conserved systems is well described by the kam theorem @xcite and its characteristic is profoundly understood . the chaos in the dissipative systems , however , has a richer structure such as strange attractors . it is really intriguing to realize strange attractors on the string - theory side and consider the physical interpretation of its gauge - theory counter part . it would also be nice to look for other string backgrounds to which melnikov s method is applicable . we have just presented one example here . it is interesting to consider some applications of melnikov s method for black hole solutions in relation to the information loss process . it may be a good issue to clarify the relation between the event horizon and the melnikov function . we hope that melnikov s function would provide a new tool in studying the gauge / gravity correspondence . we are very grateful to daisuke kawai and takeshi matsumoto for useful discussions . the work of y.a . was supported by a dublin institute for advanced studies scholarship . the work of h.k . was supported by the japan society for the promotion of science ( jsps ) . the work of k.y . was supported by the supporting program for interaction - based initiative team studies ( spirits ) from kyoto university and by a jsps grant - in - aid for scientific research ( c ) no.15k05051 . this work was also supported in part by the jsps japan - russia research cooperative program and the jsps japan - hungary research cooperative program . we shall summarize the definitions of elliptic functions and useful identities utilized in sec . [ 3:sec ] . firstly , jacobian elliptic functions are defined as follows : @xmath275\,,\quad { \operatorname{cn}}(x , k)\equiv \cos [ \operatorname{am}(x , k)]\,,\nonumber\\ { \operatorname{dn}}(x , k)&\equiv & \sqrt{1- k^2 { \operatorname{sn}}^2 ( x , k)}\,,\end{aligned}\ ] ] where @xmath276 is the jacobi amplitude function defined in ( [ jacobi - amp ] ) . furthermore , we define functions additionally as below : @xmath277 there are addition theorems for the jacobian elliptic functions . for example , @xmath278 satisfies the following equation : @xmath279 lastly , we present useful identities of jacobian elliptic functions , @xmath280 where @xmath281 . one can show these identities by using the integral expressions of the elliptic functions . we shall present here the calculation of the melnikov function in detail . let us first write down the expressions of @xmath282 and @xmath283 like @xmath284\right ) = \frac{2\sinh\frac{\omega_2(\tau-\tau_0)}{2}}{\cosh^2\frac{\omega_2(\tau-\tau_0)}{2}}\,.\end{aligned}\ ] ] then the melnikov function is given by @xmath285 where we have introduced the following new quantities : @xmath286 the integration in ( [ melnikov_spinning_apx ] ) can be evaluated by summing up the contributions coming from the poles of @xmath287 and @xmath288 . in the following , let us evaluate the poles from @xmath289 and @xmath290 , respectively . because @xmath287 is an odd function and the integration range runs from @xmath295 to @xmath296 , only the odd part of @xmath288 can contribute to the integral . hence the calculation of derivatives of @xmath294 is simplified to some extent . in addition , by employing a relation of the jacobian elliptic function @xmath297 , one finds that the odd part of @xmath298 is given by @xmath299 note here that all of the second arguments of the elliptic functions are given by @xmath300 ( but those are omitted just for simplicity ) . the jacobian elliptic function @xmath301 has two single poles inside the fundamental unit cell . two periods of @xmath301 are given by @xmath302 and @xmath303 , where @xmath304 and @xmath305 . note here that @xmath265 and @xmath306 by construction in the present case ; hence @xmath307 and @xmath308 are real numbers . depending on the value of the residue , there are two kinds of poles . the first residue is given by @xmath309 for the poles at @xmath310 the second residue is @xmath311 for the poles at @xmath312 hence the asymptotic behaviors near these poles are given by @xmath313 now that @xmath294 has quadruple and double poles at @xmath314 , it is necessary to compute the third - order and first - order derivatives of @xmath291 in order to evaluate the residue around these poles . let perform the @xmath14-integration in . to perform this integral , we need to consider a contour integral . generally , the contour depends on the behavior of the integrand . the contour is closed on the upper - half plane with a semicircle in the counterclockwise direction when the integrand of goes to zero in the limit @xmath315 . similarly , the contour should be closed on the lower - half plane with a semicircle in the clockwise direction depending on the behavior of the integrand in the limit @xmath316 . by denoting these contours by @xmath317 and @xmath318 , respectively , the melnikov function can be rewritten as @xmath319 the integration along the upper contour @xmath317 is composed of two contributions : 1 ) the contributions from the poles of @xmath287 , namely , @xmath320 and 2 ) the contributions from the poles of @xmath288 , namely , @xmath321 the former is given by @xmath322 and the latter is @xmath323 similarly , one can evaluate the contributions coming from the lower - half plane . its contributions from the poles of @xmath287 and @xmath288 are @xmath324 and @xmath325 , respectively . thus , the integration along the lower contour @xmath318 is evaluated as the same as for the former contribution thanks to the oddness of function @xmath268 , and the complex conjugate of for the latter contribution , respectively . then after summing up the integration along @xmath317 and @xmath318 , one finds that has finally been derived . when @xmath43 is sufficiently small , one can readily derive an approximate form of the melnikov function . the leading term is given by @xmath326 \right ) + o(e^2)\,.\end{aligned}\ ] ] now that the @xmath14-integration can be performed easily , the resulting expression is given by @xmath327 } { \omega_2 ^ 2 \sinh\frac{2\pi\sqrt{\omega_1 ^ 2-\kappa^2}}{\omega_2 } } + o(e^2)\,.\end{aligned}\ ] ] then it is obvious that @xmath97 has non - trivial simple zeros at @xmath328 thus the conclusion that classical chaos exists does not change even with a small energy approximation . note here that the approximation is made for the integrand before performing the @xmath14-integration . one can also reproduce the same expression by expanding the exact formula in terms of small @xmath43 . that is , the order of the @xmath14-integration and the small @xmath43 expansion is irrelevant to the final result . m. crichigno , t. matsumoto and k. yoshida , `` deformations of @xmath8 as yang - baxter sigma models , '' jhep * 1412 * ( 2014 ) 085 [ arxiv:1406.2249 [ hep - th ] ] ; `` towards the gravity / cybe correspondence beyond integrability yang - baxter deformations of @xmath8 , '' j. phys . conf . ser . * 670 * ( 2016 ) no.1 , 012019 [ arxiv:1510.00835 [ hep - th ] ] . + i. kawaguchi , t. matsumoto and k. yoshida , `` jordanian deformations of the ads@xmath329s@xmath1 superstring , '' jhep * 1404 * ( 2014 ) 153 [ arxiv:1401.4855 [ hep - th ] ] . p. basu and l. a. pando zayas , `` chaos rules out integrability of strings in ads@xmath330 , '' phys . b * 700 * ( 2011 ) 243 [ arxiv:1103.4107 [ hep - th ] ] ; `` analytic non - integrability in string theory , '' phys . d * 84 * ( 2011 ) 046006 [ arxiv:1105.2540 [ hep - th ] ] . l. a. pando zayas and c. a. terrero - escalante , `` chaos in the gauge / gravity correspondence , '' jhep * 1009 * ( 2010 ) 094 [ arxiv:1007.0277 [ hep - th ] ] . + p. basu , d. das and a. ghosh , `` integrability lost , '' phys . b * 699 * ( 2011 ) 388 [ arxiv:1103.4101 [ hep - th ] ] . + p. basu , d. das , a. ghosh and l. a. pando zayas , `` chaos around holographic regge trajectories , '' jhep * 1205 * ( 2012 ) 077 [ arxiv:1201.5634 [ hep - th ] ] . + l. a. pando zayas and d. reichmann , `` a string theory explanation for quantum chaos in the hadronic spectrum , '' jhep * 1304 * ( 2013 ) 083 [ arxiv:1209.5902 [ hep - th ] ] . + p. basu , p. chaturvedi and p. samantray , `` chaotic dynamics of strings in charged black hole backgrounds , '' arxiv:1607.04466 . a. stepanchuk and a. a. tseytlin , `` on ( non)integrability of classical strings in p - brane backgrounds , '' j. phys . a * 46 * ( 2013 ) 125401 [ arxiv:1211.3727 [ hep - th ] ] . + y. chervonyi and o. lunin , `` ( non)-integrability of geodesics in d - brane backgrounds , '' jhep * 1402 * ( 2014 ) 061 [ arxiv:1311.1521 [ hep - th ] ] . d. giataganas and k. sfetsos , `` non - integrability in non - relativistic theories , '' jhep * 1406 * ( 2014 ) 018 [ arxiv:1403.2703 [ hep - th ] ] . + x. bai , j. chen , b. h. lee and t. moon , `` chaos in lifshitz spacetimes , '' j. korean phys . * 68 * ( 2016 ) no.5 , 639 [ arxiv:1406.5816 [ hep - th ] ] . c. t. asplund , d. berenstein and d. trancanelli , `` evidence for fast thermalization in the plane - wave matrix model , '' phys . lett . * 107 * ( 2011 ) 171602 [ arxiv:1104.5469 [ hep - th ] ] ; c. t. asplund , d. berenstein and e. dzienkowski , `` large n classical dynamics of holographic matrix models , '' phys . d * 87 * ( 2013 ) 8 , 084044 [ arxiv:1211.3425 [ hep - th ] ] . c. g. callan , jr . , h. k. lee , t. mcloughlin , j. h. schwarz , i. swanson and x. wu , `` quantizing string theory in ads@xmath333s@xmath1 : beyond the pp wave , '' nucl . b * 673 * ( 2003 ) 3 [ hep - th/0307032 ] ; i. swanson , `` superstring holography and integrability in ads@xmath334s@xmath1 , '' hep - th/0505028 .

a parameterization strategy for molecular models on the basis of force fields is proposed , which allows a rapid development of models for small molecules by using results from quantum mechanical ( qm ) _ ab initio _ calculations and thermodynamic data . the geometry of the molecular models is specified according to the atom positions determined by qm energy minimization . the electrostatic interactions are modeled by reducing the electron density distribution to point dipoles and point quadrupoles located in the center of mass of the molecules . dispersive and repulsive interactions are described by lennard - jones sites , for which the parameters are iteratively optimized to experimental vapor - liquid equilibrium ( vle ) data , i.e. vapor pressure , saturated liquid density , and enthalpy of vaporization of the considered substance . the proposed modeling strategy was applied to a sample set of ten molecules from different substance classes . new molecular models are presented for iso - butane , cyclohexane , formaldehyde , dimethyl ether , sulfur dioxide , dimethyl sulfide , thiophene , hydrogen cyanide , acetonitrile , and nitromethane . most of the models are able to describe the experimental vle data with deviations of a few percent . * keywords : * molecular modeling ; modeling strategy ; vapor - liquid equilibrium ; critical properties ; iso - butane ; cyclohexane ; formaldehyde ; dimethyl ether ; sulfur dioxide ; dimethyl sulfide ; thiophene ; hydrogen cyanide ; acetonitrile ; nitromethane

0904.3223

molecular modeling and simulation is a progressive approach for describing and predicting thermophysical properties of both pure substances and mixtures of technical interest . many authors have shown the excellent capabilities of this molecular approach for different applications @xcite . unfortunately , the more widespread use of molecular methods for engineering applications is still restricted by the scarcity of suitable molecular models . for many substances , transferable molecular models have been developed , i.e. force fields for classes of compounds like alkanes or alcohols . thereby , it is assumed that the parameters for functional groups are valid for different molecular species . the main disadvantage of transferable models is that their parameters were adjusted for the whole substance class and may not be optimal for a specific substance . furthermore , they do not cover substances outside the modeled class . an overview and assessment of some commonly used transferable potentials can be found in @xcite . molecular models that were optimized for specific substances are available only for selected compounds and , particularly older ones , sometimes do not show the desired quality . therefore , from an engineering point of view , a method for the rapid development of new molecular models for specific substances is of great interest . moreover , the molecular models should be accurate , simple , and computationally efficient . in the present paper , a systematic strategy is proposed to develop such models . the present work is restricted to rigid , non - polarizable models for comparatively small molecules . the models have state - independent parameters throughout . for computational efficiency , the united - atom approach is used , i.e. hydrogen atoms bonded to carbon are not modeled explicitly . the proposed strategy uses information determined by quantum mechanical ( qm ) _ ab initio _ calculations to include physically sound molecular properties and to reduce the number of adjustable parameters . a remaining subset of model parameters typically two to four is subsequently optimized by adjustment to experimental data on vapor - liquid equilibria ( vle ) of the pure substances . the aim is to achieve deviations to experimental values for the vapor pressure , saturated liquid density , and enthalpy of vaporization in the range from triple point to critical point of below 5 , 1 , and 5% , respectively . such accurate models are known to have an excellent extrapolative and predictive power . this was recently shown , e.g. , by the prediction of 17 different thermophysical properties for ethylene oxide , covering phase equilibria , thermal , caloric , transport properties , and surface tension @xcite , or for ammonia including structural quantities @xcite . molecular models that were developed on the basis of qm calculations stand between strictly _ ab initio _ models and fully empirical models . the present strategy is based on the idea to include substantial _ ab initio _ information without giving up the freedom to reasonably optimize the model to important macroscopic thermodynamic properties . thus , for the modeling process some experimental data is needed for optimization . all three chosen properties , mentioned above , have the advantage to be well available for numerous engineering fluids and to represent dominant features of the fluid state . the parameters of a molecular model can be separated into three groups . firstly , geometric parameters specify the locations of the different interaction sites of the molecular model . secondly , electrostatic parameters define the interactions of static polarities of the single molecules . and finally , dispersive and repulsive parameters determine the attraction by london forces and the repulsion by overlaps of the electronic orbitals . here , the lennard - jones 12 - 6 ( lj ) potential @xcite was used to assure straightforward compatibility with the overwhelming majority of the molecular models in the literature . to describe the intermolecular interactions , a varying number of lj sites and superimposed ideal point dipoles and/or ideal linear point quadrupoles were used . point dipoles or quadrupoles were employed for the description of the electrostatic interactions to reduce the computational effort significantly . a point dipole may , e.g. when a simulation program does not support this interaction site type , be approximated by two point charges @xmath0 separated by a distance @xmath1 . limited to small @xmath1 , one is free to choose this distance as long as @xmath2 holds . analogously , a point quadrupole can be approximated by three collinear point charges @xmath3 , @xmath4 , and @xmath3 separated by @xmath1 each , where @xmath5 . the total intermolecular interaction energy writes as @xmath6 \right . + \\ \left . \sum_{c=1}^{s_i^\mathrm{e } } \sum_{d=1}^{s_j^\mathrm{e } } \frac{1}{4 \pi \epsilon_0 } \left [ \frac{\mu_{ic } \mu_{jd}}{r_{ijcd}^3 } \cdot f_1(\bm{\omega}_i , \bm{\omega}_j ) + \frac{\mu_{ic } q_{jd } + q_{ic } \mu_{jd}}{r_{ijcd}^4 } \cdot f_2(\bm{\omega}_i , \bm{\omega}_j ) + \frac{q_{ic } q_{jd}}{r_{ijcd}^5 } \cdot f_3(\bm{\omega}_i , \bm{\omega}_j ) \right ] \right\rbrace,\end{gathered}\ ] ] where @xmath7 , @xmath8 , @xmath9 are the distance , the lj energy parameter , and the lj size parameter , respectively , for the pair - wise interaction between lj site @xmath10 on molecule @xmath11 and lj site @xmath12 on molecule @xmath13 . @xmath14 is the permittivity of vacuum , whereas @xmath15 and @xmath16 denote the dipole moment and the quadrupole moment of the electrostatic interaction site @xmath17 on molecule @xmath11 , and so forth . @xmath18 are expressions for the dependency of the electrostatic interactions on the orientations @xmath19 and @xmath20 of the molecules @xmath11 and @xmath13 , cf . finally , the summation limits @xmath21 , @xmath22 , and @xmath23 denote the number of molecules , the number of lj sites , and the number of electrostatic sites , respectively . interactions between lj sites of different type are determined by applying the standard lorentz - berthelot combining rules @xcite @xmath24 and @xmath25 in a recent publication , sandler et al . @xcite gives a brief overview on the use of qm for the calculation of thermophysical properties . by numerically solving schrdinger s equation , it is nowadays possible to calculate different molecular properties for technically relevant components in a quite standardized way . many different qm codes are available for this task . for license reasons , the open source code gamess(us ) @xcite was used in the present work . all geometric data of the molecular models , i.e. bond lengths , angles , and dihedrals , were directly taken from qm calculations . therefore , a geometry optimization , i.e. an energy minimization , was initially performed using gamess(us ) @xcite . the hartree - fock level of theory was applied with a relatively small ( 6 - 31 g ) basis set . alternatively , density functional theory ( dft ) methods , e.g. bly3p , can be used , as they are known to give reasonable results for the molecular structure @xcite . the resulting configuration of the atoms was taken without subsequent modification to specify the position of the lj sites in space , except for the hydrogen atoms . as the united atom approach was used to obtain computationally efficient molecular models , the hydrogen atoms were modeled together with the carbon atom they are bonded to . for the methylene ( ch@xmath26 ) and methyl ( ch@xmath27 ) united atom sites , the lj potential was located at the geometric mean of the nuclei , while the methine ( ch ) united atom site was located at 0.4 of the distance between carbon and hydrogen atom , cf . figure [ fig_united_atom ] . these empirical offsets are in good agreement with the results of ungerer et al . @xcite , which were found by optimization of transferable molecular models for n - alkanes . intermolecular electrostatic interactions mainly occur due to static polarities of single molecules that can well be obtained by qm . here , the mller - plesset 2 level was used that considers electron correlation in combination with the polarizable 6 - 311g(d , p ) basis set . the purpose of the present work is the development of effective pair potentials with a state - independent set of model parameters . obviously , the electrostatic interactions are stronger in the liquid state than in the gaseous state due to the higher density . furthermore , the mutual polarization raises their magnitude in the liquid . thus , for the calculation of the electrostatic moments by qm a liquid - like state should be considered . this was done here by placing the molecule within a dielectric continuum and assigning the experimental dielectric constant of the liquid to it , as in the cosmo method @xcite . from the resulting electron density distribution for the small symmetric molecules regarded here , ideal point dipoles and ideal linear point quadrupoles were estimated by simple integration over the orbitals . magnitudes and orientations of these electrostatic interaction sites were used in the molecular models without any modification . for other , more complex molecules , more sophisticated methods like chelp @xcite , chelpg @xcite , or the distributed multipole analysis @xcite are available in the literature . these methods adjust a set of partial charges or higher order electrostatic sites to the electrostatic potential around the molecule calculated by qm . although they are able to reflect the electrostatic interactions with higher accuracy , they are not considered here . they always yield a larger number of interaction sites if they are not co - located with other sites and would thus lead to computationally more expensive molecular models . it would be highly desirable to also calculate the dispersive and repulsive interactions using _ ab initio _ methods as well . this approach was followed by different authors in the past , e.g. for neon @xcite , argon @xcite , krypton @xcite , nitrogen @xcite , carbon dioxide @xcite , hydrogen chloride @xcite , acetonitrile @xcite , methanol @xcite , acetylene @xcite , and methanethiol @xcite . however , from an engineering point of view , this leads to difficulties . for an estimation of dispersive and repulsive interactions at least two molecules must be taken into account . to properly scan the energy hyper surface , many qm calculations at different distances and orientations of the molecules have to be performed . as the dispersive , and partly also the repulsive , interactions are just a very small fraction of the total energy calculated by qm , highly accurate methods like coupled cluster ( cc ) with large basis sets or even extrapolations to the basis set limit must be used for this task @xcite . due to the fact that this is computationally too expensive for engineering purposes , we used the parameters for the dispersive and repulsive interactions for an initial model from similar sites of other molecular models . some of these parameters were subsequently fitted in the optimization process to yield the correct vle behavior of the modeled substance . the optimization was performed using a newton scheme as proposed by stoll @xcite . the applied method has many similarities with the one published by ungerer et al . @xcite and later on modified by bourasseau et al . it relies on a least - square minimization of a weighted fitness function @xmath28 that quantifies the deviations of simulation results from a given molecular model compared to experimental data . the weighted fitness function writes as @xmath29 wherein the @xmath30-dimensional vector @xmath31 represents the set of @xmath30 model parameters @xmath32 to be optimized . the deviations of results from simulation @xmath33 to experimental data @xmath34 are weighted with the expected simulation uncertainties @xmath35 . equation ( [ eq_fitness ] ) allows simultaneous adjustment of the model parameters to different thermophysical properties @xmath36 ( saturated liquid densities @xmath37 , vapor pressures @xmath38 , and enthalpies of vaporization @xmath39 at various temperatures in the present work ) . the unknown functional dependence of the property @xmath36 on the model parameters is approximated by a first order taylor series developed in the vicinity of the initial parameter set @xmath40 @xmath41 therein , the partial derivatives of @xmath36 with respect to each model parameter @xmath42 , i.e. the sensitivities , are calculated from difference quotients @xmath43 assuming a sound choice of the model parameter variations @xmath44 , i.e. small enough to ensure linearity and large enough to yield differences in the simulation results significantly above the statistical uncertainties , this method allows a step - wise optimization of the molecular model by minimization of the fitness function @xmath28 . experience shows that an optimized set of model parameters was usually found within a few iterative steps when starting from a reasonable initial model . correlations for vapor pressure , saturated liquid density , and enthalpy of vaporization , taken from the dippr database @xcite , were used as `` experimental data '' for model adjustment and evaluation . this was done even in cases where the correlation is based on no or only few true experimental data points , as the correlations were regarded best practice . the comparison between simulation results and experiment was done by applying fits to the simulation data according to lotfi et al . @xcite . the relative deviation between fit and correlation was calculated in steps of 1 k from 55 to 97% of the critical temperature and is denoted by `` mean unsigned error '' in the following . vapor - liquid equilibrium simulations were performed using the grand equilibrium method by vrabec et al . @xcite , technical simulation details are given in the appendix . the selected ten molecules belong to different substance classes to show the wide applicability of the proposed strategy . in the present work , we restricted ourselves to small molecules , where the internal degrees of freedom may be neglected . thus , the molecular models are rigid , using the most stable configuration determined by qm . the optimized parameter sets of the new molecular models are summarized in table [ tab_parameter ] . table [ tab_critical ] compares the critical properties from simulation to experimental data . the critical properties from simulation were obtained through fits to vle simulation results as suggested by lotfi et al . @xcite . the estimated uncertainties of critical temperature , critical density , and critical pressure from simulation are 0.5 , 2 , and 2% , respectively . a very good agreement between simulation and experiment was reached , being predominantly within the combined error bars . in the following sections , substance specific details are briefly discussed . furthermore , references to alternative models from the literature are given and the simulation results from the present models are compared to simulation data from the literature where available . numerical vle simulation results are given as supplemtary material . the branched alkane iso - butane and the cyclic alkane cyclohexane show only very weak static polarities . here , the main contributions to the intermolecular interaction are dispersion and repulsion . the electrostatic interactions have only a minor influence but should not be neglected completely . in the literature different molecular models for iso - butane can be found , which are mostly based on force fields for branched alkanes . the well - known opls force field by jorgensen et al . is available in two versions for iso - butane , one using the united - atom approach @xcite and one using an all - atom description @xcite . both opls force fields were optimized to liquid density and enthalpy of vaporization at 293 k only . the model of poncela et al . @xcite was adjusted to yield correct second virial coefficients . for a better applicability in a wider range of states , recent developments were optimized to experimental vle data . examples from this group are the force fields presented by nath and de pablo @xcite , martin and siepmann @xcite , bourasseau et al . @xcite , or chang and sandler @xcite . for the present model , four lj sites , one for each methyl group and one for the methine group , were used to describe dispersion and repulsion of iso - butane . the polarity was modeled by a single ( weak ) dipole ( 0.1347 d ) located in the center of mass . orientation and magnitude of the dipole were taken from the qm calculation . for the initial model , the lj parameters were taken from ungerer et al . it was sufficient to adjust a single parameter , the offset distance of the methine group . it was optimized to 0.4 of the carbon - hydrogen distance , cf . figure [ fig_united_atom ] and held constant subsequently . the parameters of the present model are given in table [ tab_parameter ] . figures [ fig_vle1_rho ] to [ fig_vle1_dhv ] show saturated densities , vapor pressure , and enthalpy of vaporization , respectively , from the present iso - butane model in comparison to experimental data @xcite . figure [ fig_vle_ibutan_dev ] shows a deviation plot between simulation and experimental data . in the deviation plot also simulation results from martin and siepmann , using their trappe force field , @xcite and from nath and de pablo @xcite are included . a very good agreement was obtained for the present model yielding mean unsigned errors in vapor pressure , saturated liquid density , and enthalpy of vaporization of 4.2 , 0.6 , and 1.8% , respectively , in the temperature range from 55 to 97% of the critical temperature , which is about 225 to 395 k. for vapor pressure , the present model yields significantly better results than the trappe force field , while no simulation data is available from nath and de pablo for this property . for saturated liquid density , all three models yield comparable results within 1% deviation to experimental data . no comparison between the models was possible for enthalpy of vaporization due to missing numerical data in @xcite . for cyclohexane different molecular models are available in the literature @xcite which all account for the internal degrees of freedom . nevertheless , the present cyclohexane model was assumed to be rigid and in its most stable configuration , i.e. the saddle shape . the molecular model consists of six lj sites , one for each methylene group . the static polarity was modeled by a single quadrupole parameterized according to qm . the two lj parameters for methylene were optimized to experimental vle data . the parameters of the present model are given in table [ tab_parameter ] . figures [ fig_vle1_rho ] to [ fig_vle1_dhv ] again show present vle simulation results in comparison to experimental data for cyclohexane . figure [ fig_vle_chex_dev ] shows the deviation plot including simulation results from bourasseau et al . @xcite . the mean unsigned errors in vapor pressure , saturated liquid density , and enthalpy of vaporization for the present model are 0.9 , 0.5 , and 5.6% , respectively . simulation results for the enthalpy of vaporization show systematic relative deviations towards the critical point , while the absolute deviation is below 2.5 kj / mol . compared to the model of bourasseau et al . improvements in the description of the saturated liquid density were achieved . the simulation results for vapor pressure agree within their assumed simulation uncertainties which were not reported by bourasseau et al . an interesting point is that both molecular models yield the same deviations from experimental data on the enthalpy of vaporization , while the dippr database reports true experimental data up to @xmath45 . for all other molecular models for cyclohexane mentioned above , no numerical vle simulation data is reported in the literature . thus , no comparison is made here . the present molecular model for formaldehyde consists of two lj sites , one for the oxygen atom and one for the methylene group , as well as one dipole . the dipole is located in the center of mass and its moment was specified according to qm results , cf . table [ tab_parameter ] . all four lj parameters were adjusted to experimental vle data and are given in table [ tab_parameter ] . figures [ fig_vle3_rho ] to [ fig_vle3_dhv ] compare simulation results and experimental vle data for formaldehyde , figure [ fig_vle3_dev ] shows the relative deviations . mean unsigned errors in vapor pressure , saturated liquid density , and enthalpy of vaporization are 4.3 , 0.9 , and 8.4% , respectively . note that , in contrast to iso - butane or cyclohexane , the available experimental data base is very weak here . in fact , for vapor pressure only a single data set from the year 1935 @xcite is available . for saturated liquid density and enthalpy of vaporization , respectively , a single data point at 254 k is `` accepted '' by the dippr database @xcite . thus , no further optimization of the molecular model was attempted although the desired quality seems not to be fully reached . hermida - ran and ros @xcite published a molecular model based on qm calculations of formaldehyde dimers and trimers . they applied their model to liquid phase simulations but report no results on vle properties . thus , no comparison is presented here . dimethyl ether was modeled with three lj sites in the present work , one for the oxygen atom and one for each methyl group . a dipole was located in the center of mass and oriented along the symmetry axis of the molecule , where the dipole moment was again taken from qm calculation . for an initial model , the same lj parameters of the methyl groups were taken as for the iso - butane model , i.e. those by ungerer et al . an adjustment of the two lj parameters of the oxygen site was sufficient to reach the desired quality . alternative models for dimethyl ether are given in @xcite . the simulation results for the present model of dimethyl ether in comparison to the experimental vle data are shown in figures [ fig_vle1_rho ] to [ fig_vle1_dhv ] . figure [ fig_vle_dme_dev ] shows the relative deviations between simulation and experiment also including simulation results from stubbs et al . @xcite and very recent results from ketko and potoff @xcite . for dimethyl ether a good experimental data base is available for optimization of the molecular model . the simulation data of the present model is in very good agreement with the correlations of experimental data . the mean unsigned errors in vapor pressure , saturated liquid density , and enthalpy of vaporization are 2.6 , 0.4 , and 1.0% , respectively . for vapor pressure , both molecular models specifically adjusted to dimethyl ether , i.e. the present model and the model by ketko and potoff , yield better results than the transferable molecular model by stubbs et al . , while for saturated density all models perform similarly . for enthalpy of vaporization , the present model and the model by ketko and potoff also yield comparable results . note that the parameters for the electrostatic interactions of the model by ketko and potoff were adjusted to experimental vle data as well . it can be summarized that the model by ketko and potoff @xcite and the present dimethyl ether model are of similar quality and outperform the transferable model by stubbs et al . @xcite . while ketko and potoff adjusted four lj parameters and the point charge magnitudes for their electrostatic interactions , following the proposed modeling strategy an optimization of only two lennard - jones parameters was sufficient to reach the same quality . for sulfur dioxide , a molecular model was published by sokoli et al . @xcite which was optimized to total energy and pressure in the liquid state . it was recently reviewed by ribeiro @xcite . alternatively , the commercial force field compass @xcite reports parameters for sulfur dioxide . for the present molecular model , the intermolecular interactions of sulfur dioxide were modeled with three lj sites , i.e. one per atom , plus one dipole and one quadrupole . the electrostatic sites are located in the center of mass and parameterized according to the results of qm calculation . the four parameters of the lj sites , i.e. @xmath46 , @xmath47 , @xmath48 , and @xmath49 , were adjusted to experimental vle data . all parameters of the molecular model are given in table [ tab_parameter ] . present simulation results for sulfur dioxide are compared to experimental vle data in figures [ fig_vle2_rho ] to [ fig_vle2_dhv ] . figure [ fig_vle3_dev ] shows the relative deviations between simulation and experiment for the present model , while no numerical vle simulation data for sulfur dioxide from other authors was available to us . mean unsigned errors in vapor pressure , saturated liquid density , and enthalpy of vaporization are 4.0 , 0.9 , and 1.6% , respectively . the good experimental data base , with more than 60 individual experimental vle data points , allows a thorough optimization of the molecular model to the desired quality . literature models for dimethyl sulfide are given in @xcite . the present dimethyl sulfide model consists of three lj sites , one for the sulfur atom and one for each methyl group . the electrostatic interactions are modeled by one dipole and two quadrupoles oriented perpendicularly to each other . this description of the electrostatics was chosen , as qm yields a charge distribution , which can not properly be described with a lower number of electrostatic sites . the lj parameters of the methyl groups were assumed to be the same as for iso - butane and dimethyl oxide , while the parameters of the sulfur group were adjusted to experimental vle data . figures [ fig_vle2_rho ] to [ fig_vle2_dhv ] show the present simulation results for dimethyl sulfide in comparison to experimental vle data . figure [ fig_vle_dms_dev ] shows the relative deviations between simulation and experiment for the present model and for the model of lubna et al . note that simulation results on enthalpy of vaporization were not included in @xcite . for the present model of dimethyl sulfide , mean unsigned errors in vapor pressure , saturated liquid density , and enthalpy of vaporization are 4.0 , 0.7 , and 3.8% , respectively . particularly the vapor pressure is better described than by the model of lubna et al . also thiophene was described in the present work by a rigid model in its most stable conformation , as for cyclohexane . five lj sites , one for each of the four methylene groups and one for the sulfur atom , as well as one dipole and one quadrupole were used . the electrostatic parameters were directly passed on from qm . a total of four lj parameters , i.e. @xmath50 , @xmath51 , @xmath46 , and @xmath47 , were adjusted to experimental vle data . the optimized parameters are given in table [ tab_parameter ] . alternative molecular models for thiophene can be found in the literature @xcite figures [ fig_vle2_rho ] to [ fig_vle2_dhv ] compare simulation results to experimental vle data for thiophene , while the relative deviations are shown in figure [ fig_vle_thiophen_dev ] . in the deviation plot also simulation results from lubna et al . @xcite , jurez - guerra et al . @xcite , and prez - pellitero et al . @xcite are included . mean unsigned errors of the present model in vapor pressure , saturated liquid density , and enthalpy of vaporization are 3.8 , 1.2 , and 3.2% , respectively . the thiophene model by jurez - guerra et al . @xcite shows significant deviations in both vapor pressure and saturated liquid density while no simulation results for the enthalpy of vaporization were given by the authors . the anisotropic united atoms ( aua ) potential by prez - pellitero et al . @xcite overpredicts the vapor pressure over the complete temperature range by up to 20% . simulation results for saturated liquid density and enthalpy of vaporization are in very good agreement with the dippr correlation for low temperatures but give noticeably higher values than the dippr correlation for temperatures above @xmath52 . the trappe force field by lubna et al . @xcite yields vapor pressure results that agree very well with the dippr correlation and the simulation results within their scatter . saturated densities are higher than the correlation towards the critical point for all four molecular models . for enthalpy of vaporization no numerical data were given by lubna et al . it should be noted that for thiophene experimental vapor pressure data is available for temperatures up to around @xmath53 , experimental saturated liquid densities and enthalpies of vaporization are only available up to approximately @xmath54 . thus , an assessment of the different molecular models regarding density and enthalpy of vaporization above @xmath55 on the basis of the dippr correlations is questionable . hydrogen cyanide was modeled in the present work with two lj sites , one for the methine group and one for the nitrogen atom . electrostatic interactions were modeled by one dipole and one quadrupole oriented along the symmetry axis , where the parameters were passed on from qm . all four lj parameters were adjusted to experimental vle data . the optimized parameters can be found in table [ tab_parameter ] . for hydrogen cyanide no other molecular models were found in the literature . figures [ fig_vle2_rho ] to [ fig_vle2_dhv ] compare simulation results and experimental vle data for hydrogen cyanide . figure [ fig_vle3_dev ] shows the relative deviations . unfortunately , the dippr database @xcite contains no true experimental data on the enthalpy of vaporization for hydrogen cyanide . consequently , in the optimization process , only minor attention was paid to the enthalpy of vaporization . mean unsigned errors in vapor pressure , saturated liquid density , and enthalpy of vaporization are nominally 7.2 , 1.0 , and 12.2% , respectively . several molecular models for acetonitrile are available in the literature . jorgensen and briggs @xcite , price et al . @xcite , gurdia et al . @xcite , and nikitin and lyubartsev @xcite present models that were optimized to the liquid density and enthalpy of vaporization at 293 k. hloucha and deiters @xcite give a polarizable molecular model for simulations in the liquid state while hloucha et al . @xcite published a model that is based on _ ab initio _ calculations . finally , wick et al . @xcite proposed an extension of their trappe force field that was optimized to vle data to cover acetonitrile . in the present work , acetonitrile was modeled using three lj sites , one for the methyl group , one for the central carbon atom , and one for the nitrogen atom . the electrostatic interactions were modeled by one dipole and one quadrupole , located in the center of mass and parameterized strictly according to qm results . the parameters of the lj sites of the methyl group and the nitrogen atom were adjusted to experimental vle data . the lj parameters of the central carbon atom were taken from unpublished work on carbon dioxide and excluded from optimization , only a very weak sensitivity of the vle simulation results on these parameters was found . all parameters of the molecular model are given in table [ tab_parameter ] . simulation results for acetonitrile are compared to experimental vle data in figures [ fig_vle3_rho ] to [ fig_vle3_dhv ] . the relative deviations between simulation and experiment are shown in figure [ fig_vle_acetonitril_dev ] next to results of the opls - ua force field by jorgensen and briggs @xcite and the trappe force field by wick et al . , which were reported for both models in @xcite . despite the good experimental data base , the desired quality was not achieved by the present optimization . only a fair description of the experimental vle was reached . mean unsigned errors in vapor pressure , saturated liquid density , and enthalpy of vaporization are 19.7 , 0.9 , and 5.4% , respectively . the large relative deviations in vapor pressure result from systematic underestimations at low temperatures . in this region also difficulties were encountered in the simulative calculation of the chemical potential in the liquid phase to determine the phase equilibrium . this was even the case when the more sophisticated gradual insertion method @xcite was used , as described in the appendix in greater detail . the opls - ua force field by jorgensen and briggs @xcite significantly overestimates the vapor pressure of acetonitrile and shows deviations in the saturated liquid density up to 4% , cf . figure [ fig_vle_acetonitril_dev ] . simulation results for vapor pressure obtained with the trappe force field by wick et al . @xcite show a very good agreement with the dippr correlation , while the saturated density is slightly overpredicted towards the critical point . no comparison of the literature models for acetonitrile is possible regarding the enthalpy of vaporization due to the lack of numerical simulation results . the present molecular model for nitromethane consists of four lj sites , one for the methyl group , one for the nitrogen atom , and one for the two oxygen atoms each . the electrostatic interactions were modeled by one dipole and one quadrupole oriented along the symmetry axis , where the parameters were specified according to qm results . all six lj parameters were adjusted to experimental vle data . alternative models are available in the literature @xcite . figures [ fig_vle2_rho ] to [ fig_vle2_dhv ] compare the present simulation results for nitromethane with experimental vle data . figure [ fig_vle_nitromethan_dev ] shows the relative deviations obtained with the present model , the opls - aa force field by price et al . @xcite , and the trappe force field by wick et al . for the present model of nitromethane , mean unsigned errors in vapor pressure , saturated liquid density , and enthalpy of vaporization are 18.7 , 0.2 , and 7.0% , respectively . again , a systematic underprediction of the vapor pressure at low temperatures was found , leading to the high relative deviations , as for acetonitrile . the opls - aa force field overpredicts the vapor pressure by about 80% and underpredicts the saturated liquid density by up to 6% . the trappe force field of wick et al . that was optimized to experimental vle data , yields very good results in vapor pressure for low temperatures while the result for the highest simulated temperature deviates by + 34% from the dippr correlation . regarding saturated liquid density , strong scatter and large statistical uncertainties of the simulation results are observed for both literature models . a strategy was proposed for the rapid development of molecular models for engineering applications . the strategy relies on results from qm _ ab initio _ calculations to include physically sound molecular properties and to reduce the number of adjustable parameters . dispersive and repulsive interactions were modeled by lj sites . thus , the lj interaction sites were located according to atom positions obtained by qm energy minimization on hartree - fock level . for the parameterization of the electrostatic interactions , qm calculations were performed using the mller - plesset 2 level of theory and the cosmo method . the resulting electron density distribution was reduced to ideal point dipoles and ideal linear point quadrupoles located in the center of mass . the moments and orientations of the dipoles and quadrupoles were passed on to the molecular models without any modification . a united - atom approach was used for methine , methylene , and methyl groups to reduce the total number of interaction sites . the parameters of the lj sites were initially taken from similar sites of other molecular models and were subsequently optimized to reproduce experimental vle data , i.e. vapor pressure , saturated liquid density , and enthalpy of vaporization . it was aimed to achieve deviations between simulation and experiment of below 5 , 1 , and 5% in vapor pressure , saturated liquid density , and enthalpy of vaporization , respectively . the new modeling strategy was successfully applied to ten molecules from different substances classes , i.e. iso - butane , cyclohexane , formaldehyde , dimethyl ether , sulfur dioxide , dimethyl sulfide , thiophene , hydrogen cyanide , acetonitrile , and nitromethane . simulation results for the different substances agree well with correlations of experimental data taken from the dippr database @xcite , with noticeable deviations in the vapor pressure at low temperatures for acetonitrile and nitromethane . for the two elongated molecules acetonitrile and nitromethane , the reduction of the electrostatic interactions to sites located in the center of mass seems to be an over - simplification , as the optimization of the molecular models was not fully successful . also an adjustment of the electrostatic parameters , while keeping the ratio of the polar moments constant ( not reported here in detail ) , did not yield significant improvements in the quality of the molecular models . a further study , e.g. using two or more spatially distributed electrostatic sites , is beyond the scope of this paper . for all other molecules a significant improvement compared to available molecular models from the literature was achieved . through an optimization to substance specific experimental vle data , a very good description of the phase equilibria was obtained . furthermore , in case of dimethyl ether it was shown that with the proposed modeling strategy it was possible to reach the same model quality by optimization of just two model parameters compared to five optimized parameters for the model by ketko and potoff @xcite . the authors gratefully acknowledge financial support by deutsche forschungsgemeinschaft , schwerpunktprogramm 1155 `` molecular modeling and simulation in process engineering '' . the simulations were performed the national super computer nec sx-8 at the high performance computing center stuttgart ( hlrs ) under the grant mmhbf and on the hp xc6000 super computer at the steinbuch centre for computing , karlsruhe under the grant mmstp . the authors want to thank inga utz for her modeling work on the sulfur and nitrogen containing compounds . the grand equilibrium method @xcite was used to calculate vle data at seven to thirteen temperatures from 50 to 97% of the critical temperature during the optimization process.for the liquid , molecular dynamics simulations were performed in the isobaric - isothermal ( @xmath56 ) ensemble using isokinetic velocity scaling @xcite and anderson s barostat @xcite . there , the number of molecules was @xmath57 throughout and the time step was @xmath58 to @xmath59 fs depending on the molecular weight and the magnitude of the intermolecular interactions . the initial configuration was a face centered cubic lattice , the fluid was equilibrated over @xmath60 time steps with the first @xmath61 time steps in the canonical ( @xmath62 ) ensemble . the production run time span was @xmath63 to @xmath64 time steps with a membrane mass of @xmath65 kg / m@xmath66 . widom s insertion method @xcite was used to calculate the chemical potential by inserting up to @xmath67 test molecules every production time step . in cases where widom s insertion method yielded large statistical uncertainties for the chemical potential , i.e. at high densities for strongly interacting molecules , monte carlo simulations were performed in the @xmath56 ensemble for the liquid . then , the chemical potential was calculated by the gradual insertion method @xcite . the number of molecules was @xmath68 . starting from a face centered cubic lattice , @xmath69 monte carlo cycles were performed for equilibration and @xmath70 for production , each cycle containing @xmath68 translation moves , @xmath68 rotation moves , and @xmath58 volume move . every @xmath71 cycles , @xmath72 fluctuating state change moves , @xmath72 fluctuating particle translation / rotation moves , and @xmath73 biased particle translation / rotation moves were performed , to determine the chemical potential . these computationally demanding simulations yield the chemical potential in dense and strong interacting liquids with high accuracy , leading to reasonable uncertainties in the vle . for the corresponding vapor , monte carlo simulations in the pseudo-@xmath74 ensemble were performed . the simulation volume was adjusted to lead to an average number of @xmath68 molecules in the vapor phase . after @xmath75 initial @xmath62 monte carlo cycles , starting from a face centered cubic lattice , @xmath76 equilibration cycles in the pseudo-@xmath74 ensemble were performed . the length of the production run was @xmath70 cycles . one cycle is defined here to be a number of attempts to displace and rotate molecules equal to the actual number of molecules plus three insertion and three deletion attempts . the cut - off radius was set to @xmath77 throughout and a center of mass cut - off scheme was employed . lennard - jones long - range interactions beyond the cut - off radius were corrected employing angle averaging as proposed by lustig @xcite . electrostatic interactions were approximated by a resulting molecular dipole and corrected using the reaction field method @xcite . statistical uncertainties in the simulated values were estimated by a block averaging method @xcite . schmidt , m.w . ; baldridge , k.k . ; boatz , j.a . ; elbert , s.t . ; gordon , m.s . ; jensen , j.h . ; koseki , s. ; matsunaga , n. ; nguyen , k.a . ; shujun , s. ; windus , t.l . ; dupuis , m. ; montgomery , a.m. _ j. comput . chem . _ * 1993 * , _ 14 _ , 13471363 . rowley , r.l . ; wilding , w.v . ; oscarson , j.l . ; yang , y. ; zundel , n.a . ; daubert , t.e . ; danner , r.p . _ dippr^^ data compilation of pure compound properties _ ; design institute for physical properties , aiche : new york , 2006 . lccccccccc interaction & @xmath78 & @xmath79 & @xmath80 & @xmath81 & @xmath82 & @xmath83 & @xmath84 & @xmath85 & @xmath86 + site & & & & & k & @xmath87 & @xmath87 & d & b + + * ch * & 0 & 0 & 0.8179 & 3.360 & 51.00 & & & & + * ch@xmath27*(1 ) & 1.7302 & 0 & -0.1893 & 3.607 & 120.15 & & & & + * ch@xmath27*(2 ) & -0.8651 & 1.4984 & -0.1893 & 3.607 & 120.15 & & & & + * ch@xmath27*(3 ) & -0.8651 & -1.4984 & -0.1893 & 3.607 & 120.15 & & & & + dipole & 0 & 0 & 0 & & & 0 & 0 & 0.1347 & + quadrupole & 0 & 0 & 0 & & & 0 & 0 & & 0.7236 + + * ch@xmath26*(1 ) & 0.0210 & -0.3118 & 1.8052 & 3.497 & 87.39 & & & & + * ch@xmath26*(2 ) & 1.5318 & 0.2989 & 0.8863 & 3.497 & 87.39 & & & & + * ch@xmath26*(3 ) & -1.5528 & 0.2983 & 0.9986 & 3.497 & 87.39 & & & & + * ch@xmath26*(4 ) & 1.5318 & -0.2989 & -0.8863 & 3.497 & 87.39 & & & & + * ch@xmath26*(5 ) & -1.5528 & -0.2983 & -0.9986 & 3.497 & 87.39 & & & & + * ch@xmath26*(6 ) & 0.0210 & 0.3118 & -1.8052 & 3.497 & 87.39 & & & & + quadrupole & 0 & 0 & 0 & & & 90 & 90 & & 0.8179 + + * o * & 0 & 0 & 0.6721 & 3.010 & 112.61 & & & & + * ch@xmath26 * & 0 & 0 & -0.7682 & 3.422 & 77.42 & & & & + dipole & 0 & 0 & 0.0480 & & & 180 & 0 & 2.6668 & + _ continued on next page _ lccccccccc _ continued from previous page _ + interaction & @xmath78 & @xmath79 & @xmath80 & @xmath81 & @xmath82 & @xmath83 & @xmath84 & @xmath85 & @xmath86 + site & & & & & k & @xmath87 & @xmath87 & d & b + + * o * & 0 & 0 & 0.6427 & 2.727 & 89.57 & & & & + * ch@xmath27*(1 ) & 1.4041 & 0 & -0.3086 & 3.607 & 120.15 & & & & + * ch@xmath27*(2 ) & -1.4041 & 0 & -0.3086 & 3.607 & 120.15 & & & & + dipole & 0 & 0 & 0 & & & 180 & 0 & 1.7040 & + + * s * & 0 & 0 & 0.3757 & 3.312 & 139.23 & & & & + * o*(1 ) & 1.2790 & 0 & -0.3653 & 3.106 & 43.18 & & & & + * o*(2 ) & -1.2790 & 0 & -0.3653 & 3.106 & 43.18 & & & & + dipole & 0 & 0 & 0 & & & 0 & 0 & 1.9980 & + quadrupole & 0 & 0 & 0 & & & 90 & 0 & & -5.3340 + + * s * & 0 & 0 & 0.2819 & 3.398 & 207.57 & & & & + * ch@xmath27*(1 ) & 1.1583 & 0 & -0.3868 & 3.607 & 120.15 & & & & + * ch@xmath27*(2 ) & -1.1583 & 0 & -0.3868 & 3.607 & 120.15 & & & & + dipole & 0 & 0 & 0 & & & 90 & 180 & 2.3610 & + quadrupole & 0 & 0 & 0 & & & 90 & 0 & & 3.0740 + quadrupole & 0 & 0 & 0 & & & 0 & 0 & & 2.7600 + + * s * & 0 & 0 & 0.7359 & 4.292 & 95.65 & & & & + * ch*(1)-s & 1.2513 & 0 & 0.2067 & 3.590 & 48.49 & & & & + * ch*(2)-s & -1.2513 & 0 & 0.2067 & 3.590 & 48.49 & & & & + * ch*(3)-ch & 0.7734 & 0 & -1.2643 & 3.590 & 48.49 & & & & + * ch*(4)-ch & -0.7734 & 0 & -1.2643 & 3.590 & 48.49 & & & & + dipole & 0 & 0 & 0 & & & 180 & 0 & 1.8120 & + quadrupole & 0 & 0 & 0 & & & 90 & 0 & & 6.5389 + _ continued on next page _ lccccccccc _ continued from previous page _ + interaction & @xmath78 & @xmath79 & @xmath80 & @xmath81 & @xmath82 & @xmath83 & @xmath84 & @xmath85 & @xmath86 + site & & & & & k & @xmath87 & @xmath87 & d & b + + * n * & 0 & 0 & 0.6380 & 3.233 & 39.69 & & & & + * ch * & 0 & 0 & -0.9671 & 3.445 & 102.44 & & & & + dipole & 0 & 0 & 0.0589 & & & 180 & 0 & 3.4084 & + quadrupole & 0 & 0 & 0.0589 & & & 0 & 0 & & 2.1800 + + * n * & 0 & 0 & 1.1507 & 3.368 & 53.00 & & & & + * c * & 0 & 0 & 0.0507 & 2.810 & 10.64 & & & & + * ch@xmath27 * & 0 & 0 & -1.2868 & 3.835 & 163.04 & & & & + dipole & 0 & 0 & 0 & & & 180 & 0 & 4.1186 & + quadrupole & 0 & 0 & 0 & & & 0 & 0 & & -3.1373 + + * n * & 0 & 0 & 0.2199 & 3.321 & 34.90 & & & & + * o*(1 ) & 1.1045 & 0 & 0.7858 & 3.060 & 45.17 & & & & + * o*(2 ) & -1.1045 & 0 & 0.7858 & 3.060 & 45.17 & & & & + * ch@xmath27 * & 0 & 0 & -1.5135 & 3.501 & 158.79 & & & & + dipole & 0 & 0 & 0.2535 & & & 180 & 0 & 3.9901 & + quadrupole & 0 & 0 & 0.2535 & & & 90 & 0 & & -4.7903 + .critical properties : present simulation results compared to recommended experimental data . the numbers in parentheses indicate the experimental uncertainty in the last digits . [ cols="<,^,^,^,^,^,^,^ " , ] c|ccccccc @xmath88 & @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath93 & @xmath94 & @xmath95 + k & mpa & mpa & mol / l & mol / l & mol / l & kj / mol & kj / mol + + 150 & & & 12.15(1 ) & 12.149 & & & + 200 & & & 11.28(1 ) & 11.325 & & & + 270 & 0.14(1 ) & 0.140 & 9.98(1 ) & 10.050 & 0.067(6 ) & 20.80(3 ) & 20.811 + 300 & 0.37(2 ) & 0.370 & 9.37(1 ) & 9.420 & 0.164(7 ) & 19.10(3 ) & 19.006 + 320 & 0.59(2 ) & 0.633 & 8.88(2 ) & 8.955 & 0.254(8 ) & 17.80(4 ) & 17.591 + 335 & 0.87(2 ) & 0.906 & 8.54(2 ) & 8.575 & 0.37(1 ) & 16.72(4 ) & 16.375 + 350 & 1.30(3 ) & 1.258 & 8.10(3 ) & 8.159 & 0.58(1 ) & 15.16(6 ) & 14.980 + 365 & 1.74(3 ) & 1.701 & 7.69(3 ) & 7.688 & 0.79(1 ) & 13.76(6 ) & 13.325 + 380 & 2.21(3 ) & 2.252 & 7.05(5 ) & 7.113 & 1.03(2 ) & 11.89(9 ) & 11.240 + 390 & 2.77(3 ) & 2.689 & 6.59(9 ) & 6.609 & 1.44(2 ) & 9.9(1 ) & 9.408 + + 300 & & & 9.13(0 ) & 9.174 & & & + 330 & 0.02(1 ) & 0.046 & 8.80(0 ) & 8.831 & 0.013(3 ) & 31.49(1 ) & 31.339 + 360 & 0.12(1 ) & 0.121 & 8.45(0 ) & 8.476 & 0.041(4 ) & 29.89(1 ) & 29.589 + 390 & 0.27(2 ) & 0.267 & 8.09(1 ) & 8.101 & 0.088(5 ) & 28.19(2 ) & 27.646 + 415 & 0.51(2 ) & 0.467 & 7.77(1 ) & 7.767 & 0.165(7 ) & 26.50(3 ) & 25.828 + 440 & 0.75(2 ) & 0.763 & 7.41(1 ) & 7.404 & 0.240(7 ) & 24.78(3 ) & 23.758 + 460 & 1.08(2 ) & 1.085 & 7.12(1 ) & 7.083 & 0.343(6 ) & 23.18(4 ) & 21.853 + 480 & 1.48(2 ) & 1.496 & 6.76(3 ) & 6.723 & 0.478(7 ) & 21.27(5 ) & 19.647 + 500 & 2.01(2 ) & 2.012 & 6.38(2 ) & 6.305 & 0.68(1 ) & 18.99(7 ) & 17.027 + 515 & 2.44(2 ) & 2.478 & 5.97(3 ) & 5.935 & 0.85(1 ) & 16.92(9 ) & 14.675 + 535 & 3.29(2 ) & 3.223 & 5.44(6 ) & 5.280 & 1.31(2 ) & 13.1(1 ) & 10.566 + + 250 & 0.103(1 ) & 0.085 & 27.81(3 ) & 27.50 & 0.053(1 ) & 22.41(4 ) & 23.30 + 300 & 0.582(2 ) & 0.549 & 24.83(4 ) & 24.40 & 0.273(3 ) & 19.49(5 ) & 20.80 + 330 & 1.327(6 ) & 1.271 & 22.69(4 ) & 22.30 & 0.617(7 ) & 17.22(5 ) & 18.90 + 350 & 1.99(7 ) & 2.058 & 21.11(5 ) & 20.60 & 0.95(7 ) & 15.6(2 ) & 17.30 + 370 & 3.03(7 ) & 3.180 & 18.92(7 ) & 18.70 & 1.53(8 ) & 13.2(2 ) & 15.30 + 385 & 4.12(7 ) & 4.298 & 16.7(1 ) & 16.90 & 2.24(9 ) & 10.9(2 ) & 13.20 + 390 & 4.49(9 ) & 4.732 & 15.8(2 ) & 16.20 & 2.5(1 ) & 10.2(3 ) & 12.20 + _ continued on next page _ c|ccccccc _ continued from previous page _ + @xmath88 & @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath93 & @xmath94 & @xmath95 + k & mpa & mpa & mol / l & mol / l & mol / l & kj / mol & kj / mol + + 160 & & & 18.25(1 ) & 18.209 & & & + 200 & & & 17.17(1 ) & 17.189 & & & + 280 & 0.32(3 ) & 0.335 & 14.79(1 ) & 14.844 & 0.15(1 ) & 19.90(1 ) & 19.896 + 300 & 0.64(3 ) & 0.623 & 14.11(1 ) & 14.156 & 0.29(1 ) & 18.57(2 ) & 18.623 + 320 & 1.09(2 ) & 1.065 & 13.39(1 ) & 13.401 & 0.49(1 ) & 17.11(2 ) & 17.124 + 335 & 1.53(4 ) & 1.523 & 12.80(2 ) & 12.772 & 0.70(1 ) & 15.86(2 ) & 15.806 + 350 & 2.18(3 ) & 2.110 & 12.11(3 ) & 12.067 & 1.02(1 ) & 14.29(4 ) & 14.262 + 365 & 2.83(3 ) & 2.843 & 11.27(3 ) & 11.242 & 1.36(2 ) & 12.65(5 ) & 12.391 + 380 & 3.86(3 ) & 3.742 & 10.31(7 ) & 10.196 & 2.09(3 ) & 10.1(1 ) & 9.950 + 390 & 4.57(3 ) & 4.444 & 9.6(2 ) & 9.215 & 2.69(4 ) & 8.3(1 ) & 7.625 + + 240 & 0.04(1 ) & 0.032 & 23.945(3 ) & 23.664 & 0.018(4 ) & 26.40(1 ) & 26.197 + 255 & 0.09(1 ) & 0.069 & 23.327(3 ) & 23.085 & 0.043(4 ) & 25.53(1 ) & 25.375 + 270 & 0.15(1 ) & 0.136 & 22.694(3 ) & 22.488 & 0.069(4 ) & 24.71(1 ) & 24.521 + 285 & 0.24(1 ) & 0.247 & 22.043(4 ) & 21.869 & 0.106(6 ) & 23.85(1 ) & 23.623 + 300 & 0.40(4 ) & 0.417 & 21.357(8 ) & 21.222 & 0.17(1 ) & 22.88(1 ) & 22.667 + 330 & 1.03(3 ) & 1.013 & 19.88(1 ) & 19.816 & 0.44(1 ) & 20.57(1 ) & 20.511 + 360 & 2.09(3 ) & 2.098 & 18.17(2 ) & 18.190 & 0.90(2 ) & 17.83(3 ) & 17.867 + 390 & 3.85(4 ) & 3.873 & 15.98(4 ) & 16.166 & 1.80(3 ) & 14.20(7 ) & 14.354 + 405 & 5.17(3 ) & 5.095 & 14.53(6 ) & 14.855 & 2.66(3 ) & 11.50(7 ) & 11.967 + + 250 & & & 14.385(3 ) & 14.390 & & & + 302 & 0.070(2 ) & 0.0747 & 13.448(7 ) & 13.492 & 0.028(1 ) & 26.43(1 ) & 27.607 + 333 & 0.210(5 ) & 0.2082 & 12.865(8 ) & 12.906 & 0.080(2 ) & 25.02(2 ) & 25.917 + 363 & 0.47(1 ) & 0.4647 & 12.233(8 ) & 12.288 & 0.169(4 ) & 23.41(2 ) & 24.086 + 383 & 0.74(1 ) & 0.7352 & 11.809(7 ) & 11.841 & 0.261(5 ) & 22.27(2 ) & 22.726 + 403 & 1.11(2 ) & 1.1092 & 11.32(1 ) & 11.355 & 0.387(7 ) & 20.95(2 ) & 21.216 + 428 & 1.74(2 ) & 1.7584 & 10.68(1 ) & 10.673 & 0.609(7 ) & 19.07(3 ) & 19.034 + 453 & 2.59(2 ) & 2.6616 & 9.95(2 ) & 9.857 & 0.93(1 ) & 16.85(5 ) & 16.336 + 478 & 3.68(3 ) & 3.8885 & 9.01(2 ) & 8.770 & 1.39(2 ) & 14.01(8 ) & 12.580 + + 270 & 0.0009(1 ) & 0.0023 & 12.83(2 ) & 12.967 & 0.0004(1 ) & 34.13(7 ) & 35.956 + 285 & 0.0033(2 ) & 0.0054 & 12.62(2 ) & 12.763 & 0.0014(1 ) & 33.46(6 ) & 35.270 + 300 & 0.0101(8 ) & 0.0115 & 12.398(6 ) & 12.555 & 0.0041(3 ) & 32.72(2 ) & 34.562 + 325 & 0.029(2 ) & 0.0333 & 12.068(6 ) & 12.199 & 0.0110(9 ) & 31.67(2 ) & 33.329 + 348 & 0.076(5 ) & 0.0753 & 11.732(6 ) & 11.859 & 0.027(2 ) & 30.59(2 ) & 32.127 + 383 & 0.19(3 ) & 0.2089 & 11.257(8 ) & 11.315 & 0.063(9 ) & 29.06(2 ) & 30.148 + 418 & 0.53(2 ) & 0.4755 & 10.699(6 ) & 10.728 & 0.166(5 ) & 27.02(2 ) & 27.939 + 464 & 1.18(2 ) & 1.1389 & 9.88(1 ) & 9.866 & 0.359(6 ) & 24.09(3 ) & 24.532 + 493 & 1.94(2 ) & 1.8132 & 9.33(2 ) & 9.238 & 0.599(8 ) & 21.76(5 ) & 21.928 + 530 & 3.17(2 ) & 3.0679 & 8.42(3 ) & 8.264 & 1.03(1 ) & 18.06(7 ) & 17.653 + 551 & 4.20(3 ) & 4.0312 & 7.86(5 ) & 7.530 & 1.46(2 ) & 15.2(1 ) & 14.240 + _ continued on next page _ c|ccccccc _ continued from previous page _ + @xmath88 & @xmath89 & @xmath90 & @xmath91 & @xmath92 & @xmath93 & @xmath94 & @xmath95 + k & mpa & mpa & mol / l & mol / l & mol / l & kj / mol & kj / mol + + 273 & 0.030(1 ) & 0.0350 & 26.21(3 ) & 26.548 & 0.0141(4 ) & 28.78(4 ) & 27.778 + 295 & 0.078(3 ) & 0.0877 & 25.11(4 ) & 25.354 & 0.035(1 ) & 27.30(4 ) & 27.060 + 315 & 0.169(5 ) & 0.1798 & 24.01(5 ) & 24.217 & 0.074(2 ) & 25.76(5 ) & 26.336 + 358 & 0.66(1 ) & 0.6400 & 21.52(8 ) & 21.536 & 0.292(6 ) & 21.86(7 ) & 24.451 + 388 & 1.43(3 ) & 1.3255 & 19.52(7 ) & 19.367 & 0.69(1 ) & 18.41(8 ) & 22.697 + 406 & 2.09(4 ) & 1.9655 & 17.9(8 ) & 17.859 & 1.08(2 ) & 16.0(1 ) & 21.323 + 420 & 2.69(3 ) & 2.6261 & 16.3(1 ) & 16.504 & 1.44(3 ) & 14.1(1 ) & 19.953 + 430 & 3.43(3 ) & 3.2072 & 15.5(1 ) & 15.373 & 2.05(3 ) & 12.1(1 ) & 18.690 + 435 & 3.70(3 ) & 3.5377 & 14.0(3 ) & 14.726 & 2.31(3 ) & 10.7(2 ) & 17.909 + + 270 & 0.0015(1 ) & 0.0029 & 19.71(1 ) & 19.636 & 0.0007(1 ) & 38.87(8 ) & 34.302 + 300 & 0.0074(3 ) & 0.0132 & 18.90(2 ) & 18.871 & 0.0032(1 ) & 36.78(7 ) & 32.952 + 327 & 0.029(1 ) & 0.0396 & 18.20(3 ) & 18.151 & 0.0112(6 ) & 34.88(7 ) & 31.641 + 360 & 0.088(3 ) & 0.119 & 17.25(2 ) & 17.223 & 0.033(1 ) & 32.42(6 ) & 29.886 + 400 & 0.294(5 ) & 0.352 & 16.05(2 ) & 16.003 & 0.109(2 ) & 29.09(6 ) & 27.459 + 420 & 0.51(1 ) & 0.558 & 15.47(4 ) & 15.341 & 0.190(4 ) & 27.27(8 ) & 26.078 + 436 & 0.76(3 ) & 0.782 & 14.78(1 ) & 14.777 & 0.289(9 ) & 25.31(5 ) & 24.866 + 450 & 1.00(4 ) & 1.032 & 14.25(2 ) & 14.253 & 0.38(1 ) & 23.98(5 ) & 23.705 + 464 & 1.27(9 ) & 1.341 & 13.65(3 ) & 13.693 & 0.48(3 ) & 22.58(5 ) & 22.427 + 490 & 2.08(4 ) & 2.101 & 12.42(3 ) & 12.515 & 0.85(2 ) & 19.20(9 ) & 19.603 + 505 & 2.69(3 ) & 2.673 & 11.52(5 ) & 11.707 & 1.18(2 ) & 16.7(1 ) & 17.549 + 510 & 2.73(5 ) & 2.889 & 11.06(6 ) & 11.405 & 1.16(2 ) & 16.5(1 ) & 16.753 + 518 & 3.14(4 ) & 3.263 & 10.37(7 ) & 10.871 & 1.37(2 ) & 15.0(1 ) & 15.308 + + 290 & 0.001(1 ) & 0.0031 & 18.82(2 ) & 18.669 & 0.0003(1 ) & 44.05(4 ) & 38.511 + 353 & 0.026(1 ) & 0.0503 & 17.25(2 ) & 17.239 & 0.0092(4 ) & 39.62(5 ) & 35.812 + 390 & 0.105(3 ) & 0.1602 & 16.32(3 ) & 16.331 & 0.035(1 ) & 36.78(7 ) & 33.983 + 430 & 0.33(1 ) & 0.4354 & 15.27(2 ) & 15.270 & 0.107(3 ) & 33.47(5 ) & 31.716 + 471 & 0.92(2 ) & 1.0033 & 14.13(3 ) & 14.058 & 0.301(6 ) & 29.29(8 ) & 28.931 + 500 & 1.56(2 ) & 1.6696 & 13.12(4 ) & 13.083 & 0.525(8 ) & 25.98(9 ) & 26.518 + 530 & 2.70(4 ) & 2.6942 & 11.93(5 ) & 11.899 & 0.99(2 ) & 21.5(1 ) & 23.347 + 560 & 4.09(6 ) & 4.2110 & 10.4(1 ) & 10.345 & 1.64(3 ) & 16.7(2 ) & 18.696 +

we present and analyze _ far ultraviolet spectroscopic explorer _ ( fuse ) observations of six solar analogs . these are single , main - sequence g05 stars selected as proxies for the sun at several stages of its main - sequence lifetime from myr up to gyr . the emission features in the fuse 9201180 wavelength range allow for a critical probe of the hot plasma over three decades in temperature : k for the h i lyman series to k for the coronal fe xviii line . using the flux ratio c iii/ as diagnostics , we investigate the dependence of the electron pressure of the transition region as a function of the rotation period , age and magnetic activity . the results from these solar proxies indicate that the electron pressure of the stellar-k plasma decreases by a factor of between the young , fast - rotating ( d ) magnetically active star and the old , slow - rotating ( d ) inactive star . also , we study the variations in the total surface flux for specific emission features that trace the hot gas in the stellar chromosphere ( c ii ) , transition region ( c iii , o vi ) , and corona ( fe xviii ) . the observations indicate that the average surface fluxes of the analyzed emission features strongly decrease with increasing stellar age and longer rotation period . the emission flux evolution with age or rotation period is well fitted by power laws , which become steeper from cooler chromospheric ( k ) to hotter coronal ( k ) plasma . the relationship for the integrated ( 9201180 ) fuse flux indicates that the solar far - ultraviolet ( fuv ) emissions were about twice the present value 2.5 gyr ago and about 4 times the present value 3.5 gyr ago . note also that the fuse / fuv flux of the zero - age main sequence sun could have been higher by as much as 50 times . our analysis suggests that the strong fuv emissions of the young sun may have played a crucial role in the developing planetary system , in particular through the photoionization , photochemical evolution and possible erosion of the planetary atmospheres . some examples of the effects of the early sun s enhanced fuv irradiance on the atmospheres of earth and mars are also discussed .

astro-ph0305279

the sun s magnetic activity is expected to have greatly decreased with time ( skumanich 1972 ; simon , boesgaard , & herbig 1985 ; dorren & guinan 1994 ; guinan , ribas , & harper 2002 ) as the solar rotation slows down because of angular momentum loss in the stellar wind and resultant reduction in magnetic dynamo - related activity . the study of the young sun s far - ultraviolet ( fuv ) fluxes using solar proxies provides important diagnostics for the state of the younger solar system and the physics of the much more active early sun . the comprehensive `` sun in time '' project , begun in 1988 ( dorren & guinan 1994 ) , focuses on the study of the long - term evolution of the outer atmosphere of an early g star , from the zero - age main sequence to the terminal - age main sequence . a crucial component of the program is a carefully - selected sample of nearby solar analogs ( gdel , guinan , & skinner 1997 ) with spectral types confined between g0 - 5 v and with well - determined physical properties ( including temperatures , luminosities , and metal abundances ) . we have obtained extensive photometry ( @xmath8 ) of these stars from ground - based observatories for over a decade to determine their rotation periods , investigate starspots and possible activity cycles . in addition , we have been able to estimate the stellar ages by making use of their memberships in clusters and moving groups , rotation period age relationships , and , for the older stars , fits to stellar evolution models . the sample of solar proxies within the `` sun in time '' program ( see guinan , ribas & harper 2002 ) contains stars that have masses close to 1 m@xmath9 and cover most of the sun s main sequence lifetime from @xmath0130 myr to @xmath09 gyr . basically , the program stars have similar convective - zone depths and chemical abundances to those of the sun and vary only by their age and rotation periods ( @xmath10 ) , and hence dynamo - generated magnetic activity , i.e. , chromospheric , transition region ( tr ) , and coronal emissions . in essence , we use these one solar mass stars with @xmath10 between 2.7 and @xmath035 days as laboratories to study and test solar and stellar dynamo theories by varying essentially only one parameter : rotation period . the `` sun in time '' is a comprehensive and multi - frequency program that addresses a variety of topics : study of short- and long - term magnetic evolution ; physics and energy transfer mechanisms of the chromosphere , transition region ( tr ) and corona ; evolution of the spectral irradiances of the sun and their effects on paleo - planetary environments and atmospheres . to this end we utilize observational data spanning almost the entire electromagnetic spectrum , which allows us to probe the structure of stellar / solar atmospheres . the fuv has not been readily accessible until the successful launch of the _ far ultraviolet spectroscopic explorer _ ( fuse ) . fuse allows a critical probe of the hot plasmas over nearly three decades in temperature : e.g. , @xmath1 k for the h i lyman series ( ly@xmath11 , ly@xmath12 , ... ) through o vi @xmath131032 , 1038 at @xmath14 k , and the recently identified fe xviii @xmath3975 coronal emission feature at @xmath2 k. several of the strongest emission features , such as c iii @xmath13977,1176 and o vi @xmath131032,1038 , originate in tr plasmas , and are pivotal for understanding the mechanisms of chromospheric and coronal heating . thus , spectrophotometry with fuse fills the energy gap between iue / hst ( 11603200 ) and euve ( 80720 ) to yield a complete and comprehensive picture of a solar - type star s atmosphere at different ages and rotation periods . in this paper we present and analyze fuse observations of six bright solar analogs that span a wide range of rotation periods ( and ages ) from 2.7 days ( @xmath0130 myr ) to @xmath035 days ( @xmath09 gyr ) . the stars , whose properties are listed in table [ tabprop ] along with the sun s , are : ek dra , @xmath15 uma , @xmath16 cet , @xmath11 com , @xmath11 hyi , and 16 cyg a. the present study focuses on two particular aspects of the `` sun in time '' program that can be effectively addressed with fuv observations : _ 1 . ) _ the flux ratios between the c iii @xmath31176 and c iii @xmath3977 lines ( both within the fuse spectral range ) yield empirical measures and estimates of the electron pressure ( @xmath17 ) of the tr ; _ 2 . ) _ integrated fluxes for selected emission lines can be used to study the evolution of the solar fuv irradiances at different ages and rotation periods . fuv observations of the six targets discussed here were obtained with fuse . the fuse instrument consists of four spectroscopic channels ( lif1 , lif2 , sic1 , and sic2 ) that cover a combined wavelength range 9051187 . detailed information on the fuse mission and its on - orbit performance are provided by moos et al . ( 2000 ) and sahnow et al . ( 2000 ) . the observations of the targets were secured during fuse cycles 1 , 2 and 3 . spectra of three stars ( @xmath16 cet , @xmath11 com , and @xmath11 hyi ) were taken as part of the guest investigator program a083 ( p.i . guinan ) , one star ( @xmath15 uma ) was observed as part of the program b078 guinan ) , and to additional stars ( ek dra and 16 cyg a ) were acquired within program c102 ( p.i . the large aperture ( lwrs ) was employed in all cases to minimize photon losses due to telescope misalignments , and this caused a fairly severe contamination of day - time spectra from geocoronal emission lines and scattered solar radiation . information on the dates and exposure times of the observations is provided in table [ taball ] . the fuse data were retrieved from the multimission archive at space telescope ( _ mast _ ) , and then re - calibrated with calfuse 2.0.5 . for each star , the time - tagged data for the sub - exposures were combined into a single dataset . the spectral extraction windows were individually aligned to ensure correct extraction of the stellar spectra . we employed the calfuse 2.0.5 event burst screening and default background subtraction . the data were then calibrated and screened separately for orbital night and day , allowing us to estimate any contamination of the stellar spectra from `` airglow '' and scattered solar light . the scattered light was found to be negligible compared to the stellar source , except for the faint fuv target 16 cyg a , which could suffer from some contamination especially in the o vi features . the calibrated spectra are oversampled so they were re - sampled by binning over four wavelength intervals , thereby increasing the signal - to - noise ratio . a comparison of the emission line fluxes from different channels , e.g. , c iii 977 sic2a and sic1b , provides an independent check on data artifacts , and whether the star was in both apertures for the same length of time . for illustration we show the extracted spectrum of @xmath16 cet in figure [ figsp ] . note that we only show the night - time data that presents less contamination by terrestrial airglow and geocoronal emission . thus , all features in figure [ figsp ] are of stellar origin , although the h i lyman series still has some geocoronal contamination . the fuv spectra of the other four targets are qualitatively similar , and only the spectrum of ek dra and @xmath15 uma are somewhat noisier . for these two stars the increased noise arises from the failure of the sic2a and lif2a observations and the consequent use of the less sensitive segment b of detector 1 ( sic1b and lif1b ) for those parts of the spectra . as can be seen in figure [ figsp ] , the continuum level is negligible and the flux is in the form of emission lines . individual channels are represented in the figure because the fuse effective area is heavily dominated by only one channel at each wavelength . in this case , sic2a , lif1a , and lif2a spectra are shown in the upper , central , and lower panels , respectively . studies of fuse observations of late - type stars , ab dor and capella , have been presented by ake et al . ( 2000 ) and young et al . ( 2001 ) , respectively . the authors carried out thorough investigations of the chromospheric and tr emission features in the fuse wavelength range and provided identification spectra . a comparison with our sample spectrum in figure [ figsp ] indicates a striking resemblance to these spectra , except for small to moderate changes in the relative strengths of some features . we have thus been able to identify the most significant features in our spectra by using the results of ake et al . ( 2000 ) and young et al . ( 2001 ) . among the strongest fuv emission lines in coronal spectra of late - type stars are the c iii singlet at 977.020 , the c iii multiplet around 1176 , and the o vi doublet at 1031.925 and 1037.614 . we focus here primarily on the analysis of stellar features that are not significantly affected by geocoronal contamination and day - time spectra can be used without any loss of accuracy . we thus combined the night - time and day - time data by using a exposure - time weighting scheme . all of our fuse spectra show clean profiles for these lines that permit reliable integrated flux measurements , which were carried out by means of noao / iraf tasks . different methods ( plain addition , gaussian fitting ) were employed to measure the integrated line fluxes , study the associated uncertainties , to check the line profiles , and also to assess the possible presence of interstellar absorption ( see below for further discussion ) . the total integrated fluxes for each star and their intrinsic uncertainties ( random errors ) are provided in table [ taball ] . the ratio of the fluxes for the c iii @xmath31176 and @xmath3977 emission features is provided in table [ tabr ] . this ratio , which yields an excellent electron pressure / density diagnostic , is the topic of discussion for [ secne ] . the ratio of the two o vi features ( @xmath31038/@xmath31032 ) is not provided in table [ tabr ] but it is in all cases very close to 1:2 . this is the value expected for an effectively - thin plasma where the flux ratio is proportional to the ratio of the collision strengths for the two transitions , which for o vi is very close to 1:2 ( zhang , sampson , & fontes 1990 ) . plasma model fits to extensive x - ray observations of the three youngest stars in the sample ( ek dra , @xmath15 uma , and @xmath16 cet ) indicate the presence of high - temperature coronal material ( gdel et al . 1997 ) . as discussed by feldman & doschek ( 1991 ) , young et al . ( 2001 ) and redfield et al . ( 2003 ) , coronal emission lines of highly - ionized ions arising from forbidden transitions are expected in the fuse wavelength range . a coronal fe xviii forbidden line at 974.85 was identified for the first time in fuse spectra of capella ( young et al . fuse spectra of ek dra , @xmath15 uma and @xmath16 cet also clearly show this emission feature ( see figure [ figfe ] for an example ) . very recently , redfield et al . ( 2002 ) have reported the detection of a coronal fe xix @xmath31118 emission line in the fuse spectra of several nearby cool stars . the fuse spectra of ek dra , @xmath15 uma and @xmath16 cet show weak fe xix @xmath31118 features barely above the noise level . as pointed out by redfield et al . ( 2003 ) , this line can be corrupted by a nearby c i multiplet . because of the poor s / n of the flux measurements for fe xix @xmath31118 and concerns with nearby line contamination have focused only on the clean fe xviii @xmath3975 line as a tracer of the hot coronal plasma . integrated flux measurements have been carried out using the same method described above and are reported in table [ taball ] . no trace of this feature is detected for the older stars in the sample down to the noise level of the spectra , and flux upper limits are therefore provided in table [ taball ] . in addition to integrating the fluxes , we also obtained line centroid positions and fwhm measurements for the selected features in table [ taball ] . the motivation for such measurements was the search for possible line shifts or broadenings that could arise from plasma outflows or extended material . a critical point when measuring line shifts is an accurate wavelength scale . aperture centering problems caused by misalignments may result in systematic velocity shifts among the different channels . redfield et al . ( 2002 ) used ism features to derive the absolute wavelength scale of their spectra . in our case ism absorption features are not measurable and we followed a different approach . namely , we aligned the o vi @xmath131032,1038 doublet , which is available in all four channels , using the lif1 channel as the reference ( this is the channel used for guiding ) . this procedure should ensure relative velocities accurate to 510 km s@xmath18 in all channels . the velocity corrections were generally quite relevant with values up to 40 km s@xmath18 . with a corrected wavelength scale , we measured the line centroids and velocity shifts for several features . the results are given in table [ tabrv ] . no significant line shifts with respect to the photospheric velocity were observed down to the precision of the measurements . the expected @xmath010 km s@xmath18 redshift of tr features ( e.g. , redfield et al . 2002 ) is only observed in the most active star of the sample , ek dra . the rest of the stars actually exhibit slightly blueshifted tr features , albeit with very low statistical significance . fwhm measurements for all the emission lines in table [ taball ] yielded relatively small widths of @xmath19 , which are equivalent to velocities of @xmath20 km s@xmath18 . only ek dra , the most active of our targets , was found to have broader line profiles . in this case , the measured fwhm was about 0.35 or 110 km s@xmath18 . ( the actual rotational velocity of this star is @xmath21 km s@xmath18 . ) a closer inspection of the c iii @xmath3977 and o vi @xmath31032 line profiles revealed extended line wings in some of the targets . this is a well - known phenomenon ( see , e.g. , linksy & wood 1994 ) that is usually modeled by assuming double gaussians with components of different widths . for the stars in our sample , only ek dra shows unambiguous evidence of a broad component , whereas @xmath15 uma and @xmath16 cet show somewhat extended line wings but the two - component fits yielded inconclusive results . two - gaussian fits to the o vi @xmath31032 line of ek dra indicate a ratio between the broad component flux and the total flux of 0.38 . thus , extended line profiles appear to correlate with stellar activity ( and age ) , in agreement with the results of redfield et al . ( 2002 ) and wood , linsky , & ayres ( 1997 ) . the presence of broad and narrow components in high - temperature lines is still not well understood . a plausible model ( wood et al . 1997 ) suggests that these may arise from high - velocity nonthermal motions during magnetic reconnection events ( microflares ) . before analyzing and modeling the line integrated fluxes , the possible effects of interstellar medium ( ism ) absorption must be investigated . the c iii @xmath3977 line and the h i lyman series lines are the most sensitive to attenuation effects because of possible superimposed ism absorption . as shown in table [ tabprop ] , the stars in the sample are nearby and lie within 734 pc of the sun and have negligibly small values of @xmath22 from ism dust . however , even for the small ism column densities at these close distances ( @xmath23 @xmath24 ; redfield & linsky 2000 ) , one expects some absorption in the lines mentioned above because they are among the strongest ism features in the entire spectrum . unfortunately , both the resolution and s / n of our fuse spectra are not sufficiently high for reliable direct estimations of the ( expected weak ) ism fuv absorptions . we therefore adopted an alternative procedure and carried out indirect determinations of the flux corrections for the individual spectral features . to estimate the ism column densities in the line of sight of our targets we made use of the local ism observations and model by redfield & linsky ( 2002 ) . these authors analyzed high - resolution observations of fe , mg and ca features to determine column densities of these elements toward stars within 100 pc of the sun . one of our targets , @xmath16 cet , is included in redfield & linsky s study and thus direct ism measurements are available . for the other five targets we adopted the local ism characteristics of neighboring stars with direct measurements : dk uma for ek dra and @xmath15 uma , hz 43 for @xmath11 com , @xmath25 dor for @xmath11 hyi , and @xmath26 cyg for 16 cyg a. the c abundance was computed from the observed column densities of mg ii and ca ii and local ism log abundances ( relative to h ) of @xmath27 , @xmath28 and @xmath29 for c , mg ii , and ca ii , respectively , from wood et al . ( 2002b ) and redfield & linsky ( 2000 ) . the c iii column density was subsequently obtained by adopting the local ism ionization ratio of c iii / c ii@xmath300.02 from wood et al . ( 2002b ) . ( note that c ii is the dominant ionization species in the ism . ) the ism absorption equivalent widths were estimated from the c iii column densities and the transition oscillator strengths through the curve of growth method ( see spitzer 1978 ) . for the c iii @xmath3977 transition , despite its strength , the ism absorption features for all our targets were found to be relatively weak ( unsaturated ) because of the small column densities . then , using the ism component velocities and the observed c iii line profile , we calculated the fraction of stellar flux absorbed by the ism . our results indicate that the flux corrections for the c iii @xmath3977 line are very small ( 3 - 7% ) for ek dra , @xmath15 uma , @xmath16 cet , and @xmath11 com , and somewhat larger ( 11 - 13% ) for @xmath11 hyi and 16 cyg a. in addition to the features mentioned above , we attempted an estimation of the total integrated fluxes within the fuse bandpass ( 9201180 ) . however , even when restricting to night - time spectra , contamination of the stellar h i lyman features by geocoronal emission is very significant . this is because fuse observations in this program were obtained through the large aperture to ensure , where possible , that the important c iii 977 emission line was detected in the sic channels . as a consequence , the lyman geocoronal emission is much broader than for the medium aperture , reaching about @xmath31 km s@xmath18 near ly@xmath11 . this emission overlies and in most cases completely dominates the stellar emission profile . the stellar signal is expected to be a self - reversed emission feature with a dark ism absorption core . to estimate the stellar fluxes for ly@xmath11 we took the difference between the night and day spectra , and assumed that the shape of geocoronal emission remains constant . a fraction of this geocoronal emission was then subtracted from the night spectra so that the residual flux at the location of the ism feature is consistent with the flux predicted by a simulation of the stellar profile attenuated by the ism . if emission peaks remained in the subtracted spectrum we assummed that the remaining signal was stellar . only @xmath16 cet and @xmath15 uma had any significant signal to satisfy these requirements . to estimate the shape of the underlying stellar emission profile , required to estimate the total stellar flux , we used the functional form of the `` universal '' profile for strong partially coherent scattered lines derived by gayley ( 2002 ) . this allowed us to use the observed flux in the wings , which is relatively uncontaminated by geocoronal emission , to extrapolate towards line center and provide a total flux estimate . the model stellar profile was then attenuated by an ism model ( described above ) and convolved with a psf of @xmath32 to mimic the fuse spectral resolution . from the fits we obtained total ly@xmath11 fluxes of 4.0@xmath3310@xmath34 erg s@xmath18 @xmath24 and 2.4@xmath3310@xmath34 erg s@xmath18 @xmath24 for @xmath15 uma and @xmath16 cet , respectively . given the uncertainty of the spectra , the adopted ism model and the assumed stellar profile shapes , we expect that the accuracy of estimated stellar flux is no better than a factor of 1.52 . future observations of h ly@xmath35 and mdrs observations of ly@xmath11 are warranted and required to improve the quantification of the total stellar fuv emission . a correction to account for the rest of the h i lyman series was derived by comparison with soho sumer spectra for the sun . we roughly estimate that the total flux contributed by the h i lyman lines within the fuse bandpass is about 1.6 times the ly@xmath11 flux . then , the total flux in the range 9201180 was calculated by adding the contributions from the individual features discussed above and the integrated h i lyman series . we obtained values of 10.5@xmath3310@xmath34 erg s@xmath18 @xmath24 and 8.7@xmath3310@xmath34 erg s@xmath18 @xmath24 for @xmath15 uma and @xmath16 cet , respectively . our integrations indicate that the h i lyman features in this wavelength range ( ly@xmath11 , ly@xmath12 , etc ) are important contributors ( @xmath04060% ) to the total flux of both @xmath15 uma and @xmath16 cet . this is also presumably the case of the remaining solar - type stars in the sample but the weak stellar signal and strong geocoronal contamination of the fuse observations prevented us from carrying out reliable measurements . one straightforward application of the measured line fluxes presented in table [ taball ] is the analysis of their evolution along the main sequence of a one solar mass star . to compare the absolute irradiances , we scaled the observed fluxes to surface fluxes at a radius of 1 r@xmath36 . to estimate the actual radius of each target we made use of the stellar luminosity ( computed from the hipparcos distances and apparent magnitudes ) and effective temperature ( obtained from spectroscopic measurements in the literature ) . the radii of the targets are included in table [ tabprop ] and using these we calculated the equivalent flux for a star of 1 r@xmath36 in radius . as can be seen in table [ tabprop ] , the radii of all stars but @xmath11 hyi and 16 cyg a are very close to 1 r@xmath36 ( within 5 - 8% ) and the correction was typically very small . the situation is somewhat different for @xmath11 hyi because of its advanced evolutionary stage and its mass being about @xmath01.1 m@xmath36 . in this case the flux correction was quite significant . the adopted final stellar surface fluxes for the four emission features studied are provided in table [ tabsurf ] . also in this table are the total integrated fuv fluxes in the fuse wavelength range ( 9201180 ) for two of the targets ( @xmath15 uma and @xmath16 cet ) . all fluxes have been corrected for ism absorption ( see [ secism ] and [ lyman ] ) . before proceeding with a numerical analysis of these data , an illustrative visual impression of the surface flux changes for the targets stars can be seen in figure [ figovi ] . we have plotted details of the spectra ( flux corrected to the stellar surface ) in a narrow 10- wavelength interval that contains the o vi doublet at @xmath131032,1038 . a rough comparison of the line variations as a function of stellar age does reveal a clear and definite trend or decreasing strength with increasing age . for a more quantitative analysis we used the integrated surface fluxes in table [ tabsurf ] , which are plotted in figure [ figirr ] as a function of the stellar rotation period . the features represented are c ii @xmath31037 , c iii @xmath3977 , o vi @xmath31032 , and fe xviii @xmath3975 . the typical peak formation temperatures of the c ii , c iii , o vi , and fe xviii ions are approximately @xmath37 k , @xmath38 k , @xmath39 k , and 6 mk , respectively ( arnaud & rothenflug 1985 ) . thus , these lines probe a wide interval of plasma temperatures . as can be seen , the newly - detected coronal line of fe xviii is very important to our investigation because it extends the coverage to the hot plasma component . note the much smaller surface fluxes of @xmath11 hyi with respect to the younger solar analogs for all the features studied . roughly , @xmath11 hyi s flux levels are comparable to , or slightly smaller than , those of today s sun . in addition , figure [ figirr ] also depicts the total integrated fuv fluxes in the fuse bandpass for @xmath15 uma and @xmath16 cet . the random measurement errors affecting the fluxes in figure [ figirr ] are discussed in [ secan ] . however , additional systematic uncertainties can be caused by the intrinsic stellar activity cycle . from an analogy with the sun , we expect peak - to - peak cycle amplitudes in the fuv of about 3050% . since the target stars are observed at different stages of their activity cycles , these activity variations can be a source of scatter in the irradiance plot . fortunately , this activity - related scatter is rendered negligible by the large relative differences between the fluxes of the targets ( with factors of over 30 in flux ) . the fluxes of figure [ figirr ] can be fit to good accuracy with power laws of different slopes . these are listed in table [ tabslopes ] , which contains the power - law slopes using both rotation period ( @xmath10 ) and age as independent variable . note that only two fuse measurements are available for the total flux within 9201180 . to constrain the fit we used the flux values for the quiet sun obtained through integration of the solar reference spectrum by heroux & hinteregger ( 1978 ) in the fuse wavelength window . the resulting surface flux was 2.4@xmath3310@xmath40 erg s@xmath18 @xmath24 . interestingly , the slopes not only change from line to line , but also they show a clear trend . the power law becomes steeper as we move from cooler to hotter material . the most extreme trend is that of the flux for the coronal ion fe xviii . the results indicate that the flux is reduced by a factor of over 1000 when the rotation period increases ten - fold . to serve as a reference , plasma of similar ( or maybe slightly lower ) temperature is probed when analyzing fluxes in the x - ray domain . gdel et al . ( 1997 ) studied x - ray fluxes ( 0.12.4 kev ) of about a dozen solar analogs covering also a wide period range . a power law fit to the data yielded a slope of @xmath41 , which is in very good agreement with our value for the decrease of the fe xviii flux . the results indicate that the emission from hotter plasma decreases more rapidly than that from cooler plasma . analogous behavior was observed by gdel et al . ( 1997 ) when comparing the emission measures derived from observations in hard and soft x - rays . a possible scenario to explain the steeper flux decrease for hotter plasma could be the weakening of the stellar dynamo as the stars spin down with age and the corresponding decline in the efficiency of heating mechanisms ( e.g. flares ) and in the strength of magnetic fields that confine the gas . it should be noted that our program stars , except for age and rotation period , have similar properties to the sun ( mass , radius , effective temperature , metal abundance ) . importantly , these stars should all have similar convection zone depths , which together with rotation , is an important parameter in most modern magnetic dynamo theories . thus , the stars in our sample can serve as laboratories for testing the generation ( and dissipation ) of magnetic dynamo - related energy and heating in solar - type stars where rotation ( angular velocity ) is the only important variable . the observed power law dependencies on rotation of the various emission line fluxes should be important as inputs for testing and constraining stellar / solar dynamo theories . the c iii emission line at 977 and the c iii multiplet at 1176 occur in the fuse wavelength region . the ratio of the c iii @xmath31176/@xmath3977 emission line fluxes has long been recognized to be a sensitive diagnostic for measuring the electron pressure ( @xmath17 ) of hot ( @xmath060,000 k ) , optically - thin plasmas ( e.g. , dupree , foukal , & jordan 1976 ; keenan & berrington 1985 ) . fuse offers an excellent opportunity to exploit the full potential of this important diagnostic ( @xmath42 ) because both c iii features can be measured simultaneously . furthermore , unlike the uv c iii ] @xmath31909 line , these fuv features are not contaminated by the stellar continuum or nearby line emissions . details of the c iii @xmath3977 and @xmath31176 emission features for two of the targets in the sample are shown in figure [ figcomp ] . to compute @xmath43 ratios we adopted a simple c iii model which includes the first 10 fine - structure levels . the energy levels are taken from moore ( 1970 ) , the highest level in the model being @xmath44 at @xmath45 . the electric dipole oscillator strengths are taken from the compilation of allard et al . ( 1990 ) , except for transitions calculated by tachiev & froese fischer ( 1999 ; 2002 , priv . comm . ) and fleming , hibbert , & stafford ( 1984 ) . the electric quadrupole and magnetic dipole oscillator strengths are from nussbaumer & storey ( 1978 ) and tachiev & froese fischer ( 1999 ) and the @xmath46 and @xmath47 magnetic quadrupole strengths are from shorer & lin ( 1977 ) and tachiev & froese fischer ( 1999 ) , respectively . the electron collision strengths are taken from berrington ( 1985 ) and berrington et al . the proton collision rates for transitions within @xmath48 are from ryans et al . ( 1998 ) and within @xmath49 are from doyle , kingston , & reid ( 1980 ) . we solved the statistical equilibrium equations , under the assumption that the plasma is in steady state , with no photo - excitation from external sources . the ratios presented here are computed under the assumption that all the c iii transitions are optically thin . we have also computed the ratios under optically thick conditions using escape probabilities with hydrogen columns of @xmath50 and assuming solar abundances ( grevesse & sauval 1998 ) . we find that @xmath43 is quite insensitive to optical depth effects under these conditions , in agreement with the findings of bhatia & kastner ( 1992 ) . the @xmath31176/@xmath3977 flux ratio of be - like c iii is well known to be sensitive to both electron density and temperature and is therefore sensitive to the _ shape of the emission measure distribution _ between @xmath51 and 4.9 . the density sensitivity arises from collisional thermalization of the lower levels of the 1176 multiplet ( @xmath48 ) which connect to the ground state ( @xmath52 ) by an intercombination electric dipole and a weak magnetic quadrupole transition at 1908.7 and 1906.7 , respectively . the contribution functions for 1176 and 977 contain the product of the collisional excitation which increases with @xmath53 , the ionization balance of c iii which peaks near @xmath54 ( arnaud & rothenflug 1985 ) , and the shape of the emission measure distribution between @xmath55 and 4.9 which is a strong declining function of @xmath53 , e.g. , jordan et al . the peak contribution is expected to lie below @xmath56 . for example for the sun , judge et al . ( 1995 ) find a peak as low as @xmath55 for the 977- transition , while macpherson & jordan ( 1999 ) find @xmath57 . for the solar analog transition regions , we consider that the c iii emission lines are formed at a constant electron pressure , and we derive the ratio @xmath43 as a function of electron pressure rather than at a ( less realistic ) single temperature . the emission measure distribution at c iii temperatures may change as a function of stellar magnetic activity . however , as pointed out by jordan ( 2000 ) , both the analyses of main sequence stars by jordan et al . ( 1987 ) and the similarity of the c iv / c ii uv emission line ratios for a wide range of active coronal stars ( oranje 1986 ) , indicate that the shape of the emission measure distribution in the c iii forming region approximately follows the shape of the inverse of total radiative loss curve between @xmath58 and 5.3 . if the density sensitivity of the ionization balance and abundance gradients can be neglected , then the relative electron pressures derived for our sample should be reliable . to quantify the effects of assuming differential emission measure distributions ( dem ) with different slopes , we adopt a form @xmath59 where @xmath60 and @xmath61 are the electron and hydrogen densities , respectively , and @xmath62 is the radial coordinate . here we assume that the shape of dem is similar to the emission measure over the c iii 977 formation region . note that @xmath63 is a reasonable approximation to the radiative power - loss function for this temperature range . for each pressure we integrate over temperature from @xmath64 to 5.3 , where the fractional abundance of c iii exceeds 0.01 , for each dem to derive the total fluxes and derive @xmath43 . in figure [ figpres ] we show the ratio @xmath43 for @xmath35=1.5 , 2.0 , and 2.5 . this figure shows that the resulting ratio is not too sensitive to uncertainties in the shape of the emission measure distribution . indeed , uncertainties in the collision strengths lead to greater uncertainties in the electron pressure . the empirical c iii ratios @xmath43 computed for the six stars in the sample are given in table [ tabr ] . also provided are the @xmath43 values corrected for ism absorption . from these one can estimate the electron pressures through the relationship discussed above . when doing so , we obtain @xmath17 values between @xmath65 and @xmath66 @xmath67 k , approximately . the actual electron pressures computed for each of the stars in the sample are given in table [ tabr ] . both the @xmath43 values and the derived plasma electron pressures are plotted in in figure [ figrobs ] as a function of the stellar rotation period . for comparison , the value for the quiet sun was determined to be @xmath43=0.29 ( dupree , foukal , & jordan 1976 ) , and the active sun ( measured in solar active regions such as sunspots ) has a c iii ratio of @xmath43=0.44 ( noyes et al . 1985 ; doyle et al . note that the @xmath43 values determined for our stars correspond to the integrated stellar disk average , thus including both active and inactive regions , weighted towards region of large areal filling factor and high @xmath60 . interestingly , the observed data for the target stars suggest a strong correlation between the activity level of the star (= age or @xmath10 ) and @xmath43 or @xmath17 . the elusive density of the tr has been probed for the first time along the evolutionary path of a solar - type star , from zams to tams . our results show that the electron pressure of the tr decreases monotonically as the star spins down ( and magnetic dynamo activity decreases ) . from the fit in figure [ figrobs ] we infer a power - law relationship with a slope of @xmath68 . this suggests that the electron pressure of the solar tr may have decreased by about a factor of @xmath040 since the beginning of its main sequence evolution . a likely explanation for the greater plasma electron pressure for the young and active ( rapidly rotating ) stars is a stronger magnetic confinement of the emitting material . most notably , the power - law slope for the electron pressure and the c iii 977 flux are very similar , namely @xmath69 . note , however , that the flux power - law relationship has been computed from the integrated flux and it thus represents an average over the stellar surface . if , however , the emitting tr plasma only occupies a fraction ( @xmath70 ; @xmath71 ) of the surface , known as filling factor , then the specific flux emitted by the active regions will be correspondingly larger . the pressure power - law does not depend on the area factor if only one atmospheric component dominates . thus , assuming that the same fraction of c iii 977 is emitted over the same @xmath72 interval for our sample then we can adopt @xmath73 where @xmath74 represents the surface integrated flux . using the power - law relationships derived above , one can write @xmath75 as a function of the rotation period . thus , with empirically measured filling factors for the solar proxies or with a reliable relationship between @xmath70 and @xmath10 ( or rossby number ) , which is proportional to the rotation period of the star and inversely proportional to the turnover time at the base of the convective zone , is commonly used as a measure of the activity level of a star . for our targets , which have very similar masses and spectral types , one can safely assume that the rossby number is simply proportional to @xmath10 . ] ) the relationship above can yield an estimate of the dependence of the electron temperature gradient with age . for example , using the empirical fit of montesinos & jordan ( 1993 ) that relates the magnetic filling factor with the rossby number , we can write for our solar proxies @xmath76 and thus , the temperature gradient can be expressed as @xmath77 a representation of the relative variations of the temperature gradient , the electron pressure and the specific c iii @xmath3977 flux ( i.e. , flux divided by filling factor ) are provided in figure [ figphys ] . note that these relationships are only represented for rotation periods shorter than that of the sun , where the filling factor is better defined . strikingly , the specific c iii @xmath3977 flux remains relatively constant with rotation period : as the filling factor increases , the total stellar flux increases correspondingly . however , since the electron density increases with decreasing rotation period , the temperature gradient must also increase to keep the specific flux relatively constant . it is worth noting that , given the large change in the implied temperature gradient across the sample , one or more of the several assumptions on which eq . [ eqffact ] is based may no longer hold . the relationships obtained from the solar proxies indicate that the total solar fuse / fuv flux was about twice the present value 2.5 gyr ago and about 4 times the present value about 3.5 gyr ago . note also that the fuse / fuv flux of the zero - age main sequence sun could have been stronger than today by as much as 50 times . as discussed by canuto et al . ( 1982 , 1983 ) , luhmann & bauer ( 1992 ) , ayres ( 1997 ) , guinan , ribas & harper ( 2002 ) and others , the strong fuv and uv emissions of the young , more active sun could have played a major role in the early development and evolution of planetary atmospheres especially those of the terrestrial planets . the expected strong x - ray uv irradiance of the young sun can strongly influence the photochemistry and photoionization ( and possible erosion ) of the early planetary atmospheres and even surfaces ( in the case of mercury , moon and mars ) and also may play a role in the origin and development of life on earth as well as possibly on mars . for example , canuto et al . ( 1982 , 1983 ) discuss the photochemistry of o@xmath78 , o@xmath79 , co@xmath78 , h@xmath78o , etc , in the presumed outgassed co@xmath78-rich early atmosphere of the earth . ayres ( 1997 ) discusses the effect of the young sun s increased ionizing x - ray uv flux , and possible accompanying enhanced solar wind , on the erosion of the early atmosphere of mars about 34 gyr ago . also , lammer et al . ( 2003 ) utilize irradiance data from the `` sun in time '' project and solar - wind estimates by wood et al . ( 2002a ) to evaluate the mechanisms for loss of water from mars and study the implications for the oxidation of the martian soil . similarly , our data can also provide insights into the so - called faint sun paradox . the paradox arises from the fact that standard stellar evolutionary models show that the zero - age main sequence sun had a luminosity of @xmath070% of the present sun . this should have led to a much cooler earth in the past while geological and fossil evidence indicate otherwise . a solution to the faint sun paradox proposed by sagan & mullen ( 1972 ) was an increase of the greenhouse effect for the early earth . the gases that have been suggested to account for this enhanced greenhouse effect are carbon dioxide ( co@xmath78 ) , ammonia ( nh@xmath79 ) or methane ( ch@xmath80 ) . however , recent results for atmospheric composition of the early earth ( rye , kuo , & holland 1995 ) are in conflict with the high levels of co@xmath78 and h@xmath78o needed to explain the stronger greenhouse effect . moreover , ammonia is a likely candidate except that it is quickly and irreversibly photodissociated . however , sagan & chyba ( 1997 ) proposed that hydrocarbon - based aerosols could shield the ammonia from damaging incoming solar radiation . another alternative explanation for the enhanced greenhouse effect is from atmospheric methane as discussed by pavlov et al . methane is a very strong greenhouse gas ( 10@xmath8110@xmath40 times stronger than co@xmath78 ) and the necessary amount of ch@xmath80 could have been provided by methanogenic bacteria in the proposed early earth s reducing atmosphere . in addition to the stronger greenhouse effect , the expected higher x - ray uv irradiance of the young sun should heat up the upper atmosphere of the early earth . although the stronger high - energy solar radiation can not by itself explain the faint sun paradox , the photoionization and photodissociation reactions triggered could play a major role in what greenhouse gases are available . for example , the high levels of fuv - uv radiation of the young sun could strongly influence the abundances of ammonia and methane in the pre - biotic and archean planetary atmosphere some 2 - 4 gyr ago . similarly , the photochemistry and abundance of ozone ( o@xmath79 ) is of great importance to study life genesis in the earth . ozone is an efficient screening mechanism for the enhanced uv radiation of the young sun , thus protecting the emerging life on the earth s surface . to fully evaluate the influence of fuv radiation in the developing planetary system one must first estimate the flux contribution of the strong h i ly@xmath35 feature , which is not included in the fuse wavelength range . preliminary estimates using hst / stis spectra indicate that ly@xmath35 @xmath31216 contributes a significant fraction ( up to 90% ) of the total fuv flux . further observations and measurements in the near future will allow us to obtain this missing piece of information and complete the fuv irradiance study . in this study we have utilized fuv spectra acquired with the fuse satellite to investigate the fuv emission characteristics of solar analogs . our six targets are well - known g05 solar - type stars especially selected to serve as proxies for the sun at different ages , nearly covering the entire solar main sequence lifetime from 130 myr to @xmath09 gyr . here we have focused on two aspects of the `` sun in time '' program which are the study of tr plasma electron pressures ( using c iii @xmath31176/@xmath3977 line ratio diagnostics ) for stars that differ only in rotation period ( and age ) and the evolution of irradiances for specific features covering emitting plasma with temperatures from @xmath1 k to @xmath7 k. to analyze the plasma density of the tr we have used a theoretical relationship between the ratio of c iii @xmath31176 and @xmath3977 emission line fluxes and the electron pressure ( @xmath17 ) of the material responsible for the emission . fortunately , both c iii transitions are within the fuse wavelength range so that they can be measured simultaneously . our results indicate a power - law relationship between the electron pressure and the stellar rotation period and overall magnetic - related activity ( both related to age ) . the slope of this relationship has been found to be @xmath0@xmath821.7 , which suggests that the electron pressure of @xmath010@xmath4-k material in the sun has decreased significantly since the beginning of its main sequence evolution . the higher values of electron pressure found for the more active stars are best explained by stronger magnetic confinment of the tr plasma . the measured fluxes for four emission features c ii @xmath31037 , c iii @xmath3977 , o vi @xmath31032 , and fe xviii @xmath3975 were referred to a radius of 1 @xmath83 and also corrected for ism absorption whenever necessary . the typical formation temperatures of the studied features are @xmath37 k , @xmath38 k , @xmath39 k , and 6 mk , respectively . our analysis indicates that the evolution of these fluxes with stellar rotation period or age can be accurately fit with power - law relationships of different slopes . interestingly , the slopes not only change from line to line , but also they show a clear trend : the power law becomes steeper as we move from cooler to hotter plasmas , with the most extreme trend being that of the coronal ion fe xviii . also , the evolution total integrated flux in the fuse wavelength range ( 9201180 ) can be fit with a power - law relationship with a slope of @xmath0@xmath821.7 , which indicates a factor of @xmath050 flux decrease along the solar main sequence evolution as the magnetic dynamo activity decreases . the high levels of fuv line emission fluxes ( and related high - energy emission ) of the early sun could have played a crucial role in the photochemistry and photoionization of terrestrial and planetary atmospheres . to address this point we are completing spectral irradiance tables covering 1 to 3200 for our program stars that can serve as input data for evolution and structure models of the paleo - atmospheres of the solar system planets . the fuse observations fill a critical wavelength and energy gap in the `` sun in time '' program and complement observations of the same stars in the x - ray and euv regions ( corona ) made with rosat , sax , asca , xmm , chandra , and euve , and in the uv ( tr and chromosphere ) made using iue and hst . the only missing piece of information is the evolution of the irradiance of the strong chromospheric h i ly@xmath35 @xmath31216 fuv feature . we are currently carrying out this part of the study and we expect to complete it shortly . full spectral irradiance tables for five solar proxies are thus nearing completion and will be made available in a forthcoming publication ( ribas & guinan 2003 , in preparation ) . we anticipate that these irradiance results will be important for the study of paleo - atmospheres of the solar system planets . in particular , preliminary analyses indicate that the high x - ray and euv emission fluxes of the early sun could have produced significant heating of the planetary exospheres and upper - atmospheres thus enhancing processes such as thermal escape . the early sun s strong fuv and uv fluxes penetrate further into the atmosphere and probably influenced the photochemistry of , e.g. , methane and ammonia , which are important greenhouse gases . thus , this study has strong implications for the evolution of the pre - biotic and archean atmosphere of the earth as well as for the early development of life on earth and possibly on mars . we thank seth redfield for making available to us the data from the colorado lism model , which we have used to correct the measured fluxes for ism absorption . the referee , tom ake ( johns hopkins univ . ) , is thanked for helpful comments and suggestions that led to significant improvements . we acknowledge with gratitude the support for the `` sun in time '' program from nasa - fuse grants nag 5 - 08985 , nag 5 - 10387 , nag 5 - 12125 and also from nsf - rui grant ast-00 - 71260 . g.m.h.s research was funded by nasa grant nag5 - 4808 ( ltsa ) . this research has made use of the simbad database , operated at cds , strasbourg , france . ake , t. b. , dupree , a. k. , young , p. r. , linsky , j. l. , malina , r. f. , griffiths , n. w. , siegmund , o. h. w. , & woodgate , b. e. 2000 , , 538 , l87 allard , n. , artru , m .- c . , lanz , t. , & le dourneuf , m. 1990 , , 84 , 563 arnaud , m. , & rothenflug , r. 1985 , , 60 , 425 ayres , t. r. 1997 , jgr , 102 , 1641 berrington , k. a. 1985 , j. phys . phys . , 18 , l395 berrington , k. a. , burke , p. g. , dufton , p. l. , & kingston , a. e. 1985 , atomic nucl . data tables , 33 , 195 bhatia , a. k. , & kastner , s. o. 1992 , , 79 , 139 canuto , v. m. , levine , j. s. , augustsson , t. r. , & imhoff , c. l. 1982 , nature , 296 , 816 canuto , v. m. , levine , j. s. , augustsson , t. r. , imhoff , c. l. , & giampapa , m. s. 1983 , nature , 305 , 281 dorren , j. d. , & guinan , e. f. 1994 , in iau coll . 143 , the sun as a variable star , eds . j. m. pap , c. frlich , h. s. hudson , & s. k. solanki ( cambridge : cup ) , 206 doyle , j. g. , kingston , a. e. , & reid , r. h. g. 1980 , , 90 , 97 doyle , j. g. , raymond , j. c. , noyes , r. w. , & kingston , a. e. 1985 , , 297 , 816 dupree , a. k. , foukal , p. v. , & jordan , c. 1976 , , 209 , 621 feldman , u. , & doschek , g. a. 1991 , , 75 , 925 fleming , j. , hibbert , a. , & stafford , r. p. 1984 , physica scripta , 49 , 316 gayley , k. g. 2002 , , 565 , 545 grevesse , n. , & sauval , a. j. 1998 , space science reviews , 85 , 161 gdel , m. , guinan , e. f. , & skinner , s. l. 1997 , , 483 , 947 guinan , e. f. , ribas , i. , & harper , g. m. 2002 , in asp conf . 264 , continuing the challenge of euv astronomy : current analysis and prospects for the future , eds . s. b. howell , j. dupuis , d. golombek , f. m. walter , & j. cullison ( san francisco : asp ) , 129 heroux , l. , & hinteregger , h. e. 1978 , jgr , 83 , 5305 jordan , c. 2000 , plasma phys . control fusion , 42 , 415 jordan , c. , ayres , t. r. , brown , a. , linsky , j. l. , & simon , t. 1987 , , 225 , 903 judge , p. g. , woods , t. n. , brekke , p. , & rottman , g. j. 1995 , , 455 , l85 keenan , f. p. , & berrington , k. a. 1985 , solar phys . , 99 , 25 lammer , h. , lichtenegger , h. , kolb , c. , ribas , i. , guinan , e. f. , & bauer , s. j. 2003 , icarus , in press linsky , j. l. , & wood , b. e. 1994 , , 430 , 342 luhmann , j. g. , & bauer , s. j. 1992 , in venus and mars : atmospheres , ionospheres , and solar wind interactions , proceedings of the chapman conference ( washington , dc : american geophysical union ) , 417 macpherson , k. p. , & jordan , c. 1999 , , 308 , 510 montesinos , b. , & jordan , c. 1993 , , 264 , 900 moore , c. e. 1970 , nsrds - nbs 3 nbs , washington , d.c . moos , h. w. , cash , w. c. , cowie , l. l. , et al . 2000 , , 538 , l1 nussbaumer , h. & storey , p. j. 1978 , , 64 , 139 noyes , r. w. , raymond , j. c. , doyle , j. g. , & kingston , a. e. 1985 , , 297 , 805 oranje , b. j. 1986 , , 154 , 185 pavlov , a. a. , kasting , j. f. , brown , l. l. , rages , k. a. , & freedman , r. 2000 , jgr , 105 , 11981 redfield , s. , & linsky , j. l. 2000 , , 534 , 825 redfield , s. , & linsky , j. l. 2002 , , 139 , 439 redfield , s. , ayres , t. r. , linsky , j. l. , ake , t. b. , dupree , a. k. , robinson , r. d. , & young , p. r. 2003 , , 585 , 993 redfield , s. , linsky , j. l. , ake , t. b. , ayres , t. r. , dupree , a. k. , robinson , r. d. , wood , b. e. , & young , p. r. 2002 , , 581 , 626 ryans , r. s. i. , foster - woods , v. j. , reid , r. h. g. , keenan , f. p. , & copeland , f. 1990 , , 336 , 393 rye , r. , kuo , p. h. , & holland , h. d. 1995 , nature , 378 , 603 sagan , c. , & mullen , g. 1972 , science , 177 , 52 sagan , c. , & chyba , c. 1997 , science , 276 , 1217 sahnow , d. j. , moos , h. w. , ake , t. b. , et al . 2000 , , 538 , l7 shorer , p. , & lin , c. d. 1977 , phys . a. , 16 , 2068 simon , t. , boesgaard , a. m. , & herbig , g. 1985 , , 293 , 551 skumanich , a. 1972 , , 171 , 565 spitzer , l. 1978 , physical processes in the interstellar medium ( new york : john wiley ) tachiev , g. , & froese fischer , c. 1999 , j. phys . b. , 32 , 5805 wood , b. e. , linsky , j. l. , & ayres , t. r. 1997 , , 478 , 745 wood , b. e. , mller , h .- r . , zank , g. , & linsky , j. l. 2002a , , 574 , 412 wood , b. e. , redfield , s. , linsky , j. l. , & sahu , m. s. 2002b , , 581 , 1168 young , p. r. , dupree , a. k. , wood , b. e. , redfield , s. , linsky , j. l. , ake , t. b. , & moos , h. w. 2001 , , 555 , l121 zhang , h. l. , sampson , d. h. , & fontes , c. j. 1990 , atomic data nucl . data tables , 44 , 31 lrlrcccccl ek dra & 129333 & g0 v & 33.9 & 5818&1.07&0.95&2.75 & 0.13 & pleiades str . + @xmath15 uma & 72905 & g1.5 v&14.3 & 5840&0.98&0.97&4.68 & 0.3 & uma str . + @xmath16 cet & 20630 & g5 v & 9.2 & 5700&1.01&0.94&9.2 & 0.65 & @xmath10-age rel . + @xmath11 com & 114710 & g0 v & 9.2 & 5950&1.11&1.08&12.4 & 1.6 & @xmath10-age rel . + sun & & g2 v & 1 au&5777&1.00&1.00&25.4 & 4.6 & isotopic dating + @xmath11 hyi & 2151 & g2 iv & 7.5 & 5800&1.09&1.88&@xmath028&6.7 & isochrones + 16 cyg a & 186408 & g2 v & 21.6 & 5740&0.99&1.27&@xmath035&@xmath09 & isochrones + lclccrrrrr ek dra & c1020501 & 2002 may 14 & 24.2&11.1@xmath840.7 & 7.6@xmath840.4 & 7.3@xmath840.3 & 3.4@xmath840.2 & 0.47@xmath840.07 & 0.81@xmath840.19 + @xmath15 uma & b0780101 & 2001 dec 5 & 16.5&14.7@xmath840.7 & 9.3@xmath840.5 & 9.6@xmath840.3 & 4.8@xmath840.2 & 0.74@xmath840.09 & 0.65@xmath840.26 + @xmath16 cet & a0830301 & 2000 sep 10 & 13.0&16.4@xmath840.7 & 10.7@xmath840.4 & 12.4@xmath840.4 & 6.2@xmath840.2 & 1.07@xmath840.12 & 0.44@xmath840.14 + @xmath11 com & a0830401 & 2001 jan 26 & 15.0&11.0@xmath840.5 & 5.5@xmath840.2 & 5.9@xmath840.2 & 2.8@xmath840.1 & 0.63@xmath840.08&@xmath850.06 + @xmath11 hyi & a0830101 & 2000 jul 1 & 18.0&15.1@xmath840.5 & 5.8@xmath840.2 & 6.1@xmath840.2 & 2.8@xmath840.1 & 1.02@xmath840.09&@xmath850.04 + 16 cyg a & c1020101 & 2002 jul 30 & 35.0&0.66@xmath840.14&0.20@xmath840.09&0.43@xmath840.05&0.20@xmath840.03&@xmath850.05&@xmath850.04 + lrrrrrrrrr ek dra & @xmath8230.5&@xmath8223 & 8 & @xmath8215 & 16 & @xmath8225 & 6 & @xmath8226 & 5 + @xmath15 uma & @xmath8212.0&@xmath8223 & @xmath8211 & @xmath8224 & @xmath8212 & @xmath8219 & @xmath827 & @xmath8251 & @xmath8239 + @xmath16 cet & 19.9 & 12 & @xmath828 & 16 & @xmath824 & 16 & @xmath824 & 1 & @xmath8219 + @xmath11 com & 6.1 & 3 & @xmath823 & 0 & @xmath826 & 4 & @xmath822 & & + @xmath11 hyi & 22.7 & 6 & @xmath8217 & 20 & @xmath823 & 16 & @xmath827 & & + 16 cyg a & @xmath8225.6&@xmath8240&@xmath8214 & @xmath8229 & @xmath823 & @xmath8222 & 4 & & + lllllr ek dra & 1.17@xmath840.18 & 28.4@xmath841.7 & 18.2@xmath840.8 & 2.02@xmath840.51 & + @xmath15 uma & 0.32@xmath840.04 & 6.52@xmath840.33 & 4.10@xmath840.12 & 0.28@xmath840.11 & 45@xmath8420 + @xmath16 cet&0.20@xmath840.02 & 3.30@xmath840.15 & 2.32@xmath840.07 & 0.082@xmath840.025&16.4@xmath843.8 + @xmath11 com & 0.090@xmath840.011&1.64@xmath840.07 & 0.842@xmath840.030&@xmath850.0085 & + @xmath11 hyi & 0.032@xmath840.003&0.543@xmath840.019&0.190@xmath840.007&@xmath850.0012 & + 16 cyg a & @xmath850.03 & 0.423@xmath840.090&0.246@xmath840.028&@xmath850.02 & + lcll fuv ( 9201180 ) & @xmath86&@xmath821.74&@xmath821.03 + c ii & @xmath87 & @xmath821.49&@xmath820.88 + c iii & @xmath88 & @xmath821.61&@xmath820.95 + o vi & @xmath89 & @xmath821.82&@xmath821.07 + x - rays ( 0.12.4 kev)&@xmath90 & @xmath822.64&@xmath91 + fe xviii & @xmath92 & @xmath823.2 & @xmath821.9 + lccl ek dra & 0.68@xmath840.05&0.67@xmath840.05&@xmath9316.0 + @xmath15 uma & 0.63@xmath840.05&0.61@xmath840.04&@xmath94 + @xmath16 cet&0.65@xmath840.04&0.60@xmath840.04&@xmath95 + @xmath11 com & 0.50@xmath840.03&0.48@xmath840.03&@xmath96 + @xmath11 hyi & 0.38@xmath840.02&0.33@xmath840.02&@xmath97 + 16 cyg a & 0.30@xmath840.15&0.27@xmath840.13&@xmath98 +

we deal with the elastic scattering by a large number of rigid bodies of arbitrary shapes with maximum radius with constant lam coefficients and . we show that , when these rigid bodies are distributed arbitrarily ( not necessarily periodically ) in a bounded region of where their number is and the minimum distance between them is with in some appropriate range , as , the generated far - field patterns approximate the far - field patterns generated by an equivalent ( possibly variable ) mass density . this mass density is described by two coefficients : one modeling the local distribution of the small bodies and the other one by their geometries . in particular , if the distributed bodies have a uniform spherical shape then the equivalent mass density is isotropic while for general shapes it might be anisotropic . in addition , we can distribute the small bodies in such a way that the equivalent mass density is negative . finally , if the background density is variable in and in , then if we remove from appropriately distributed small bodies then the equivalent density will be equal to unity in , i.e. the obstacle characterized by is approximately cloaked . * keywords * : elastic wave scattering , small - scatterers , effective mass density .

1504.06947

let @xmath13 be @xmath0 open , bounded and simply connected sets in @xmath5 with lipschitz boundaries , containing the origin . we assume that their sizes and lipschitz constants are uniformly bounded . we set @xmath14 to be the small bodies characterized by the parameter @xmath15 and the locations @xmath16 , @xmath17 . assume that the lam coefficients @xmath2 and @xmath3 are constants satisfying @xmath18 and the mass density @xmath10 to be a constant that we normalize to a unity . let @xmath19 be a solution of the navier equation @xmath20 , @xmath21 . we denote by @xmath22 the elastic field scattered by the @xmath0 small bodies @xmath23 due to the incident field @xmath19 . we restrict ourselves to the scattering by rigid bodies . hence the total field @xmath24 satisfies the following exterior dirichlet problem of the elastic waves @xmath25 @xmath26 with the kupradze radiation conditions ( k.r.c ) @xmath27 where the two limits are uniform in all the directions @xmath28 . also , we denote @xmath29 to be the longitudinal ( or the pressure or p ) part of the field @xmath22 and @xmath30 to be the transversal ( or the shear or s ) part of the field @xmath22 corresponding to the helmholtz decomposition @xmath31 . the constants @xmath32 and @xmath33 are known as the longitudinal and transversal wavenumbers , @xmath34 and @xmath35 are the corresponding phase velocities , respectively and @xmath36 is the frequency . the scattering problem ( [ elaimpoenetrable]-[radiationcela ] ) is well posed in the hlder or sobolev spaces , see @xcite for instance , and the scattered field @xmath37 has the following asymptotic expansion : @xmath38 uniformly in all directions @xmath39 . the longitudinal part of the far - field , i.e. @xmath40 is normal to @xmath41 while the transversal part @xmath42 is tangential to @xmath41 . we set @xmath43 . as usual , we use plane incident waves of the form @xmath44 , where @xmath45 is any direction in @xmath41 perpendicular to the incident direction @xmath46 , @xmath47 are arbitrary constants . the functions @xmath48 and @xmath49 for @xmath50 are called the p - part and the s - part of the far - field pattern respectively . [ def1 ] we define 1 . @xmath51,$ ] 2 . @xmath52 @xmath53 . we assume that @xmath54 and @xmath55 is given . 3 . @xmath56 as the upper bound of the used frequencies , i.e. @xmath57 $ ] . 4 . @xmath4 to be a bounded domain in @xmath5 containing the small bodies @xmath58 . we assume that @xmath59 , with the same diameter @xmath60 , are non - flat lipschitz obstacles , i.e. @xmath61 s are lipschitz obstacles and there exist constants @xmath62 $ ] such that @xmath63 where @xmath64 are assumed to be uniformly bounded from below by a positive constant . in @xcite , we have shown that there exist two positive constants @xmath65 and @xmath66 depending only on the size of @xmath4 , the lipschitz character of @xmath67 , @xmath55 and @xmath56 such that if @xmath68 then we have the following asymptotic expansion for the p - part , @xmath69 , and the s - part , @xmath70 , of the far - field pattern : @xmath71+m(m-1)\left[\frac{a^3}{d^{2\alpha}}+\frac{a^4}{d^{4-\alpha}}+ \frac{a^4}{d^{5 - 2\alpha}}\right]+m(m-1)^2\frac{a^4}{d^{3\alpha}}\right ) \right ] , \label{x oustdie1 d_m farmainp - near}\\ & \hspace{-1cm}u^\infty_s(\hat{x},\theta)= & \frac{1}{4\pi\,c_s^{2}}(\rm \textbf{i}- \hat{x}\otimes\hat{x})\left[\sum_{m=1}^{m}e^{-i\frac{\omega}{c_s}\hat{x}\cdot\,z_m}q_m\right.\nonumber\\ & & \hspace{-2cm}\left.+o\left(m\left[a^2+\frac{a^3}{d^{5 - 3\alpha}}+\frac { a^4}{d^{9 - 6\alpha}}\right]+m(m-1)\left[\frac{a^3}{d^{2\alpha}}+\frac{a^4}{d^{4-\alpha}}+ \frac{a^4}{d^{5 - 2\alpha}}\right]+m(m-1)^2\frac{a^4}{d^{3\alpha}}\right ) \right ] , \label{x oustdie1 d_m farmains - near } \end{aligned}\ ] ] where @xmath72 , @xmath73 , is a parameter describing the relative distribution of the small bodies . the vector coefficients @xmath74 , @xmath75 are the solutions of the following linear algebraic system @xmath76 for @xmath77 with @xmath78 denoting the kupradze matrix of the fundamental solution to the navier equation with frequency @xmath36 , @xmath79 and @xmath80 is the solution matrix of the integral equation of the first kind @xmath81 with @xmath82 the identity matrix of order 3 . consider now the special case @xmath83 and @xmath84 with @xmath85 , @xmath86 , @xmath87 and @xmath88 are positive . then the asymptotic expansions ( [ x oustdie1 d_m farmainp - near]-[x oustdie1 d_m farmains - near ] ) can be rewritten as @xmath89,\quad \label{x oustdie1 d_m farmainp - near*}\\ u^\infty_s(\hat{x},\theta)&= & \frac{1}{4\pi\,c_s^{2}}(\rm \textbf{i}- \hat{x}\otimes\hat{x})\big[\sum_{m=1}^{m}e^{-i\frac{\omega}{c_s}\hat{x}\cdot\,z_m}q_m \nonumber \\ & & + o\left(a^{2-s}+a^{3-s-5t+3t\alpha}+a^{4-s-9t+6t\alpha}+a^{3 - 2s-2t\alpha}+a^{4 - 3s-3t\alpha}+a^{4 - 2s-5t+2t\alpha}\right ) \big].\quad \label{x oustdie1 d_m farmains - near * } \end{aligned}\ ] ] as the diameter @xmath60 tends to zero the error term tends to zero for @xmath8 and @xmath90 such that @xmath91 in @xcite , we have shown that @xmath92 , then we have the upper bound @xmath93 hence if the number of obstacles is @xmath94 and @xmath8 satisfies ( [ general - condition - s - t ] ) , @xmath95 , then from ( [ x oustdie1 d_m farmainp - near * ] , [ x oustdie1 d_m farmains - near * ] ) , we deduce that @xmath96 this means that this collection of obstacles has no effect on the homogeneous medium as @xmath9 . let us consider the case when @xmath97 . we set @xmath4 to be a bounded domain , say of unit volume , containing the obstacles @xmath98 . given a positive and continuous function @xmath99 , we divide @xmath4 into @xmath100 $ ] subdomains @xmath101 $ ] , each of volume @xmath102}{k(z_m)+1}$ ] , with @xmath103 as its center and contains @xmath104 $ ] obstacles , see fig [ distribution - obstacles ] . we set @xmath105 , hence @xmath106}_{j=1}[k(z_m)+1]\leq k_{max}[a^{-1}]=o(a^{-1})$ ] . [ distribution - obstacles ] [ equivalent - medimu ] let the small obstacles be distributed in a bounded domain @xmath4 , say of unit volume , with their number @xmath6 and their minimum distance @xmath107 , @xmath108 , as @xmath95 , as described above . 1 . if the obstacles are distributed arbitrarily in @xmath4 , i.e. with different capacitances , then there exists a potential @xmath109 with support in @xmath4 such that @xmath110 where @xmath111 is the farfield corresponding to the scattering problem + @xmath112 + @xmath113 + with the radiation conditions @xmath114 2 . if in addition @xmath115 is in @xmath116 , @xmath117 $ ] and the obstacles have the same capacitances @xmath118 , then @xmath119 where @xmath120 in @xmath4 and @xmath121 in @xmath122 . assume that the lam coefficients @xmath2 and @xmath3 are constants satisfying @xmath18 and the mass density @xmath10 to be a measurable and bounded function which is equal to a constant that we normalize to a unity outside of a bounded domain @xmath4 . we set @xmath123 to be the upper bound of @xmath10 . in this case , the total field @xmath124 satisfies the following exterior dirichlet problem of the elastic waves @xmath125 @xmath126 with the kupradze radiation conditions ( k.r.c ) @xmath127 where the two limits are uniform in all the directions @xmath28 and @xmath128 and @xmath129 are respectively the p - part and s - part of the scattered field @xmath130 the scattering problem ( [ elaimpoenetrable - rho]-[radiationcela - rho ] ) is well posed in the hlder or sobolev spaces , see @xcite for instance , and the scattered field @xmath37 has the following asymptotic expansion : @xmath131 uniformly in all directions @xmath39 . the longitudinal part of the far - field , i.e. @xmath132 is normal to @xmath41 while the transversal part @xmath42 is tangential to @xmath41 . we set @xmath133 . as in the case of constant background mass density , there exist two positive constants @xmath65 and @xmath66 depending only on the size of @xmath4 , the lipschitz character of @xmath67 , @xmath55 , @xmath56 , @xmath123 and @xmath123 such that if @xmath134 then we have the following asymptotic expansion for the p - part , @xmath135 , and the s - part , @xmath136 , of the far - field pattern : @xmath137+m(m-1)\left[\frac{a^3}{d^{2\alpha}}+\frac{a^4}{d^{4-\alpha}}+ \frac{a^4}{d^{5 - 2\alpha}}\right]+m(m-1)^2\frac{a^4}{d^{3\alpha}}\right ) \right ] , \label{x oustdie1 d_m farmainp - near - rho}\\ & \hspace{-1cm}u^\infty_{\rho , s}(\hat{x},\theta)= & v^\infty_{\rho , s}(\hat{x},\theta)\;+~ \left[\sum_{m=1}^{m}g_{\rho , s}^{\infty}(\hat{x } , z_m ) q_{\rho , m}\right.\nonumber\\ & & \hspace{-2cm}\left.+o\left(m\left[a^2+\frac{a^3}{d^{5 - 3\alpha}}+\frac { a^4}{d^{9 - 6\alpha}}\right]+m(m-1)\left[\frac{a^3}{d^{2\alpha}}+\frac{a^4}{d^{4-\alpha}}+ \frac{a^4}{d^{5 - 2\alpha}}\right]+m(m-1)^2\frac{a^4}{d^{3\alpha}}\right ) \right ] , \label{x oustdie1 d_m farmains - near - rho } \end{aligned}\ ] ] where @xmath138 and @xmath139 are the p - part and s - part of the farfields of the green s function @xmath140 , of the operator @xmath141 in the whole space @xmath5 , evaluated in the direction @xmath142 and the source point @xmath143 . the vector coefficients @xmath144 , @xmath75 are the solutions of the following linear algebraic system @xmath145 for @xmath77 with @xmath146 is the total field satisfying @xmath147 and the scattered field @xmath148 the kupradze radiation conditions ( k.r.c ) . [ equivalent - medimu - rho ] let the small obstacles be distributed in a bounded domain @xmath4 , say of unit volume , with their number @xmath6 and their minimum distance @xmath107 , @xmath108 , as @xmath95 , as described above . 1 . if the obstacles are distributed arbitrarily in @xmath4 , i.e. with different capacitances , then there exists a potential @xmath109 with support in @xmath4 such that @xmath149 where @xmath150 is the farfield corresponding to the scattering problem + @xmath151 + @xmath152 + with the radiation conditions . 2 . if in addition @xmath115 is in @xmath116 , @xmath117 $ ] and the obstacles have the same capacitances , then @xmath153 where @xmath120 in @xmath4 and @xmath121 in @xmath122 . the main contribution of this work is to have shown that by removing from a bounded region of an elastic background , modeled by constant lam coefficients @xmath2 and @xmath3 and a possibly variable density @xmath10 , a number @xmath154 small and rigid bodies of radius @xmath60 distant from each other of at least @xmath155 , @xmath156 , then the perforated medium behaves , as @xmath9 , as a new elastic medium modeled by the same lam coefficients @xmath2 and @xmath3 but with a mass density - like @xmath157 . the coefficient @xmath158 models the local distribution ( or the local number ) of the bodies while the coefficient @xmath159 , coming from the capacitance of the bodies , describes the geometry of the small bodies as well as their elastic directional diffusion properties ( i.e. the anisotropy character ) . in addition , we provide explicit error estimates between the far - fields corresponding to the perforated medium and the equivalent one . from this result we can make the following conclusions : 1 . assume that the removed bodies have spherical shapes . for these shapes the corresponding elastic capacitance @xmath118 is of the form @xmath160 ( i.e. a scalar multiplied by the identity matrix ) . in section [ structure - capacitance ] , we describe a more general set of shapes satisfying this property . hence the equivalent mass density @xmath161 is isotropic while for general shapes it might be anisotropic . to achieve anisotropic densities , a possible choice of the shapes might be an ellipse . 2 . if we choose the local number of bodies @xmath158 large enough or the shapes of the reference bodies , @xmath162 , @xmath163 , having a large capacitance ( i.e. a relative large radius ) so that @xmath164 , then we design elastic materials having negative mass densities . 3 . assume that the background medium is modeled by variable mass density @xmath165 in @xmath4 . if we remove small bodies from @xmath4 with appropriate @xmath158 and/ or capacitance @xmath159 so that @xmath166 , then the new elastic material will behave every where in @xmath5 as the background medium . hence the new material will not scatter the sent incident waves , i.e. the region @xmath4 modeled by @xmath10 will be cloaked . the equivalent behavior between a collection of , appropriately dense , small holes and an extended penetrable obstacle modeled by an additive potential was already observed by cioranescu and murat @xcite and also the references therein , where the coefficient @xmath158 is reduced to zero since locally they have only one hole . their analysis is based on the homogenization theory for which they assume that the obstacles are distributed periodically , see also @xcite and @xcite . in the results presented here , we do not need such periodicity and no homogenization is used . instead , the analysis is based on the invertibility properties of the algebraic system ( [ fracqcfracmain ] ) and the precise treatment of the summation in the dominant terms of ( [ x oustdie1 d_m farmainp - near])-([x oustdie1 d_m farmains - near ] ) . this analysis was already tested for the acoustic model in @xcite . compared to @xcite , here , in addition to the difficulties coming from the vector character of the lam system , we improved the order of the error estimate , i.e. @xmath167 instead of @xmath168 which , for @xmath169 for instance , reduce to @xmath170 and @xmath171 respectively . let us finally mention that a result similar to ( [ b - rho ] ) , for the acoustic model , is also derived by ramm in several of his papers , see for instance @xcite , but without error estimates . compared to his results , and as we said earlier in addition to the vector character of lame model , we provide the approximation by improved explicit error estimates without any other assumptions while , as shown in @xcite for instance , in addition to some formal arguments , he needs extra assumptions on the distribution of the obstacles . the rest of the paper is organized as follows . in section 2 , we give the detailed proof of theorem [ equivalent - medimu ] . in section 3 , we describe the one of corollary [ equivalent - medimu - rho ] by discussing the main changes one needs to make in the proof of theorem [ equivalent - medimu ] . finally , in section 4 , we discuss some invariant properties of the elastic capacitance to characterize the shapes that have a scalar capacitance . the kupradze matrix @xmath172 of the fundamental solution to the navier equation is given by @xmath173,\end{aligned}\ ] ] where @xmath174 denotes the free space fundamental solution of the helmholtz equation @xmath175 in @xmath5 . the asymptotic behavior of kupradze tensor at infinity is given as follows @xmath176 with @xmath177 and @xmath82 being the identity matrix in @xmath178 , see @xcite for instance . as mentioned in @xcite , can also be represented as @xmath179 from which we can get the gradient @xmath180.\end{aligned}\ ] ] using the formulas and we can have the following estimates , for @xmath181 , see @xcite ; @xmath182,\qquad&&%\\ \left|\nabla_y\gamma^\omega(x , y)\right| \leq \frac{1}{4\pi}\left[\frac{c_9}{\vert\;x - y\vert^2}+c_{10}\right],\label{gradkupradzeten1dfnt}\end{aligned}\ ] ] with @xmath183,\qquad c_9:=3\left(\frac{1}{c_s^2}+\frac{1}{c_p^2}\right),\ ] ] @xmath184 @xmath185 and @xmath186 $ ] where we assume the wave number @xmath36 and the lam parameters @xmath2 and @xmath3 to satisfy the condition @xmath187 . the estimates ( [ gradkupradzeten1dfnt ] ) can be written as @xmath188 for different points @xmath189 , where @xmath190 the following observation will be useful for the proof of theorem [ equivalent - medimu ] . for @xmath17 fixed , we distinguish between the obstacles @xmath191 , @xmath192 by keeping them into different layers based on their distance from @xmath61 . let us first assume that @xmath193 for every @xmath143 . hence each @xmath194 has the ( same ) volume @xmath60 and contains only one obstacle @xmath61 . without loss of generality , we can take the @xmath194 s as cubes . hence we can suppose that these cubes are arranged in cuboids , for example unit rubics cube , in different layers such that the total cubes upto the @xmath195 layer consists @xmath196 cubes for @xmath197 $ ] , and @xmath194 is located on the center , see fig [ distribution - obstacles ] . hence the number of obstacles located in the @xmath195 , @xmath198 layer will be @xmath199=24 n^2 + 2 $ ] and their distance from @xmath61 is more than @xmath200 . observe that , @xmath201 . now , we come back to the case where @xmath202 . first observe that @xmath203}{k(z_m)+1}\leq 1 $ ] . hence with such @xmath194 s , the total cubes located in the @xmath204th layer @xmath204 consists of at most the double of @xmath199 $ ] , i.e. @xmath205 . we start with the following lemma on the uniform bounds of the elastic capacitances , see ( * ? ? ? * lemma 3.1 ) and the references there in . [ capacitance - eig - single ] let @xmath206 and @xmath207 be the minimal and maximal eigenvalues of the elastic capacitance matrices @xmath208 , for @xmath209 . denote by @xmath210 the capacitance of each scatterer in the acoustic case , , @xmath211 and @xmath80 is the solution of the integral equation of the first kind @xmath212 , see @xcite . ] then we have the following estimate ; @xmath213 for @xmath214 . the constant @xmath215 is the acoustic capacitance of the reference body @xmath162 which can be estimated above and below by the lipschitz character of @xmath162 , @xcite . the following lemma provides us with the needed estimate on the invertibility of the algebraic system whose coefficient matrix @xmath216 is given by ; @xmath217 [ mazyawrkthmela]the matrix @xmath216 is invertible and the solution vector @xmath218 of satisfies the estimate : @xmath219 if @xmath220 . in addition , @xmath221 since @xmath222 , then @xmath223 makes sense if @xmath224 as @xmath225 is proportional to the radius of @xmath162 , then ( [ algebraic - system - invertible - condition ] ) will be satisfied if @xmath36 and the lam parameters @xmath2 and @xmath3 satisfy the condition @xmath226 recalling that @xmath227 is defined in ( [ constant - final - estimates - green ] ) . finally , let us observe that the right hand side of ( [ upper - bound - omega - lame - coe ] ) depends only @xmath88and the lipschitz character of the reference obstacles @xmath162 s . _ proof of lemma [ mazyawrkthmela ] . _ we start by factorizing @xmath216 as @xmath228 where @xmath229 , @xmath230 is the identity matrix and @xmath231 . we have @xmath232 , so it is enough to prove the injectivity in order to prove its invetibility . for this purpose , let @xmath233 are vectors in @xmath234 and consider the system @xmath235 we denote by @xmath236 and @xmath237 the real and the imaginary parts of the corresponding complex vector / matrix . we also set @xmath238 by @xmath239 . from we derive the following two identities : @xmath240 and then @xmath241 summing up and we obtain @xmath242 the right - hand side in can be estimated as @xmath243 let us now consider the right hand side in . first we have @xmath244 where @xmath245 and @xmath246 if @xmath247 and @xmath248 for @xmath249 . hence from ( [ kupradzeten1 ] ) @xmath250 . from the observation before lemma [ capacitance - eig - single ] , we deduce that @xmath251}_{n=1}\hspace{-.1cm}2[(2n+1)^3-(2n-1)^3]\frac{{\mathring{c}}^2}{(4\pi)^2 n^2 \left(\frac{a^{\frac{1}{3}}}{2}\right)^{2 } } \leq 13 { \mathring{c}}^2 m \frac{a^{-\frac{2}{3}}}{\pi^2}\sum^{[2a^{-\frac{1}{3}}]}_{n=1}1 = \frac{26 m_{max}{\mathring{c}}^2 a^{-2}}{\pi^2}\end{aligned}\ ] ] or @xmath252 and then @xmath253 using lemma [ capacitance - eig - single ] , we deduce that @xmath254 from ( [ systemsolve1-small - sub1 - 2-ac-3 ] ) , ( [ systemsolve1-small - sub1 - 2-ac-4 ] ) , ( [ b_n - estimation-1 ] ) and ( [ lower - bound - c - i ] ) , we deduce that @xmath255 if @xmath256 and then the matrix @xmath216 in algebraic system is invertible . from the function @xmath158 , we define a bounded function @xmath257 as follows : @xmath258 . \end{array}\right.\ ] ] hence each @xmath194 contains @xmath259 $ ] obstacles and @xmath260 . let @xmath261 be the @xmath262 matrix having entries as piecewise constant functions such that @xmath263 for all @xmath17 and vanishes outside @xmath4 . here , @xmath208 are the capacitances of @xmath162 s . from @xcite , we can observe that @xmath208 are defined through defined through @xmath264 , and are independent of @xmath60 . we set @xmath265 consider the lippmann - schwinger equation @xmath266 and set the lam potential @xmath267 the coefficients @xmath268 and @xmath269 are uniformly bounded . the next lemma concerns the mapping properties of the lam potential . these properties are proved for the scalar poisson potential in @xcite , for instance . similar arguments are applicable for the lam potential as well , so we omit to give the details . [ lip - schw ] the operator @xmath270 is well defined and it is a bounded operator for any bounded domain @xmath271 in @xmath5 , i.e. there exists a positive constant @xmath66 such that @xmath272 we have also the following lemma . [ invertibility - of - vc ] there exists one and only one solution @xmath273 of the lippmann - schwinger equation ( [ fracqcfracmain - effect - int ] ) and it satisfies the estimate @xmath274 where @xmath275 being a large bounded domain which contains @xmath276 . _ proof of lemma [ invertibility - of - vc ] _ using the lemma [ lip - schw ] , we see that @xmath277 is fredholm with index zero and then we can apply the fredholm alternative to @xmath278 . the uniqueness is a consequence of the uniqueness of the scattering problem corresponding to the model @xmath279 where @xmath280 and @xmath281 satisfies the kupradze radiation conditions and @xmath282 is an incident field . the estimate ( [ est - lipm - sch ] ) can be derived , as it is done in @xcite for the acoustic case , by coupling the invertibility of @xmath277 and the @xmath283 interior estimates of the solutions of the system @xmath284 . the capacitances of the obstacles @xmath285 , i.e. @xmath286 are bounded by their lipschitz constants , see @xcite , and we assumed that these lipschitz constants are uniformly bounded . hence @xmath269 is bounded in @xmath287 and then there exists a function @xmath159 in @xmath288 ( actually in every @xmath289 ) such that @xmath269 converges weakly to @xmath159 in @xmath287 . now , since @xmath158 is continuous hence @xmath268 converges to @xmath290 in @xmath291 and hence in @xmath288 . then we can show that @xmath292 converges to @xmath293 in @xmath288 . since @xmath294 is bounded in @xmath295 , then from the invertibility of the lippmann - schwinger equation and the mapping properties of the lam potential , see lemma [ invertibility - of - vc ] , we deduce that @xmath296 is bounded and in particular , up to a sub - sequence , @xmath297 tends to @xmath298 in @xmath288 . from the convergence of @xmath299 to @xmath300 and the one of @xmath297 to @xmath298 and ( [ fracqcfracmain - effect - int ] ) , we derive the following equation satisfied by @xmath301 @xmath302 this is the lippmann - schwinger equation corresponding to the scattering problem @xmath303 , @xmath304 , and @xmath37 satisfies the kupradze radiation conditions . as the corresponding farfields are of the form @xmath305 and the ones of @xmath306 are of the form @xmath307 we deduce that @xmath308 if we assume that @xmath310 $ ] , then we have the estimate @xmath311 , @xmath312 . since the capacitances of the obstacles are assumed to be equal , we set @xmath159 to be a constant in @xmath4 and @xmath121 in @xmath313 . recall that @xmath314 and @xmath315 are solutions of the lippmann - schwinger equations @xmath316 and @xmath317 from the estimate @xmath311 , @xmath312 , we derive the estimate @xmath318 for each @xmath17 , we rewrite the equation as follows @xmath319 } \gamma^{\omega}(z_m , z_j)k_a(z_j)\bar{c}_j u_a(z_j ) vol(\omega_j)\nonumber\\ & & + \sum_{\substack{j=1 \\ j\neq m}}^{[a^{-1 } ] } \gamma^{\omega}(z_m , z_j)k_a(z_j)\bar{c}_j u_a(z_j ) vol(\omega_j)-\int_{\omega } \gamma^{\omega}(z_m , y)k_a(y){\bold{c}_a}(y ) u_a(y ) dy.\end{aligned}\ ] ] let us estimate the following quantities : @xmath320 } \gamma^{\omega}(z_m , z_j)k_a(z_j)\bar{c}_j u_a(z_j ) vol(\omega_j)-\int_{\omega } \gamma^{\omega}(z_m , y)k_a(y){\bold{c}_a}(y ) u_a(y ) dy\ ] ] and @xmath321 } \gamma^{\omega}(z_m , z_j)k_a(z_j)\bar{c}_j u_a(z_j ) vol(\omega_j).\ ] ] by the decomposition of @xmath4 , @xmath323}_{l=1}$ ] , we have @xmath324}\int_{\omega_l } \gamma^{\omega}(z_m , y){k_a}(y){\bold{c}_a}(y ) u_a(y ) dy.\ ] ] @xmath325 } \left [ \gamma^{\omega}(z_m , z_j)k_a(z_j)\bar{c}_j u_a(z_j ) vol(\omega_j ) -\int_{\omega_j } \gamma^{\omega}(z_m , y)k_a(y){\bold{c}_a}(y ) u_a(y ) dy \right].\label{rearranging - a}\end{aligned}\ ] ] for @xmath326 , we have @xmath327 dy . \end{aligned}\ ] ] we set @xmath328 then every component @xmath329 of @xmath330 satisfies @xmath331 where @xmath332\,d\beta\nonumber\\ & = & \int_0 ^ 1\sum^3_{j=1}\left[\nabla_y\gamma_{i , j}^{\omega}(z_m , y-\beta(y - z_l))\right ] u_{a , j}(y-\beta(y - z_l))\,d\beta\nonumber\\ & & + \int_0 ^ 1\sum^3_{j=1}\gamma_{i , j}^{\omega}(z_m , y-\beta(y - z_l))\left[\nabla_y u_{a , j}(y-\beta(y - z_l))\right]\,d\beta . \end{aligned}\ ] ] from and from section [ the - relative - distribution - of - the - small - bodies ] , we derive for @xmath333 @xmath334 where @xmath227 depends only on @xmath36 and some universal constants . then @xmath335 then , for @xmath326 , and and observing that @xmath336 is a constant matrix in @xmath337 , imply the estimate @xmath338 & \vert y - z_l\vert dy \nonumber\\ & & + \frac{\mathring{c } \bar{c}_l k_a(z_l)}{2\pi\;n a^{\frac{1}{3 } } } \int_{\omega_l } \left[\int_0 ^ 1 { \vert \nabla_y u_{a}(y-\beta(y - z_l))\vert}d\beta \right]\vert y - z_l\vert dy \nonumber\\ \substack{\leq\\ \eqref{est - lipm - sch}}&c_1\frac{\textcolor{black}{[{k_a}(z_l)]}\bar{c}_l}{n^2 a^{\frac{2}{3}}}\,a^\frac{4}{3 } \,\substack{\leq\\ \eqref{sup - cap - all}}\ , c_1\frac{\textcolor{black}{k_{\max}}\mathcal{c}}{n^2 } \,a^\frac{2}{3},\qquad \end{aligned}\ ] ] for a suitable constant @xmath339 . regarding the integral @xmath340 we do the following estimates : @xmath341\right)}_{=:lm(r , a)}\frac{\mathring{c}}{4\pi } c_2\bar{c}_m{k_a}&(z_m)\nonumber\\ \leq\,&\frac{\mathring{c}}{4\pi}c_2{k_a}(z_m)\bar{c}_m~lm(r^c , a),\qquad\qquad\nonumber\\ & & \mbox{here $ r^c$ is the value of $ r$ where $ lm(r , a)$ attains maximum}.\nonumber\\ & & { \partial_r lm(r , a)=0 \rightarrow 4\pi r-\frac{a}{r^2}-\frac{8}{3}\pi r=0\rightarrow r_c=\left(\frac{3}{4}\pi a\right)^\frac{1}{3}}\nonumber\\ & & { \begin{array}{ccc } lm(r_c , a)&=&2\pi\left(\frac{3}{4}\pi\right)^\frac{2}{3 } a^\frac{2}{3}+\left(\frac{4}{3\pi } \right)^\frac{1}{3}a^\frac{2}{3}-\frac{4}{3}\pi\left(\frac{3}{4}\pi \right)^\frac{2}{3}a^\frac{2}{3}\\ & & \\ & = & \left[\frac{2}{3\pi}\left(\frac{3}{4}\pi\right)^\frac{2}{3}+\left(\frac{4}{3\pi } \right)^\frac{1}{3 } \right]a^\frac{2}{3}=\frac{3}{2}\left(\frac{4}{3\pi } \right)^\frac{1}{3}a^\frac{2}{3 } \end{array } } \nonumber\\ \leq\,&\frac{3}{8\pi}\mathring{c}c_2k_{max}\mathcal{c}\left(\frac{4}{3\pi } \right)^\frac{1}{3}a^\frac{2}{3}.\qquad\qquad\end{aligned}\ ] ] from , we can have @xmath342 } \left [ \vert \gamma^{\omega}(z_m , z_j)k_a(z_j)\bar{c}_j u_a(z_j ) vol(\omega_j)-\int_{\omega_j } \gamma^{\omega}(z_m , y)k_a(y){\bold{c}_a}(y ) u_a(y ) dy \vert_\infty\right].\end{aligned}\ ] ] which we can estimate by @xmath343}_{n=1}2[(2n&+1)^3-(2n-1)^3 ] \bigg [ \vert \gamma^{\omega}(z_m , z_j)k_a(z_j)\bar{c}_j u_a(z_j ) vol(\omega_j)\\ & -\int_{\omega_j } \gamma^{\omega}(z_m , y)k_a(y){\bold{c}_a}(y ) u_a(y ) dy \vert_\infty\bigg]+\vert \int_{\omega_m } \gamma^{\omega}(z_m , y)k_a(y){\bold{c}_a}(y ) u_a(y ) dy \vert_\infty.\end{aligned}\ ] ] and then @xmath344.\ ] ] finally @xmath345 @xmath347 } \gamma^{\omega}(z_m , z_j)k_a(z_j)\bar{c}_j u_a(z_j ) vol(\omega_j ) \\ = \sum_{\substack{l=1 \\ l\neq m\\ z_l \in \omega_m}}^{\textcolor{black}{[k_a(z_m)]}}\gamma^{\omega}(z_m&,z_l)\bar{c}_l u_a(z_l ) a+ \sum_{\substack{j=1 \\ j\neq m}}^{[a^{-1 } ] } \sum_{\substack{l=1 \\ z_l \in \omega_j}}^{\textcolor{black}{[k_a(z_j)]}}\gamma^{\omega}(z_m , z_l)\bar{c}_l u_a(z_l ) a- \sum_{\substack{j=1 \\ j\neq m}}^{[a^{-1 } ] } \gamma^{\omega}(z_m , z_j)k_a(z_j)\bar{c}_j u_a(z_j ) vol(\omega_j ) \\ = \bar{c}_m l\neq m\\ z_l \in \omega_m}}^{\textcolor{black}{[k_a(z_m)]}}\gamma^{\omega}&(z_m , z_l ) u_a(z_l ) + \sum_{\substack{j=1 \\ j\neq m}}^{[a^{-1}]}\bar{c}_j a\big[\big(\sum_{\substack{l=1 \\ z_l \in \omega_j}}^{\textcolor{black}{[k_a(z_j)]}}\gamma^{\omega}(z_m , z_l ) u_a(z_l)\big)- \gamma^{\omega}(z_m , z_j)\textcolor{black}{[k_a(z_j ) ] } u_a(z_j)\big ] , \end{aligned}\ ] ] since @xmath348}}{k_a(z_j)}\ ; \mbox { and } \bar{c}_l=\bar{c}_j , \mbox { for } l=1 , ... , \ ; k_a(z_j)$ ] . we write , @xmath349}}\gamma^{\omega}(z_m , z_l ) u_a(z_l)\end{aligned}\ ] ] and @xmath350}}\gamma^{\omega}(z_m , z_l ) u_a(z_l)\big)- \gamma^{\omega}(z_m , z_j)\textcolor{black}{[k_a(z_j ) ] } u_a(z_j)\big]\nonumber\\ & = & \sum_{\substack{l=1 \\ z_l \in \omega_j}}^{\textcolor{black}{[k_a(z_j)]}}\big(\gamma^{\omega}(z_m , z_l ) u_a(z_l)- \gamma^{\omega}(z_m , z_j ) u_a(z_j)\big).\end{aligned}\ ] ] we need to estimate @xmath351 and @xmath352}\bar{c}_j ae^j_2 $ ] . + now by writing @xmath353 . for @xmath354 , using taylor series , we can write @xmath355 with @xmath356 by doing the computations similar to the ones we have performed in ( [ taylorremind1]-[taylorremind - effect ] ) and by using lemma [ invertibility - of - vc ] , we obtain @xmath357}\bar{c}_j a e^j_2 \vert \leq c_4\mathcal{c}k_{max } a^{\frac{1}{3 } } \label{este2j1}\end{aligned}\ ] ] one can easily see that , @xmath358 substitution of in and using the estimates and associated to @xmath322 and the estimates and associated to @xmath346 gives us @xmath359 we rewrite the algebraic system ( [ fracqcfracmain ] ) as @xmath360 where we set @xmath361 , recalling that @xmath362 . taking the difference between and produces the algebraic system @xmath363 comparing this system with and by using lemma [ mazyawrkthmela ] , we obtain the estimate @xmath364 for the special case @xmath365 with @xmath366 , we have the following approximation of the far - field from the foldy - lax asymptotic expansion and from the definitions @xmath367 and @xmath264 , for @xmath17 : @xmath368 } \end{aligned}\ ] ] consider the far - field of type : @xmath369 corresponding to the scattering problem ( [ elas - scat - equi ] ) and set @xmath370 and @xmath371 taking the difference between and we have : @xmath372}\int_{\omega_j } e^{-i\frac{\omega}{c_p}\hat{x}\cdot{y}}{k_a}(y){\bold{c}_a}(y ) u_a(y)\cdot \hat{x}dy -\sum_{j=1}^{[a^{-1 } ] } \sum_{\substack{l=1 \\ z_l \in \omega_j}}^{\textcolor{black}{[k_a(z_j)]}}e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_l}\bar{c}_lu_{a , l}\cdot \hat{x}a\nonumber\\ \nonumber & & \hspace{0cm}+o\left(a \hspace{-.03cm}+\hspace{-.03cm}a^{2 - 5t+3t\alpha}\hspace{-.03cm}+ \hspace{-.03cm}a^{3 - 9t+6t\alpha}\hspace{-.03cm}+\hspace{-.03cm}a^{1 - 2t\alpha}\hspace{-.03cm}+\hspace{-.03cm}a^{1 - 3t\alpha}\hspace{-.03cm}+ \hspace{-.03cm}a^{2 - 5t+2t\alpha}\right ) \\ \nonumber & = & \sum_{j=1}^{[a^{-1}]}{k_a}(z_j)\bar{c}_j\int_{\omega_j } \left[e^{-i\frac{\omega}{c_p}\hat{x}\cdot{y } } u_a(y)\cdot \hat{x } - e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j}u_a(z_j)\cdot \hat{x}\right]dy\nonumber\\ \nonumber \\ \nonumber & & + \sum_{j=1}^{[a^{-1}]}\bar{c}_ja \left[\sum_{\substack{l=1 \\ z_l \in \omega_j}}^{\textcolor{black}{[k_a(z_j)]}}\left(e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j } u_a(z_j)\cdot \hat{x}-e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_l}u_a(z_l)\right)+\sum_{\substack{l=1 \\ z_l \in \omega_j}}^{\textcolor{black}{[k_a(z_j ) ] } } e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_l } \left(u_a(z_l)-u_{a , l}\right)\cdot \hat{x}\right]\nonumber\\ \nonumber & & \hspace{0cm}+o\left(a\hspace{-.03cm}+\hspace{-.03cm}a^{2 - 5t+3t\alpha}\hspace{-.03cm}+\hspace{-.03 cm } a^{3 - 9t+6t\alpha}\hspace{-.03cm}+\hspace{-.03cm}a^{1 - 2t\alpha } \hspace{-.03cm}+\hspace{-.03cm}a^{1 - 3t\alpha}\hspace{-.03cm}+ \hspace{-.03cm}a^{2 - 5t+2t\alpha}\right ) \\ \nonumber & = & \sum_{j=1}^{[a^{-1}]}\int_{\omega_j}{k_a}(z_j)\bar{c}_j \left[e^{-i\frac{\omega}{c_p}\hat{x}\cdot{y } } u_a(y)\cdot \hat{x } - e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j}u_a(z_j)\cdot \hat{x}\right]dy\nonumber\\ & & + \sum_{j=1}^{[a^{-1}]}\bar{c}_ja \sum_{\substack{l=1 \\ z_l \in \omega_j}}^{\textcolor{black}{[k_a(z_j)]}}\left(e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j } u_a(z_j)\cdot \hat{x}-e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_l}u_a(z_l)\cdot \hat{x}\right)+\sum_{j=1}^{m}e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j}\bar{c}_ja \left[u_a(z_j)-u_{a , j}\right]\cdot \hat{x}\nonumber\\ \nonumber & & \hspace{0cm}+o\left(a\hspace{-.03cm}+\hspace{-.03 cm } a^{2 - 5t+3t\alpha}\hspace{-.03cm}+\hspace{-.03cm}a^{3 - 9t+6t\alpha } \hspace{-.03cm}+\hspace{-.03cm}a^{1 - 2t\alpha}\hspace{-.03cm}+ \hspace{-.03cm}a^{1 - 3t\alpha}\hspace{-.03cm}+\hspace{-.03 cm } a^{2 - 5t+2t\alpha}\right)\\ & \substack{= \\ \eqref{mazya - fnlinvert - small - ac-3-effect - dif } } & \sum_{j=1}^{[a^{-1}]}{k_a}(z_j)\bar{c}_j\int_{\omega_j } \left[e^{-i\frac{\omega}{c_p}\hat{x}\cdot{y } } u_a(y)\cdot \hat{x } - e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j}u_a(z_j)\cdot \hat{x}\right]dy\nonumber\\ & & + \sum_{j=1}^{[a^{-1}]}\bar{c}_ja \sum_{\substack{l=1 \\ z_l \in \omega_j}}^{\textcolor{black}{[k_a(z_j)]}}\left(e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j } u_a(z_j)\cdot \hat{x}-e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_l}u_a(z_l)\cdot \hat{x}\right)+o\left(\mathcal{c}^2 k_{max } ( a^{\frac{1}{3}}+a^{1-t})\right)\nonumber\\ & & + o\left(a\hspace{-.03cm}+\hspace{-.03cm}a^{2 - 5t+3t\alpha}\hspace{-.03cm}+ \hspace{-.03cm}a^{3 - 9t+6t\alpha}\hspace{-.03cm}+\hspace{-.03 cm } a^{1 - 2t\alpha}\hspace{-.03cm}+\hspace{-.03cm}a^{1 - 3t\alpha } \hspace{-.03cm}+\hspace{-.03cm}a^{2 - 5t+2t\alpha}\right ) . \end{aligned}\ ] ] now , let us estimate the difference @xmath373}{k_a}(z_j)\bar{c}_j\int_{\omega_j } \left[e^{-i\frac{\omega}{c_p}\hat{x}\cdot{y } } u_a(y ) - e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j}u_a(z_j)\right]dy$ ] . write , @xmath374 . using taylor series , we can write @xmath375 with @xmath376 \right]_k\,d\beta\nonumber\\ & = & \int_0 ^ 1\left[\nabla_ye^{-i\frac{\omega}{c_p}\hat{x}\cdot(y-\beta(y - z_j))}\right]_k u_{a}(y-\beta(y - z_j))\,d\beta\nonumber\\ % & & + \int_0 ^ 1e^{-i\frac{\omega}{c_p}\hat{x}\cdot(y-\beta(y - z_j))}\left[\nabla_yc(y-\beta(y - z_j))\right ] y(y-\beta(y - z_j))\,d\beta\nonumber\\ & & + \int_0 ^ 1e^{-i\frac{\omega}{c_p}\hat{x}\cdot(y-\beta(y - z_j))}\left[\nabla_y u_{a}(y-\beta(y - z_j))\right]_k\,d\beta . \end{aligned}\ ] ] we have @xmath377 then @xmath378 using we get the estimate @xmath379}{k_a}(z_j)\bar{c}_j\int_{\omega_j } \left[e^{-i\frac{\omega}{c_p}\hat{x}\cdot{y}}(y ) u_a(y ) - e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j}u_a(z_j)\right]dy\right \vert_\infty \leq\ ] ] @xmath380}{k_a}(z_j)\bar{c}_j \left(\frac{\omega}{c_p } \int_{\omega_j}\vert y - z_j\vert_\infty\int_0 ^ 1 |u_{a}(y-\beta(y - z_j))|\ , d\beta\,dy\right)\,\ ] ] @xmath381}{k_a}(z_j)\bar{c}_j \left(\int_{\omega_j}\vert y - z_j\vert\int_0 ^ 1\vert\nabla_y u_{a}(y-\beta(y - z_j))\vert_\infty \,d\beta\,dy\right)\ ] ] @xmath382}{k_a}(z_j)\bar{c}_jc_1\ , a\,a^\frac{1}{3}\,\left(\frac{\omega}{c_p } + c_5\right ) \leq k_{max } \mathcal{c}c_1\left(\frac{\omega}{c_p } + c_5\right)\,a^{\frac{1}{3}}.\ ] ] in the similar way , using ( [ mazya - fnlinvert - small - ac-3-effect - dif ] ) , we have , @xmath383}\bar{c}_ja \sum_{\substack{l=1 \\ z_l \in \omega_j}}^{\textcolor{black}{[k_a(z_j)]}}\left(e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_j } u_a(z_j)-e^{-i\frac{\omega}{c_p}\hat{x}\cdot z_l}u_a(z_l)\right)\right\vert_\infty & \leq & o\left(k_{max } \mathcal{c}(a^{\frac{1}{3}}+a^{1-t})\right ) . \end{aligned}\ ] ] using the estimates and in , we obtain @xmath384 since @xmath385 is of order @xmath386 , and @xmath387 is of the order @xmath388 , we should have @xmath389 . hence @xmath390 hence for @xmath391 , we have @xmath392 equating @xmath393 , we find that @xmath394 and then @xmath395 and @xmath396 . in addition , since @xmath397 , then @xmath398 . hence , the error is @xmath399\cdot \hat{x } & = & o\left ( a^{\frac{1}{3 } } + a^{\frac{3}{2}-3t}\right),\;~~ \frac{1}{3 } \leq t \leq \frac{7}{12}. \end{aligned}\ ] ] combining the estimates ( [ elastic - difference - farfield - effect-1**1 - 1 ] ) and ( [ appro-0-a ] ) , we deduce that @xmath400\cdot \hat{x}=o(a^{\min \{\gamma,\ ; \frac{1}{3},\ ; \frac{3}{2}-3t\}}),\ ; a<<1 , \;~~ \frac{1}{3 } \leq t \leq \frac{7}{12}\ ] ] uniformly in terms of @xmath401 . for an obstacle @xmath402 of radius @xmath403 , @xmath404 and @xmath405 . similarly , we set @xmath406 and @xmath407 . we see that @xmath408 satisfies @xmath409 with the kupradze radiation conditions . since @xmath410 , @xmath411 is bounded in @xmath289 , for @xmath412 , by interior estimates , we deduce that @xmath413 , @xmath411 is bounded in @xmath414 , for @xmath412 , and hence , in particular , the normal traces are bounded in @xmath415 . then we can show that the norms of the operators @xmath416 and @xmath417 are of the order @xmath418 at least . 1 . using these properties and arguing as in @xcite , we derive the asymptotic expansions ( [ x oustdie1 d_m farmainp - near - rho])-([x oustdie1 d_m farmains - near - rho ] ) . indeed , apart from the computations done in @xcite , the main arguments needed to extend those results to the case of variable density is the fredholm alternative for the corresponding integral operators and the application of the neumann series expansions . after splitting @xmath419 as @xmath420 , these two arguments are applicable as soon as we have ( [ single - layer - rho])-([double - layer - rho ] ) . 2 . the justification of the invertibility of the algebraic system ( [ fracqcfracmain ] ) depends only on ( 1 ) the distribution of small bodies and ( 2 ) the background medium through the singularities of the fundamental solution ( of the form @xmath421 ) . however , this type of singularity is true for general background elastic media . with the singularity of @xmath78 and the smoothness of @xmath422 then the same arguments can be used to justify the invertibiliy of the algebraic system ( [ fracqcfracmain - rho ] ) . using the above mentioned decomposition of the green s function @xmath419 , the properties of the lippmann schwinger integral equation are also valid replacing @xmath78 by @xmath419 , and hence the results in section [ subsection - piecewise - constant ] are valid . finally , and again using the decomposition of @xmath419 , the computations in section [ the approximation - by - as ] can be carried out using @xmath419 . we start with the following lemma on the symmetry structure of the elastic capacitance . _ proof of lemma [ lemma - cap - adj ] . _ we know that the matrix @xmath427 solves the invertible integral equation @xmath428 or precisely @xmath429 , where @xmath430 and @xmath431 is the @xmath432 column of @xmath433 . let @xmath60 be any constant vector in @xmath5 , then the vector @xmath434 satisfies @xmath435 we set @xmath436 . then @xmath437 satisfies the problem @xmath438 , in @xmath424 and @xmath439 on @xmath440 . in addition , we have the jump relation @xmath441 on @xmath440 where @xmath442 is the elastic conormal derivative . hence @xmath443 now , let @xmath60 and @xmath444 be arbitrary constant vectors in @xmath5 . to both @xmath60 and @xmath444 , we correspond @xmath437 and @xmath445 as above . using the green formulas inside and outside of @xmath424 , we deduce that @xmath446 recalling that every quantity here is real valued . [ capacitance - rotation ] let @xmath447 be a unitary transform in @xmath448 , @xmath424 be bounded lipschitz domain in @xmath448 , @xmath449 and @xmath450 . let @xmath118 and @xmath451 be the corresponding elastic capacitance matrices due to the denisty matrices @xmath452 and @xmath453 , as defined in , respectively . then we have @xmath454 . _ proof of lemma [ capacitance - rotation ] . _ first recall the relation @xmath455 , see ( * ? ? ? * lemma 6.11 ) . from , we have that @xmath456 now from the uniqueness of solutions of , we deduce that @xmath457 and then @xmath458 from the definition of the capacitance , see ( [ barqcimsurfacefrm1main ] ) , and , we have @xmath459 [ capacitance - rotation-2drltn ] let @xmath447 be a unitary transform in @xmath448 , @xmath460 , and let @xmath424 be a bounded lipschitz domain in @xmath448 , @xmath449 , and @xmath450 . let @xmath118 and @xmath451 be the corresponding elastic capacitance matrices due to the denisty matrices @xmath452 and @xmath453 , as defined in , respectively . 1 . @xmath461-case . if the shape of @xmath424 is rotationally invariant for any rotation by one angle @xmath462 , then @xmath118 is a scalar multiplied by the identity matrix . 2 . @xmath463-case . if the shape of @xmath424 is rotationally invariant for any two of the rotations around the @xmath464 or @xmath465 axis by one angle @xmath462 and @xmath466 respectively , then @xmath118 is a scalar multiplied by the identity matrix . _ proof of proposition [ capacitance - rotation-2drltn ] . _ in @xmath467 case , the rotation matrix by an angle @xmath468 is given by @xmath469 as the shape is invariant by this rotation then @xmath470 . since @xmath471 is unitary then @xmath472 and then implies @xmath473 we deduce from ( [ cap - rotation-1 ] ) and the symmetry of matrix @xmath118 the following relations : @xmath474 @xmath475 @xmath476 we rewrite ( [ relations-2d-1 ] ) and ( [ relations-2d-3 ] ) respectively as @xmath477 @xmath478 taking @xmath479 as a multiplicative factor in ( [ relations-2d-1 - 1 ] ) and ( [ relations-2d-3 - 1 ] ) we see that if @xmath462 , i.e. @xmath480 , then we have @xmath481 let us now consider the 3d case . first , let us assume that the shape is invariant under the rotation about the @xmath482 and an angle @xmath462 . this rotation matrix is given by @xmath483 since @xmath471 is unitary , @xmath472 , then gives us ; @xmath484 @xmath485 we observe the equality of the @xmath486 matrices : @xmath487 to show that @xmath118 is scalar multiplied by the identity matrix we need to prove that @xmath489 , for instance , and @xmath490 . for this purpose , we use another rotation . taking the rotation around the @xmath465-axis axis . ] by one angle @xmath491 and proceeding as we did for the rotation about the @xmath492-axis , we show that @xmath493 1 . for the spherical shapes , in particular , the capacitance is a scalar multiplied by the identity matrix . ellipsoidal shapes are invariant only under rotations with angle @xmath494 ( or trivially @xmath495 ) . for these shapes , the capacitance might not be a scalar multiplied by the identity matrix but a diagonal matrix instead . to justify this property , the arguments in @xcite can be useful .

we propose a new framework of quantum field theory for an arbitrary observer in curved spacetime , defined in the spacetime region in which each point can both receive a signal from and send a signal to the observer . multiple motivations for this proposal are discussed . we argue that radar time should be applied to slice the observer s spacetime region into his simultaneity surfaces . in the case where each such surface is a cauchy surface , we construct a unitary dynamics which evolves a given quantum state at a time for the observer to a quantum state at a later time . we speculate on possible loss of information in the more general cases and point out future directions of our work . + + _ keywords _ : quantum field theory in curved spacetime , observer - dependence , simultaneity , notion of particles , dynamics of quantum states , information loss . .... .... * towards quantum field theory in curved spacetime * * for an arbitrary observer * 1.5 hui yao ` hy255@cam.ac.uk ` _ damtp , university of cambridge _ .... .... = 11 @page=3 = 12 .... .... = 11 @page=0 = 12

0907.2481

the universe is ultimately observer - participatory , each part of which is communicating with the others and never at rest . existence are not simply `` things just out there '' , but reality for an observer is his dynamical construction . both quantum theory and relativity , the two pillars of our modern conceptual understanding of nature , taught us so , although the great union of the two remains a mystery still . we shall hence aim at a further understanding of the observer s quantum description of matter in a general relativistic background . we first discuss the multiple motivations of this paper . 1 . quantum field theory in curved spacetime is the necessary first step towards a conceptual understanding of a quantum theory of gravity . even though such a framework is ultimately incomplete , we expect some features of the theory to remain . we shall treat the quantum field like a `` test field '' with the background spacetime completely classical and unaffected by the matter field . 2 . we view physics as a theory of an arbitrary observer s dynamical description of his system . therefore any physical theory should be formulated explicitly in terms of an observer s own physical quantities at a fundamental level . there have been great conceptual advancements in understanding the observer - dependent aspects of quantum field theory [ 1 - 6 ] . rather than treating observer - dependence as an emergent feature , we now push these developments further by taking observer - dependence to be a cornerstone of our formulation of quantum field theory in curved spacetime . 3 . working at a semi - classical level , we shall understand an observer as a history of spacetime events , in other words , a timelike curve in a classical curved spacetime . a quantum state is a quantum description of the physical system for the observer at some proper time of his . we consider dynamics as the updating of his description of the system rather than as the `` changing of things in themselves '' . above all , we would like to construct a theory specifying a one - parameter family of quantum states along a timelike worldline . dynamics is the specification of a one - parameter family of physical states , whether they be points in classical phase spaces or vectors in hilbert spaces or some other mathematical representation . therefore intrinsic to dynamics is a notion of `` time '' . a notion of time is usually only available in special cases . one may restrain oneself to the notion of asymptotic past / future only , but this does not define a parameter of physical time . there is a `` natural '' choice of time if the spacetime manifold admits a special symmetry , like the killing time for stationary spacetimes or the usual cosmic time of friedmann - robertson - walker ( frw ) cosmology . however , the last two choices are rather mathematical and only apply to special cases . we would instead like a single physically - motivated definition applicable to all situations . the only possibility of such a definition is the proper time on an observer s clock . such a choice in fact forces one to adopt an observer - dependent description of physics , which is precisely what we have been aiming for . 5 . in this paper we shall take a hilbert space approach in that we consider hilbert spaces to be the most fundamental objects mathematically representing our quantum description of physical systems . quantum states are vectors or more generally density matrices in these spaces . such a mathematical framework is especially suitable for discussing quantum state evolution , where a two - parameter mapping @xmath0 relates a quantum state in a hilbert space at some time @xmath1 for an observer to another state in a possibly different hilbert space at time @xmath2 . the hilbert space approach is also particularly suitable for discussing any possible loss of information in the form of a pure state evolving to a mixed density matrix . there are two important conceptual lessons that we should draw from hawking s semi - classical analysis of black hole information loss . firstly , hawking has insisted that any quantum state for the observer outside the black hole must be physically meaningful for him : to obtain the correct description of the quantum system according to the observer , we must trace out all degrees of freedom that he can not causally access . as we shall see , our quantum theory is precisely defined over the spacetime region causally connected to the given observer . secondly , that there is a loss of information when black hole evaporates indicates that the loss of information is rooted in the `` evolution '' of a horizon rather than simply the presence of a horizon . to make precise sense of evolution one again needs a notion of time , which we have chosen to be the proper time of an observer . 7 we shall speculate that there will be a non - unitary evolution and hence information loss for an observer , precisely when his surface of simultaneity evolves from a cauchy surface to a non - cauchy surface for the spacetime region to which the observer has causal access . instead of resolving information loss as a paradox , we propose to forcefully carry hawking s argument through and speculate that the possibility of information loss is a fundamental feature of quantum field theory in curved spacetime rather than special to black hole evaporation . 8 . to obtain physical quantities from the point of view of an observer , one normally has to study the response of a model particle detector following the observer s worldline . we would instead like to have a fundamental theory in which the particle content at any time can be simply read off from the mathematical representation of the physical state . furthermore a quantum state , which completely encodes all information about the possible measurement outcomes and their probabilities of occurrence , tells much more than simply the particle spectrum , in particular whether the quantum state is pure or mixed . wald has forcefully argued that there is no natural choice of fock space of particles in quantum field theory in a most general spacetime , and that this is in analogy to choosing a coordinate system on a manifold in general relativity which can not be physical . but for a given observer , as we shall show , there indeed exists a natural notion of `` particles '' . there have been serious difficulties in attempting to define a notion of particles in a most general spacetime , but this does not imply that the idea of `` particle '' itself is not fundamental , as long as one takes an observer - dependent viewpoint of quantum field theory . a notion of particles is usually only available in special cases such as if the spacetime admits a special symmetry or special asymptotic behaviours , or if the physical situation admits an adiabatic approximation . however generalisation is essential , as elements particular to special cases can obscure what the fundamental features are . the formulation which we will present can be applied rather generally to a wide class of observers without requiring any special symmetries or asymptotic behaviours of the spacetime . unlike in e.g. , here we shall make no fundamental distinction between particle creation due to the motion of the observer and that due purely to gravitational fields . we view spacetime as a geometric and causal background in which the observer s worldline is defined . all observers are regarded as completely equivalent at the fundamental level of quantum field theory in curved spacetime . ashtekar and magnon have constructed a one - parameter family of fock spaces in their formulation of quantum field theory in curved spacetime . the authors found their construction to depend on a choice of timelike vector field and hence of the corresponding integral curves . they were therefore forced to conclude that their theory depends on a field of observers . rather than be led to a conclusion of observer - dependence , we have taken as our conceptual starting point the construction of a quantum field theory with observer - dependence . one might interpret ashtekar and magnon s construction , which depends on a congruence of timelike curves , to be for a family of observers . however , we insist any physical theory should be formulated for an arbitrary single observer . mathematically , as we shall show , the construction of the family of fock spaces depends only on a choice of scalar function @xmath3 and a corresponding foliation . physically , if one starts with a single observer , in a most general situation it is far from clear how to choose a family which he is a member of . furthermore , specifying an arbitrary family of observers has no direct physical interpretation as one can never set up an experiment with an uncountably many number of detectors which trace out a congruence of curves covering the entire spacetime . finally , one expects that different observers within one family , even if there is such a preferred grouping , would differ in their description of a physical system . one therefore would like this difference to be naturally accounted for and built into the fundamental theory . 14 . this work in some sense parallels einstein s construction of special relativity . just as the principle of relativity is the guiding principle of special relativity , it is our conceptual starting point that the same laws of quantum field theory should apply to any arbitrary observer , although the observers dynamical quantum descriptions may differ . secondly , just as einstein recognised the inseparable connection between time and the signal velocity , we shall apply radar time , which is operationally defined using light signal communication , to formulating quantum field theory for an arbitrary observer in a general spacetime . the plan of the paper is given as follows . in section 2 , we shall construct a one - parameter family of fock spaces based on the formalism of ashtekar and magnon . we shall show that the formalism requires a choice of scalar function @xmath3 . in section 3 , we shall apply radar time to operationally define this function @xmath3 for each point which a given observer can both send a signal to and receive a signal from . we regard defining quantum field theory in the spacetime region causally connected to a given observer as an axiom of our framework . in section 3.1 we discuss the importance of cauchy surfaces in the formulation of a unitary theory and the preservation of information . in section 4 , we construct the dynamics of quantum states in our theory . we define quantum evolution in terms of a two - parameter mapping from one fock space to another , each associated with a time for an observer , and we show this mapping satisfies certain necessary physical conditions . finally in section 5 , we summarise our main results and consider possible directions in which our work might be developed . we shall use natural units throughout this paper . the sign convention in general relativity is the same as that of , in particular , @xmath4 . in this section , we shall construct a one - parameter family of fock spaces of the free real scalar field based on the work of ashtekar and magnon . definitions and notations are introduced which will be used throughout the subsequent sections . we shall summarise the formalism in a form most simple and ready for our purpose of the hilbert space approach as motivated in section 1 . in particular , we shall not start from the * -algebra of abstract field operators of . let @xmath5 be the vector space of all well - behaved are smooth and induce , on any spacelike cauchy surface , initial data sets of compact support . the assumption about compact support is used to ensure convergence of various integrals and to discard various surface terms when integrating by parts . ] real - valued solutions of klein - gordon equation ( ^a_a - m^2)= 0 on a given globally hyperbolic spacetime . let @xmath6 be an arbitrary spacelike cauchy surface of the given spacetime , with arbitrary coordinates @xmath7 and unit future - directed normal @xmath8 . the induced metric on @xmath6 is @xmath9 . a symplectic form @xmath10 on @xmath5 is defined as ( , ) = ( _ a- _ a)n^ad^3x . [ symplectic ] let @xmath11 be a complex structure on the real vector space @xmath5 , i.e. an automorphism on @xmath5 which satisfies @xmath12 . @xmath11 endows @xmath5 with the structure of a complex vector space , which we shall denote as @xmath13 . we shall use @xmath14 to distinguish an element @xmath15 in @xmath13 from its counterpart @xmath16 in @xmath5 . hence in our notation we have , for example , @xmath17 . an inner - product @xmath18 can be defined on @xmath13 as | = ( , j ) + ( , ) . [ inner_prod ] this indeed defines an inner - product if and only if the complex structure @xmath11 is compatible with the symplectic form @xmath10 , i.e. ( , ) = ( j , j ) , , v [ symplectic1 ] ( , j ) > 0 v \\{0}.[symplectic2 ] the cauchy completion of the complex inner - product space @xmath19 is a hilbert space which we shall denote as @xmath20 . we now define the @xmath21-particle space to be the hilbert space @xmath22 , i.e. the @xmath21th - rank symmetric tensor over @xmath20 . the space of all quantum states is then the symmetric fock space @xmath23 based on the hilbert space @xmath20 : ( _ j ) = _ n=0^^n_s_j . creation and annihilation operators are defined as mappings on this @xmath23 in the usual way ; see e.g. . we shall denote the creation and annihilation operators associated with @xmath15 as @xmath24 , where @xmath25 is for creation and @xmath26 is for annihilation .. although it should be understood that @xmath24 depends on @xmath27 rather than on @xmath28 . ] the use of the index notation @xmath29 will become clear in section 4 . one can show from their definitions these operators satisfy the following properties : ( i ) @xmath30 ; ( ii ) each creation / annihilation operator is complex - linear / anti - linear in its argument ; and ( iii ) = [ c^- ( ) , c^- ( ) ] = 0 , = | . [ commutations ] to summarise , we have constructed a fock space for each choice of a complex structure @xmath11 on @xmath5 which is compatible with the symplectic form @xmath10 in the sense of ( [ symplectic1 ] ) and ( [ symplectic2 ] ) . to finish the construction , it remains to specify a one - parameter family of complex structures @xmath31 satisfying ( [ symplectic1 ] ) and ( [ symplectic2 ] ) . we now summarise , in a slightly different form , the construction of @xmath31 due to ashtekar and magnon . let @xmath3 be a scalar function on the spacetime such that each constant @xmath3 hypersurface @xmath32 is a spacelike cauchy surface and the set @xmath33 foliates the given spacetime . let @xmath34 be the unit future - directed normal to @xmath32 , where n = . to construct @xmath31 , we introduce a @xmath3-dependent hamiltonian operator @xmath35 on @xmath36 defined by on @xmath36 is to remind us that it is constructed out of @xmath31 as described previously . ] h_t|_t : = -i|_t _ t = -|j_t _ t _ t where @xmath37 is a real - linear operator on @xmath5 which is defined as follows . if @xmath16 has on @xmath32 the cauchy data @xmath38 , @xmath39 , then @xmath40 is the solution with cauchy data @xmath41 , @xmath42 on the same cauchy surface , where : = -n^2 h^ab d_a d_b - n h^ab ( d_a n ) d_b + m^2 n^2 ; @xmath9 is the induced metric on @xmath32 ; and @xmath43 is the covariant derivative on @xmath44 . it is easy to see that @xmath35 is complex - linear on @xmath36 if and only if @xmath37 is a real - linear operator on @xmath5 and @xmath37 commutes with @xmath31 , i.e. @xmath45 = 0 $ ] . it is proven that there exists a unique complex structure @xmath31 satisfying ( [ symplectic1 ] ) and ( [ symplectic2 ] ) such that @xmath46 is real for any @xmath47 . if @xmath16 is the solution with cauchy data @xmath48 on @xmath32 as defined before , then @xmath49 is the solution with cauchy data @xmath50 on the same cauchy surface . one can check that @xmath31 does indeed commute with @xmath37 : @xmath51 = 0 $ ] . furthermore , the following relation holds automatically | h_t | _ t = t_abnn^ad^b , where the energy - momentum tensor @xmath52 is given by t_ab = _ a _ b - g_ab ( ^c _ c + m^2 ^2 ) . in applying the above formalism to obtain the quantum theory as motivated and outlined in section 1 , we need to solve two more problems . first of all , we would like to interpret the above mathematical formalism physically for an arbitrary single observer in a given spacetime . in the above construction of the one - parameter family of fock spaces , there is an ambiguity in the choice of the scalar function @xmath3 . we need to specify the scalar @xmath3 and understand its operational meaning for a given observer . this we shall discuss in the next section . secondly , we would like to understand the dynamics of the theory . a differential form of the dynamics has been discussed in , however we will not follow that approach . the question we would like to answer is : given a quantum state for an observer at his proper time @xmath1 , what is the evolved quantum state at another time @xmath2 ? in other words , we need to construct a two - parameter mapping from @xmath53 to @xmath54 satisfying certain properties , which we shall discuss in section 4 . in the previous section , we have constructed a one - parameter family of fock spaces . the construction relies on a choice of scalar function @xmath3 and a corresponding slicing of the spacetime @xmath33 . in this section , we would like to understand this @xmath3 for a spacetime event @xmath55 as the `` time '' of @xmath55 for a given arbitrary observer , and @xmath32 as the set of all events `` simultaneous '' to the observer at his proper time @xmath3 . we would like to understand how the events of a given spacetime are directly related to the local physical quantities of the observer in an operational way . to understand the physical meaning of this `` time '' and `` simultaneity '' , we first look at an inertial observer @xmath56 in minkowski spacetime . let @xmath57 be the earliest time when @xmath56 can receive a signal from an event @xmath55 . let @xmath58 be the latest time for @xmath56 to send a signal to reach @xmath55 . we define @xmath59 and @xmath60 analogously for another event @xmath61 . then the events @xmath55 and @xmath61 are simultaneous for the inertial observer @xmath56 if and only if the relation @xmath62 holds . this captures the intuition that if @xmath55 and @xmath61 are simultaneous then the difference in inquiring time and response time would equal , and this difference is purely due to any difference in the distances from @xmath55 and @xmath61 to the observer . this construction , known as _ radar time _ , has been advocated by bondi . applications of radar time to an observer - dependent particle interpretation in quantum field theory has recently been pioneered by dolby and gull . ( 150,130 ) ( 0,65)(1,0)150 ( 75,5)(0,1)125 ( 20,65)(1,1)55 ( 20,65)(1,-1)55 ( 105,65)(-1,1)30 ( 105,65)(-1,-1)30 ( 15,57)@xmath55 ( 107,57)@xmath61 ( 107,77)@xmath63 ( 145,57)@xmath64 ( 80,125)@xmath56 ( 65,123)@xmath57 ( 65,7)@xmath58 ( 65,90)@xmath59 ( 65,30)@xmath60 the above characterisation of simultaneity is readily generalisable to an arbitrary observer in a curved spacetime . let an observer @xmath56 be a timelike curve parametrised by his proper time @xmath3 and consider a spacetime event @xmath55 . let @xmath65 be the earliest intersects an observer s worldline multiple times . for example , in cylindrical spacetime with metric @xmath66 , a null geodesic from a point @xmath55 intersects a constant @xmath67 observer infinitely many times . in this case , our definition of @xmath65 is the first future point of intersection . similar remarks hold for @xmath68 . ] proper time of the observer such that there is a future - directed null geodesic originating from @xmath55 to @xmath69 , and let @xmath68 be the latest time such that there exists a future - directed null geodesic originating from @xmath70 to @xmath55 . if there exist both @xmath65 and @xmath68 , then we say that @xmath55 is in causal contact with the observer . following , we shall call the set of all points in causal contact with the observer @xmath56 the _ diamond _ of @xmath56 . the diamond is thus defined to be the set of all points which can both receive light signals from and send light signals to the observer @xmath56 , in other words , those points which can exchange information with @xmath56 . equivalently , the diamond is defined to be the intersection of the past of @xmath56 and the future of @xmath56 . we may also talk about the horizon for an observer @xmath56 , most generally , as the boundary of the diamond of the observer . we define the scalar function @xmath71 for each point @xmath55 in the diamond of an observer to be t(p ) : = ( _ p + _ p^)/2 , [ prop_time ] and we say @xmath55 is simultaneous to the observer at @xmath71 . in the case when the point @xmath55 is lying on @xmath56 , i.e. @xmath72 for some @xmath73 , we define @xmath74 . given a time @xmath73 , we define the _ surface of simultaneity _ @xmath75 to be the set of all points @xmath55 in the diamond for which @xmath76 . since each point @xmath55 of the diamond belongs to exactly one @xmath77 , we obtain a foliation @xmath33 of the entire diamond . furthermore , we have a timelike vector field @xmath78 everywhere orthogonal to @xmath33 . the simultaneity of our definition depends on the observer s entire history ; that is , the surface of simultaneity @xmath75 at time @xmath73 depends on the observer s trajectory @xmath56 for all @xmath3 . this should not be taken as a flaw of our definition . firstly , our definition using light signal communication has a clear operational meaning . secondly , in minkowksi spacetime with the usual coodinates @xmath79 , the constant @xmath3 surfaces are the usual planes of simultaneity for an inertial observer with constant @xmath80 , and the specification of the entire trajectory of constant @xmath80 is essential . thirdly , if two observers locally coincide around a common point , their surfaces of simultaneity through that point will also be locally the same . one could instead naively define simultaneity surface @xmath75 to be the set of all points lying in geodesics orthogonal to the observer s trajectory @xmath56 at @xmath73 . this definition of @xmath75 depends only on the point @xmath81 and its tangent vector , and for an arbitrary observer @xmath56 in minkowski spacetime it gives the usual simultaneity plane for the inertial observer tangent to @xmath56 at @xmath81 . however , the physical meaning of these surfaces for an arbitrary observer becomes unclear far from the worldline . furthermore , these surfaces may intersect within the observer s diamond , so that the `` time '' of the point of intersection becomes ambiguous under this definition . our definition of simultaneity does not suffer from the same problem , as each point in the diamond of the observer belongs to exactly one simultaneity surface . we have defined a scalar function @xmath3 and a corresponding foliation @xmath33 of the diamond region of an arbitrary observer . application of this idea to the formalism summarised in section 2 implies that we should treat the diamond as the physically relevant part of spacetime for the observer , on which the vector space @xmath5 of real klein - gordon solutions is defined . operationally , only those events in the diamond of an observer are physically relevant to him . if there exists a future - directed null geodesic originating from a point in @xmath56 to a point @xmath55 , then one may naively say that the observer can send a signal to @xmath55 . however this has no causal effect on him if there is no future - directed null geodesic from @xmath55 to @xmath56 . the only way that the observer is able to operationally determine whether @xmath55 has received his signal is if in principle he can receive a reply from @xmath55 . similarly , any event @xmath61 to which there is no future - directed null geodesic from @xmath56 connecting is physically irrelevant . it is in fact a more or less usual practice in quantum field theory that one interprets only the degrees of freedom in the region causally connected to the observer as the physically relevant ones . in the example of the rindler observers , only the field in the rindler wedge is considered relevant . for observers staying outside of a black hole , only the field outside the black hole horizon is relevant . and for an inertial minkowski observer of a finite - lengthed worldline , only the field in the observer s diamond region is relevant . rather than interpreting the physically relevant spacetime region for a given observer on a case by case basis , we here regard defining quantum field theory on only the diamond for an arbitrary observer as an axiom of our framework . the formalism in section 2 requires that each surface @xmath32 of the foliation be a cauchy surface . our slicing of the diamond for an arbitrary observer defined earlier in this section , however , does not sastisfy this requirement in general . it may indeed be the case that all simultaneity surfaces of the observer are cauchy surfaces for his diamond , as in the case of the rindler observer . but it may also be that at least one but not all simultaneity surfaces are cauchy surfaces , as in the case of an inertial minkowski observer with a finite worldline ; or that none of the simultaneity surfaces are cauchy surfaces , such as a co - moving observer in the milne universe , even though the spacetime is globally hyperbolic . in the next section , we shall construct a unitary quantum theory in the case that all simultaneity surfaces are cauchy surfaces . the construction for the more general cases has to remain future work . however , we argue here that the limitation of the current formulation should not be regarded as a fundamental flaw of our conceptual approach . we instead interpret this restriction as revealing the importance of cauchy surfaces in the formulation of a unitary theory and the preservation of information . we shall now speculate on the qualitative features of the theory for the more general cases . let @xmath1 and @xmath82 be an observer s proper time with @xmath83 . generally , when surface @xmath84 is not contained in the future cauchy development @xmath85 of surface @xmath86 , one can not determine the data on @xmath84 and hence the physical state at @xmath2 from the data on @xmath86 , and therefore the dynamics of the theory is non - deterministic . conversely , if all of @xmath84 lies in @xmath85 but not all of @xmath86 lies in @xmath87 , where @xmath87 is the past cauchy development of @xmath84 , then evolution of the field from @xmath86 to @xmath84 is deterministic but non - invertible and hence there is a loss of information . there will exist timelike curves crossing @xmath86 but not @xmath84 , which physically means that there will exist worldlines of objects leaving the observer s horizon before he reaches @xmath2 . in the case where only one of the simultaneity surfaces @xmath75 is a cauchy surface , such as the example in , one can determine the field throughout the entire diamond from @xmath75 . one can predict the future of @xmath75 , but information is lost as the field evolves from @xmath75 to a future simultaneity surface . one can also `` postdict '' at @xmath73 what happened before that time , although before @xmath73 he can not predict what happens at @xmath73 : there is more for him to learn , but in retrospect nothing is surprising . we now consider the implications of the above discussion for the example of a gravitationally collapsing black hole without evaporation . as shown in figure 2 ( a ) , an arbitrary observer that does not fall into the black hole has a diamond corresponding to spacetime outside the black hole horizon . if the causal structure and the simultaneity surfaces are correctly depicted in our figure where all simultaneity surfaces are cauchy surfaces , then we see that the observer s quantum field theory is unitary and that no information is lost . this is consistent with the intuition that an observer remaining outside the black hole never in his finite proper time sees an object falling behind the horizon and no information carried by the in - falling object could ever disappear from the sight of the observer . even though the observer can not access spacetime behind the horizon , that region is physically irrelevant to him : he does not know everything , but at least he knows what he knew . in the spacetime of black hole with evaporation , however , we speculate that not all simultaneity surfaces will be cauchy surfaces for the diamond of the observer staying outside of the black hole , as depicted in figure 2 ( b ) . we expect that in the figure , @xmath84 will lie in @xmath85 , but @xmath86 will not be contained in @xmath87 . hence it follows that evolution will be deterministic , but information will be lost , a profound and long renowned proposal due to hawking . wald has already pointed out that there should generally be a loss of information associated with evolution of a cauchy surface to a non - cauchy surface . here we have further developed the idea . with our one - parameter family of hypersurfaces @xmath33 constructed previously , we are able to provide wald s `` evolution of surfaces '' a precise mathematical meaning . physically , we give these `` surfaces '' an operational meaning as the simultaneity surfaces of a given observer . moreover , the question of whether or not a surface is a cauchy surface is now addressed with respect to the observer s diamond . we have emphasised that the observer should play a fundamental role in the discussion of a possible non - unitary quantum theory . we have proposed that hawking s original insight that the laws of physics may allow for loss of information can be much generalised . information can be lost not only in black hole evaporation but precisely whenever one s surface of simultaneity evolves from a cauchy surface to a non - cauchy surface , whether this be due to the background spacetime or to the observer s motion in that background . we emphasise that because our formulation reduces to the standard quantum field theory for inertial observers in minkowski spacetime , allowing for a non - unitary theory for general observers in curved spacetime can not be in violation of the usual laws of physics in flat spacetime . and as we have argued that quantum field theory should be formulated in an observer - dependent way , the notion that information loss is observer - dependent does not mean that it is not fundamental . here we do not intend to claim a resolution to the paradox of `` black hole information loss '' . rather , we have merely speculated on the qualitative features that a general quantum field theory for an arbitrary observer might possess . to consolidate our ideas about information loss , we would need a mathematical scheme for tracing out field degrees of freedom along the worldline of the observer and a corresponding detailed analysis . moreover , we have adopted a semi - classical approximation , although we expect some features of the current theory , in particular the importance of causal structure on possible loss of information , to remain in a more complete theory . we have aimed at constructing a quantum theory for an arbitrary observer . to compare physical descriptions between arbitrary observers it is therefore very important to study quantum state transformation : specifically given a quantum state for an observer @xmath88 at time @xmath73 , we would like to determine the corresponding state for another observer @xmath89 at time @xmath90 . it is conceivable that when the simultaneity surface @xmath91 for observer @xmath89 is a subset of @xmath92 for observer @xmath88 , the state transformation from @xmath88 at @xmath73 to @xmath89 at @xmath90 should be defineable . furthermore , one would expect tracing out the extra degrees of freedom on @xmath92 to be necessary , so that a pure state for @xmath88 would in general be a mixed state for @xmath89 . finally , we require such a state transformation to be consistent with quantum state evolution . we will not however develop these any further . we have constructed a one - parameter family of fock spaces . the scalar function @xmath3 on which this construction depends is the time of each event for the observer defined in section 3 , and the corresponding foliation of spacetime is the simultaneity surfaces of constant @xmath3 . we have also argued that the vector space v of real solutions of klein - gordon equation should be defined on the diamond of the observer , which is the set of all points that can both receive light signals from and send light signals to the observer . in this section , we turn to constructing the dynamics of quantum states for our theory : given a quantum state for an observer at his proper time @xmath1 , what is the evolved quantum state at another time @xmath2 ? in other words , we need to construct a two - parameter mapping @xmath93 such that : a. @xmath0 is an isomorphism , i.e. a complex - linear bijection ; [ a ] b. the inner product of @xmath94 and @xmath95 is equal to @xmath96 , for all @xmath97 and @xmath98 in @xmath53 ; and [ b ] c. @xmath99 . [ c ] it then follows from these properties that @xmath100 on @xmath101 and that @xmath102 . for notational simplicity , we will concentrate on some arbitrary choice of @xmath103 and @xmath104 . we use subscript @xmath3 on various symbols to denote its dependence on @xmath3 and/or @xmath31 , and subscript @xmath105 for the case of @xmath106 . in addition , we denote @xmath107 as @xmath108 , although it should be understood that @xmath108 implicitly depends on a choice of @xmath109 . consider a state of the form @xmath110 , where @xmath111 is the vacuum in @xmath112 . we define our @xmath108 by u_t ( c_0^+(_1 ) c_0^+(_n ) |0_0 ) = ( u_t c_0^+(_1 ) ) ( u_t c_0^+(_n ) ) ( u_t |0_0 ) , [ 3 ] where we have used the same symbol @xmath108 to denote the mapping from @xmath112 to @xmath101 as well as the mapping from the complex vector space _0 : = \ { c_0^+ ( ) + c_0 ^ - ( ) : , , , v } [ aaa ] to @xmath113 defined similarly to ( [ aaa ] ) . the nature of @xmath108 , i.e. whether it is on @xmath112 or @xmath114 , should be clear from the argument on which it is acting . once we have defined @xmath108 on both @xmath114 and @xmath111 , generalising the action of @xmath108 to the entire fock space @xmath112 via complex - linearity and to density matrices should be straightforward . we shall then show that the mapping is well - defined and satisfies the three properties ( [ a ] ) , ( [ b ] ) , and ( [ c ] ) . + + we now construct the map and @xmath115 are mappings on different fock spaces and hence can not be directly related by `` = '' . ] from @xmath114 to @xmath113 . our construction is guided by the following intuition : there is a `` natural '' change of creation / annihilation operators under a change of @xmath31 ; and there is a change of creation / annihilation operators induced by the evolution of their underlying classical fields . we now capture the contribution of @xmath31 to @xmath108 . define _t^:= \{c^_t ( ) : v } , [ 4.3 ] where @xmath116 . we have @xmath117 , where @xmath118 denotes isomorphism . the subspace @xmath119 is naturally isomorphic to @xmath120 via c^+_t ( ) |_t where recall @xmath120 is the complex vector space constucted out of @xmath121 as in section 2 . similarly , the subspace @xmath122 is isomorphic to @xmath123 , the conjugate of @xmath120 . observe that there is also a natural isomorphism from @xmath120 to the @xmath124 @xmath31-eigensubspace of @xmath125 . we shall denote this eigensubspace as @xmath126 , and clearly @xmath127 where @xmath128 is the projection from @xmath129 onto @xmath126 . this isomorphism is given by |_t p^+_t . similarly , @xmath123 is isomorphic to @xmath130 , the @xmath131 eigensubspace of @xmath129 , via @xmath132 . above all , we have a natural isomorphism between @xmath113 and @xmath129 . we denote this isomorphism as @xmath133 , that is : f_tc^_t ( ) = p^_t , v,= 1 . [ 4 ] this is the equation we have been looking for , capturing the intuition that different @xmath31 corresponds to different notions of `` + '' and `` - '' , embodied as different projections in our formalism . it is important to notice that @xmath129 is independent of @xmath31 . this disentangles the contribution of @xmath31 from the contribution of the classical evolution to @xmath108 , which we now describe . let @xmath134 be a two - parameter automorphism on @xmath5 . we denote its complexification also by @xmath135 . it is obvious that this complexification defines an automorphism on @xmath129 . again , we denote @xmath136 as @xmath137 for some arbitrary @xmath109 , @xmath104 . to capture the contribution of @xmath137 to @xmath108 , combined with that of @xmath31 , we now define the map @xmath138 to be @xmath139 : rcl _0 & & _t + & & + v^ & & v^ + [ diagram ] having defined @xmath108 , we now determine the image of operators in @xmath114 under this map . since @xmath140 is identity on @xmath129 : rl u_tc^_0 ( ) & = f_t^-1 ( p^+_t + p^-_t ) u_t f_0 c^_0 ( ) + + & = f_t^-1(p^+_t u_t p^_0 ) + f_t^-1(p^-_t u_t p^_0 ) , where we used ( [ 4 ] ) . it is then straightforward to show that the action of @xmath108 on @xmath114 is u_t c^_0 ( ) = _ c^_t(^_|t ) , [ 5 ] where ^_|t = ( u_t - j_t u_t j_0 ) , [ soln ] with all greek indices taking the value @xmath141 . we shall henceforth drop the summation sign in any equation with repeated @xmath142 or @xmath143 indices , with one index up and the other down . we have defined @xmath138 and determined its action on @xmath114 as given by ( [ 5 ] ) . to finish the construction of @xmath108 on @xmath112 , it remains to specify how vacuum evolves . denote @xmath144 to be the state evolved from vacuum : @xmath145 . we shall only impose that @xmath144 is a uniquely defined unit vector in @xmath101 and that it satisfies the condition ( u_t c^-_0())|_t = c^_t(^-_|t ) |_t = 0 v. [ 6 ] before specifying @xmath144 and @xmath137 , and before showing that our construction satisfies ( [ a ] ) , ( [ b ] ) , and ( [ c ] ) , we first say a few words about vacuum and particle creation . from ( [ 5 ] ) , we see that @xmath146 will evolve to a pure creation operator if and only if @xmath147 and that @xmath148 will evolve to a pure annihilation operator if and only if @xmath149 . but @xmath150 , so the two conditions are equivalent . if no @xmath151 gets mixed , we have @xmath152 for all @xmath28 , i.e. j_t = u_t j_0 u_t^-1 . [ 7 ] in this case , @xmath153 . additionally , ( [ 6 ] ) reduces to c^-_t(u_t)|_t = 0v . the vector @xmath144 is then fixed to be @xmath154 , the vacuum in @xmath101 , since @xmath137 is an automorphism on @xmath5 . we then have u_t|_1 _n_0 = |(u_t_1)(u_t_2) (u_t_n)_t . [ vac_evol2 ] that is , a @xmath21-particle state in @xmath155 will evolve to a state in @xmath156 for all @xmath21 : particles are not created . the mapping @xmath108 in ( [ vac_evol2 ] ) is indeed complex - linear , as illustrated by rl u_t i|_1_2 _n_0 & = u_t |(j_0_1)_2 _n_0 + & = |(u_t j_0 _ 1)(u_t_2) (u_t_n)_t + & = |(j_t u_t _ 1)(u_t_2) (u_t _ n)_t + & = i |(u_t _ 1)(u_t_2) (u_t _ n)_t + & = i u_t |_1_2 _n_0 therefore , we see that particle creation is precisely due to @xmath31 failing to evolve in a way satisfying ( [ 7 ] ) . + + there are two conditions that the automorphism @xmath137 must satisfy . we now discuss the first condition motivated by condition ( b ) on @xmath108 . to begin with , we have that | _ 0 = 0| [ c^-_0 ( ) , c^+_0 ( ) ] |0_0 , |_0 , |_0 _ 0 , [ inn_prod ] and that the inner product between @xmath157 and @xmath158 is given by |(u_t c^-_0 ( ) ) ( u_t c^+_0())| _ t , [ 9 ] where we have used the property ( u_t c^_0())^= u_t(c^_0())^. [ 8 ] to see this property , notice that the right - hand side of ( [ 8 ] ) is rl u_t c^-_0 ( ) & = c^-_t(^-_-|t ) + & = c^-_t(^_|t ) + & = ( c^_t(^_|t))^ , where the final expression is simply the left - hand side of ( [ 8 ] ) . condition ( b ) then implies 0| [ c^-_0 ( ) , c^+_0 ( ) ] |0_0=|(u_t c^-_0 ( ) ) ( u_t c^+_0())| _ t. [ star ] using ( [ 6 ] ) , we can write the right - hand side of ( [ star ] ) as | | _ t. [ 10 ] noticing that @xmath159 and that by ( [ commutations ] ) the commutator in ( [ 10 ] ) is a multiple of @xmath160 on @xmath101 , we see ( [ star ] ) is equivalent to = u_t([c^-_0 ( ) , c^+_0 ( ) ] ) [ 11 ] where we have defined on the right - hand side u_t _ 0 = _ t. [ def1 ] on the other hand , for ( [ 3 ] ) to be well - defined , it is necessary that = 0 [ 12 ] that @xmath108 preserves the commutation relationship on all @xmath114 then follows from ( [ 8 ] ) , ( [ 11 ] ) , ( [ 12 ] ) , and complex - linearity . we now derive the condition on @xmath108 for this commutation to be preserved . a straightforward calculation using ( [ inner_prod ] ) , ( [ symplectic1 ] ) , and ( [ commutations ] ) shows that = i ( p^_t , p^_t ) _ t , [ 13 ] where we have extended the symplectic form @xmath10 given by ( [ symplectic ] ) to @xmath129 via complex - linearity . applying ( [ 5 ] ) and using the fact that both sides of ( [ 13 ] ) are bilinear , we have rl [ u_t c^_0 ( ) , u_t c^_0 ( ) ] & = + & = i ( p^_t ^_|t , p^_t ^_|t ) _ t . since rl p^_t ^_|t & = f_tc^_t(^_|t ) + & = f_tu_tc^_0 ( ) + & = u_tf_0 c^_0 ( ) + & = u_t p^_0 , where we have used ( [ 4 ] ) , ( [ diagram ] ) , and ( [ 5 ] ) , we have = i ( u_t p^_0 , u_t p^_0 ) _ t . hence = u_t([c^_0 ( ) , c^_0 ( ) ] ) [ comm ] will be satisfied if and only if ( u_t p^_0 , u_t p^_0 ) = ( p^_0 , p^_0 ) . [ 14 ] summing ( [ 14 ] ) over @xmath161 , one can show that this in turn is equivalent to ( u_t , u_t ) = ( , ) , v. [ 15 ] that is , @xmath108 preserves the commutation relation on @xmath114 if and only if @xmath137 preserves the symplectic form @xmath10 on @xmath5 ( or equivalently on @xmath129 ) . with @xmath137 satisfying ( [ 15 ] ) , it is easy to generalise the proof that @xmath108 satisfies condition ( b ) for states in @xmath162 to the general case of all states in @xmath112 . hence , in order for definition ( [ 3 ] ) of @xmath108 to be well - defined and for @xmath108 to satisfy condition ( [ b ] ) , the first condition on @xmath137 which we impose is that it satisfies ( [ 15 ] ) . the second condition , motivated by condition ( [ c ] ) , which we shall impose on @xmath137 is : u(t_3 , t_1 ) = u(t_3 , t_2 ) u(t_2 , t_1 ) . [ 17 ] + + with a choice of automorphism @xmath137 on v satisfying ( [ 15 ] ) and ( [ 17 ] ) , and a choice of @xmath144 satisfying ( [ 6 ] ) , we shall now show our @xmath108 on @xmath112 satisfies ( [ a ] ) , ( [ b ] ) , and ( [ c ] ) . first of all , we have shown ( [ b ] ) is satisfied . this in particular implies @xmath108 is injective . complex - linearity is obvious . hence to show condition ( [ a ] ) is satisfied , it remains to show that @xmath108 is onto . we now show @xmath108 is onto . we first show there exists a @xmath163 such that @xmath164 , the vacuum in @xmath101 . let @xmath163 be @xmath165 where @xmath166 is defined precisely in the same way as @xmath107 except for a change of `` @xmath105 '' and `` @xmath3 '' . then by construction using ( [ 6 ] ) , ( u(0 , t ) c^-_t())| _ 0 = 0 v. [ 4.33 ] since by using ( [ 6 ] ) and ( [ comm ] ) we have u_t ( c^_0(_)|_0 ) = ( u_t c_0^(_))u_t |_0 _ v , |_0 ( _ 0 ) , [ 16 ] we may apply @xmath108 on both sides of ( [ 4.33 ] ) to obtain ( u_t u(0 , t ) c^-_t())(u_t |_0 ) = 0 . [ 18 ] observe that @xmath167 on @xmath113 is f^-1_t u(t , 0 ) f_0 f_0 ^ -1 u(0 , t ) f_t = _ t where @xmath168 on @xmath5 follows for our proof . ] from ( [ 17 ] ) and the fact that @xmath169 is an automorphism on @xmath5 . equation ( [ 18 ] ) now becomes c^-_t ( ) ( u_t |_0 ) = 0 v with a unique solution u_t|_0 = |0_t . [ 19 ] we now conclude @xmath108 on @xmath112 is onto , using ( [ 19 ] ) , ( [ 16 ] ) , and the fact that @xmath108 from @xmath114 to @xmath113 is an isomorphism . furthermore on @xmath101 we have u(0 , t ) = u_t^-1 . [ inv ] finally , we prove that condition ( [ c ] ) is satisfied . applying @xmath170 for an arbitrary @xmath171 to ( u_t c^-_0 ( ) ) u_t |0_0 = 0 v which is simply ( [ 6 ] ) , we have that ( u(t^ , t ) u_t c^-_0 ( ) ) u(t^ , t ) u_t |0_0 = 0 , where we have used ( [ 16 ] ) again . observe that , using ( [ diagram ] ) , the condition ( [ 17 ] ) on @xmath137 is equivalent to condition ( c ) as a mapping on @xmath114 . we have that ( u_t^ c^-_0 ( ) ) u(t^ , t ) u_t |0_0 = 0 . [ 20 ] apply @xmath172 to ( [ 20 ] ) and use ( [ 16 ] ) to obtain c^-_0()(u(0 , t^ ) u(t^ , t ) u_t|0_0)=0 , v. hence , u(0 , t^ ) u(t^ , t ) u_t|0_0=|0_0 . using ( [ inv ] ) again , this is simply u(t^ , t ) u_t|0_0= u_t^ |0_0 . [ 21 ] we finally conclude that condition ( [ c ] ) is satisfied over all @xmath112 , using ( [ 21 ] ) , ( [ 16 ] ) , and the fact that condition ( [ c ] ) is satisfied as mappings on @xmath114 . + + we now specify the evolution of vacuum from @xmath111 to @xmath173 satisfying ( [ 6 ] ) . to do so , we shall first extend the definition of the evolution map @xmath174 to sums of products of operators in @xmath114 via u_t ( _ i c^_i_0(_i ) ) : = _ i u_t c^_i_0(_i ) [ def ] and complex linearity . clearly this mapping is well defined , following from definition ( [ def1 ] ) and that @xmath108 is linear on @xmath114 and preserves commutation ( [ comm ] ) . furthermore , this extended @xmath108 is invertible , satisfies the desired composition as in condition ( c ) at the beginning of this chapter , and commutes with taking the adjoint of its argument as in ( [ 8 ] ) . the evolution of vacuum is then given by applying ( [ def ] ) . observe that the vacuum density matrix can _ formally _ be written as |0 0|_0 = _ n=1^(_0 - n_0/n ) [ formal ] where the total number operator @xmath175 on @xmath112 is defined by @xmath176 for all @xmath177 . this can be seen by noticing that @xmath178 and @xmath179 are true for both sides of ( [ formal ] ) , for @xmath177 and @xmath180 with at least one of @xmath181 or @xmath21 being non - zero . the evolved state is then given by _ t:=u_t(|0 0|_0 ) = _ n=1^(_t - ( u_tn_0)/n ) . [ vac_evol ] the operator @xmath182 is also defined using ( [ def ] ) . we write @xmath175 as n_0 = _ k c^+_0(_k ) c^-_0(_k ) where summation is over all @xmath183 such that @xmath184 is an orthonormal basis for @xmath162 . then we have that u_tn_0 = _ k ( u_tc^+_0(_k))(u_t c^-_0(_k ) ) . one can show that this definition of @xmath182 is independent of a choice of an orthonormal basis @xmath184 for @xmath162 . in general , ( [ vac_evol ] ) does not define a finite - trace operator on @xmath101 , in which case strictly speaking our construction of unitary dynamics fails mathematically , although physically meaningful predictions can still be made . this , our only problem with infinity , resembles the renormalisation problem in the usual field theory of quantum operators . in a more complete theory , the dimension of the physical hilbert space should probably be replaced by a finite one , so we believe this difficulty does not indicate that our formulation is fundamentally towards a wrong direction . in the case when ( [ vac_evol ] ) does converge to a finite - trace operator , however , we shall show that @xmath185 defined in ( [ vac_evol ] ) is a pure density matrix . first of all , ( [ def ] ) maps hermitian operators to hermitian operators , using ( [ 8 ] ) . it then follows @xmath186 . secondly , we have @xmath187 , since u_t ( |0 0|_0 ) u_t ( |0 0|_0 ) = u_t ( |0 0||0 0|_0 ) = u_t ( |0 0|_0 ) where we have used ( [ def ] ) . it follows from these two properties that @xmath185 is a projector with eigenvalues 1 or 0 , so that we can write @xmath185 in an orthonormal basis as k=1^n |k k| where @xmath21 is the trace of @xmath185 if it is finite . we shall now show in this case @xmath188 and that @xmath185 defined in ( [ vac_evol ] ) is a pure density matrix . evolving the identity @xmath189 for an arbitrary @xmath190 between 1 and @xmath21 by @xmath191 we obtain |0 0|_0 u(0,t ) ( |i i| ) = u(0,t ) ( |i i| ) where we have used ( [ def ] ) . similarly , u(0,t ) ( |i i| ) |0 0|_0=u(0,t ) ( proportional to @xmath192 , that is , @xmath193 is proportional to @xmath194 for all @xmath190 . this is only possible if @xmath188 , i.e. @xmath195 for some pure state @xmath144 as desired . we have specified the evolution of vacuum as in ( [ vac_evol ] ) , and we have demonstrated that it indeed defines a pure state @xmath144 . it remains to check that @xmath144 satisfies condition ( [ 6 ] ) , i.e. @xmath196 . this is the case if and only if c^_t(^-_|t ) | |_t c^-_t(^-_|t ) = 0 . but by ( [ def ] ) the left hand side of this is u_t c^-_0 ( ) u_t(|0 0|_0 ) u_t c^+_0()=u_t ( c^-_0 ( ) |0 0|_0 c^+_0 ( ) ) which is simply 0 as desired . we say a final word about particle creation . if the quantum state was vacuum at time zero , then the expected particle number at time @xmath3 is @xmath197 , where @xmath198 is the total number operator for @xmath101 defined in the same way as @xmath175 for @xmath112 . for the @xmath199 particle number operator @xmath200 , using that @xmath108 is trace - preserving , we have ( ||_t c^+_t ( ) c^-_t ( ) ) = ( |0 0 |_0 ( u(0,t ) c^+_t ( ) c^-_t ( ) ) ) . it is then straightforward to calculate that the expected @xmath199 particle number seen by the observer at his proper time @xmath3 is given by the square of the norm of @xmath201 , where ^-_+|0 : = ( u_t^-1+j_0 u_t^-1 j_t ) . see ( [ soln ] ) for a comparison of definitions . + + finally it remains to specify the classical evolution @xmath137 , the automorphism on @xmath5 , satisfying ( [ 15 ] ) and ( [ 17 ] ) . one natural choice seems to be the following . for each @xmath28 , we define @xmath202 to be the unique solution with cauchy data in the definition of cauchy data which is slightly different from the convention in section 2 . ] on @xmath32 rl u_t |__t = & |__0 + n^a_a(u_t)|__t = & n^a_a|__0 where we have used a natural point identification map from @xmath203 to @xmath32 given by the integral curves of @xmath204 . clearly this mapping is defined independent of the labelling of the 3-parameter family of the integral curves of @xmath204 . it can be shown that it indeed is an automorphism on @xmath5 satisfying ( [ 15 ] ) and ( [ 17 ] ) . however , it remains future work to check whether this choice of @xmath137 , although self - consistent , leads to the same physical conclusions as those in the various well - known cases . [ aside . one may attempt to define the classical evolution by `` pulling '' cauchy data back from @xmath32 to @xmath203 along @xmath78 , instead of `` pushing '' cauchy data as above . more precisely , one may define @xmath205 such that rl _ t |__0 = & |__t + n^a_a(_t)|__0 = & n^a_a|__t . + it can be shown that @xmath205 is also an automorphism on @xmath5 and satisfies ( [ 15 ] ) . however , it does not satisfy ( [ 17 ] ) . in fact @xmath206 , hence @xmath207 implies that @xmath208 , which is not the physically desired composition law . ] we summarise the main results of our discussion here . the goal of this paper is to propose a quantum field theory for an arbitrary observer in curved spacetime . to this end , we constructed a one - parameter family of fock spaces based on the formalism of ashtekar and magnon . each fock space is based on a hilbert space constructed from the vector space @xmath5 of real - valued klein - gordon solutions and a parameter - dependent complex structure @xmath31 . this construction requires a choice of scalar function @xmath3 such that each constant @xmath3 hypersurface @xmath32 is a spacelike cauchy surface . we then applied this mathematical formalism for an arbitrary observer to the region of spacetime which the observer can both send signals to and receive signals from . following , we have used the terminology `` diamond '' to refer to this region . we argue that radar time should be applied for the above function @xmath3 used in the construction of the fock spaces . physically this means @xmath32 , the set of points in the diamond with radar time @xmath3 , is the set of all events `` simultaneous '' to the observer at his proper time @xmath3 . our definition using light signal communication has a clear operational meaning and directly relates each point to the observer s local physical quantities . our slicing applies to all observers and reflects the causal structure of the underlying spacetime . although the diamond of an observer in general may not cover the entire spacetime , it is operationally the only region physically relevant to the observer . we therefore define the vector space @xmath5 of real klein - gordon solutions and our quantum field theory on the diamond of a given observer . we regard this as an axiom of our framework . in the case where all simultaneity surfaces of an observer are cauchy surfaces of his diamond , we have constructed a unitary dynamics where no information is lost : given a quantum state for an observer at his proper time @xmath1 , we constructed a two - parameter mapping @xmath0 from @xmath53 to @xmath54 that will give us the evolved state at time @xmath2 . we require our mapping to satisfy three conditions : that @xmath108 is an isomorphism ; that the mapping preserves the inner product ; and that the mapping satisfies @xmath209 . the action of @xmath108 on @xmath112 is specified by its action on the vector space @xmath114 of creation and annihilation operators given by ( [ aaa ] ) , as well as its action on the vacuum in @xmath112 . the construction is guided by the intuition that there is an evolution of operators in @xmath114 according to a change in the complex structure @xmath210 , as well as according to an evolution @xmath137 of the operators underlying classical fields , i.e. an automorphism on @xmath129 . the evolution of vacuum state is given by ( [ vac_evol ] ) and satisfies ( [ 6 ] ) . we have deduced that particle creation will take place if and only if @xmath31 fails to evolve according to @xmath211 . we have also speculated on features that the theory covering the more general cases might have . we have further developed wald s insight that there should generally be a loss of information when there is evolution of a cauchy surface to a non - cauchy surface . our one - parameter foliation of the diamond gives this `` evolution of surfaces '' a precise mathematical meaning . physically , we give these `` surfaces '' an operational meaning as the simultaneity surfaces of a given observer . moreover , the question of whether or not a surface is a cauchy surface is now addressed with respect to the observer s diamond . above all , we speculate that information will be lost precisely whenever an observer s surface of simultaneity evolves from a cauchy surface to a non - cauchy surface , whether this be due to the background spacetime or to the observer s motion in that background . this generalises hawking s original insight and takes the information loss of black hole evaporation as just a special case . + + there is much scope for future generalisations and applications of this work . 1 . we would like to generalise our current formulation to the cases when not all surfaces of simultaneity of an observer are cauchy surfaces for his diamond . as we evolve data from a cauchy surface to a non - cauchy surface , we need a general scheme to trace out the field degrees of freedom lost along the worldline of the observer . such a scheme would also consolidate our speculations about information loss . it is important to apply our framework to various examples , especially to see how our results would compare with experiments and with the results obtained by other formulations , such as with canonical quantization or with model particle detectors . + in particular , it would be very important to explore quantum field theory in frw cosmology using our formalism , where the simultaneity surfaces for a co - moving observer are not the usual surfaces of constant energy density . this offers the opportunity to test our theory with cosmological observations . we would like to generalise our formalism to higher - spin fields and interacting fields . one important extension of our work is to study quantum state transformation between arbitrary observers and its operational meaning . given a quantum state for an observer @xmath88 at time @xmath73 , we would like to know the corresponding state for another observer @xmath89 at time @xmath90 . we require such a transformation to be consistent with the state evolution we have already constructed . we constructed a unitary dynamics relating a quantum state at some time of an observer to a state at a later time . however we have not discussed , for an arbitrary observer in curved spacetime , how a quantum measurment projects quantum states nor how the notion of `` wavefunction collapse '' should be understood . one of the most intriguing aspects of quantum field theory in curved spacetime is the yet to be understood deep relationship between causal horizons and thermodynamics . firstly , it has been argued that the ultimate significance of the thermodynamics of black hole horizons hangs on the issue of its generalisation . in our framework , the diamond for an arbitrary observer provides a natural and most general notion of causal horizons , and by defining quantum theory over this region , our formalism naturally establishes a link between quantum theory and causal horizons . + secondly , thermodynamics of causal horizons can be studied using a statistical mechanics approach . this requires a concept of particles , and our formulation indeed provides a notion of particles for each observer . on the other hand , it would also be very enlightening from our formalism to obtain an expression directly relating thermodynamic quantities and properties of causal horizons without needing to consider any particle spectrum . + finally , it has been argued that entropy is an observer - dependent quantity . our observer - dependent quantum field theory may be an ideal framework for a further investigation of the observer - dependence of thermodynamic quantities . we propose that observer - dependence as a fundamental feature of quantum field theory should be taken much further . since the quantum states of particles in general depend on the observer , therefore different observers will have different @xmath212 , the expectation values of the energy - momentum associated with their respective quantum states . by einstein s field equations , the quantum field s back - reaction on the spacetime metric will also be observer - dependent and hence so will the spacetime metric itself . indeed , the possibility that spacetime itself may be observer - dependent has been suggested by gibbons and hawking . such a profound suggestion merits further investigation which our framework may be well - suited to pursue , and this investigation may indeed be the correct path leading to a complete and consistent union of quantum theory and relativity . i would like to thank rex liu for discussions , david wiltshire and steffen gielen for comments and jonathan oppenheim for criticisms .

we study effective lattice actions describing the polyakov loop dynamics originating from finite - temperature yang - mills theory . starting with a strong - coupling expansion the effective action is obtained as a series of-invariant operators involving higher and higher powers of the polyakov loop , each with its own coupling . truncating to a subclass with two couplings we perform a detailed analysis of the statistical mechanics involved . to this end we employ a modified mean field approximation and monte carlo simulations based on a novel cluster algorithm . we find excellent agreement of both approaches concerning the phase structure of the theories . the phase diagram exhibits both first and second order transitions between symmetric , ferromagnetic and anti - ferromagnetic phases with phase boundaries merging at three tricritical points . the critical exponents and at the continuous transition between symmetric and anti - ferromagnetic phases are the same as for the 3-state potts model .

hep-lat0605012

in two seminal papers svetitsky and yaffe have tentatively linked the finite - temperature phase transitions in ` hot ' gauge theories to the simpler order - disorder phase transitions of spin models @xcite . in the general case their conjecture may be stated as follows : the effective theory describing finite - temperature yang - mills theory with gauge group @xmath3 in @xmath4 dimensions is a spin model in @xmath5 dimensions with a global symmetry group given by the centre @xmath6 of the gauge group . a somewhat stronger version of the conjecture can be formulated if the phase transitions in questions are of second order . in this case yang - mills theory and spin model fall into the same universality class and the critical exponents coincide . this has been convincingly demonstrated for @xmath7 @xcite in @xmath8 . however , continuous phase transitions in hot gauge theories are not generic and hence the universality statement is almost empty at least in 3 + 1 dimensions @xcite . on the other hand , the more general version of the statement has already been used by svetitsky and yaffe to argue that the phase transition for @xmath9 yang - mills theory must be first order for @xmath8 as there is no @xmath0 rg fixed point in this case . since then this has been firmly established by a number of lattice calculations @xcite . the conjecture implies that the effective theories may be formulated as ` polyakov loop models ' @xcite . for @xmath3 this means that the macroscopic dynamical variables have to reflect the complete gauge invariant information contained in the ( untraced ) polyakov loop , which in lattice notation is given by @xmath10 \equiv \prod_{t=1}^{n_{\tau } } u_{t,{\mbox{\scriptsize\boldmath$x$}};0 } \ ; .\ ] ] this is a temporal holonomy winding around the compact euclidean time direction of extent @xmath11 . for gauge groups with nontrivial centre the traced polyakov loop , @xmath12 with the trace being taken in the defining representation for the ease of later notation . ] serves as an order parameter for the deconfinement phase transition . the phase transition goes along with spontaneous breaking of the centre symmetry resulting from _ non - periodic _ gauge transformations under which @xmath13 the deconfined broken - symmetry phase at sufficiently large wilson coupling @xmath14 is characterised by @xmath15 ( see @xcite for a review ) . recently it has been found that gauge groups with trivial centre may also lead to a deconfinement transition depending on the _ size _ of the gauge group @xcite . at this point the choice of dynamical variables needs to be addressed . under _ periodic _ gauge transformations @xmath16 the holonomy ( [ eq : utrpoldef ] ) transforms as @xmath17 which leaves its eigenvalues and , as a consequence , its trace ( [ eq : trpoldef ] ) invariant . the eigenvalues are permuted arbitrarily by gauge transformations corresponding to weyl reflections @xcite . this invariance is taken into account by constructing symmetric polynomials in the @xmath18 eigenvalues . from unimodularity the product of the eigenvalues is @xmath19 and there are only @xmath20 independent polynomials , for example the traces @xmath21 for @xmath22 . these in turn are in one - to - one correspondence with the characters of the @xmath23 fundamental representations ( see below ) . hence , for @xmath7 @xmath24 ( which is real ) is sufficient while @xmath9 requires @xmath24 and @xmath25 , the latter being a linear combination of @xmath26 and @xmath27 . only for @xmath28 traces of higher powers of @xmath29 are needed as independent dynamical variables @xcite . in two recent papers @xcite we have studied polyakov loop models on the lattice for the simplest non - abelian gauge group @xmath7 . the models have been derived using strong - coupling techniques at small wilson coupling @xmath30 and a newly developed inverse monte carlo ( imc ) method which works for arbitrary values of @xmath30 . the latter method allows for a mapping of yang - mills theory at a certain value of @xmath30 to any appropriately chosen polyakov loop model of the form @xmath31 where the summation is over nearest neighbours and group representations @xmath32 , @xmath33 . the group character @xmath34 is the trace of the polyakov loop in representation @xmath32 , @xmath35 \equiv { \text{tr}}_i \ , { \mathfrak{p}_{{\mbox{\footnotesize\boldmath$x$ } } } } \ ; .\ ] ] all characters @xmath34 are polynomials in the characters corresponding to the fundamental representations . for @xmath7 the simplest model is of ising type , @xmath36 as first suggested in @xcite ( see also @xcite ) . the output of the imc routines are the @xmath30-dependent effective couplings @xmath37 . even without particular knowledge of these one may study the models ( [ eq : polmodel ] ) in their own right as statistical field theories . the svetitsky - yaffe conjecture may then be utilised to deduce information about the yang - mills phase transition . for @xmath7 we have been able to show that the mean - field analysis of the effective models yields a surprisingly good agreement with the monte carlo analysis of both the model itself and the underlying yang - mills dynamics . it should be stressed at this point that the models ( [ eq : polmodel ] ) are not just simple scalar field theories with a linear target space as all characters ( [ eq : chars ] ) take values in a compact space parametrised by the @xmath38 fundamental characters . hence , in the lattice path integral each lattice site is endowed with the reduced haar measure on the gauge group rather than a lebesgue measure . this paper presents our first steps to generalise the results of @xcite to the gauge group @xmath9 . we label the characters by two integers @xmath39 and @xmath40 which count the numbers of fundamental and conjugate representations ( ` quarks ' and ` antiquarks ' in the @xmath9 flavour language ) required to construct the representation @xmath41 . equivalently , these integers characterise the horizontal extensions of @xmath9 young tableaux ( see fig . [ fig : pq ] ) . character labels and @xmath9 young tableaux . ] under the @xmath0 centre transformations the characters transform according to the rule @xmath42 so that the most general centre - symmetric effective action with nearest - neighbour interaction may be written as @xmath43 = \sum_{\substack{{\langle}{\mbox{\scriptsize\boldmath$x$ } } { \mbox{\scriptsize\boldmath$y$}}{\rangle } , \ , pq , \ , p'q ' \\ p + p ' = q + q ' \ , \mathrm{mod } \ , 3 } } \lambda_{pq , p'q ' } \ , \big ( \chi_{pq } ( { \mbox{\boldmath$x$ } } ) \ , \chi_{p'q ' } ( { \mbox{\boldmath$y$ } } ) + \text{h.c . } \big ) \\ + \sum_{\substack{{\mbox{\scriptsize\boldmath$x$ } } , \ , pq \\ \mathrm{mod } \ , 3 } } \lambda_{pq,00 } \ , \big ( \chi_{pq } ( { \mbox{\boldmath$x$ } } ) + \text{h.c . } \big ) \ ; .\end{gathered}\ ] ] this coincides with the ansatz suggested by dumitru et al.@xcite . the first sum in the effective action consists of hopping terms involving monomials of the form @xmath44 or @xmath45 ( and h.c . ) while the second sum is a ` potential ' term containing only powers @xmath46 ( and h.c . ) localised at single sites . the remainder of the paper is organised as follows . in section [ sec : sce ] we confirm the ansatz ( [ eq : ssu3 ] ) by means of a strong coupling ( small-@xmath30 ) expansion for @xmath47 $ ] . for a restricted set of couplings ( and hence representations ) we investigate the resulting @xmath0 models by minimising the classical action ( section [ sec : qca ] ) followed by an improved mean field analysis in section [ sec : mf ] . in agreement with the svetitsky - yaffe conjecture we find a first order phase transition from the symmetric to a ferromagnetic phase . our improved mean field analysis already reveals an interesting phase structure with four different phases : two distinct ferromagnetic phases , one symmetric and one anti - ferromagnetic phase . besides the first order transitions we detect second order transitions from the symmetric to an anti - ferromagnetic and to a ferromagnetic phase . the continuous transition from the symmetric to the anti - ferromagnetic phase is to be expected since the models reduce to the @xmath48-state potts model for polyakov - loops having values in the centre of the gauge group . section [ sec : mcsim ] contains the results of our extensive monte carlo simulations performed on a linux cluster where we have implemented the powerful package jenlatt . similarly as for @xmath7 the mean field and numerical results are in surprisingly good agreement . this is presumably due to the existence of ( three ) tricritical points . depending on the order of the transition we localise the critical lines with either a metropolis , a multicanonical or a modified cluster algorithm . in addition we have checked that the critical exponents @xmath1 and @xmath2 at the second order transition to the anti - ferromagnetic phase agree with those of the @xmath48-state potts model in @xmath48 dimensions . finally , in section [ sec : so ] we wrap up with discussion and conclusions . in this section we briefly recapitulate the strong - coupling ( small-@xmath30 ) expansion for the @xmath9 wilson action at finite temperature @xcite . it is known that the leading order result ( @xmath49 ) stems from ladder diagrams that wind around the temporal lattice extension and corresponds to an ising type model analogous to ( [ eq : su2ising ] ) . by going beyond the leading order we will encounter higher group representations / characters and hence have an independent confirmation of the ansatz ( [ eq : ssu3 ] ) for the effective action . our starting point is the standard wilson action , @xmath50 where the summation over plaquettes contains both temporal and spatial links . the effective action @xmath51 $ ] is introduced as usual by inserting an appropriate ( group valued ) delta function in the path integral , @xmath52 } \\ & \equiv \int { \mathfrak{dp}}\ , e^{-{s_{\text{\tiny eff}}}[{\mathfrak{p } } ] } \ ; . \end{split}\ ] ] while it is not known how to perform the final integration analytically for the full wilson action one can straightforwardly integrate order by order in @xmath30 . thus we expand the boltzmann weight @xmath53 and integrate separately over temporal and spatial links , @xmath54 . adopting temporal gauge we set all temporal links equal to unity apart from the links in the first timeslice which according to ( [ eq : utrpoldef ] ) may be identified with the polyakov loop . integrating out all spatial link variables we obtain the partition function @xmath55 \right ) \ ; , \end{split}\ ] ] where we have introduced the operators @xmath56 . the effective action may finally be written as @xmath57 where the coefficients @xmath58 are related to the operators @xmath59 via the linked - cluster theorem . in the remainder of this section we are going to determine the explicit form of the effective operators @xmath58 . we first rewrite the wilson - boltzmann weight ( [ eq : boltzmann1 ] ) as @xmath60 and expand the single - plaquette contribution in terms of @xmath9 characters , @xmath61 with @xmath62 ( see fig . [ fig : pq ] ) . all @xmath30 dependence now resides in the generalised fourier coefficients @xmath63 which accordingly may be further expanded , @xmath64 an explicit computation of the @xmath65 shows that these vanish whenever the representation labels become sufficiently large , namely if @xmath66 @xcite . this yields the important intermediate result that to any given order @xmath67 in the strong - coupling expansion only a _ finite _ number of characters contributes , @xmath68 the integrations over the spatial links are standard group integrals which can be found in the texts @xcite . the upshot is that only connected link arrangements ( ` polymers ' ) wrapping around the temporal extent of the lattice yield nonvanishing contributions . the leading term is a ladder diagram consisting of @xmath69 plaquettes each of which contributes a factor of @xmath30 implying a total contribution of @xmath70 . the associated operator is explicitly found to be @xmath71 a typical operator of order @xmath72 is given by @xmath73 in terms of which the wilson - boltzmann weight ( [ eq : boltzmann2 ] ) becomes @xmath74 expanding this to next - to - leading order ( @xmath75 ) yields the simple expression @xmath76 note , however , that some care has to be taken in interpreting products such as @xmath77 which by also contain disconnected pieces . upon expanding the logarithm by means of the linked - cluster theorem we are led to keep only connected contributions of the form @xmath78 making use of the character reduction formula @xmath79 products of characters at the same site may be reduced to single characters . as a consequence , the connected part of ( [ eq : s2n ] ) takes the explicit form @xmath80 for what follows it is useful to introduce the short - hand notation @xmath81 with the obvious term - by - term identifications as compared to ( [ eq : effactionlong ] ) . our conventions are such that all couplings in ( [ eq : effactionlong ] ) and ( [ eq : effactionshort ] ) are real functions of @xmath30 , the single leading one being @xmath82 ( as noted already in @xcite ) while the subleading ones are @xmath83 . it is straightforward to include higher - order terms the number of which increases rapidly . at order @xmath84 , for instance , there are already 11 terms so that we refrain from going beyond next - to - leading order in @xmath30 . for later purposes it is useful to express the operators appearing in ( [ eq : effactionshort ] ) in terms of the fundamental loops @xmath85 and @xmath86 . octet and sextet characters ( @xmath87 and @xmath88 , respectively ) are eliminated via the standard reduction identities @xmath89 which are equivalent to the character relations ( recall that @xmath90 ) @xmath91 making use of the latter the different terms in ( [ eq : effactionshort ] ) become @xmath92 from these expressions it is obvious that each operator @xmath93 is manifestly real . under charge conjugation @xmath94 , hence @xmath95 , whereupon all terms in @xmath96 are charge conjugation invariant as required @xcite . note that the orders included correspond to terms that are quadratic , cubic and quartic in @xmath85 and/or @xmath86 . higher powers will arise upon taking into account higher representations . thus , in this respect @xmath97 is somewhat singled out being of only cubic order . this will become important in a moment . it has already been pointed out by svetitsky and yaffe @xcite that effective actions with @xmath0 centre symmetry are closely related to the 3-state potts model which shows a first order phase transition from a symmetric to a ferromagnetic phase @xcite . to make the relation manifest we restrict the polyakov loop to the centre elements @xmath98 introduced in ( [ eq : centre ] ) . setting @xmath99 we find the general formula @xmath100 where @xmath101 denotes the dimension of the representation @xmath41 , @xmath102 applying this to the effective action in ( [ eq : effactionshort ] ) we find , up to an additive constant , @xmath103 = \lambda \sum _ { { \langle}{\mbox{\scriptsize\boldmath$x$ } } { \mbox{\scriptsize\boldmath$y$ } } { \rangle}}\cos \left ( \frac{2\pi}{3}(k_{{\mbox{\scriptsize\boldmath$x$ } } } - k_{{\mbox{\scriptsize\boldmath$y$ } } } ) \right ) \ ; , \quad k_{{\mbox{\scriptsize\boldmath$x$}}}\in\{0,1,2\}\ ; , \label{eq : pottsm}\ ] ] with effective coupling @xmath104 the action ( [ eq : pottsm ] ) is precisely the one of the 3-state potts model @xcite . we thus expect that our effective polyakov loop models will have a phase structure generalising the one of the 3-state potts model . the latter is known to have a ferromagnetic phase for large negative @xmath105 and an anti - ferromagnetic phase for large positive @xmath105 @xcite . as a preparation for the discussion of later sections it is hence useful to obtain some qualitative understanding of the phase structure of the polyakov loop models viewed as generalisations of the 3-state potts model . for the reason mentioned at the end of the previous section we choose as a minimal generalisation the following effective action , @xmath106 which in terms of the fundamental loops may be written explicitly as @xmath107 note that there are also quadratic contributions stemming from @xmath97 . the action ( [ eq:2cmodel ] ) is manifestly @xmath0 centre symmetric under @xmath108 . it is important to realise that ( [ eq:2cmodel ] ) differs from the standard lattice actions for scalar fields in several respects . first , the field @xmath24 is dimensionless , being the trace of a unitary matrix . this allows for the presence of cubic hopping terms connecting neighbouring sites . even more important is the fact that the target space of @xmath24 is compact . introducing the eigenvalues of @xmath109 via @xmath110 and writing @xmath111 we find for the real and imaginary part of @xmath85 , @xmath112 the target space of @xmath85 may then be sketched in the complex @xmath85-plane ( see fig . [ fig : domain ] ) . the boundary corresponds to the points with @xmath113 , the singular ` corners ' being given by the three centre elements @xmath114 . target space of the polyakov loop @xmath85 in the complex @xmath85-plane . the corners represent the three centre elements . the intermediate points ( denoted anti - centre elements ) will also become relevant for the discussion of the phase structure . ] let us try to get some first rough idea of the phase structure associated with the two - coupling model ( [ eq:2cmodel ] ) in the @xmath115-@xmath116 plane by looking at the extrema of the classical action . if we vary the couplings these will trace out a certain ( possibly discontinuous ) trajectory in the target space given by the triangle of fig . [ fig : domain ] . as we argued earlier , for centre - valued polyakov loops the effective action ( [ eq:2cmodel ] ) reduces to the action of the potts model ( [ eq : pottsm ] ) with coupling @xmath117 . thus we expect a ferromagnetic phase ( f ) for large negative @xmath118 and an anti - ferromagnetic phase ( af ) for @xmath118 large and positive . in a region around the origin in the coupling plane entropy dominates energy and we can not expect to actually obtain the correct phase - structure in this region by purely classical reasoning based on minimising the energy . qualitatively we expect a symmetric phase in a neighbourhood of the origin . this is represented schematically in fig . [ fig : clphases ] by the central rectangle . in order to study the ordered phases ( in particular af ) we divide the lattice in two sub - lattices ( denoted ` even ' and ` odd ' ) where the polyakov loop takes values @xmath119 and @xmath120 , respectively . two nearest neighbours belong to different sub - lattices . the absolute minima of the classical action @xmath121 will then be located at certain values @xmath122 and @xmath123 of the polyakov loop which are identified with its ` expectation values ' . we trust this reasoning as long as we are sufficiently far from the origin of the coupling plane i.e. from the disordered , entropy - dominated phase . any _ ferromagnetic _ ordering will be characterised by a minimum with @xmath124 while in an _ anti - ferromagnetic _ phase @xmath125 . quite interestingly we find two distinct ferromagnetic phases , one for which the polyakov loop is near a centre element or @xmath126 in the vicinity of @xmath127 and a different ferromagnetic phase with @xmath126 taking values near the intermediate points marked by triangles in fig . [ fig : domain ] . we call this an _ anti - centre _ phase ( ac ) . we expect a phase transition line separating the ferromagnetic and anti - ferromagnetic phases at vanishing potts - coupling @xmath118 . the resulting qualitative phase diagram is depicted in fig . [ fig : clphases ] . qualitative prediction of the phase diagram in the coupling constant plane for the effective action ( [ eq:2cmodel ] ) . the ferromagnetic ( f ) , anti - centre ( ac ) and anti - ferromagnetic ( af ) phases are obtained by looking for classical minima . the symmetric , disordered phase ( s ) is located where entropy is expected to dominate over energy . ] to discuss the ferromagnetic phases it suffices to minimize the action ( [ eq : smin ] ) with @xmath128 , in which case @xmath129 this can be done analytically . to localise the anti - ferromagnetic phase we have calculated the absolute minima of ( [ eq : smin ] ) on the target space depicted in fig . [ fig : domain ] numerically . the combined analytical and numerical results are summarised as follows . for negative @xmath116 we have a transition @xmath130 while for positive @xmath116 there is a richer structure , @xmath131 behaviour of the order parameter @xmath132 defined in ( [ eq : defprojector ] ) as a function of @xmath115 for fixed @xmath133 . for comparison we have added the result from the mean field analysis to be developed in the next section . ] the behaviour of a suitably projected order parameter @xmath132 ( the precise definition of which will only be needed later on ) for positive @xmath133 is shown in fig . [ fig : classicafer ] . upon inspection one notes that for @xmath115 sufficiently negative the system starts out with the polyakov loop at a centre element . increasing @xmath115 beyond @xmath134 the order parameter drops monotonically until , at a critical coupling @xmath135 , there is a jump to the ac phase with @xmath126 near a anti - centre element . the jump of @xmath132 is due to a centre - transformation and does not imply that the polyakov loop itself jumps . indeed , @xmath136 changes smoothly and arrives at an anti - center element for @xmath137 . the system stays there until @xmath138 , where it jumps again , this time to the af phase . as expected , we see no symmetric phase in a purely classical analysis . actually , for @xmath139 there is just no symmetric phase . the next step of refinement to be presented in this section is a mean field ( mf ) analysis of the effective action ( [ eq:2cmodel ] ) . this will serve as a basis for a comparison with results from direct monte carlo simulations to be discussed later on . due to the peculiarities of the model as compared to standard scalar field theories the application of the mf approximation is not entirely straightforward . for the benefit of the reader we will set the stage by giving a brief outline of the necessary modifications . for further details the reader is referred to our earlier paper @xcite . to keep the discussion sufficiently general we will first treat the effective action ( [ eq : effactionshort ] ) with five couplings focussing on simpler examples later on . we are interested in expectation values which are computed by evaluating integrals of the form @xmath140 } \int \mathcal{d}{\mathfrak{p}}\ , e^{-{s_{\text{\tiny eff}}}[{\mathfrak{p } } ] } \ , a[{\mathfrak{p } } ] \ ; , \\ \mathcal{d } { \mathfrak{p}}&\equiv \prod_{{\mbox{\scriptsize\boldmath$x$ } } } d\mu({\mathfrak{p}}_{{\mbox{\scriptsize\boldmath$x$ } } } ) \ ; , \end{split}\ ] ] which apparently extend over the whole group manifold employing the haar measure @xmath141 . however , due to the gauge invariance of both action and measure the integrals can be reduced to the coset space of conjugacy classes which we ( somewhat symbolically ) denote by @xmath142 . hence we integrate with the reduced haar measure by replacing @xmath143 thus , is equivalent to @xmath144 } \int \mathcal{dp } \ , e^{-{s_{\text{\tiny eff}}}[\mathcal p ] } \ , a[\mathcal p ] \ ; , \\ \mathcal{dp } & = \prod_{{\mbox{\scriptsize\boldmath$x$ } } } d\mu_{\rm red}(\mathcal p_{{\mbox{\scriptsize\boldmath$x$}}})\;. \end{split}\ ] ] the probability measure @xmath145 $ ] is characterized as the unique solution to the variational problem @xmath146 = -w[0 ] \;.\ ] ] the expectation value on the left - hand side has to be taken with respect to the integration measure @xmath147 $ ] with @xmath148 $ ] denoting the probability density of @xmath149 . from this point of view the mf approximation is nothing else but the restriction of the permissible densities to product form , @xmath150 \to p_{\mathrm{mf}}[\mathcal p ] \equiv \prod_{{\mbox{\scriptsize\boldmath$x$ } } } p_{{\mbox{\scriptsize\boldmath$x$ } } } ( \mathcal p_{{\mbox{\scriptsize\boldmath$x$}}})\ ; .\ ] ] expectation values can now be simply computed site by site via factorisation , @xmath151 our goal is to compute the effective potential as function of the mean characters @xmath152 . for that purpose we solve the variational problem ( [ eq : definfimum ] ) on the space of product measures with fixed expectation values of the characters . this is done by introducing appropriate lagrange multipliers @xmath153 . for ferromagnetic systems one may assume that the weight functions @xmath154 at each site are identical , @xmath155 . this assumption corresponds to a translationally invariant ground state . according to the discussion of the previous section we expect anti - ferromagnetic phases and hence we must refine our choice for @xmath154 . we therefore introduce different weight functions on the even and odd sub - lattices , respectively , @xmath156 defining the sign of a lattice point as @xmath157 as a consequence , expectation values of characters will subsequently have two values depending on the sub - lattice where they are evaluated , @xmath158 the sources are taken constant as well when restricted to the even and odd sub - lattices , @xmath159 or @xmath160 , respectively . like the characters the sources are complex . the action couples only nearest - neighbour sites so that its expectation value entering ( [ eq : definfimum ] ) may be written as @xmath161 with @xmath162 denoting the lattice volume in @xmath5 spatial dimensions . the logarithm in ( [ eq : definfimum ] ) decomposes as @xmath163 it is convenient to drop the common volume factor @xmath164 and consider densities instead . the variation of finally yields the weight function for the even sub - lattice , @xmath165 and a completely analogous expression for the odd sub - lattice . here we have introduced @xmath166 as a short - hand for @xmath167 and the single - site partition function @xmath168 the sources @xmath169 are eliminated by inverting the relations @xmath170 to be evaluated separately on both sublattices . the schwinger function @xmath171 is defined as usual , @xmath172 introducing the legendre transform of according to @xmath173 the solution of is finally obtained as the mf potential ( density ) as a function of even and odd mean fields , @xmath174 from this expression one can easily derive relations between the sources @xmath175 and mean characters @xmath176 . for instance , by varying @xmath177 with respect to @xmath178 we obtain @xmath179 where we have used that the current is given as @xmath180 the first term in ( [ eq : distrsolution ] ) derives directly from the effective action ( [ eq : effactionshort ] ) and hence contains four couplings to which the potential coupling @xmath181 has to be added . a complete mf analysis of this system becomes very awkward . in what follows , we therefore specialise to effective actions with only one or two couplings . the action ( [ eq : s10 ] ) defines what we call the ` minimal model ' with one coupling only , @xmath182 identifying @xmath183 and inserting ( [ eq : sgp ] ) the mf potential simplifies to @xmath184 it is useful to define the generic mf potential @xmath185 so that ( [ eq : simplesolution_2 ] ) can be rewritten as @xmath186 this expression clearly shows that for negative @xmath105 configurations with @xmath187 are favoured making the distinction between even and odd sub - lattices obsolete . the remaining unique expectation value , @xmath188 thus serves as an order parameter for the ferromagnetic phase transition . on the other hand , if @xmath189 @xmath119 and @xmath120 will cease to be positively correlated and we expect an anti - ferromagnetic phase transition . in this case , a reasonable order parameter is given by @xcite @xmath190 let us also introduce the absolute values of the order parameters , henceforth denoted by @xmath191 if we assume for the moment that the occurring phase transitions are second order the corresponding mf critical couplings can be computed analytically as we are going to demonstrate next . the consistency condition reduces for the case at hand to @xmath192 as both @xmath119 and @xmath120 may be assumed small near the critical coupling @xmath193 the corresponding sources will also be small . thus , we may expand @xmath194 which again holds separately on each sublattice , @xmath195 and @xmath196 . plugging this into yields @xmath197 and hence the critical couplings @xmath198 for arbitrary order parameters we have solved the gap equations ( [ eq : gap ] ) numerically . the result is depicted in the following figure . the transition s - af is second order , the one to the phase f first order . thus , the anti - ferromagnetic transition must be at @xmath199 . the first order ferromagnetic transition is slightly above @xmath200 . mean field results for the minimal model with one coupling . the first order ferromagnetic transition s - f is at @xmath201 and the second order transition s - af at @xmath202 . ] our mf analysis thus confirms the qualitative results from the preceding section that already in the simplest model there is both a ferromagnetic and an anti - ferromagnetic transition . this is qualitatively consistent with the phase diagram fig . [ fig : clphases ] restricted to the horizontal axis . even for the simple model of the previous subsection no explicit expression is known for @xmath203 as the @xmath9 group integrals can not be evaluated in closed form ( a fact well known from strong - coupling expansions , see e.g.@xcite ) . things naturally become worse if additional couplings are turned on . already for two couplings i.e. the action ( [ eq:2cmodel ] ) , the only way to proceed is by means of numerical methods . in order to obtain the mf version of the phase diagram fig . [ fig : clphases ] we have employed the following algorithm : 1 . at the extremal points of all sources @xmath204 occurring in can be eliminated in favour of the expectation values @xmath205 as in . since the character target space is compact it can be easily discretised defining measures @xmath206 and @xmath207 at each point . using these measures expectation values @xmath208 can now be computed which in general will differ from @xmath205 . hence , we first look for local minima of @xmath209 with norm @xmath210 . these minima , however , do not correspond to exact solutions yet but rather serve as the starting points of a recursion . 2 . we now solve the equation @xmath211 for @xmath212 and @xmath213 by newton iteration using the local minima of the previous step as initial input . in this way we end up with multiple solution vectors @xmath205 each extremising . 3 . the solution vector with minimal @xmath177 contains the desired ground state expectation values . with the help of this algorithm we are able to compute the expectation values of @xmath85 and @xmath214 . we conclude this section with some remarks concerning our choice of the sources @xmath175 . allowing them to be complex yields an 8-dimensional parameter space for the observables of model ( [ eq:2cmodel ] ) with two characters , @xmath215 and @xmath216 . for this large parameter space it would actually be simpler to perform a high - precision monte carlo simulation than to find a good approximation for the global minima of . for this reason , we have chosen real sources as an input to our mf approximation . we expect this to be a very good approximation as long as the peak of the probability distribution for @xmath215 is concentrated near the real axis in fig . [ fig : domain ] . comparing with the classical analysis we note that our real - source assumption is justified for all couplings which are located away from the boundary between the phases f and ac . the f - ac transition should be second order according to the analysis of section [ sec : qca ] while the mf approximation with real sources predicts a first order transition ( see fig . [ fig : classicafer ] ) . the contradiction will be finally resolved by monte carlo simulations showing that near the f - ac transition the peaks of the probability distribution for @xmath215 are off the real axis . apart from this fairly small region in parameter space we find a remarkable agreement between the mf results and the monte carlo results of the next section . this provides the ultimate justification for our simplifying choice of real sources in the mf approximation . the previous two sections have provided us with a good deal of information on the phase structure of the effective model in the most interesting regions of parameter space . based on this we have performed a large number of monte carlo simulations to quantitatively check the mf predictions and obtain a precise picture of the critical behaviour . as before , to avoid excessive complexity , we have concentrated on the action ( [ eq:2cmodel ] ) with two couplings @xmath115 and @xmath116 . at the beginning of section [ sec : mf ] we have already noted that both the action and the reduced haar measure depend only on the conjugacy class of the ( untraced ) polyakov loop . hence , one can choose the basic field variables either as the traces in the fundamental representations ( @xmath85 , @xmath86 ) and powers thereof or as a suitable parametrisation in terms of the eigenvalues as introduced in ( [ eq : pdiag ] ) . it turns out that for a numerical treatment the latter proves to be more appropriate and so we use ( [ eq : pdiag ] ) to represent the conjugacy class according to @xmath217 & \equiv \mathcal p_x ( \boldsymbol \phi ) = \text{diag } \big ( e^{i\phi_1 } , e^{i\phi_2 } , e^{-i(\phi_1+\phi_2 ) } \big ) \ ; , \\ \quad \boldsymbol\phi & \equiv ( \phi_{1},\phi_{2 } ) \ ; . \end{split}\ ] ] here , the following restrictions should be imposed such that the angular coordinates cover each class only once , @xmath218 these restrictions are somewhat awkward to implement in a simulation code . it is much more convenient to let @xmath219 take values in the full square @xmath220 which covers the fundamental domain given by ( [ eq : funddom ] ) six times . due to the residual gauge symmetry of the system it is clear that expectation values will be unaffected by this over - counting . in the coordinates ( [ eq : defangulardecomp ] ) the reduced haar measure becomes @xmath221 where the normalization is such that the measure integrates to unity over the square @xmath220 . all characters can be expressed in terms of @xmath222 and @xmath223 . using ( [ eq : redhaar ] ) the full measure may be straightforwardly expressed in terms of the angular coordinates as @xmath224 } = \prod_x d\mu_{\mathrm{red } } ( \boldsymbol \phi_x ) e^{-{s_{\text{\tiny eff}}}[\boldsymbol \phi ] } \ ; .\ ] ] the focus of our numerical studies has been the phase diagram in the @xmath225 plane . there , we have scanned through the region @xmath226 \times[-0.22\ldots 0.16]$ ] with a resolution of @xmath227 points , which were in total @xmath228 different monte carlo simulations . before we present our results a few words on our numerical techniques are in order . for the investigation of phase transitions , in particular their order , histogram methods are widely used and accepted . this approach , however , requires large statistics and thus tends to consume a lot of computer time . in addition , we are interested in a fairly large range of coupling constants . for these reasons , the updating algorithm for the polyakov loop models has to be fast and versatile . it turns out that the standard metropolis algorithm favourably matches both requirements if we aim at an accuracy of about @xmath229 . on the other hand , because of the highly nontrivial probability measure involved , a heat bath algorithm does not seem applicable or , in any case , would be too time consuming . in addition , its local nature should not yield any enhancement of statistics near a first order phase transition . we have thus refrained from implementing a heat bath update scheme but rather decided to optimise the metropolis algorithm as described in the following two paragraphs . when a system undergoes a first order phase transition , the histogram associated with the order parameter will generically display a multi - peak structure . depending on the total volume of the system the peaks can be very pronounced . in other words , the configuration space is decomposed into distinct sectors between which local algorithms can hardly mediate . one way to overcome the resultant failure in sampling the total configuration space is to make use of the multicanonical algorithm , see e.g.@xcite . the crucial improvement step consists in the replacement of the measure used in ( [ eq : haarphi ] ) according to @xmath230 the new improved measure on the right - hand side has a weight @xmath231 which depends on the ( modulus of the ) order parameter . one chooses the particular form @xmath232 where @xmath233 denotes the probability density of the order parameter . this choice leads to an enhancement of configurations that would otherwise be suppressed and thus allows for a much improved ergodic behaviour of the algorithm . histogram of @xmath234 with @xmath235 on a @xmath236-lattice sampled against the histogram for @xmath237 . ] the effect due to the altered measure is illustrated in fig . [ fig : multicanhist ] where two typical distributions are plotted . in the original distribution one clearly recognises two well - separated peaks . any local algorithm will fail to sample such a distribution properly once it is trapped at one of its peaks . the distribution actually used ` closes the gap ' enabling transitions between different peak regions during the simulation . in the end , of course , one has to correct for the change in the measure by reweighting with @xmath238 , @xmath239 where we have denoted expectation values taken with respect to the modified measure with a subscript ` mult ' . a slight problem with this approach , however , still has to be overcome . one actually needs right at the beginning what one set out to compute originally , namely the distribution @xmath240 . a practicable strategy is e.g. the following . from a small lattice volume , say @xmath241 , where peaks are usually less pronounced , one obtains an approximate distribution function @xmath242 which is then used on a slightly bigger lattice , say of volume @xmath243 . this simple trick can be further refined if on the larger lattice one first computes @xmath244 using @xmath245 and subsequently repeats all measurements employing @xmath181 . in practice , this procedure is iterated several times to make larger and larger lattices available . going beyond a volume of @xmath246 requires additional knowledge of the scaling behaviour of @xmath233 . [ fig : multcanlog ] shows that , to a very good approximation , the scaling depends linearly on the volume , @xmath247 , @xmath248 plot of @xmath249 against volume @xmath250 for the minimal model with @xmath251 confirming the scaling behaviour . the coupling is very close to critical . ] in summary , the multicanonical algorithm yields substantial improvements compared to the standard metropolis algorithm and allows for very accurate simulations . lattices with volumes up to @xmath252 could thus be studied near the first order phase transition . however , as the implementation is very involved we applied it only to the minimal model ( @xmath253 ) . in principle , the generalisation to incorporating further couplings is straightforward , being merely a matter of having sufficient computing time available . the multicanonical algorithm is best suited for studying first order transitions . this is no longer true for second order transitions where it is outperformed by cluster algorithms at least if there is one available . for our purposes , none of the algorithms on the market could be immediately put to use . we therefore decided to modify the well - known wolff algorithm @xcite , an extension of the swendsen - wang @xcite algorithm originally proposed for discrete spin systems . before actually describing our modifications let us briefly recall the main idea of @xcite . suppose the action ( and the measure ) are invariant under a certain global symmetry which acts on the fields according to @xmath254 typically , the symmetry operator @xmath255 will depend on some parameters which we collectively denote by @xmath256 ( continuous or discrete ) so that @xmath257 . it is important to note that the operator @xmath255 has to be idempotent , @xmath258 , in order to ensure detailed balance . the algorithm then works according to the following list . 1 . fix some parameter @xmath259 and hence some transformation @xmath260 . randomly choose a lattice point @xmath261 and apply the symmetry transformation @xmath262 which may be viewed as flipping the field variable @xmath263 . the point @xmath261 is checked and added to the cluster . [ it : flipnn ] repeat the following for all unchecked neighbours @xmath264 of @xmath261 : + let @xmath265 denote the contribution to the action from the link @xmath266 and compute @xmath267 which may be rewritten as @xmath268 since @xmath269 is already invariant under @xmath255 . the decision to add @xmath264 to the cluster is subject to an accept / reject step so that the probability to flip @xmath270 becomes @xmath271 if @xmath272 is flipped , check the point @xmath264 . 3 . go back to step ( [ it : flipnn ] ) for all sites added to the cluster in the previous step . note that measures whether it is advantageous to flip @xmath270 once @xmath263 has been flipped . from the very construction of the algorithm it should be already clear that the clusters will increase with the correlation length of the system . in this way one suppresses the phenomenon of critical slowing down observed with local algorithms . this makes the cluster algorithm particularly suited for the study of second order phase transitions . the first step in adapting the cluster algorithm to our needs is to find suitable symmetries of the action . being a polyakov loop model the symmetry in question is the discrete @xmath0 symmetry , @xmath273 . in addition , the action has to be real which implies constraints on the way complex conjugation acts . from these symmetries one can construct three operators @xmath274 , @xmath275 , acting on the polyakov loop according to @xmath276 as required the @xmath274 square to unity . furthermore , it is easy to see that both the operators @xmath93 appearing in and the domain of the polyakov loop @xmath85 are left invariant by the action of @xmath274 . the latter is illustrated in fig . [ fig : reflections ] for a particular value of @xmath85 . illustration of the @xmath0 reflections @xmath274 used in our modified cluster algorithm . ] for the actual algorithm the transformations are not immediately applicable since the simulation is based on the angular variables @xmath277 introduced in ( [ eq : defangulardecomp ] ) . however , it is just a matter of a little algebra to show that the @xmath274 act on the angles @xmath277 via @xmath278 whereas in the original cluster algorithm flipping along randomly chosen lattice sites is sufficient to guarantee ergodicity this is no longer true in the present case . hence , we have to augment the update scheme by standard metropolis sweeps to make the algorithm ergodic . for one monte carlo step our cluster algorithm finally can be summarised as follows : 1 . choose a random number @xmath279 between @xmath280 and @xmath250 . 2 . do @xmath279 standard metropolis sweeps at randomly drawn lattice points . 3 . for a suitable fixed number @xmath281 repeat the steps for building a cluster as described above . do @xmath282 additional metropolis sweeps , again at randomly chosen lattice sites . from several test runs we have found that the number @xmath281 should be chosen such that total number of flipped sites after performing step ( 3 ) is approximately half the number of all lattice sites . thus , if @xmath283 denotes the typical size of a cluster , the following equation should hold ( at least approximately ) , @xmath284 one of the most interesting questions of course is the gain in performance compared to e.g. the standard metropolis algorithm . we have found that the autocorrelation time for the order parameter @xmath285 is independent of both the lattice extent ( at least in the range @xmath286 ) and the coupling constant ( if close to the second order phase transition ) implying a dynamical critical exponent of @xmath287 . moreover , for an optimal choice of @xmath281 the autocorrelation time is of order unity . as a result , our cluster algorithm outperforms the metropolis algorithm even for small lattices . on the largest lattices we have considered the cluster algorithm reduces autocorrelation times by two orders of magnitude as compared to metropolis updating this improvement comes at the cost of a slightly increased complexity , specifically a factor of 1.5 in computing time which clearly is negligible . on the other hand , in line with our expectations , no significant improvement has been found near the first order transition . we present the results of our monte carlo simulations in the same two steps as for the mf approximation . hence , we first report on the minimal model ( [ eq : sgp ] ) and switch on @xmath116 later on . with only @xmath115 different from zero it is reasonably cheap to perform highly accurate measurements so that a precise quantitative comparison with the 3-state potts model is possible . for the model ( [ eq:2cmodel ] ) with two couplings we determine the phase diagram in the @xmath115@xmath116 plane and compare with our expectations as laid out in the previous two sections . we conclude with a careful study of the nature of the phase transitions , in particular their continuity properties . to determine the ferromagnetic phase transition for @xmath288 we have used standard techniques which need no further explanation . it suffices to note that in order to study the first order ( s - f ) transition we performed @xmath289 sweeps on a @xmath290-lattice in the multicanonical ensemble for each value of the coupling constant . this led to highly accurate statistics until we approached the close vicinity of the critical coupling itself . in this regime our statistics is restricted to @xmath291 independent samples due to large correlation times of the order of @xmath292 sweeps . as fig . [ fig : ferroptmc16 ] shows this is sufficient to demonstrate that the transition is first order . phase transition of fig . [ fig : ferroptmc16 ] when calculated within mf approximation . ] ferromagnetic phase transition computed with the multicanonical algorithm on a @xmath290-lattice . the expectation value of @xmath234 is plotted against its probability distribution given by the area shaded in grey . the latter clearly shows the correct discontinuous behaviour . note that , since we measure the modulus , statistical fluctuations manifest themselves in a ( small ) positive value of the order parameter even in the symmetric phase . ] in fig . [ fig : ferromf ] we compare our monte carlo result for @xmath293 with its mf approximation . the figure basically zooms into that part of fig . [ fig : meanscan ] where the first order ferromagnetic transition is located . again , the first order nature of the transition is corroborated . in addition to the coupling dependence of the order parameter we have determined both the critical coupling and the discontinuity @xmath294 at @xmath295 . the results are given in table [ tab : crit1 ] together with the mf prediction . again we find a surprisingly good _ quantitative _ agreement between simulations and mf approximation . moreover , if we consider the ratio of the two critical couplings , say @xmath296 , we are able to compare this with results of the 3-state potts model . the actual figures turn out to be fairly close , namely @xmath297 for the minimally coupled polyakov model and @xmath298 for the 3-state potts model @xcite . .[tab : crit1 ] critical couplings for the s - f and s - af phase transitions and jump @xmath294 . for the first order transition ( s - f ) there is excellent agreement between mf and monte carlo data . even the values for @xmath294 agree within 10% . for the second order transition ( s - af ) the critical couplings agree within approximately 20% . [ cols="<,^,^,^",options="header " , ] the uncertainty in @xmath299 quoted in table [ tab : crit1 ] is mainly due to the fact that the different cumulants do not precisely meet in a single intersection point ( cf . again fig . [ fig : bindercumulant ] ) . the error in the critical exponents is estimated from a least square fit to the logarithm of ( [ eq : critexpdef ] ) and . to cross - check our results we have measured the expectation value of @xmath300 and its probability distribution @xmath301 in analogy with the ferromagnetic transition already discussed . as in the former case , the expectation value of the order parameter alone does not suffice to decide on the order of the transition . however , the probability distribution shown in fig . [ fig : aferrmcs ] is quite different from the one in fig . [ fig : ferroptmc16 ] . no discontinuous behaviour is observed now which provides further ( numerical ) evidence for a second order phase transition . equally important , the critical coupling obtained is compatible with the results presented in table [ tab : crit1 ] . for the sake of completeness fig . [ fig : aferrmf ] shows the mf prediction for the transition s - af . comparing with earlier results on the 3-state potts model @xcite we draw the important conclusion that the minimal polyakov loop model is in the same universality class . fundamental domain @xmath302 of the order parameter @xmath85 obtained by identifying @xmath0 copies according to the depicted arrows . ] having discussed the minimal model at length let us continue by switching on the second coupling @xmath116 in order to analyse the phase diagram in the coupling constant plane . this requires a suitably chosen ` indicator ' to distinguish the ( at least ) four phases ( s , f , ac and af ) we expect in accordance with our mf analysis of section [ sec : mf ] . while @xmath234 and @xmath303 clearly are order parameters for the minimal model they are numerically less suited for the model with two couplings . it turns out advantageous to construct a new observable denoted @xmath132 which may be obtained from @xmath234 by the following procedure . we first divide the domain of @xmath85 into six distinct parts as shown in fig . [ fig : rotatedobs ] . the light - shaded region represents the preferred locus of the polyakov loop in the ferromagnetic phase f , whereas the dark - shaded region corresponds to the anti - centre ferromagnetic phase ac . to eliminate the ( numerically superfluous ) @xmath0 symmetry the first step in our projection is to identify the regions as indicated by the arrows in fig . [ fig : rotatedobs ] . in this way we end up with a fundamental domain @xmath302 for the @xmath0 symmetry centred along the real axis . every @xmath85 is mapped into @xmath302 by a centre transformation . to finally obtain a real observable we project the transformed @xmath85 onto the real axis . this projection results in the variable @xmath132 the sign of which clearly distinguishes between the two ferromagnetic phases . @xmath304 indicates the ac phase , @xmath305 the ferromagnetic phase f , while @xmath306 in the symmetric phase s. mathematically , the projection of @xmath85 to @xmath132 is given by @xmath307 to detect the af phase we simultaneously measure @xmath303 so that we can finally discriminate between all possible phases . our main results for the phase diagram are shown in fig . [ fig : phasediagmf ] obtained by the modified mf approximation explained earlier and fig . [ fig : phasediagmc8 ] which displays the monte carlo data and hence constitutes the most faithful representation of the phase structure . the monte carlo simulations were carried out on a @xmath308-lattice . the number of sweeps was chosen such as to reduce the jackknife error in the estimate of @xmath132 below @xmath309 . the independence of our samples was ensured by demanding that the autocorrelation time @xmath310 associated with the observable @xmath132 was less then one percent of the total number of sweeps . as a result our simulations included at least @xmath311 sweeps far away from the critical regions and more than @xmath289 sweeps in their vicinity . it is reassuring to note that the qualitative phase diagram fig . [ fig : clphases ] predicted from energy - entropy arguments is quantitatively confirmed by both figs . [ fig : phasediagmf ] and [ fig : phasediagmc8 ] . comparing the latter in some detail it is once more remarkable how good the mf approximation works . within large regions of parameter space it agrees with the ` real ' data within 10% or less . interestingly , the observable @xmath132 also seems to be sensitive to the af phase ( see lower right part of figs . [ fig : phasediagmf ] and [ fig : phasediagmc8 ] ) . however , any further discrimination between phases f and af by means of @xmath132 is impossible . to lift this degeneracy one clearly needs the af order parameter @xmath303 as an additional input . it remains to be discussed why @xmath132 is sensitive at all to af ordering . in what follows we will provide a heuristic answer in the context of the 3-state potts model . by definition , all spins are ( anti)aligned in the ( anti)ferromagnetic ground state . in the @xmath312 symmetric ising model with only two spin states each possible ground state is two - fold degenerate , independent of the particular ordering ( f or af ) . the counting of degeneracies , however , is totally different for systems with more spin states ( hence higher symmetry ) . in the af phase the additional freedom of choice between two ( or more ) states anti - aligned with a given one leads to an enormous degeneracy of the ground state _ in energy_. as a consequence , _ entropy _ will be the sole judge deciding what is to be observed in a measurement . for the @xmath0 polyakov loop model both mf approximation and the monte carlo simulations tell us that the most probable ground state corresponds to a preferred direction for the polyakov loop on one of the two sub - lattices and an equal distribution of the two remaining directions on the other sub - lattice . although we do not have an analytical justification for this statement the numerical evidence is compelling . based on the latter , the sensitivity of @xmath132 to af ordering can be explained by a net expectation value which for the case of the 3-state potts model is easily computed as @xmath313 phase boundaries and orders of transitions as obtained via monte carlo simulation on a @xmath308-lattice . we observe a mixture of both first and second order transitions depending on the particular values for the couplings . the symmetric phase is enclosed by ordered phases as already expected from the discussion of section [ sec : qca ] . for further details on the simulation see the main text . ] phase diagram of fig . [ fig : phasestructure ] with oriented trajectories ( marked by arrows ) used for the histograms of figs . [ fig : histoone ] to [ fig : histoeight ] . the curves are directed from the symmetric to the broken phases intersecting the critical lines vertically . we have chosen a representative subset from a total of twenty such curves which were analysed to determine the order of the transitions . ] we conclude this discussion with an overview of the resulting phases , their boundaries and the order of the transitions in between presented in fig . [ fig : phasestructure ] . our reasoning here is based on additional measurements carried out along parametrised curves @xmath314 as depicted in fig . [ fig : histopositions ] . these simulations were exclusively focused on the order of the phase transition measuring the histograms of the observables @xmath85 and @xmath214 . as before all measurements were carried out on @xmath308-lattices each trajectory @xmath315 being sampled with twenty points . for every such point @xmath316 @xmath289 monte carlo sweeps were performed . to improve the statistics of the histograms we made use of all previously discussed symmetries ( @xmath0 , complex conjugation and exchange of even and odd sub - lattice ) by binning @xmath85 or @xmath214 together with their centre images @xmath317 , @xmath318 , @xmath319 , @xmath320 and their complex conjugates . this amounts to using each measured value of @xmath85 six times and of @xmath214 even twelve times . in total we have recorded twenty such runs four of which are depicted in fig . [ fig : histopositions ] . to illuminate the order of the transitions corresponding to the four directed line crossings displayed in fig . [ fig : histopositions ] we present six ( out of twenty ) histograms in figs . [ fig : histoone ] to [ fig : histoeight ] which display the distribution of the observable @xmath85 respectively @xmath214 in the complex plane . the first figure in this series , fig . [ fig : histoone ] , corresponds to the ferromagnetic transition , s - f . the histograms displayed may be viewed as a ` movie sequence ' starting out in the symmetric phase with the distribution @xmath321 located at the origin . recalling the relation between probability distribution @xmath321 and the constraint effective potential @xcite , @xmath322 , this situation corresponds to a unique minimum of @xmath323 at @xmath324 . as the couplings change three further maxima which are @xmath0 copies of each other arise in addition to the one at the origin . hence , we observe coexistence of ordered and disordered phases . the new maxima are separated from the original unique maximum at the origin by a finite amount . the associated discontinuity together with coexistence clearly shows that the transition is first order . in contrast to this the situation depicted in fig . [ fig : histotwo ] is quite different . first of all the distribution is much more delocalised as compared to fig . [ fig : histoone ] implying a much flatter effective potential . moreover , as soon as the maxima of the broken ( ordered ) phase emerge the maximum at the origin has dissolved . hence , there is no coexistence of different phases as in the previous case but rather a continuous appearance of new maxima , branching away from the origin towards the corners of the fundamental domain . these features characterise a second order phase transition . analogous behaviour can be observed in fig . [ fig : histothree ] with the maxima now moving in roughly the opposite directions indicating a transition to the ac phase . still different is the s - af transition of fig . [ fig : histoeight ] . by the same reasoning as given above this transition is of second order as well . however , the probability function emerging in the broken phase does not display the usual three - fold symmetry we have observed for the previous transitions . although the figures seemingly look rotationally invariant one actually finds a six - fold degeneracy rather than a continuous one . the symmetry enhancement by a factor of two has a simple explanation . note that fig . [ fig : histoeight ] shows the probability distribution of the af order parameter @xmath214 which is defined in terms of even and odd sub - lattices . interchanging the latter accounts for the additional factor of two . let us conclude this section with a short remark concerning figs . [ fig : histotwo ] and [ fig : histothree ] . we have seen that our numerical data seem to indicate the possibility of second order transitions from the disordered phase s to f or ac ordering . this is at variance with the folklore claiming the absence of a @xmath0 universality class in @xmath8 , see e.g. @xcite . it is not excluded that the observed second order characteristics near the tricritical point ( s - f - ac ) derive from artifacts due to our algorithms . at the moment , however , we can not make a definite statement about this issue . obviously , the puzzle deserves further investigation in the future . as outlined in the introduction the original motivation for polyakov loop models lies in their status as effective theories for finite - temperature gluodynamics . however , our investigation of the statistical mechanics involved should have made it clear that they are interesting in their own right . in support of this statement we mention the 3-fold @xmath0 symmetry shared with the 3-state potts model , the nontrivial complex target space and the enormously rich phase structure resulting thereof . as we have seen , the latter can be understood qualitatively by simple energy - entropy considerations which predict a symmetric phase s close to the origin in coupling space ` surrounded ' by broken phases . these can exhibit ferromagnetic ( f ) , anti - ferromagnetic ( af ) or anti - centre ( ac ) ordering . this picture has been confirmed quantitatively both within a mean - field ( mf ) approximation and by extensive monte carlo simulations . both methods find a first order transition for s - f and a second order one for s - af . in order to capture the details of the phase diagram a total number of 8000 simulations ( corresponding to roughly 3000 hours of cpu time ) was required . naturally , this has added further refinements to our classical and mf analysis like precise values for critical couplings and exponents . the agreement between mf and monte carlo result is much better than naively expected for a statistical model in @xmath8 . the s - f critical couplings agree to within 1% , the discontinuity @xmath325 within 10% and the s - af critical coupling within 20% . the precise agreement for the s - f transition suggests that the point where the s - f and s - ac critical lines merge is actually a _ tricritical _ point where two first order transitions merge into a second order one . it is known that the upper critical dimension for a tricritical point is three @xcite so that in @xmath8 the mf approximation actually is exact apart from logarithmic corrections @xcite . this observation is corroborated by the fact that the polyakov loop model with two couplings is somewhat similar to the spin - one blume - capel @xcite model which apart from an ising term has an addition @xmath326 , @xmath327 , with 3 spin states . the model is known to have a tricritical point and is closely related to the 3-state potts model @xcite . the second order phase transition line is characterised by 3@xmath5 ising critical exponents . obviously , one should test the analogous exponents for our ac - f transition at the very left of our phase diagrams . one may also address the spatial localisation of the phases . on the 3@xmath5 lattice the coexisting phases should be separated by interfaces the tension of which can be measured ( see e.g.@xcite and references therein ) . in the end one is of course interested in matching the effective couplings to the underlying microscopic theory , i.e. finite - temperature yang - mills . the resulting curve @xmath328 in the space of effective couplings should stay in the symmetric and ferromagentic phases . we have already solved the matching - problem for the ( simpler ) case of @xmath7 using inverse monte carlo methods based on schwinger - dyson equations . for @xmath9 this is considerably more difficult ( at least technically ) due to the increased complexity of the haar measure . the problem is actually similar to one encountered in strong - coupling expansions where @xmath7 group integrals can be performed analytically which is no longer true for @xmath9 . nevertheless , we have been able to derive the relevant schwinger - dyson equations and hope to report on their applications in the near future . we thank guy buss for his contributions to the strong coupling expansion and georg bergner , wolfhard janke and erhard seiler for valuable hints . tk acknowledges support by the konrad - adenauer - stiftung e.v . , cw by the studienstiftung des deutschen volkes and th by the plymouth particle theory group .

in joint work with yang huang , we defined a canonical absolute grading on heegaard floer homology by homotopy classes of oriented 2-plane fields . a similar grading was defined on embedded contact homology by michael hutchings . in this paper we show that the isomorphism between these homology theories defined by colin - ghiggini - honda preserves this grading .

1403.2875

let @xmath0 be a closed , connected , oriented three - manifold . the embedded contact homology ( ech ) and the heegaard floer homology of @xmath0 are invariants that have been studied and computed for many manifolds . ech was defined by hutchings using a contact form on @xmath0 , see @xcite , and heegaard floer homology was defined in @xcite by ozsvth - szab using a heegaard decomposition of @xmath0 . these two homology theories have very distinct flavors , but they have recently been shown to be isomorphic by colin - ghiggini - honda @xcite . more specifically , they construct an isomorphism @xmath1 . here @xmath2 is a version of heegaard floer homology of @xmath0 with the opposite orientation . one could instead consider the heegaard floer cohomology of @xmath0 with the right orientation . one can define a canonical absolute grading on ech and heegaard floer homology by homotopy classes of oriented 2-plane fields on @xmath0 . we now explain why this is a natural grading set to consider . first note that the orientation of @xmath0 induces a bijection between the homotopy classes of oriented 2-plane fields and the homotopy classes of nonvanishing vector fields . let @xmath3 denote the set of homotopy classes of nonvanishing vector fields on @xmath0 . one also defines @xmath4 to be the set of equivalence classes of vector fields on @xmath0 , where two vector fields are said to be equivalent if they are homotopic in the complement of an embedded three - dimensional ball . let @xmath5 denote the quotient map . we observe that @xmath6 acts on @xmath3 as follows . let @xmath7\in{\text{vect}}(y)$ ] and let @xmath8 be a small embedded ball in @xmath0 . fix a trivialization of @xmath9 . so we can see @xmath10 as a map @xmath11 . without loss of generality , we can assume that @xmath10 is constant in @xmath8 . let @xmath12 be the vector field given by the composition of the quotient map @xmath13 with the hopf map @xmath14 . one can check that the homotopy class @xmath15 $ ] does not depend on the choice of @xmath8 or @xmath10 . so @xmath7 + 1 $ ] is defined to be @xmath15 $ ] . by induction , one defines an action of @xmath6 on @xmath3 . it follows from the definition that @xmath16 is invariant under this action . moreover , one can check that for @xmath17 , any two elements in @xmath18 differ by @xmath19 , for some @xmath20 . in fact , the set @xmath18 is an affine space over @xmath21 , where @xmath22 is the divisibility of @xmath23 . one can prove all these facts using the pontryagin - thom construction , for example . now , since all versions of ech and heegaard floer homology split acording to spin@xmath24 structures and they are relatively graded on @xmath21 for the appropriate @xmath22 for each spin@xmath24 structure , the set @xmath3 is a possible choice for an absolute grading set . in fact , absolute gradings with values on @xmath3 were defined without any extra choices on ech and heegaard floer homology @xcite . for @xmath25 , let @xmath26 and @xmath27 denote the subgroups of @xmath2 and @xmath28 , respectively , consisting of all elements of grading @xmath29 . [ mainthm ] let @xmath1 be the isomorphism constructed by colin - ghiggini - honda . then @xmath30 maps @xmath26 to @xmath27 for all @xmath29 . we recall that both @xmath2 and @xmath28 admit a map @xmath31 whose mapping cone is denoted by @xmath32 and @xmath33 , respectively . the latter groups are simpler than the former and they are also isomorphic . in fact , in order to show that @xmath30 is an isomorphism , colin , ghiggini and honda first construct an isomorphism @xmath34 . they also show that the following diagram commutes . @xmath35^{\widehat{\phi } } \ar[r]^{\iota _ * } & hf(-y ) \ar[d]^{\phi}\\ \widehat{ech}(y ) \ar[r]^{\iota _ * } & ech(y ) } \label{eq : comm}\ ] ] here the horizontal maps @xmath36 denote the natural maps given by the mapping cone construction . in order to show that @xmath30 preserves the absolute grading , it is enough to prove that both maps @xmath36 and @xmath37 do . the map @xmath37 is defined as a composition @xmath38 as follows . @xmath39 here @xmath40 is the homology of a chain group computed from the page of an open book decomposition @xmath41 of @xmath0 and @xmath42 is the homology of a subcomplex of the ech chain complex @xmath43 generated by orbits sets that intersect @xmath44 exactly @xmath45 times , where @xmath46 is a certain contact form on @xmath47 . the maps @xmath48 , @xmath49 and @xmath50 are all isomorphisms and we will show that all of them preserve the grading . this paper is organized as follows . in section [ sec : hf ] , we review the definition heegaard floer homology , using lipshitz s cylindrical reformulation , and the absolute grading on it . we explain how to compute its `` hat''-version from a page of an open book decomposition , following @xcite . we also prove that @xmath48 preserves the grading . in section [ sec : ech ] , we recall the definition of ech and its absolute grading . we explain its `` hat''-version and the relationship between ech and open book decompositions . we show how to construct the map @xmath50 and we prove that it preserves the grading . in section [ sec : thm ] , we recall some of the steps to construct the isomorphism @xmath49 and we finish the proof of theorem [ mainthm ] . in this section , we will recall the definition of heegaard floer homology using the cylindrical reformulation from @xcite and the absolute grading on it from @xcite . a heegaard diagram is a triple @xmath51 , where @xmath52 is a closed oriented surface of genus @xmath53 and the tuples @xmath54 and @xmath55 are @xmath53-tuples of disjoint circles on @xmath52 which are linearly independent in @xmath56 . given such a heegaard diagram , one can construct a closed oriented three - manifold by considering @xmath57 $ ] , attaching disks to it along the circles @xmath58 and @xmath59 and filling the rest with two three - dimensional balls . any closed oriented three - manifold @xmath0 can be obtained this way from a heegaard diagram . a pointed heegaard diagram is a quadruple @xmath60 where @xmath51 is a heegaard diagram and @xmath61 is a point on @xmath52 in the complement of all of the circles @xmath62 and @xmath63 . given a pointed heegaard diagram @xmath60 , an intersection point is a @xmath53-tuple @xmath64 , where @xmath65 and @xmath66 is a permutation of @xmath67 . the chain complex @xmath68 is the @xmath6-module generated by the intersection points . the differential @xmath69 is defined as follows . consider the manifold @xmath70\times\sigma$ ] with the symplectic form @xmath71 , where @xmath72 and @xmath73 are the coordinates on the @xmath74 and @xmath75$]factors , respectively , and @xmath76 is an area form on @xmath52 . if @xmath64 and @xmath77 , one defines a moduli space @xmath78 , which is basically the moduli space of holomorphic curves in @xmath79 with boundary on the union of the cylinders @xmath80 and @xmath81 , and that converge to the union of chords @xmath75\times x_i$ ] as @xmath82 and to the union of @xmath75\times y_j$ ] as @xmath83 . for details , see @xcite . now @xmath84 is defined to be the signed count of certain embedded holomorphic curves in @xmath78 that do not intersect @xmath85\times\{z\}$ ] . it turns out that @xmath86 , so one defines @xmath87 to be the homology of this chain complex . we now recall the definition of the other versions of heegaard floer homology . the complex @xmath88 is defined to be the @xmath6-module generated by @xmath89 $ ] , where @xmath90 is an intersection point and @xmath20 . the differential is defined by @xmath91=\sum_{{\mathbf{y}},a } c({\mathbf{x}},{\mathbf{y}},a ) [ { \mathbf{y}},n - n_z(a)],\ ] ] where @xmath92 is a relative homology class in @xmath79 , the number @xmath93 is the signed count of certain holomorphic curves in @xmath94 in @xmath92 and @xmath95 is the intersection number of @xmath92 and @xmath85\times\{z\}$ ] . one can show that @xmath86 . one can now define @xmath96 to be the subgroup of @xmath88 generated by @xmath89 $ ] , for @xmath97 . this is a subcomplex because @xmath98 , whenever @xmath92 can be represented by a holomorphic curve . one also defines @xmath99 to be the quotient of @xmath88 by @xmath96 . the homologies of these various complexes are denoted , respectively , by @xmath100 the @xmath31 map is defined by @xmath101=[{\mathbf{x}},m-1]$ ] . so there is a short exact sequence @xmath102 this sequence induces an exact triangle . @xmath103^{u } & & hf^+(\sigma,{\bm{\alpha}},{\bm{\beta}},z ) \ar[dl ] \\ & \ar[lu ] \widehat{hf}(\sigma,{\bm{\alpha}},{\bm{\beta}},z ) & } \label{tri : hf}\end{aligned}\ ] ] we will now recall the absolute grading on all these homology groups . let @xmath104 be a pair consisting of a self - indexing morse function @xmath105 on @xmath0 and a gradient - like vector field @xmath106 , i.e. @xmath107 , whenever @xmath108 . we assume that @xmath105 has exactly one index 0 and one index 3 critical points . we also assume that all stable and unstable manifolds intersect transversely . for each index 1 critical point @xmath109 , let @xmath110 denote the unstable manifold containing @xmath109 and , for each index 2 critical point @xmath111 , let @xmath112 denote the stable manifold containing @xmath111 . the pair @xmath104 is said to be compatible with the heegaard diagram @xmath113 if * @xmath114 , * @xmath115 and @xmath116 , for all @xmath117 . an intersection point @xmath90 determines @xmath53 flow lines @xmath118 connecting the points @xmath109 to the points @xmath111 . the basepoint @xmath61 determines a flow line @xmath119 from the index 0 critical point to the index 3 critical point . outside the union of small neighborhoods of @xmath120 , the vector field @xmath106 does not vanish . the absolute grading @xmath121 is the homotopy class of an appropriate extension of this nonvanishing vector field to the union of these small neighborhoods , as we briefly explain . figure [ nbhd](a ) illustrates two transverse vertical sections of the vector field @xmath106 in a small neighborhood of @xmath122 , for some @xmath123 and figure [ nbhd](b ) illustrates a vertical section of @xmath106 in a small neighborhood of @xmath119 . now we substitute @xmath106 in these neighborhoods by the vector fields illustrated on figure [ defn ] . we note that in the neighborhood of @xmath119 , the vector field on figure [ defn](b ) has a circle of zeros . we modify the vector field in neighborhood of this circle so that it rotates clockwise on the @xmath124-plane . then we define @xmath121 to be the homotopy class of this vector field . for more details on this construction , see @xcite . nbhd.eps ( 10,-3)@xmath125-plane ( 40,-3)@xmath126-plane ( 27.5,-7)(a ) ( 86,-7)(b ) defn.eps ( 10,-3)@xmath125-plane ( 40,-3)@xmath126-plane ( 27.5,-7)(a ) ( 86,-7)(b ) now , for an intersection point @xmath90 and @xmath20 , we define @xmath127)={\text{\textnormal{gr}}}({\mathbf{x}})+2n$ ] . the following was the main theorem of @xcite . [ hf : absgrad ] for a pointed heegaard diagram @xmath60 , @xmath128 defines an absolute grading on any version of heegaard floer homology @xmath129 , satisfying the following properties : 1 . if @xmath90 and @xmath130 are in the same spin@xmath24 structure , then @xmath131 , where @xmath22 is the divisibity of the first chern class of the given the spin@xmath24 structure . 2 . if @xmath132 is a contact structure on @xmath0 and @xmath133 is the contact class in @xmath32 , then @xmath134 is the homotopy class of the reeb vector field @xmath135 for any contact form @xmath46 for @xmath132 . the grading @xmath128 is invariant under heegaard moves , so it is does not depend on the pointed heegaard diagram . the grading @xmath128 respects the cobordism maps . it follows from the definition that the map @xmath136 in the exact triangle has degree 0 . in this subsection , we will recall how to compute @xmath32 from the page of an open book decomposition and we will show how to adapt the absolute grading to it . an _ open book _ is a pair @xmath41 , where @xmath44 is a compact oriented surface with boundary and @xmath66 is a diffeomorphism of @xmath44 which is the identity on @xmath137 . we will always assume that @xmath137 is connected . given such a pair , one constructs a three - manifold as follows . let @xmath138 be the mapping cylinder of @xmath66 , i.e. the quotient of @xmath139 $ ] by the equivalence relation given by @xmath140 . we obtain a closed three - manifold by further quotienting @xmath138 by the equivalence relation given by @xmath141 for all @xmath142 and @xmath143 . for every closed oriented three - manifold @xmath0 , there exists an open book @xmath41 giving rise to @xmath0 by the above construction . so @xmath41 is called an open book decomposition of @xmath0 . let @xmath41 be an open book decomposition of @xmath0 . by ( * ? ? ? * lemma 2.1.1 ) , we can assume that there exists a diffeomorphism of a neighborhood of @xmath137 in @xmath44 to @xmath144\times \partial s$ ] such that the monodromy @xmath66 is given by @xmath145 in this neighborhood . then @xmath41 gives rise to a heegaard decomposition as follows . the heegaard surface is @xmath146 , where @xmath147 denotes @xmath148 . if we denote the genus of @xmath44 by @xmath53 , then @xmath52 has genus @xmath45 . we choose a set of properly embedded arcs @xmath149 of @xmath44 such that @xmath150 is a disk . one can then let @xmath151 , where @xmath152 is seen as an arc in @xmath153 and @xmath154 is its copy in @xmath155 . one also lets @xmath156 , where @xmath157 is the simplest arc in @xmath155 which is isotopic to @xmath154 and extends @xmath158 to a smooth curve in @xmath52 , see figure [ fighf ] ( ) . hence @xmath51 is a heegaard diagram for @xmath0 . so @xmath159 is a heegaard diagram for @xmath160 . for each @xmath161 , the circle @xmath154 intersects @xmath157 in @xmath155 at three points . we label them @xmath162 , as in figure [ fighf ] ( ) . we fix a basepoint @xmath163 . one defines @xmath164 to be the subcomplex of @xmath165 generated by @xmath45-tuples of intersection points contained in @xmath153 . one also defines @xmath166 to be the quotient @xmath167 , where two @xmath45-tuples of intersection points in @xmath153 are equivalent if they differ by substituting @xmath168 by @xmath169 . there is an induced differential on @xmath166 . it is shown in ( * ? ? ? * theorem 4.9.4 ) that the homology of this chain complex is isomorphic to @xmath170 . the absolute grading on the complex @xmath166 is well - defined as the following simple lemma shows . if @xmath90 is a @xmath45-tuple of intersection points in @xmath153 containing @xmath168 and @xmath171 , then @xmath172 . therefore @xmath128 is well - defined on the quotient @xmath166 . we observe that @xmath90 and @xmath173 are in the same spin@xmath24 structure and that @xmath174 . therefore , by theorem [ hf : absgrad](a ) , @xmath175 moreover , by definition , the map @xmath176 preserves the absolute grading . therefore the isomorphism @xmath177 preserves the absolute grading . [ hfobd1 ] [ hfobd2 ] we observe that @xmath0 can also be decomposed as @xmath178 , where @xmath179 is glued to @xmath180 . this decomposition can also be obtained by considering the open book @xmath181 , where @xmath182 is an annulus and @xmath183 is an extension of @xmath66 to @xmath184 which is the identity in a small neighborhood of @xmath185 . let @xmath46 be a contact form on @xmath0 which is _ adapted _ to @xmath41 , i.e. the reeb vector field @xmath135 is a positively transverse to @xmath186 for all @xmath73 and positively tangent to the binding @xmath187 . for each generator @xmath188 , we will now construct a vector field @xmath189 in the homotopy class of @xmath121 which differs from @xmath135 only in a small set . the inclusion @xmath190 can be extended to a diffeomorphism @xmath191 , as in figure [ fighf ] ( ) . we will use this description of @xmath52 for the rest of this section . let @xmath192 and @xmath193 be the sets of @xmath194 and @xmath195 curves under this diffeomorphism . note that @xmath196 and that @xmath197 . we can also assume that @xmath198 and that @xmath199 and @xmath157 coincide outside a small neighborhood of @xmath200 , see figure [ fighf ] ( ) . we isotope @xmath61 from @xmath201 to @xmath202 in the complement of the @xmath194 and @xmath195 circles . we now choose a morse function @xmath105 and a gradient - like vector field @xmath106 for @xmath105 such that @xmath104 is compatible with @xmath159 and such that the index two and one critical points are contained in @xmath203 and @xmath204 , respectively . fix @xmath205 and let @xmath206\cup[1-{\varepsilon},1])$ ] . since @xmath106 and @xmath135 are positively transverse to @xmath44 on @xmath207 , we can assume that @xmath208 in @xmath207 . for each @xmath209 , let @xmath210 be a small neighborhood of @xmath211 in @xmath212 containing @xmath213 and let @xmath214 be a small thickening of it . we can see @xmath215 and @xmath216 as subsets of @xmath184 . we can assume that @xmath104 is chosen so that @xmath217 $ ] contains @xmath218 , where @xmath110 and @xmath219 denote the unstable and stable manifolds of the index one and two critical points corresponding to @xmath220 and @xmath62 , respectively . for @xmath221 , we define @xmath222\big)\setminus m_0 \subset y. \label{def : m}\ ] ] let @xmath223 be a small neighborhood of the flow line passing through @xmath61 and let @xmath224 be the vector field obtained by modifying @xmath106 in @xmath223 as in figure [ defn](b ) . we note that @xmath224 does not vanish in @xmath225 . for a generator @xmath188 , its grading @xmath121 is the homotopy class of a vector field obtained by modifying @xmath224 in @xmath226 . we will now show how to homotope @xmath224 in the complement @xmath226 to obtain a vector field which mostly coincides with @xmath135 . [ lem : iso ] the vector field @xmath224 is homotopic to a nonvanishing vector field @xmath227 in @xmath228 relative to its boundary such that @xmath227 coincides with @xmath135 in @xmath229 . the proof of this lemma is essentially the last paragraph of the proof of ( * ? ? ? * theorem 1.1(b ) ) . we will rewrite it here for the reader s sake . let @xmath230 . it follows from the definition of the arcs @xmath231 that @xmath232 is topologically a disk with @xmath233 boundary punctures . so , @xmath8 is diffeomorphic to a three - ball with @xmath233 interior punctures . we can choose this diffeomorphism so that @xmath234 is contained on the @xmath124-plane in @xmath235 , as shown in figure [ hfmod ] ( ) . under this diffeomorphism , @xmath236 is a union of very small punctured balls centered at the @xmath233 punctures from above . the vector field @xmath224 is depicted in figure [ hfmod ] ( ) . a horizontal section of @xmath106 can be seen in figure [ hfmod ] ( ) , where the green punctured disks represent @xmath236 . we can isotope @xmath224 so the closed orbit contained on this horizontal section coincides with the binding @xmath237 outside @xmath238 , see figure [ hfmod ] ( ) . the vector field obtained by this isotopy can be chosen to coincide with @xmath135 in @xmath239 . so we let @xmath227 be this vector field . [ hf1 ] [ hf2 ] [ hf3 ] [ lj ] [ hf4 ] now we choose disjoint neighborhoods of the flow lines corresponding to each @xmath240 which are contained in @xmath226 . we define @xmath189 to be the vector field obtained by modifying @xmath227 in these neighborhoods as in [ sec : absgrad ] . it follows from lemma [ lem : iso ] that @xmath241={\text{\textnormal{gr}}}({\mathbf{x}})\in{\text{vect}}(y)$ ] . in this section , we will recall the definition of the ech chain complex and its absolute grading . for more details , see @xcite . let @xmath0 be a closed , oriented three - manifold , let @xmath46 be a contact form on @xmath0 and let @xmath242 . the ech chain complex @xmath43 is generated by finite orbit sets @xmath243 , where @xmath122 are distinct embedded reeb orbits , @xmath244 are positive integers , and @xmath245 whenever @xmath122 is hyperbolic . the chain complex @xmath43 splits by homology classes as @xmath246 where @xmath247 is the subcomplex generated by the orbit sets @xmath248 whose total homology class @xmath249=\gamma\in h_1(y)$ ] . now consider two orbit sets @xmath248 and @xmath250 whose total homology classes equal @xmath251 and let @xmath252 be a 2-chain in @xmath253 $ ] such that @xmath254 let @xmath255 denote a trivialization of @xmath132 over all of the reeb orbits @xmath122 and @xmath256 and let @xmath257 be the relative first chern class of @xmath258 with respect to @xmath255 . let @xmath259 be the relative self - intersection number of @xmath252 with respect to @xmath255 , as explained in @xcite , and let @xmath260 where @xmath261 denotes the conley - zehnder index of the reeb orbit @xmath262 with respect to the trivialization @xmath255 . in @xcite , hutchings defined the _ ech index _ to be @xmath263 one can show that @xmath264 does not depend on @xmath255 and only depends on the relative homology class of @xmath252 . if we change the relative homology class of @xmath252 , the ech index @xmath265 changes by a multiple of @xmath22 , where @xmath22 is the divisibility of @xmath266 . so , @xmath267 is well defined in @xmath21 . hence @xmath247 is relatively graded by @xmath21 . the differential @xmath268 is defined as follows . let @xmath269 be the symplectization of @xmath0 , where @xmath72 denotes the @xmath74-coordinate , and let @xmath270 be a cylindrical almost - complex structure on @xmath271 . the coefficient of @xmath272 in @xmath268 is the signed count of @xmath270-holomorphic curves @xmath273 in @xmath271 that have ech index @xmath274 , and that converge to @xmath275 as @xmath82 and @xmath272 as @xmath83 , modulo translation . it is shown in @xcite that for a generic @xmath270 the differential @xmath69 is well - defined and @xmath86 . so one defines @xmath276 to be the homology of @xmath247 for a given @xmath270 . it follows from @xcite that @xmath276 does not depend on the choice of @xmath270 , so we will omit @xmath270 from the notation . we will also denote by @xmath277 the homology of @xmath43 . it follows from the definition of @xmath69 that @xmath278 moreover , by taubes @xcite , for two contact forms @xmath279 , there exists an isomorphism @xmath280 . therefore one can write @xmath28 . we note that there are additional structures on @xmath277 that do depend on the contact form , e.g. ech capacities . we now recall the definition of the absolute grading on ech from @xcite . this grading takes values on the set of homotopy classes of oriented 2-plane fields on @xmath0 . the orientation of @xmath0 induces an isomorphism from this set to the set of homotopy classes of nonvanishing vector fields @xmath3 . let @xmath248 be an orbit set . the absolute grading @xmath281 is the homotopy class of the vector field obtained by modifying the reeb vector field in disjoint neighborhoods of the reeb orbits @xmath122 , as follows . for each @xmath161 , fix a small tubular neighborhood of @xmath122 and choose a braid @xmath282 with @xmath244 strands in that neighborhood . let @xmath283 be the union of the braids @xmath282 . a trivialization @xmath284 of @xmath132 over each @xmath122 , induces a framing @xmath284 on each @xmath282 . let @xmath255 denote this framing on @xmath283 . now , for each component @xmath285 of @xmath283 , let @xmath286 denote a small neighborhood of @xmath285 in @xmath0 . we can choose these neighborhoods so that @xmath286 and @xmath287 do not intersect for different components @xmath285 and @xmath288 . the framing on @xmath285 induces a diffeomorphism @xmath289 and a trivialization of @xmath290 , identifying @xmath291 and @xmath292 . using the previous identifications , one can define a vector field @xmath293 on @xmath286 as @xmath294 one now constructs a vector field on @xmath0 by defining it to be given by @xmath293 in each neighborhood @xmath286 and to equal the reeb vector field in the complement of the union of the neighborhoods @xmath286 . let @xmath295 be the homotopy class of this vector field . now define @xmath296 here @xmath297 denotes the writhe of @xmath282 with respect to @xmath284 and @xmath298 one can check that @xmath281 does not depend on the choice of @xmath255 or @xmath283 . in @xcite , hutchings proved the following proposition . let @xmath275 and @xmath272 be orbit sets with @xmath299=[\sigma]\in h_1(y)$ ] and let @xmath22 denote the divisibity of @xmath300)\in h^2(y;{{\mathbb z}})$ ] . then @xmath301 for @xmath29 , we denote by @xmath302 the submodule of @xmath277 consisting of the elements of grading @xmath262 . it follows from @xcite that for two contact forms @xmath279 , the isomorphism @xmath280 restricts to an isomorphism @xmath303 for every @xmath29 . so we can write @xmath27 for @xmath29 . the @xmath31 map is defined similarly to the differential . for an orbit set @xmath275 , one defines @xmath304 to be the signed count of @xmath270-holomorphic curves @xmath273 in @xmath271 with @xmath305 that go through a fixed basepoint in @xmath271 . the @xmath31 map is a degree @xmath306 chain map @xmath307 . the chain complex @xmath308 is defined to be the mapping cone of @xmath31 . the homology of @xmath308 is denoted by @xmath309 . again , it follows from @xcite that the @xmath31 map in homology does not depend on @xmath270 or @xmath46 so one can write @xmath33 . we obtain an exact triangle , as follows . @xmath310^{u } & & ech(y ) \ar[dl ] \\ \label{tri : ech } & \ar[lu ] \widehat{ech}(y ) & } \end{aligned}\ ] ] we define the absolute grading on @xmath308 so that the map @xmath311 has degree 0 . hence for @xmath29 , we can write @xmath312 . we note that the map @xmath313 has degree 1 . in this subsection , we will show that the cobordisms maps in ech defined by hutchings and taubes in @xcite preserve the absolute grading . this fact will be used in the next subsection . let @xmath46 be a contact form on @xmath0 . the symplectic action of an orbit set @xmath314 is defined to be @xmath315 . for @xmath316 , the filtered ech chain complex @xmath317 is defined to be the subcomplex of @xmath43 generated by all orbit sets @xmath275 with @xmath318 . since the differential decreases the action , the subgroup @xmath317 is indeed a subcomplex . its homology is denoted by @xmath319 and it is independent of the almost - complex structure by ( * ? ? ? * theorem 1.3(a ) ) . for @xmath320 , let @xmath321 be a 3-manifold with contact form @xmath322 . an exact symplectic cobordism from @xmath323 to @xmath324 is a pair @xmath325 , where @xmath79 is a compact 4-manifold , @xmath326 is a symplectic form , @xmath327 and @xmath328 for @xmath320 . according to ( * ? ? ? * theorem 1.9 ) , such corbordisms induce maps @xmath329 the maps @xmath330 are constructed by taking the composition of the corresponding map in seiberg - witten floer homology and the isomorphism from ech to seiberg - witten floer homology . [ lem : cob ] let @xmath331,d\lambda)$ ] be an exact cobordism from @xmath332 to @xmath333 . then , for every @xmath316 , the map @xmath334,\lambda)$ ] preserves the absolute grading , i.e. @xmath334,\lambda)$ ] maps @xmath335 to @xmath336 for every @xmath29 . the maps @xmath334,\lambda)$ ] are defined as a composition of maps @xmath337 here @xmath338 and @xmath339 are appropriate filtered seiberg - witten floer cohomology groups , as explained in @xcite . the second map in is a filtered version of the cobordism maps defined in @xcite . now it follows from the definition of these maps that if an element of @xmath338 has grading @xmath340\in{\text{vect}}(y)$ ] , then its image in @xmath341 is the sum of elements of ( possibly different ) gradings @xmath342 $ ] such that for each such @xmath343 there exists an almost - complex structure @xmath270 on @xmath253 $ ] satisfying @xmath344 now by considering @xmath345 , we conclude that @xmath346 and @xmath347 are homotopic . hence @xmath348 . so the second map in preserves the absolute grading . now , the first and third maps in preserve the grading by @xcite . therefore @xmath334,\lambda)$ ] preserves the grading . we now recall from @xcite how to compute @xmath349 and @xmath277 from an open book decomposition . we fix an open book decomposition @xmath41 of @xmath0 and write @xmath350 , as in [ sec : hfobd ] . let @xmath351 and let @xmath73 denote the @xmath75$]-coordinate in @xmath47 as in [ sec : hfobd ] . we again assume that @xmath66 satisfies @xmath145 in a neighborhood of @xmath137 . let @xmath46 be a contact form on @xmath47 such that @xmath135 is parallel to @xmath352 in @xmath47 . hence the torus @xmath353 is foliated by reeb orbits . up to a small isotopy of @xmath66 , we can assume that all the reeb orbits in the interior of @xmath47 are nondegenerate and that @xmath353 is a negative morse - bott torus . after a morse - bott perturbation , we obtain a pair of reeb orbits @xmath354 on @xmath353 . let @xmath355 be the chain complex generated by orbit sets contructed from reeb orbits in the interior of @xmath47 and @xmath356 and let @xmath357 be the chain complex generated by orbit sets contructed from reeb orbits in the interior of @xmath47 and @xmath354 . the differential in both cases counts morse - bott buildings of ech index 1 . let @xmath358 and @xmath359 denote the homology of these chain complexes . now let @xmath360 and @xmath361 denote the quotients of @xmath358 and @xmath359 by the equivalence relations generated by @xmath299\simeq [ e\gamma]$ ] , respectively . in @xcite , colin , ghiggini and honda proved that @xmath360 and @xmath361 are well - defined and they constructed isomorphisms @xmath362 we will now show that @xmath360 and @xmath361 have well - defined gradings and that and preserve the gradings . we start by recalling the construction of the isomorphism @xmath363 . we write @xmath364)$ ] where @xmath106 is a tubular neighborhood of the binding @xmath365 , which is again a solid torus . let @xmath366 be a contact form on @xmath106 which is nondegenerate in the interior of @xmath106 such that the reeb vector field of @xmath366 is positively transverse to the interior of the pages and positively tangent to the binding and such that @xmath367 is a _ positive _ morse - bott torus . the precise construction of @xmath366 will not be necessary here and we refer the reader to @xcite . we denote by @xmath368 and @xmath369 the elliptic and hyperbolic orbits obtained after a morse - bott pertubation . let @xmath370 be an increasing sequence such that @xmath371 . following @xcite , we can choose a family of contact forms @xmath372 on @xmath0 which equal @xmath46 in a neighborhood of @xmath47 and a positive multiple of @xmath366 in a neighborhood of @xmath106 such that @xmath373 is a morse - bott contact form and all reeb orbits in @xmath374 $ ] have action larger than @xmath375 . so as in @xcite , we can perturb @xmath372 to a sequence of contact forms @xmath376 satisfying , in particular , the following conditions : * @xmath377 coincides with @xmath373 outside neighborhoods of the morse - bott tori . * the reeb orbits of @xmath373 of action less than @xmath375 are nondegenerate and are either the reeb orbits of @xmath46 and @xmath366 in the interior of @xmath47 and @xmath106 , respectively , or one of the orbits @xmath378 , @xmath379 , @xmath368 or @xmath369 . hence @xmath380 is generated by elements of the form @xmath381 , where @xmath382 is an orbit set contructed from reeb orbits in the interior of @xmath106 and @xmath383 , and @xmath384 is a generator of @xmath357 . for @xmath316 , let @xmath385 be the subcomplex of @xmath386 generated by orbit sets @xmath275 with action @xmath387 and whose total homology class intersects a page up to @xmath388 times . following @xcite , we can define another increasing sequence @xmath389 with @xmath390 such that the maps @xmath391 below are well - defined . @xmath392 here @xmath393 is defined by the equation @xmath394 , where @xmath395 and @xmath396 are the differentials in @xmath357 and @xmath355 , respectively . it follows from ( * ? ? ? * lemma 9.7.2 ) that the maps @xmath391 are chain maps so they induce maps @xmath397 . following ( * ? ? ? 3.2.3 ) , there are chain maps @xmath398 which are given by cobordism maps as in [ sec : cob ] . so we obtain a directed system @xmath399^{\sigma_k}\ar[d]^{\iota_k } & ecc^{l_k}(y,\lambda_k')\ar[d]^{\phi_k}\\ ecc^{\flat , l_{k+1}'}_{\le k+1}(n,\lambda)\ar[r]^{\sigma_k } & ecc^{l_{k+1}}(y,\lambda_{k+1 } ' ) } \label{eq : iota}\ ] ] where @xmath400 denotes the inclusion . the maps @xmath401 induce maps in homology with respect to which one can take the direct limit @xmath402 . there is also a nondegenerate contact form @xmath403 and cobordism maps @xmath404 . it is shown in ( * ? ? ? 3.2.3 ) that the direct limit of these maps is an isomorphism @xmath405 now we note that @xmath406 . therefore the maps @xmath391 give rise to a map @xmath407 the calculations in @xcite imply that the image of the map @xmath408 given by @xmath299\mapsto[\gamma]-[e\gamma]$ ] is contained in the kernel of @xmath409 . hence we obtain a map @xmath410 it is shown in ( * ? ? ? * theorem 9.8.3 ) that @xmath363 is an isomorphism . we will now prove a lemma that will be useful to show that the absolute grading is well - defined in @xmath360 and that @xmath363 preserves the grading . [ lem : greh ] let @xmath275 be an orbit set obtained from the reeb orbits of @xmath46 in the interior of @xmath47 , respectively , and the orbits @xmath378 , @xmath379 , @xmath368 or @xmath369 . then @xmath411 is well - defined . moreover , @xmath412 to see that @xmath275 has a well - defined grading , first note that there exists @xmath413 such that @xmath414 for every @xmath415 . so we define @xmath281 using the contact form @xmath377 for some @xmath415 . it follows from lemma [ lem : cob ] that the maps @xmath401 preserve the grading . so @xmath416 is well - defined . to prove , we can restrict to the case when @xmath275 does not contain @xmath378 , @xmath379 , @xmath368 or @xmath369 . the general case is a straightforward consequence of this case . let @xmath255 be a trivialization of @xmath132 over @xmath275 and let @xmath283 be a link as in [ abs : ech ] so that @xmath417 , where @xmath418 denotes the sum of the writhes of all components of @xmath283 . let @xmath419 . we can choose a trivialization @xmath420 of @xmath421 that is trivial with respect to the morse - bott torus containing @xmath422 . let @xmath423 be a knot obtained by pushing @xmath422 in a direction which is transverse to the morse - bott torus containing @xmath422 such that @xmath423 is in the interior of @xmath424 . then @xmath425 . now let @xmath426 a the disk in @xmath424 bounding @xmath422 . it follows from ( * ? ? ? * lemma 3.4(d ) ) that @xmath427 moreover , @xmath428 therefore it follows from that holds . the module @xmath360 has a well - defined absolute grading and the isomorphism @xmath429 preserves the grading . it follows from lemma [ lem : greh ] that the grading on @xmath358 is well - defined . we recall that @xmath360 is the quotient of @xmath358 by the equivalence relation given by @xmath299\sim[e\gamma]$ ] . by lemma [ lem : greh ] that @xmath430 . so the grading on @xmath360 is well - defined . let @xmath275 be an orbit set in @xmath431 for some @xmath388 . since @xmath395 decreases the grading by @xmath432 , it follows that @xmath433 . now , by lemma [ lem : greh ] , @xmath434 . hence for all @xmath435 , @xmath436 so @xmath391 preserves the grading . now , it is tautological that the inclusion @xmath400 in preserves the grading . moreover , by lemma [ lem : cob ] , the maps @xmath401 and the isomorphism preserve the grading . hence after passing to homology and taking the direct limit we conclude that @xmath409 , and hence @xmath363 , preserve the grading . it also follows from lemma [ lem : greh ] that the gradings on @xmath359 and @xmath361 are well - defined . we now define two chain maps as follows . @xmath437 here @xmath438 and @xmath439 do not contain @xmath379 . these maps descend to homology and to the quotients @xmath361 and @xmath360 . it follows from @xcite that these maps fit into an exact triangle @xmath440 & & ech(n,\partial n,\lambda ) \ar[dl]^{\iota _ * } \\ \label{tri : echrel } & \ar[lu]^{\pi _ * } \widehat{ech}(n,\partial n,\lambda ) & } \end{aligned}\ ] ] where the map @xmath441 is a version of the @xmath31 map . moreover there exists an isomorphism @xmath442 such that @xmath363 and @xmath443 give an isomorphism from to . it follows from that @xmath36 increases the grading by @xmath432 and that @xmath444 preserves the grading . hence we obtain the following long exact sequences . @xmath445 & ech_{\rho-1}(n,\partial n)\ar[r]\ar[d]^{\psi_1 } & \widehat{ech}_\rho(n,\partial n ) \ar[r]\ar[d]^{\widehat{\psi}_1 } & ech_{\rho}(n,\partial n)\ar[r]\ar[d]^{\psi_1}&\dots \\ \dots\ar[r]^-u & ech_{\rho-1}(y)\ar[r ] & \widehat{ech}_\rho(y ) \ar[r ] & ech_{\rho}(y)\ar[r]^-u&\dots } \label{eq : commdiag}\ ] ] here @xmath446 and we dropped the dependence on @xmath46 . in fact , because of the isomorphisms @xmath363 and @xmath443 , the modules on the upper row of do not depend on @xmath46 . it follows from that @xmath443 preserves the grading . we end this section with the a discussion of @xmath447 and the grading on it . the chain complex @xmath447 is a subcomplex of @xmath357 generated by orbits sets which intersect @xmath44 exactly @xmath45 times where @xmath53 is the genus of @xmath44 . this inclusion induces an absolute grading on @xmath447 and a map in homology . by composing this map with the quotient map @xmath448 we obtain a map @xmath449 which preserves the grading . it is shown in @xcite that @xmath450 is an isomorphism . let @xmath451 . therefore @xmath452 is an isomorphism and it preserves the grading . in this section , we will prove that the absolute grading is preserved under the isomorphism from heegaard floer homology to ech defined by colin - ghiggini - honda in @xcite . we now recall the construction of the map @xmath49 on the chain level @xmath454 this map is defined by counting rigid holomorphic curves with an ech - type index equal to 0 . we now review the relevant moduli spaces and this ech - type index . throughout this section we fix an open book @xmath41 for @xmath0 satisfying the conditions given in [ sec : hfobd ] and we let @xmath455 be the mapping torus of @xmath66 . we denote by @xmath53 the genus of @xmath44 and we let @xmath46 be a contact form on @xmath0 which is adapted to @xmath41 . in order to prove that @xmath49 is an isomorphism , it is necessary to make a more specific choice of @xmath46 as it is done in @xcite , but this particular choice does not affect the absolute grading . let @xmath456 be the map @xmath457 and let @xmath458 , where @xmath459 . we also round the corners of @xmath8 . now define @xmath460 and @xmath461 , where @xmath76 is a certain area form on @xmath44 . then @xmath462 is a symplectic manifold with boundary . it has a positive end , which is diffeomorphic to @xmath463 $ ] and a negative end , which is diffeomorphic to @xmath47 . the map @xmath16 restricts to a symplectic fibration @xmath464 which admits a symplectic connection whose horizontal space is spanned by @xmath465 . now if we take a copy of @xmath466 on the fiber @xmath467 and take its symplectic parallel transport along @xmath468 , we obtain a lagrangian submanifold of @xmath462 , which is denoted by @xmath469 . for each @xmath470 we denote by @xmath471 the corresponding component of @xmath469 . we will consider @xmath270-holomorphic maps @xmath472 where @xmath473 is a riemann surface with boundary and punctures , both in the interior and on the boundary . a puncture @xmath474 is said to be positive or negative if the @xmath72-coordinate of @xmath475 converges to @xmath476 or @xmath477 , respectively , as @xmath478 . now to each generator @xmath90 of @xmath166 we can associate a subset of @xmath463 $ ] given by the union of @xmath479 $ ] , for all @xmath480 . we will still denote the union of these chords by @xmath90 . given @xmath90 , an orbit set @xmath248 in @xmath447 and an admissible almost - complex structure @xmath270 , one defines @xmath481 to be the moduli space of @xmath270-holomorphic maps @xmath472 satisfying the following conditions : 1 . @xmath482 and each component of @xmath483 is mapped to a different @xmath471 . the boundary punctures are positive and the interior punctures are negative . 3 . at each boundary puncture , @xmath484 converges to a different chord @xmath479 $ ] and every chord @xmath479 $ ] is such an end of @xmath484 . 4 . at an interior puncture , @xmath484 converges to an orbit @xmath122 with some multiplicity . for each @xmath161 , the total multiplicity of all ends converging to @xmath122 is @xmath244 . the energy of @xmath484 is bounded . let @xmath485 denote the compactification of @xmath486 obtained by compactifying @xmath74 to @xmath487 . a continuous map @xmath488 satifying ( a)(d ) above can be compatified to a map @xmath489 mapping @xmath490 to @xmath491 . two such maps @xmath484 , @xmath10 are said to be homologous if the images of @xmath492 and @xmath493 are homologous in @xmath494 . let @xmath495 denote the set of homology classes of such maps @xmath488 . for a homology class @xmath496 , one defines its ech - index @xmath497 as follows . let @xmath488 be a continuous map satifying ( a)(d ) above such that @xmath498=a$ ] and let @xmath499 be its compactification . now note that one can view @xmath500 as a sub - bundle of @xmath501 . we choose an orientation of the arcs @xmath152 , which gives rise to a nonvanishing vector field along each @xmath152 . this vector field induces a trivialization @xmath255 of @xmath500 along @xmath502 . we extend this trivialization arbitrarily along @xmath503 $ ] and along @xmath504 . let @xmath505 denote the first chern class of @xmath506 relative to @xmath255 . now let @xmath507 and @xmath508 be distinct embedded surfaces in @xmath485 given by pushing @xmath509 off along vectors field which are transverse to it and trivial with respect to @xmath255 along the boundary . for more details see @xcite . then @xmath510 is defined to be the signed count of intersections of @xmath507 and @xmath508 . now let @xmath511 be a real , rank one subbundle of @xmath500 along @xmath512 $ ] defined as follows . at @xmath513 , let @xmath514 and at @xmath515 , let @xmath516 in @xmath500 . then @xmath511 is defined by rotating counterclockwise by the minimum possible amount as we travel along @xmath512 $ ] . one defines @xmath517 to be the sum of the maslov indices of @xmath511 along each @xmath479 $ ] with respect to @xmath255 . the ech - index is defined as @xmath518 now @xmath519 is defined as follows . the coefficient of an orbit set @xmath275 in @xmath519 is the ( signed ) count of maps @xmath484 in @xmath481 with @xmath520)=0 $ ] . as explained in @xcite , for a generic @xmath270 this count is finite and all the maps that are counted are embeddings . showing that the map @xmath49 induces an isomorphism on homology is far from trivial and it is done in @xcite . we remark that @xcite , they only construct @xmath49 over @xmath521 coefficients . our construction and , in particular proposition [ prop : indiso ] , do not depend on an eventual choice of signs . we first recall a relative version of the pontryagin - thom construction . let @xmath10 and @xmath523 be nonvanishing vector fields on a closed and oriented three - manifold @xmath0 . assume that @xmath10 and @xmath523 coincide in @xmath524 , where @xmath31 is an open set in @xmath0 . let @xmath255 be a trivialization of @xmath525 and let @xmath526 be a regular value of both @xmath10 and @xmath523 seen as maps @xmath527 . the one - manifolds @xmath528 and @xmath529 inherit framings by considering the isomorphisms of the normal bundles with @xmath530 given by @xmath531 and @xmath532 along @xmath533 and @xmath534 , respectively . now if @xmath533 and @xmath534 are contained in the interior of @xmath31 and are homologous in @xmath31 , there is a link cobordism @xmath535 $ ] from @xmath533 to @xmath534 . that is , @xmath273 is a surface such that @xmath536 . the framings on @xmath533 and @xmath534 induce a framing on @xmath273 along @xmath537 which we denote by @xmath538 . [ lem : pt ] let @xmath10 and @xmath523 be nonvanishing vector fields and @xmath533 and @xmath534 the links as above . let @xmath273 be an immersed cobordism from @xmath533 to @xmath534 and let @xmath539 denote the number of self - intersections of @xmath273 . let @xmath538 denote the framing on @xmath273 along @xmath537 which is induced by the framings on @xmath533 and @xmath534 . then @xmath540-[w]=c_1(nc,\nu)+2\delta(c).\ ] ] first assume that @xmath273 is an embedded surface . we can find a framing @xmath541 of @xmath273 which coincides with @xmath538 along @xmath542 . it follows from the pontryagin - thom contruction that @xmath7-[w]$ ] equals the difference of the framings @xmath541 and @xmath538 along @xmath543 . but this difference is given by @xmath544 . now the general case follows from ( * ? ? ? * lemma 2.3 ) . we write @xmath248 . let @xmath488 be an immersion such that @xmath498=a$ ] and let @xmath545 denote its continuous compactification . we note that by rounding the corners of @xmath485 , we obtain a trivial cobordism from @xmath47 to itself which we denote by @xmath546 . here we identify @xmath547 $ ] . let @xmath283 be the union of disjoint braids @xmath282 around @xmath122 with total multiplicity @xmath244 . then by moving @xmath548 in the direction of some transverse vector field near @xmath549 , we obtain an immersed cobordism @xmath273 from @xmath550 to @xmath551 , where @xmath173 is the union of @xmath90 with segments on the arcs @xmath152 . up to an isotopy , we can assume that @xmath550 is transverse to @xmath552 . we also note that , under our identification , @xmath553)\times\{1\}$ ] . let @xmath554 be the subset defined by . let @xmath189 be the vector field defined in [ sec : hfobd ] and let @xmath555 be the vector field defined in [ sec : absgrad ] whose homotopy class is @xmath556 . let @xmath557 be a small open neighborhood of @xmath558 . then , by lemma [ lem : iso ] , we can assume that @xmath189 and @xmath555 coincide in @xmath559 . we will now use lemma [ lem : pt ] . we first choose a nonvanishing tangent vector field along the arcs @xmath152 . that induces a trivialization of @xmath560 . since @xmath561 , we obtain a trivialization of @xmath562 . we extend it arbitrarily to @xmath563 . we can now extend this trivialization to a trivialization @xmath255 of @xmath242 on @xmath564 . so @xmath565 is a trivialization of @xmath566 where the reeb vector field @xmath135 in @xmath557 is mapped to @xmath567 . we observe that @xmath568 . the framing can be calculated by considering the preimage of a vector near @xmath569 . the corresponding link gives a framing of the normal bundle @xmath570 which coincides with @xmath255 . now we compute @xmath571 . first note that this link is contained in @xmath238 . we observe that @xmath571 is a link @xmath572 which is a slight pertubation of the union of the flow lines corresponding to the points @xmath480 and the points @xmath573 from figure [ fighf ] ( ) . the framing on @xmath572 is a trivialization of @xmath574 which we denote by @xmath538 . the link @xmath572 is isotopic to @xmath173 so we can assume that @xmath575 . so we can see @xmath538 as a trivialization of @xmath576 along @xmath577 . as in lemma [ lem : pt ] , the framing on @xmath551 is also denoted by @xmath538 . by lemma [ lem : pt ] , @xmath578 the trivialization @xmath255 also gives rise to a trivialization of @xmath576 along @xmath577 and @xmath551 . we claim that @xmath579 to prove the claim , we first compute the difference @xmath580 . this difference is given by @xmath581 as framings of @xmath582 , since @xmath583 along @xmath551 . we will now identify @xmath584 , where @xmath585 is the union of the flow lines corresponding to @xmath90 and @xmath586 is the union of the flow lines corresponding to all @xmath587 . let @xmath588 be a vector field along @xmath572 which is trivial with respect to @xmath255 and which coincides with the vector field tangent to each @xmath152 as chosen above along @xmath586 . it follows from the isotopy @xmath589 and from the definition of @xmath255 that at the index two or one critical points , @xmath588 is tangent to the stable or unstable submanifolds , respectively . moreover , the vector field @xmath588 rotates a quarter of a turn counterclockwise about each component of @xmath590 as we go from an index two to an index one critical point . so along each component of @xmath586 , the trivializations @xmath538 and @xmath255 differ by a half - turn clockwise . now we compute the difference @xmath581 along each component of @xmath585 as we go from an index one to an index two critical point . if @xmath588 rotates rotates a quarter of a turn counterclockwise about a certain component , we again obtain a contribution of @xmath591 to @xmath581 . in that case , this component will contribute by @xmath592 to @xmath593 . now say that @xmath588 differs from a quarter of a turn counterclockwise rotatation by @xmath19 counterclockwise half - turns , for some @xmath20 . then we obtain a contribution of @xmath594 to @xmath581 and @xmath595 to @xmath593 . therefore the total difference @xmath581 along @xmath572 is @xmath596 and we have proven . it remains to compute @xmath597 . for that , we will use a classical construction in topology . consider a generic section of @xmath598 which is trivial with respect to @xmath255 along @xmath537 . we move @xmath273 in the direction of this section and we obtain a surface @xmath599 which intersects @xmath273 tranversely . then @xmath600 where @xmath601 denotes the signed count of intersections of @xmath273 and @xmath599 . now @xmath273 and @xmath599 can be completed to surfaces in the homology class of @xmath92 . but these surfaces are not necessarily @xmath255-trivial . in fact , the linking number of @xmath537 and @xmath602 in @xmath549 is @xmath603 . following a standard calculation in ech , see e.g. @xcite , we obtain @xmath604 recall that @xmath38 . by proposition [ prop : indiso ] , @xmath49 preserves the grading . moreover , it follows from our constructions in [ sec : hfobd ] and [ sec : echobd ] that @xmath48 and @xmath50 preserve the grading . therefore @xmath37 preserves the grading . to3em , _ the embedded contact homology index revisited _ , new perspectives and challenges in symplectic field theory , crm proc . lecture notes , vol . 49 , amer . soc . , providence , ri , 2009 , pp .

we consider the klein - gordon and sine - gordon type equations with a point - like potential , which describes the wave phenomenon in disordered media with a defect . the singular potential term yields a critical phenomenon that is , the solution behavior around the critical parameter value bifurcates into two extreme cases . finding such critical parameter values and the associated statistical quantities demands a large number of individual simulations with different parameter values . pinpointing the critical value with arbitrary accuracy is even more challenging . in this work , we adopt the generalized polynomial chaos ( gpc ) method to determine the critical values and the mean solutions around such values . first , we consider the critical value associated with the strength of the singular potential for the klein - gordon equation . we show the existence of a critical behavior with certain boundary conditions . then we expand the solution in the random variable associated with the parameter . the obtained partial differential equations are solved using the chebyshev collocation method . due to the existence of the singularity , the gibbs phenomenon appears in the solution , yielding a slow convergence of the numerically computed critical value . to deal with the singularity , we adopt the consistent spectral collocation method . the gpc method , along with the consistent chebyshev method , determines the critical value and the mean solution highly efficiently . we then consider the sine - gordon equation , for which the critical value is associated with the initial velocity of the kink soliton solution . the critical behavior in this case is that the solution _ passes _ through ( particle - pass ) , is _ trapped _ by ( particle - capture ) , or is _ reflected _ by ( particle - reflection ) the singular potential if the initial velocity of the soliton solution is greater than , equal to , or less than the critical value , respectively . due to the nonlinearity of the equation , we use the gpc mean value rather than reconstructing the solution to find the critical parameter . numerical results show that the critical value can be determined efficiently and accurately by using the proposed method . the results are also compared with the results using the monte - carlo method . critical phenomenon , klein - gordon equation , sine - gordon equation , galerkin methods , generalized polynomial chaos , spectral method , legendre polynomials , hermite polynomials , uncertainty quantification , gibbs phenomenon

1108.4720

critical phenomena of partial differential equations ( pdes ) associated with the critical parameter can be found in various physical applications . for example , in @xcite the gravitational collapse of the scalar field was considered . the critical phenomenon of the gravitational collapse was discovered ; there exists a critical initial mass such that if the initial mass is less than the critical mass , the scalar field eventually disperses . conversely , it forms a black hole if the value exceeds the critical mass . if the initial parameter is arbitrarily close to the critical value , scale - invariant behavior of the solution occurs @xcite . for pdes from nonlinear optics , a similar phenomenon has been found and studied , e.g. the sine - gordon equation or nonlinear schrdinger equation with a point - like singular potential term @xcite . for the sine - gordon equation , there exists a kink soliton solution ; this yields a critical behavior once a self - interacting singular potential is placed . this phenomenon is known as the wave phenomenon in disordered media with a defect . if the initial velocity of the kink soliton solution exceeds the critical velocity , the kink soliton eventually passes through the potential , which is known as the _ particle - pass_. if the initial velocity is less than the critical velocity , the soliton is either captured by or reflected by the potential , known as the _ particle - capture _ or _ particle - reflection _ , since the sine - gordon equation is not integrable when the singular potential term is present , no exact solution is available ; no exact value of the critical parameter is available either . there are many unknown properties of the solution behaviors in this case . for more detailed study , we refer the readers to @xcite . a similar phenomenon has been also found with nonlinear schrdinger equations @xcite . for example , in @xcite the singular potential term perturbs the soliton propagation . similar phenomena of particle - pass , particle - capture , and particle - reflection were observed for some critical parameters . in this paper , we consider the critical phenomenon that is induced by the point - like singular source term in the pdes , such as the sine - gordon equation mentioned above . in particular , we consider the klein - gordon and sine - gordon equations to be perturbed by the singular potential term . for both cases , the solution behaviors are highly sensitive to the critical value of the parameter . determination of the critical value and the mean solution is an important task because the wave or soliton interactions with the local defect should be well understood for better optical communications . in either particle - capture or particle - reflection , the optical signal is not transmitted with the associated parametric value if it is less than the critical value . the signal should be transmitted with some values of the parameter that exceed the critical value for the communication without blocking or trapping @xcite . finding the critical value is doable but challenging . there are two reasons . first , the singular potential is given by the dirac @xmath0-function , so the given pdes are not well - defined . for the sine - gordon equation , no exact solution is known with the presence of the singular potential term , which means that no analytic form of the critical value is available only the numerical or real laboratory experimental approach can render the critical value measurable . the numerical determination of the critical parameter value relies on expensive monte - carlo(mc ) simulations based on the bi - section method . moreover , it is highly expensive to determine the critical value up to an arbitrary accuracy ( for the sine - gordon case , it is not even known whether the critical value is rational or irrational ) . furthermore , the numerically obtained critical value depends on the numerical method , considered with given numerical settings such as the truncation number , domain size , final time etc . due to the singularity in the potential term , the gibbs phenomenon also appears in the finite truncated solution , resulting in a slow convergence . thus when we talk about finding the critical value , we mean the critical value defined by the numerical method used , not the _ exact _ value from the pdes . this will be explained in detail , using the klein - gordon equation in the paper . second , although the singular potential term is used to mimic the effect of the local defect , such a model is not perfect . associated parameters , such as the strength and the location of the singular potential terms , have a great degree of uncertainty . thus , determining the critical value needs to be understood within the context of the uncertainty analysis . in this paper , as the first step toward thoroughly analyzing these two issues , we propose to use the generalized polynomial chaos ( gpc ) method @xcite for finding the critical value of both the klein - gordon and sine - gordon equations and the associated statistical quantities such as mean solutions around the critical value . the gpc method has been successfully implemented to analyze the uncertainty quantification for problems such as fluid flows with uncertain initial and boundary conditions , sensitivity analysis , steady - state problems , etc . @xcite . in our proposed method , we consider the associated parameter that yields the critical phenomenon of the system as a random variable , and assume that we will try to find a critical parameter value with the given numerical setting . for the klein - gordon equation , the associated parameter is the strength or amplitude of the singular potential term ; for the sine - gordon equation , the associated parameter is the initial velocity of the soliton solution in the presence of the singular potential term that has a fixed amplitude . the original equations are then parameterized by the random variable and the solution of the parameterized equations is obtained by the gpc method . that is , the solution is expanded by the orthogonal polynomials in the random space . the associated orthogonal polynomials are determined by the probabilistic nature of the random variable , which is the critical parameter in our case . for example , if the random variable has the uniform probability density function ( pdf ) , the associated polynomial is the legendre polynomial ; if the pdf is normal distribution , the associated polynomial is the hermite polynomial @xcite . we first consider the klein - gordon equation with the singular potential and find some conditions for the critical phenomenon . the klein - gordon equation with the singular source term has been considered in various studies . for example , in @xcite , the singular source term is considered in adopting the model , the meson theory , that the particle mass varies with space and time . in @xcite , two @xmath0-functions are used as the singular potentials for schrdinger equation and the corresponding eigenfunctions are derived using the weyl - titchmarsh - kodaira ( wtk ) spectral theorem . in @xcite , the klein - gordon equation with the singular defect was considered and the chaotic behavior of the solution is studied . however , not every singular potential term yields the critical phenomenon . we consider the self - interacting singular potential term and show the existence of the critical phenomenon with a certain type of boundary condition . that is , the energy norm increases , vanishes , or decreases if the parameter value exceeds , is equal to , or is less than the critical value . the critical value of the parameter is obtained analytically in this case , which makes it possible to do an error analysis . for the critical value , we assume the uniform pdf and use the legendre polynomial for the expansion . the resulting deterministic equations are solved using the chebyshev collocation method with the consistent formulation for the dirac @xmath0-function @xcite . once the expansion coefficients are obtained , the solution is reconstructed for various values of the parameter . using the fact that the energy norm remains constant for the critical value when the time is large , the invariant point is found as the critical value . the mean solution is also found by the first gpc mode . we compare the gpc approach and the mc approach . then we apply the same method for the sine - gordon equation . the sine - gordon equation with singular potential terms is also found in much of the existing literature . the singularly perturbed solution of the sine - gordon equation shows many interesting mathematical structures , such as the fractality , self - similarity , etc . . moreover , such an equation yields the critical phenomenon associated with the critical parameter , as explained above briefly . in our work , we use the initial velocity of the soliton solution as the random variable , and try to find the critical initial velocity around which the particle - pass , particle - capture , and particle - reflection occur . for the random variable , we consider both the hermite and legendre polynomials that is , the normal and the uniform distributions as the pdf . unlike the klein - gordon equation , the reconstruction of the solution using the gpc method is not obtained directly . this is because the receiving soliton signal has two extreme cases , i.e. _ transmitted _ or _ not - transmitted _ , and because of the nonlinearity . the extreme solution behavior renders the pdf as the two @xmath0-function - like pdfs . for the particle - capture , the amplitude of the tail at the domain exit is @xmath1 ; it vanishes eventually for the particle - pass . such strong discontinuity in the solution behavior causes the gibbs phenomenon in the reconstruction . these two extreme cases are not separable in the reconstruction . in the paper , we show such non - separable reconstruction for both the hermite and legendre cases . thus instead of using the reconstruction by the gpc method , we use the gpc mean for the determination of the critical value . in this case , the legendre gpc method yields much faster convergence than the hermite gpc method . numerical results show that the legendre gpc method determines the critical value very efficiently and accurately . the determined critical value agrees well with the value found by the many direct mc simulations . to the best of our knowledge , gpc analysis for the singularly perturbed pdes by the dirac @xmath0-functions has not been thoroughly studied in existing literature . in particular , for the nonlinear optics equations with a singular defect in the disordered media , the uncertainty quantification is necessary . this is because the defect is defined in the highly localized regime and thus contains a high degree of intrinsic uncertainty . in our previous work @xcite , some linear wave equations with the singular source term were considered with the legendre pc method . our current work takes the first step to investigate the uncertainty quantification of nonlinear optics equations , with the source term confined in the highly localized regime . the paper is composed of the following sections . in section 2 , the klein - gordon equation with the singular source term is considered . in this section , we show that there exists a critical phenomenon and find the exact value of the critical parameter value . in section 3 , we briefly explain the gpc method and derive the gpc expansion of the klein - gordon equation using the legendre polynomials . in section 4 , numerical methods used for the numerical approximation of the deterministic equations derived in section 3 are explained including the consistent method for the approximation of the dirac @xmath0-function . in section 5 , numerical results of the gpc method for the klein - gordon equation are given . comparison between the mc method and the gpc method is given . in section 6 , the critical phenomenon for the sine - gordon equation is explained . the method of finding the critical initial velocity is explained using the pdf of the receiving signals and the gpc mean value . the hermite gpc method is presented . in section 7 , the legendre gpc is used for the sine - gordon equation and the deterministic equations are derived . numerical results and remark on the error are presented . in section 8 , we briefly outlined the gpc method for the case that the randomess is in the amplitue of the singular potential function and explain that the same gpc method can be applied to this case . in section 9 , concluding remark and future works are presented . we first consider the one - dimensional , inhomogeneous , klein - gordon type equation in the general form given by @xmath2 where @xmath3 , @xmath4 . the initial conditions are @xmath5 where @xmath6 @xmath7 the solution of eq . ( [ inhom_kg ] ) is given by @xmath8\right ] d\xi,\end{aligned}\ ] ] where @xmath9 is the bessel function of the first kind . the above solution can be derived using the green s function method @xcite . now suppose that the source term involves the singular function such as @xmath10 where @xmath11 is a constant and @xmath12 is the dirac @xmath0-function . then the solution becomes @xmath13\right ] d\tau\right]h(c(t-\tau)-x ) , \end{aligned}\ ] ] where @xmath14 is the heaviside function and @xmath15 . as shown in the solution form , the parameter @xmath16 only scales the solution but does not yield the critical behavior of the solution . in order to obtain the critical solution , we consider the following several different cases . we first consider the case of a possible steady - state solution with a singular source term . consider the case that @xmath18 for the bounded domain @xmath19 $ ] with the boundary conditions @xmath20 then there exists a steady - state solution that satisfies @xmath21 by using these boundary conditions , the general solution can be easily derived as @xmath22.\ ] ] if different boundary conditions are chosen such as @xmath23 then the solution becomes @xmath24e^{x } + \frac{1}{e^2}\left [ \frac{e^5}{e^4 + 1}-\frac{\eta}{2}\frac{e^2(e^2 + 1)}{e^4 + 1}\right]e^{-x } - \eta\sinh(x)h(x).\ ] ] these examples show that there exists the steady - state solution for any value of @xmath16 but we do not observe the critical phenomenon in the solution induced by the singular potential term @xmath25 . for another possible steady - state solution , we consider the self - interacting potential term given by @xmath27 where @xmath28 is the potential function . we show that some @xmath29 yield a critical phenomenon . similar equation has been used in quantum physics . when @xmath29 is given by the singular source term , the above equation represents some physical model such as one - dimensional diatomic molecule model with the singular potential well in the presence of static electric field @xcite . the boundary conditions we choose are @xmath30 with some non - vanishing initial condition @xmath31 the motivation of the given boundary condition of @xmath32 at @xmath33 is that if the potential term @xmath29 is localized inside @xmath34 and vanishes at the boundary , then the given boundary condition yields the outflow boundary condition . the potential term @xmath29 will contain the singular term such that the potential effect vanishes at the domain boundary @xmath35 . to show the criticality of the solution , we define the energy @xmath36 of the system given by @xmath37 the rate of change of energy @xmath38 with respect to time @xmath39 , @xmath40 is then simply computed by @xmath41_{-1}^{1}-\int_{-1}^{1}u_{t}u_{xx}dx . \nonumber \end{aligned}\ ] ] using the boundary conditions we have @xmath42 suppose that the potential function @xmath29 is given by @xmath43 then @xmath44 without loss of generality , we choose @xmath45 . using the continuity of @xmath46 at @xmath47 , the jump condition is given by @xmath48= \eta u(0,t ) , \quad \forall t \in \mathbb{r}^+ . \label{jump}\end{aligned}\ ] ] the steady - state solution is determined by a unique @xmath49 and @xmath50 , @xmath51 . _ proof _ : @xmath52 using the boundary conditions , we easily show that the steady - state solution @xmath53 is given by @xmath54 \\ c(1 + \alpha ) & x \in [ \alpha , 1 ] \end{array } \right . , \end{aligned}\ ] ] where the constant @xmath55 depends on the initial condition . notice that the parameter @xmath16 does not appear in the steady - state solution . using the jump condition , we obtain @xmath56= \eta u^s(\alpha).\ ] ] using eq . ( [ ss - solution ] ) , the critical value of @xmath16 , @xmath57 yielding the steady - state solution is obtained uniquely @xmath58/ u^{s}(\alpha)= { 1\over { 1 + \alpha}}.\quad \box\ ] ] there exists a critical behavior around @xmath49 obtained in lemma 1 , that is , @xmath59 for @xmath60 where @xmath61 for an arbitrary small value of @xmath62 . + _ proof _ : without loss of generality we consider @xmath45 , i.e. , the position of @xmath0-function is at the origin and suppose that we choose @xmath63 where @xmath64 ( from lemma 1 ) and @xmath65 . based on these assumptions , we look for the solution in terms of the power series in @xmath66 that is , @xmath67 where @xmath68 is the steady - state solution . plug @xmath46 into the pde given by eq . [ self_potential ] with [ delta_potential ] and then each of @xmath69 terms yields @xmath70 we choose the initial condition such that the steady - state solution becomes @xmath71 \\ 1 , & \mbox { for } x \in [ 0 , 1 ] \end{array } \right . . \nonumber\end{aligned}\ ] ] for the pde of the first order term above , the boundary conditions and the continuity condition yields @xmath72 the jump condition , @xmath73 yields @xmath74 . we choose the initial condition such that @xmath75 . then by ignoring the higher - order terms @xmath76 , @xmath77 , we obtain @xmath78 if @xmath79 the solution increases unboundedly and the energy increases with time . if @xmath80 , the solution becomes the steady - state solution and the energy is conserved . if @xmath81 , the solution decays until @xmath82 . for @xmath83 , the higher - order term becomes significant and they should be considered so that the energy further decays . @xmath84 to get the complete solution , one needs to solve the higher - order terms . it may become difficult to solve those pdes exactly but the proof is enough to show that there exists a critical behavior of the solution around @xmath49 when @xmath85 . for example , if @xmath86 , the first - order term is dominant up to @xmath87 , which is enough for the long time behavior of the solution for our purpose . the energy of the system up to the first order term is given by @xmath88 ^ 2 + \frac{7}{6}\left(\eta -1\right)^{2 } + ( 1+t)(1-\eta)^{2}\left[\eta + t(\eta -1)\right ] . \nonumber \ ] ] and the rate of change of @xmath38 with time is then given by @xmath89 + \left(\eta-1\right)^{3}\left(1+t\right ) .\ ] ] figure [ exact_solution ] shows the solutions at different times @xmath90 for different values of @xmath16 , @xmath91 around @xmath92 ( left ) and the contour of @xmath93 in @xmath16 and @xmath39 plane ( right ) . at different final times @xmath90 . right : contour plot of @xmath38 for @xmath94$],title="fig:",scaledwidth=45.0% ] at different final times @xmath90 . right : contour plot of @xmath38 for @xmath94$],title="fig:",scaledwidth=43.0% ] the same phenomenon can be obtained for the linearized sine - gordon equation with the point - like source term where @xmath95 such that @xmath96 with the same boundary conditions @xmath97 as in above , the energy , @xmath98 has the rate of change below @xmath99 we have a similar phenomenon @xmath100 that occurs around the following steady - state solution @xmath101 where @xmath55 is a constant that depends on the initial condition . the critical value of @xmath16 is @xmath102 . we use the gpc method for the solution of the klein - gordon and sine - gordon equations using the wiener - askey scheme @xcite , in which hermite , legendre , laguerre , jacobi , generalized laguerre orthogonal polynomials are used for modeling the effect of continuous random variables described by normal , uniform , exponential , beta and gamma probability distributions , respectively @xcite . these orthogonal polynomials are optimal for those pdfs since the weight function in the inner product and its support range correspond to the probability density functions for those continuous distributions . table ( 1 ) shows the set of polynomials which provide an optimal basis for different continuous pdfs @xcite . it is derived from the family of hypergeometric orthogonal polynomials known as the askey scheme , for which the hermite polynomials originally employed by wiener @xcite are a subset . [ table1 ] ccccc__distribution & _ density function & _ polynomial & _ weight function & _ support range + normal & @xmath103 & hermite @xmath104 & @xmath105 & @xmath106 + uniform & @xmath107 & legendre @xmath108 & @xmath109 & @xmath110 $ ] + beta & @xmath111 & jacobi @xmath112 & @xmath113 & @xmath110 $ ] + exponential & @xmath114 & laguerre @xmath115 & @xmath114 & @xmath116 + gamma & @xmath117 & generalized laguerre @xmath118 & @xmath119 & @xmath116 + _ _ _ _ _ _ following the standard gpc expansion , we assume that @xmath120 is sufficiently smooth in @xmath121 and has a converging expansion of the form @xmath122 where the orthonormal polynomials @xmath123 correspond to the pdf of the random variable @xmath121 and satisfy the following orthogonality relation : @xmath124 : = \int p_{k}(\xi)p_{l}(\xi)\rho(\xi)d\xi = \delta_{kl},\ ] ] where @xmath125 is the kronecker delta and @xmath126 is the weight function . note that the polynomials are normalized . we use the gpc method to find the critical value of @xmath16 numerically . assume that @xmath127 is a random variable and @xmath128 $ ] has the uniform distribution . let @xmath121 be a random variable with the uniform pdf over @xmath129 $ ] . let @xmath16 be a linear map , @xmath130 $ ] , @xmath131 . the solution @xmath132 will be expanded using the legendre polynomial in @xmath121 with time @xmath39 and the value of @xmath121 will be sought that yields the invariant solution with time when @xmath133 . such @xmath121 or @xmath16 is then determined as the critical value for the klein - gordon equation . for this , we first expand the solution @xmath46 in the legendre polynomial in the galerkin sense such as in @xcite : @xmath134 where @xmath135 are the expansion coefficients and @xmath136 the legendre polynomials @xcite . we consider the truncated version of eq . ( [ eq33 ] ) by considering the first @xmath137 terms . plugging eq . ( [ eq33 ] ) into eq . ( [ self_potential ] ) and using @xmath29 from eq . ( [ delta_potential ] ) with @xmath45 , we get @xmath138l_{l}(\xi ) = \left(\frac{b - a}{2}\xi + \frac{a+b}{2}\right)\delta(x)\sum_{l=0}^{n}\hat{u}_{l}(x , t)l_{l}(\xi).\ ] ] multiplying both sides by @xmath139 yields @xmath138l_{l}(\xi)l_{l^{'}}(\xi ) = \left(\frac{b - a}{2}\xi + \frac{a+b}{2}\right)\delta(x)\sum_{l=0}^{n}\hat{u}_{l}(x , t)l_{l}(\xi)l_{l^{'}}(\xi).\ ] ] integrating both sides over @xmath140 $ ] and using the orthogonal property of the legendre polynomials , we get , @xmath141\frac{2}{2l+1 } = \frac{2}{2l+1}\left(\frac{a+b}{2}\right)\delta(x)\hat{u}_{l}(x , t ) \nonumber \\ & + & \left(\frac{b - a}{2}\right)\delta(x)\sum_{l=0}^{n}\hat{u_{l}}(x , t)\int_{-1}^{1}\xi l_{l}(\xi)l_{l^{'}}(\xi)d\xi.\end{aligned}\ ] ] for the second term in the right hand side of the above equation , one can show the following relation @xmath142 ( for the derivation of the relation , see appendix 2 ) . then , eq . ( [ gpc_legendre1 ] ) can be rewritten as @xmath143 . \label{gpc_legendre2}\end{aligned}\ ] ] substitute eq . ( [ eq33 ] ) into the left boundary condition of eq . ( [ self_potential_bc ] ) to get , @xmath144 by using the orthogonal property of the legendre polynomials , we have @xmath145 similarly the right boundary condition of eq . ( [ self_potential_bc ] ) yields @xmath146l_{l}(\xi)=0.\ ] ] by using orthogonality we have @xmath147 we choose the initial condition as @xmath148 again using the orthogonal property , we get @xmath149 this can be written as @xmath150 so from eqs . ( [ gpc_legendre2 ] ) , ( [ gpc_legendre_bcl ] ) , ( [ gpc_legendre_bc2 ] ) , and ( [ gpc_legendre_initial ] ) we have @xmath151 equations for the gpc solution of the klein - gordon equation . the derived gpc equations for @xmath152 in section 3 will be solved numerically . for the space derivative , the chebyshev spectral collocation method is used based on the gauss - lobatto quadrature points @xmath153 where @xmath154 is the total number of the quadrature points for @xmath155 $ ] . as shown in section 3 the derived gpc equations for @xmath152 still contains the @xmath0-function . solving the pde containing the singular source term properly is the crucial part for obtaining the accurate and stable numerical solutions for @xmath152 . to deal with this problem , the @xmath0-function is approximated by the consistent formulation proposed in @xcite where the consistent formulation was proposed for the one - dimensional problems and it was shown that the method yields highly accurate results at the collocation points . it is not straightforward to extend such a formulation to higher dimensional problem , but as shown in the following sections and also shown in @xcite , this formulation yields an accurate result for some pdes and can be expanded to higher dimension by using the dimension - by - dimension approach . moreover , the method is simple and easy to implement . the underline motivation is that the approximation of the @xmath0-function is sought such that the steady - state solution can be obtained . this method is also referred as the well - balanced equation in the literature . the steady - state solution of the equation , @xmath53 in the previous section satisfies @xmath156 let @xmath157 be the approximation of @xmath0 on the @xmath154 collocation points . then the consistent formulation approximates @xmath157 as the derivative of the heaviside function @xmath158 , using the same spectral derivative operator . if @xmath159 denotes the heaviside function defined at the collocation points @xmath160 , then @xmath157 is given by the following : @xmath161 where @xmath162 is the chebyshev derivative matrix and @xmath163 given by @xmath164 . to avoid any ambiguity , we choose the odd value of @xmath165 such that for @xmath166 , @xmath167 that is , the location of the @xmath0-function is off the collocation point . then the consistent formulation of eq . ( [ steady ] ) becomes @xmath168 where @xmath169 is the @xmath170 matrix whose diagonal elements are @xmath157 and @xmath171 . as shown in section 2 , the critical value of @xmath49 is determined by the following @xmath58/ u^{s}(\alpha).\ ] ] that is , the critical value is determined by the local values of @xmath46 and @xmath172 near @xmath173 . thus the convergence of the numerically computed value of @xmath49 depends on the local convergence of @xmath46 and @xmath174 near @xmath173 . figure [ fig : steady_state ] shows how the local convergence is obtained by our numerical approximation using the consistent method . in the figure , the convergence of the solution is obtained by solving eq . ( [ linear ] ) iteratively . the left figure shows the pointwise errors with different values of @xmath165 and the right figure the pointwise errors after the spectral filter is applied . as shown in these figures , the convergence can be enhanced away from the discontinuity when the filter operation is applied . but the overall convergence is slow due to the singularity at @xmath173 , which is the gibbs phenomenon for the spectral method @xcite . this implies the slow convergence of the numerically obtained @xmath49 . . left : without the spectral filter . right : with the spectral filter . @xmath175 from top to bottom.,title="fig:",scaledwidth=47.0% ] . left : without the spectral filter . right : with the spectral filter . @xmath175 from top to bottom.,title="fig:",scaledwidth=47.0% ] to show such gibbs phenomenon we consider the exact solution of eq . ( [ steady ] ) given by @xmath176 the solution is continuous in @xmath177 but at @xmath178 it is not differentiable , @xmath179 . let the exact solution @xmath180 be expanded as the gegenbauer polynomial of degree @xmath165 : @xmath181 where @xmath182 we use the particular case of the gegenbauer polynomials for @xmath183 , which corresponds to the chebyshev polynomials.@xmath52 figure [ fig : cheb_leg ] shows the chebyshev galerkin solution ( left ) and the chebyshev collocation solution ( right ) with @xmath184 . as shown in these figures , there exist the gibbs oscillations near @xmath173 . the gibbs oscillation is seen mainly at the collocation point @xmath185 for the collocation solution . these oscillations on the collocation points yield the error in determining the critical value @xmath49 . table [ table2 ] shows the convergence of @xmath186 with @xmath165 , where @xmath186 is the numerically determined critical value of @xmath49 . the critical value we will find from the following section using the given value of @xmath165 will be those values in the table . for example , if we choose @xmath187 , the numerically computed critical value , @xmath186 is @xmath188 and our objective is to find that value as our critical value . that is , given a finite value of @xmath165 , there exists the corresponding @xmath189 uniquely and when we find @xmath190 , we mean it by finding @xmath189 instead of @xmath49 . cc__@xmath191 & _ @xmath186 + @xmath192 & @xmath193 + @xmath194 & @xmath195 + @xmath196 & @xmath197 + @xmath198 & @xmath199 + _ _ _ we use the second order centered difference in time for the time integration . for example let @xmath200 be the solution vector defined at @xmath201 with the time - step @xmath202 such as @xmath203 for @xmath204 . then the scheme is given by @xmath205 where the element of the column vector @xmath206 is @xmath207 and we apply @xmath208 twice to @xmath209 for its second derivative in @xmath210 . this is a two - step method in time . @xmath211 is given by the initial condition and we use the first - order approximation for @xmath212 . then the numerical solution of @xmath213 for each @xmath214 is obtained , by @xmath215 - \hat{u}^{n-1}_{l } , \nonumber \\ \end{aligned}\ ] ] where @xmath216 , @xmath217 , @xmath218 , and @xmath219 . the boundary conditions are @xmath220 where @xmath221 and @xmath222 . for the boundary condition at @xmath33 , we use the first order finite difference both in time and space . for the numerical example , we consider @xmath224 and the corresponding @xmath225 from table 2 . in figure [ fig : eta ] , we show @xmath40 versus time with @xmath226 . we find for @xmath227 ( blue ) , @xmath228 when @xmath229 . this shows @xmath36 is almost constant with time . another view of the steady - state solution is given in the right figure of figure [ fig : eta ] , where some wiggles are shown due to the inconsistency of the initial condition in the beginning but later the steady - state solution is obtained . note that we use the initial condition @xmath230 . to verify the claim , we also calculate @xmath231 with @xmath232 with different final time @xmath233 in figure [ fig : steady_states ] . for @xmath234 for @xmath235 the figure clearly shows that the solution remains almost constant for @xmath227 . but for @xmath236 the solution decreases monotonically and for @xmath237 the solution increases monotonically . with different values of parameter @xmath16 . right : the steady - state solution for @xmath227 , title="fig:",scaledwidth=45.0% ] with different values of parameter @xmath16 . right : the steady - state solution for @xmath227 , title="fig:",scaledwidth=50.0% ] first we compute the mean value of the solutions with time . let @xmath238 be the mean over @xmath121 . then @xmath239 we consider three mean values obtained by the following three methods . for @xmath227 ( left ) , @xmath237 ( middle ) , @xmath240 ( right ) . , title="fig : " ] for @xmath227 ( left ) , @xmath237 ( middle ) , @xmath240 ( right ) . , title="fig : " ] for @xmath227 ( left ) , @xmath237 ( middle ) , @xmath240 ( right ) . , title="fig : " ] * _ mc mean : _ * to compute the above integral we use the uniform sampling of @xmath16 such that @xmath241 and use the transformation @xmath242 then eq.([mean_int ] ) takes the form @xmath243 we use the trapezoidal rule to calculate the integration . + * _ gpc mean : _ * from eq.([mean_int ] ) we have @xmath244 which is the first pc mode . + * _ quadrature mean : _ * we consider @xmath151 gauss - legendre ( gl ) quadrature points @xmath245 between @xmath246 $ ] and weights @xmath247 . to get mean value we have @xmath248 we calculate the mean value of the solution @xmath249 by the mc method at the final time @xmath250 with different uniform sample points @xmath251 , @xmath252 , @xmath253 , @xmath254 , @xmath255 , @xmath256 , @xmath257 , @xmath258 for @xmath259 $ ] . also we calculate the gpc mean at @xmath250 . the results are shown in figure [ fig : conv_mc ] . the mc mean solution converges to the gpc mean solution but the convergence is slow . as shown in the figure , more than @xmath260 mc samples are needed to match the gpc mean . this is because the energy growth / decay rate around the critical value of @xmath49 is extremely asymmetric and the uniform sampling is obviously inefficient . the left figure of figure [ fig : mcerror ] shows the convergence of the mc simulations . we calculated the mean solutions at @xmath261 with the sample points @xmath262 where @xmath263 . we plot @xmath264 versus @xmath265 and @xmath266 . here @xmath267 is the mean solution of uniformly distributed @xmath265 samples . the result is plotted in logarithmic scale and the decay is linear . the right figure of figure [ fig : mcerror ] shows the mean solutions with the legendre gpc , mc simulation with @xmath260 equally distributed sample points and the quadrature mean with @xmath268 gauss - legendre quadrature points in @xmath269 $ ] . the solid red line represents the result by the mc simulation which deviates from the gpc mean or gl quadrature mean . the gl quadrature mean with the small number of points coincides with pc mean . using the quadrature sampling for the mean solution was similarly used in @xcite . by the mc simulations to gpc mean with different sample points . the top curve ( red dotted line ) is the gpc mean . for mc simulations the number of sample points of @xmath16 are @xmath270 .,scaledwidth=80.0% ] , where @xmath271 for @xmath272 . right : comparison of @xmath273 derived from the mc simulation with @xmath260 sample points , the gauss - legendre quadrature with @xmath268 quadrature points ( red circles ) and the legendre chaos method . it is observed that the quadrature mean and the pc mean are almost coincident . , title="fig:",scaledwidth=49.0% ] , where @xmath271 for @xmath272 . right : comparison of @xmath273 derived from the mc simulation with @xmath260 sample points , the gauss - legendre quadrature with @xmath268 quadrature points ( red circles ) and the legendre chaos method . it is observed that the quadrature mean and the pc mean are almost coincident . , title="fig:",scaledwidth=45.0% ] we first define a solution sequence @xmath274 at @xmath33 . the index @xmath275 denotes the sequence corresponding to time @xmath276 such as @xmath277 define @xmath278 , @xmath279 , @xmath280 for large @xmath39 , @xmath274 is a monotone function with respect to @xmath16 around the critical value of @xmath16 . as there exists the critical value of @xmath16 , @xmath57 , there exists @xmath16 such that @xmath281 . now we define another sequence @xmath282 such as @xmath283 as @xmath284 is monotone with respect to @xmath16 for large @xmath39 , we have @xmath285 and @xmath286 figure [ fig : crit_eta_mc ] shows the sequence of @xmath287 and @xmath288 by the mc simulations with @xmath289 sample points at different time intervals . the final time varies from @xmath268 to @xmath290 and we calculate the solution at @xmath291 , @xmath292 , @xmath293 , @xmath294 , @xmath290 . for any fixed @xmath233 , we plot the solution for different @xmath16 . if @xmath295 then the graph @xmath296 converges to zero as @xmath60 when @xmath297 and similarly @xmath296 diverges to @xmath298 as @xmath60 when @xmath299 . from figure [ fig : crit_eta_mc ] , we find that the solutions intersect at the common point @xmath300 and we claim that such @xmath301 is the critical value of @xmath16 and @xmath302 is the steady - state solution . the bottom figure shows the magnified region near @xmath301 . if we see the figure carefully , it is then observed that the solutions for @xmath303 , @xmath304 and @xmath305 intersect almost at the same point , whose coordinate is approximately equal to @xmath306 . then our estimates are @xmath307 and @xmath308 . in section 4 we found for @xmath309 , the critical value of @xmath16 is about @xmath195 , which is close to the value of @xmath310 . here note that the value of @xmath49 in table 2 was obtained by many of the individual simulations of the equation with different value of @xmath16 , which is highly expensive and @xmath49 was obtained with only @xmath311 accuracy . the error for the steady - state solution at @xmath33 is @xmath312 , which agrees our error estimation in section 5.1 and figure [ fig : eta ] . the error of @xmath313 comes from the gibbs oscillations at @xmath33 as explained . figures [ fig : crit_eta_pc ] shows how to find the critical value @xmath223 by the gpc method . in the gpc method we reconstruct the solutions by considering the first @xmath314 modes , at different time @xmath233 , where @xmath233 runs from @xmath268 to @xmath315 with the interval length @xmath268 . at every stage we plot @xmath316 , where @xmath317 $ ] and consider @xmath318 equally distributed sample points generated by the @xmath314 gpc modes . we found that the solutions intersect at the common point @xmath319 and we estimate that it is the critical value of @xmath121 for @xmath320 $ ] . we find the approximate value of @xmath321 is @xmath322 and @xmath323 . for this value of @xmath321 the corresponding @xmath223 is @xmath324 . the value of @xmath223 divides the interval @xmath325 $ ] into two subintervals @xmath326 $ ] and @xmath327 $ ] . when @xmath328 , the graph converges to zero as @xmath329 and similarly the graph diverges to @xmath298 if @xmath299 as @xmath329 . in figure [ fig : crit_eta_pc ] , the right figure is the magnified version of the left figure near @xmath321 . the graphs intersect within a very small region between @xmath330 and @xmath331 in the vertical axis and between @xmath332 and @xmath333 in the horizontal axis . the last two lines at @xmath334 and @xmath335 ( blue and black ) intersect at @xmath336 and the error in this case @xmath337 . the error in calculating the critical value of @xmath16 must be less than the difference of @xmath338 and @xmath339 which is almost equal to @xmath340 . thus we get the result with in the accuracy of @xmath341 . the difference between @xmath49 computed by the mc method and the gpc method is @xmath342 here note that for the gpc method we only use @xmath343 while for the mc method we use @xmath344 , which shows that the gpc method can determine @xmath49 efficiently . consider the sine - gordon equation @xmath345 where @xmath346 . the _ soliton ( or kink ) _ solution of the sine - gordon equation is given by @xmath347 this represents the continuous profile with @xmath348 as @xmath349 , and @xmath350 as @xmath351 . this solution propagates with the constant velocity @xmath29 . another solution can be obtained in the form @xmath352 which is known as the _ antisoliton(or antikink)_. we consider the sine - gordon equation with an impurity @xmath353 where @xmath354 is the amplitude or strength of the singular potential function . the full sine - gordon equation with an impurity eq . ( [ sg1 ] ) is , however , non - integrable and no exact solution is known . the problem we solve numerically in this work is defined by eq . ( [ sg1 ] ) in @xmath355 , where @xmath356 $ ] , @xmath357 with the initial kink velocity @xmath29 . for simplicity we assume that the singular potential exists at @xmath173 . for the numerical computation , we set @xmath358 , @xmath359 , and the initial kink front @xmath360 . the initial and boundary conditions are given below . initial conditions ( kink soliton ) @xmath361 , \nonumber \\ u_{t}(x,0 ) & = & -\frac{4v}{\sqrt{1-v^2}}\left[\frac{\exp\left(\frac{x - x_{0}}{\sqrt{1-v^2 } } \right ) } { 1+\exp\left(\frac{2(x - x_{0}}{\sqrt{1-v^2 } } \right ) } \right ] . \nonumber\end{aligned}\ ] ] boundary conditions : @xmath362 \right\rbrace . \nonumber\end{aligned}\ ] ] the above boundary condition is used for the outflow boundary condition . note that this boundary condition is not an exact outflow boundary condition because of the nonlinear potential term , @xmath363 . better boundary conditions and exact outflow boundary conditions can be found in @xcite . since the main goal of our study is to find the critical velocity , the above boundary condition does not affect the results . figure [ critical ] shows the solution behaviors around the critical value of @xmath364 , @xmath365 . the left figure shows the sub - critical solution behavior when @xmath366 and the right figure when @xmath367 . the horizontal axis denotes the time and the vertical axis the location of the kink front ( the red color shows the kink front and the @xmath1 plateau ) . these solutions are obtained by the direct numerical approximation using the chebyshev collocation method with the consistent approximation of the @xmath0-function explained in the previous section with @xmath184 . as shown in the figure , the soliton solution is captured in the potential well if @xmath368 while it eventually passes through the potential well . that is , at @xmath369 , @xmath231 becomes either @xmath1 ( particle - capture or particle - reflection ) or @xmath370 ( particle - pass ) as @xmath60 . we use this solution behavior to find the critical velocity @xmath364 . here note that the early numerical study of the discretized sine - gordon equation is found in the literature @xcite . . left : sub - critical behavior with @xmath371 . right : super - critical behavior @xmath372 @xcite.,title="fig:",scaledwidth=40.0% ] . left : sub - critical behavior with @xmath371 . right : super - critical behavior @xmath372 @xcite.,title="fig:",scaledwidth=40.0% ] in the previous section , we considered the klein - gordon type linear equation for which we use the mc sampling using the gpc expansion for constructing the solutions . however , for the nonlinear sine - gordon equation , such a mc sampling is not usable because of the two extreme pdf of the solution for large @xmath39 . the construction of the gpc for large @xmath39 yields either _ trapped _ or _ transmitted _ solution . ( 1,1 ) ( 0.5,0)(1,0)1.3 ( 0.3,0.3)@xmath373 ( 0.5,0)(0,1)0.7 ( 0.7,0)(0,1)0.6 ( 1.4,0)(0,1)0.4 ( 0.7,-0.1)@xmath370 ( 1.4,-0.1)@xmath1 ( 1.0,-0.2)@xmath374 ( 1.0,0.53)(-1,0)0.25 ( 1.05,0.5)@xmath375 ( 1.7,0.23)(-1,0)0.25 ( 1.75,0.2)@xmath376 ( 0.05,-0.3)diagram 1 : the pdf of the solution for @xmath377 at @xmath369 for uniform @xmath29 . .8 in the critical value can be computed by the weighted ratio of the two extreme pdf as shown in diagram 1 if we know the pdf of @xmath29 . the ratio is indeed the amount of how much the mean value deviates from the value of @xmath1 for the _ trapped _ solution . thus the critical value of @xmath29 can be still found from the gpc expansion using the mean mode . if @xmath378 $ ] and if @xmath379 is the critical velocity , then for @xmath380 $ ] , the solution is @xmath1 ( because the solution is _ trapped _ ) , and for @xmath381 $ ] , the solution vanishes ( because the solution is _ passed _ ) . if @xmath382 is the pdf for @xmath29 , then @xmath383 for example , if we use the legendre chaos , the distribution function is uniform and @xmath384 . for this by using eq . ( [ critical_velo_pdf ] ) , the legendre mean @xmath385 is given by @xmath386 and we have @xmath387 suppose that the input velocity , @xmath29 , is the random variable which is normally distributed with mean @xmath388 and standard deviation @xmath389 . here @xmath29 , conditional on @xmath390 , has a truncated normal distribution . we consider @xmath391 and @xmath392 where @xmath393 . the reason for choosing small @xmath389 is to keep the error @xmath341 . the pdf of @xmath29 , @xmath394 , is given by @xmath395 the density function involves @xmath396 which is the pdf of the standard normal distribution and @xmath397 , its cumulative distribution function @xcite . due to the small value of @xmath389 , @xmath398 for which see appendix 1 . we expand @xmath399 in terms of the hermite polynomials as follows @xmath400 for the numerical purpose , we consider the first @xmath137 modes . then the sine - gordon equation ( [ sg1 ] ) can be written as @xmath401_{tt } & = & [ \hat{u}_{l}(x , t)]_{xx } + \frac{1}{l!\sqrt{2\pi}}\left\lbrace \epsilon \delta(x ) -1 \right\rbrace \phi_{l}(x , t ) , \nonumber \\ \mathrm{where}\qquad \phi_{l}(x , t ) & = & \int_{-\infty}^{\infty}\sin\left(\sum_{l=0}^{n}\hat{u}_{l}(x , t)h_{l}(\gamma ) \right)h_{l}(\gamma)\omega(\gamma)d\gamma , \ ; \nonumber\end{aligned}\ ] ] and @xmath402 for the initial condition , we have @xmath403 = \sum_{l=0}^{\infty } \hat{u}_{l}(x,0)h_{l}\left(\frac{v-\mu}{\sigma}\right).\ ] ] by using the orthogonal property of the hermite polynomials we get @xmath404h_{l}(\gamma)\omega(\gamma)d\gamma.\ ] ] similarly @xmath405 . right : the first few pc modes for the hermite chaos at @xmath406 and @xmath407.,title="fig:",scaledwidth=43.0% ] . right : the first few pc modes for the hermite chaos at @xmath406 and @xmath407.,title="fig:",scaledwidth=43.0% ] the boundary conditions are given by @xmath408h_{l}(\gamma)\omega(\gamma)d\gamma , \nonumber\end{aligned}\ ] ] for @xmath409 by definition , the mean solution is given by @xmath410 ) & = & \int_{\alpha}^{\beta}f(v;\alpha,\beta,\mu,\sigma)u(x , t , v)dv \nonumber \\ & = & \frac{1}{\sqrt{2\pi}\sigma\left[\phi\left(\frac{\beta - \mu}{\sigma}\right)-\phi\left(\frac{\alpha - \mu}{\sigma}\right)\right]}\int_{\alpha}^{\beta}e^{-\frac{\left(v-\mu\right)^2}{2\sigma^{2}}}\sum_{l=0}^{\infty}\hat{u}_{l}(x , t)h_{l}\left(\frac{v-\mu}{\sigma}\right)dv \nonumber \\ & \approx & \hat{u}_{0}(x , t ) + \frac{1}{\sqrt{2\pi}\left[\phi\left(\frac{\beta - \mu}{\sigma}\right)-\phi\left(\frac{\alpha - \mu}{\sigma}\right)\right]}\sum_{l=1}^{n}\hat{u}_{l}(x , t)\int_{\frac{\alpha-\mu}{\sigma}}^{\frac{\beta-\mu}{\sigma}}e^{-\frac{\gamma^2}{2}}h_{l}(\gamma)d\gamma \nonumber\\ & \approx & \hat{u}_{0}(x , t ) + \frac{1}{\sqrt{2\pi}\left[\phi\left(\frac{\beta - \mu}{\sigma}\right)-\phi\left(\frac{\alpha - \mu}{\sigma}\right)\right]}\sum_{l=1}^{n}\hat{u}_{l}(x , t)\int_{-\infty}^{\infty}e^{-\frac{\gamma^2}{2}}h_{l}(\gamma)d\gamma \nonumber \end{aligned}\ ] ] by using the orthogonality of @xmath411 @xmath412 to calculate standard deviation we first calculate @xmath413 . @xmath414}\sum_{m=1}^{n}m!\hat{u}_{m}^{2}(x , t),\nonumber \ ] ] and @xmath415}\left[1!\hat{u}_{1}^{2}(x , t ) + 2!\hat{u}_{2}^{2}(x , t ) + \cdots + n!\hat{u}_{n}^{2}(x , t)\right].\ ] ] the standard deviation is then given by @xmath416}\sum_{m = 1}^{n}m!\hat{u}_{m}^{2}(x , t)\right ) ^{\frac{1}{2}}.\ ] ] to estimate the critical velocity we use the eq . ( [ critical_velo_pdf ] ) . here the density function @xmath394 is given by eq . ( [ hermite - pdf ] ) . by using these two equations we get , @xmath417 if @xmath418 , @xmath419 which implies @xmath420 , all solutions are _ trapped _ and if @xmath421 , @xmath422 which implies @xmath423 , all solutions are _ passed_. for numerical simulations , we consider @xmath424 we obtain @xmath425 at the final time @xmath305 . by using eq . ( [ her_critical ] ) , we get , @xmath426 + \phi\left(\frac{0.11 - 0.12}{0.01}\right).\ ] ] based on these results and by using the statistical table @xcite we estimate @xmath427 . this estimate is close to what we get from the mc simulations @xcite and the legendre chaos in the next section . the obtained value is , however , not accurate ; it matches with the value obtained by many mc simulations within two digits only . figure [ fig : hermite_sol ] shows the initial condition , the mean solution , and the standard deviation in the left figure and the first 7 pc modes ( @xmath428 ) in the right figure . suppose that @xmath378=[a , b]$ ] has the uniform distribution and @xmath429.\ ] ] we consider @xmath430 . by using the orthogonal property of the legendre polynomials and the initial and boundary conditions we have [ cols= " < " , ] in the similar way as in the previous section we have for @xmath431 @xmath432l_{m}(\xi ) = \left\lbrace \left ( \frac{b - a}{2}\xi + \frac{a+b}{2 } \right)\delta ( x)-1 \right\rbrace \sin\left(\sum_{m=0}^{n}\hat{u}_{m}(x , t)l_{m}(\xi ) \right).\ ] ] using orthogonal property of the legendre polynomials again , we get @xmath433 for @xmath434 where @xmath435 the discretized equations are given by , l @xmath436 , $ ] + + @xmath437 \right\rbrace , $ ] + + @xmath438 + + @xmath439,$ ] + for @xmath434 to get the critical value of @xmath440 , @xmath441 , we follow the same procedure as in section 6.2 . since @xmath440 is uniformly distributed in @xmath442 $ ] , we use eq . ( [ critical_legendre ] ) , but in little different way . let @xmath443 and for @xmath444 $ ] , the solutions are _ passed _ and for @xmath445 $ ] , the solution get _ trapped_. so eq . ( [ critical_legendre ] ) takes the form @xmath446 \nonumber \\ \epsilon_{c } & = & \epsilon_{b } -\left(\epsilon_{b}-\epsilon_{a } \right)\frac{\bar{u}_{leg}}{2\pi}. \label{critical_eps_leg}\end{aligned}\ ] ] figure ( [ sg_epsilon_mc ] ) shows some solutions at @xmath173 with time for different values of @xmath440 , @xmath447 . for the captured solution , the solution at @xmath173 is highly oscillatory but for the transmitted one it is close to zero . from the figure we know that @xmath448 . suppose we know for @xmath449 solution passes through the barrier and for @xmath450 solution is _ trapped _ by the barrier . by the gpc simulations ( figure [ sg_epsilon_critical ] ) for @xmath305 we have @xmath451 . using these in eq . ( [ critical_eps_leg ] ) we get @xmath452 . to get the idea of the position of @xmath441.,scaledwidth=60.0% ] where @xmath440 is stochastic variable . , scaledwidth=60.0% ] in this paper , we proposed an efficient method using the gpc method to find the critical parameter values and the related statistical quantities for the klein - gordon and sine - gordon equations with a point - like singular potential function . by assuming the unknown critical parameter values as a random variable , and expanding the solution in the orthogonal polynomials associated with the random variable , we computed the critical parameter with much less computation than the mc approach . for the klein - gordon equation , the critical parameter is associated with the amplitude of the singular potential function . using the gpc reconstruction of the solution , the invariant point is found to be the critical parameter value . for the sine - gordon equation , the initial velocity of the soliton solution is used as the random variable . for this case , both the hermite and legendre gpc methods are used . due to the nonlinearity and the @xmath0-function like pdf of the receiving signals , the reconstruction could not be used ; the gpc mean was used instead . the legrendre gpc mean converges quickly , and it efficiently and accurately determines the critical parameter value . as our numerical results show , the gpc method offers several advantages in finding the critical value as compared to the mc method . first of all , one can avoid the large number of individual simulations for different parametric values . our proposed method suggests that , with a small number of mc simulations , it is possible to make the first guess about the critical value . once the first guess is made , one can apply the gpc method to narrow down or pinpoint the critical value accurately and efficiently . moreover , regarding the gpc method for the sine - gordon equation , the stochastic variable appears only in the initial and boundary conditions , and does not appear in the main pde . to perform the integration related to @xmath29 , one can use the legendre - gauss quadrature for the legendre chaos and gauss - hermite quadrature for the hermite chaos and obtain a fast convergence . in this paper , only @xmath453 quadrature points are used . adopting the quadrature rule , one can easily speed up the whole process . during our study , we also found a limitation of the gpc method regarding the determination of the initial critical velocity or the critical amplitude of the singular potential term for the sine - gordon equation . that is , we could not use the reconstruction constructed by the gpc modes to find the pdf of the receiving solutions due to the non - linearity and non - separability related to the gibbs phenomenon of the spectral method . despite this limitation , we find that the gpc method works well for finding the critical values for the singularly perturbed klein - gordon and sine - gordon equations . since no significant gpc analysis exists for the singularly perturbed pdes by the dirac @xmath0-function , our study could be a good resource for this type of research . in our future work , we will apply a similar gpc method for different pdes , such as the nonlinear schrdinger equations with singular potential terms . we will also consider more general types of uncertainties associated with the singular potential term , for which more than one random variables are involved and the gpc expansion should be in the multi - dimensional random space . to reduce the complexity due to the dimensionality , we will attempt to use the gpc collocation method with the fast sparse grid methods @xcite . .1 in * acknowledgement : * we are grateful to bruce pitman , gino biondini , and emmanuel lorin for their useful discussions and references . some important properties of the legendre polynomials are used in the paper . legendre polynomials are defined by the solutions @xmath459 of the following sturm - liouville equation : @xmath460+n(n+1)p_{n}=0 , \quad x \in [ -1,1],\ ] ] where @xmath459 is the legendre polynomial of degree @xmath461 . first few legendre polynomials are @xmath462 @xmath463 @xmath464 @xmath465 @xmath466 . also @xmath467 , @xmath468 if @xmath461 is even and @xmath469 if @xmath461 is odd , also @xmath470 if @xmath461 is odd with the orthogonality @xmath471 from eq . ( .2 ) we have , @xmath473 multiplying both sides by @xmath474 and integrating from @xmath475 to @xmath476 , we get , @xmath477 by using the orthogonality of the legendre polynomials we get , @xmath478 by using eq . ( .3 ) we get the followings . + for @xmath461 is odd number : @xmath479 for @xmath461 is even number : @xmath480 the derivative of legendre polynomials as the finite linear sum of legendre polynomials are used in the paper and some examples are as follows : for @xmath481 , @xmath482 and for @xmath483 , @xmath484 by orthogonal property @xmath485 . so in general if @xmath165 is even and @xmath461 is odd or @xmath461 is even and @xmath165 is odd , @xmath486 + if both @xmath165 and @xmath461 are even numbers and @xmath487 then by using the orthogonal property of legendre polynomials and from eq . ( [ even_derivatives ] ) we have , @xmath488=n(n+1).\nonumber \end{aligned}\ ] ] if both @xmath165 and @xmath461 are odd numbers and @xmath487 then the orthogonal property and eq . ( [ odd_derivatives ] ) yield , @xmath489=n(n+1).\ ] ] + + k. dalbey , a. k. patra , e. b. pitman , m. i. bursik , and m. f. sheridan , input uncertainty propagation methods and hazard mapping of geophysical mass flows , _ j. geophy . research _ , 113 , 2008 , pp . b05203 : 116 . w. c. k. mak , b. a. malomed , and p. l. chu , interaction of a soliton with a local defect in a bragg grating , _ j. opt b _ 20 , 2003 , pp . . l. morales - molina and r. a. vicencio , trapping of discrete solitons by defects in nonlinear waveguides arrays , _ optics letters _ 31 , 2006 , pp . 966968 . a. trombettoni , a. smerzi , and a. r. bishop , discrete nonlinaer schrodinger equation with defects , _ phys . e _ 67 , 2003 , pp . 016607 : 111 . d. wang , j .- h . jung , and g. biondini , detailed study of numerical methods for the perturbed sine - gordon equation with impulse forcing , _ in preparation _ , 2011 .

we study one - dimensional-wave superconductors capacitively coupled to a microwave stripline cavity . by probing the light exiting from the cavity , one can reveal the electronic susceptibility of the-wave superconductor . we analyze two superconducting systems : the prototypical kitaev chain , and a topological semiconducting wire . for both systems , we show that the photonic measurements , via the electronic susceptibility , allows us to determine the topological phase transition point , the emergence of the majorana fermions , and the parity of their ground state . we show that all these effects , which are absent in effective theories that take into account the coupling of light to majorana fermions only , are due to the interplay between the majorana fermions and the bulk states of the superconductors .

1502.03082

condensed matter systems are an endless resource of emergent physical phenomena and associated quasiparticles . majorana fermions , which are particles that are their own antiparticles and which have been first proposed as particles in the context of high energy physics , emerge beautifully as zero energy excitations in condensed matter setups @xcite . specifically , they are predicted to occur as zero energy excitations in solid - state systems , such as genuine @xmath0-wave superconductors @xcite , or engineered from topological insulators @xcite , semiconductor wires in a magnetic field @xcite , or in chains of magnetic atoms @xcite , all in the proximity of @xmath1-wave superconductors . these exotic objects are robust against local perturbations and , moreover , they obey non - abelian statistics @xcite under braiding operations , thus recommending them as qubits for the implementation of topological quantum computation . electronic transport is the foremost experimental tool for investigating the majorana fermions physics but alternative , _ non - invasive _ , methods that preserve the quantum states would be highly desired to address these objects . cavity quantum electrodynamics ( cavity qed ) has been established as an extremely versatile tool to address equilibrium and out - of - equilibrium electronic and spin systems non - invasively @xcite . majorana fermions , too , have been recently under theoretical scrutiny in the context of cavity qed physics @xcite . however , most of these studies dealt with effective low energy models that involved majorana fermions only , leaving the bulk physics , which is at the heart of the majorana physics , largely unexplored . and @xmath2 , respectively , and measuring the field at the end with @xmath3 and @xmath4 . the difference between the two gives a direct access to the electronic correlation function in the wire ( see text ) . the presence of majorana end modes in the finite wire ( black curves ) is also signaled in the cavity response . ] the basic idea behind cavity qed with electronic system is that it allows one to extract various properties of the latter , such as its spectrum and its electronic distribution function , from photonic transport measurements , as opposed to electronic transport . such photonic transport is quantified by the complex transmission coefficient @xmath5 that relates the output and input photonic fields as depicted in fig . [ fig : scheme ] . in the weakly coupled limit , one finds @xcite , appendix [ inputoutput ] : @xmath6 where @xmath7 and @xmath8 are the frequency and the escape rate of the cavity , respectively , while @xmath9 is an electronic correlation function that depends on the actual coupling between the two systems , and which contains information about the spectrum of the electronic system . the phase and amplitude response of the cavity close to resonance @xmath10 are related to the susceptibility @xmath9 as follows : @xmath11 and @xmath12 , where @xmath13 , @xmath14 , and @xmath15 $ ] ( @xmath16 $ ] ) is the real ( imaginary ) part of the susceptibility . in this paper , we evaluate the function @xmath9 first for the simple case of a one - dimensional ( 1d ) @xmath0-wave superconductor described by the kitaev chain and then for more realistic model of a 1d topological semiconducting wire in proximity of a superconductor . we assume in both cases that these 1d systems are coupled to a microwave cavity , as showed schematically in fig . [ fig : scheme ] . we address various physical situations for this coupling and show that such a method allows us to ascertain the topological phase transition point , the occurrence of majorana fermions , and the parity of the ground state , all in a _ global _ and _ non - invasive _ fashion . the paper is structured as follows . in sec . [ sec2 ] , we describe our model hamiltonians for the two systems under consideration and discuss the coupling between the microwave photons and the electrons in the 1d topological systems . in sec . [ sec3 ] , we show how the optical transmission through the cavity is able to probe the topological phase transition . in [ sec4 ] , we demonstrate that the cavity allows to detect the occurrence of majorana fermions and the parity of the majorana fermionic state in a non - invasive fashion . finally , in sec . [ sec5 ] we provide a brief summary of our results . technical details of the calculations are given in the appendices . in the following we will consider various models of @xmath0-wave superconductors coupled to a microwave ( superconducting ) cavity , such as the kitaev @xmath0-wave superconductor model , and the spin - orbit coupled nanowire subjected to a magnetic field and in the proximity of an @xmath1-wave superconductor . the general hamiltonian for the one - dimensional systems we consider here is of the form : @xmath17 being the sum of the electronic hamiltonian , its capacitive coupling to the cavity , and the free photon field , respectively . while the electronic term is model specific , and it will be discussed below , the last two terms read : @xmath18 and @xmath19 in eq . , @xmath20 is the photon creation ( annihilation ) operator , respectively . @xmath21 is the electron - photon coupling constant that couples to the charge density @xmath22 . this merely acts as to shift the chemical potential . in eq . , @xmath7 is the frequency of the photonic mode ( setting @xmath23 throughout ) . such a model could be realized experimentally by coupling a spin - orbit nanowire in the presence of a zeeman field to a nearby @xmath1-wave superconductor @xcite . in the present setup , which is based on a microwave superconducting stripline cavity , the @xmath1-wave superconductor that induces superconducting correlations in the wire could be a part of the underlaying cavity . for example , the nanowire could be tunnel - coupled to the central superconducting material showed in fig . [ fig : scheme ] . we have a considered a global capacitive coupling between the electronic nanosystem and the cavity electric field . such a coupling can be justified by a full microscopic approach ( see appendix [ sec : effective ] for details and also ref . @xcite that provides a microscopic description of the electric coupling between electrons in a nanocircuit and cavity photons ) . by solving the equation of motion @xmath24 $ ] for the photonic field iteratively up to second order in @xmath21 with respect to the cavity frequency @xmath7 @xcite , we find for the correlation function @xmath9 in eq . in the time domain @xmath25\rangle\ , , \label{susc_time}\ ] ] being the _ total _ charge susceptibility of the electronic system ( which can be here a 1d @xmath26wave superconductor or a topological 1d wire ) . in eq . , we introduced @xmath27 , with @xmath22 being the total number of electrons operator and @xmath28 the evolution operator for the electronic system . we assume zero temperature limit ( @xmath29 ) so that the average @xmath30 is taken over the superconducting ground state . note that @xmath31 and that @xmath32 in the absence of superconductivity ( @xmath33 ) , i.e. there are no effects from such a coupling for a wire in the normal state . we detail below the models for both topological 1d systems we consider in this paper . the simplest model of a @xmath0-wave superconductor that hosts majorana fermions is the kitaev chain @xcite . therefore , we first consider for the electronic part in eq . , the kitaev hamiltonian @xmath34 that reads : @xmath35 where @xmath36 is the hopping parameter , @xmath37 is the @xmath0-wave superconducting pairing potential , @xmath38 is the chemical potential , and @xmath39 is the total number of sites . also , @xmath40 is the creation ( annihilation ) electronic operator at the site @xmath41 . note that the electronic operators are spinless , and the electronic density is given by @xmath42 . in the present setup , which is based on a microwave superconducting stripline cavity , the @xmath1-wave superconductor that induces superconducting correlations in the wire could be a part of the underlaying cavity . for example , the nanowire could be tunnel - coupled to the central ( super-)conductor showed in fig . [ fig : scheme ] . the fact that microwave photons effectively couple only to electrons of the kitaev chain is accounted for in appendix [ sec : effective ] . the kitaev hamiltonian in eq . can be easily diagonalized . the susceptibility in this case can be found by simply substituting the expression for the density in eq . with the one corresponding to the kitaev model . we will discuss its physical content in sec . [ sec3 ] and sec . [ sec4 ] . a realistic system that can emulate , in some limits , the kitaev chain consists of a nanowire with a spin - orbit interaction , subjected to an external magnetic field , and coupled by proximity effect to an @xmath1-wave superconductor @xcite . the entire system is then assumed to be ( capacitively ) coupled to the microwave cavity . the tight - binding hamiltonian @xmath43 for the nanowire with spin - orbit ( so ) interaction in the presence of the magnetic field reads @xcite @xmath44 where , as before , @xmath36 and @xmath38 are the hopping amplitude and the chemical potential , respectively , @xmath45 is the spin - flip hopping amplitude ( or the spin - orbit coupling ) , @xmath37 is the @xmath1-wave pairing potential induced by proximity , @xmath46 is the zeeman splitting energy ( @xmath47 , with @xmath48 and @xmath49 being the @xmath48-factor and external magnetic fields , respectively ) . also , @xmath50 ( @xmath51 ) are the annihilation ( creation ) operators for electrons at site @xmath41 and spin @xmath52 , and @xmath53 , with @xmath54 are the pauli matrices that act in the spin space . this model accounts thus for spinfull electrons . note that we assumed the spin - orbit field and the magnetic field to be orthogonal . the coupling to the cavity is again capacitive , and the density reads in this case @xmath55 . in order to find the susceptibility , we need to substitute this expression for the electronic density in eq . , and we will discuss the various cases in the following sections . next we will show that the topological phase transition can be inferred from the cavity response from the transmission @xmath56 via the susceptibility @xmath9 . this function can be calculated straightforwardly in the case of a closed ring , i.e. for periodic boundary conditions ( pbcs ) , so that @xmath57 for the kitaev chain ( @xmath58 for the so nanowire ) . for pbcs , we can utilize the fourier description for the electronic operators : @xmath59 , with @xmath60 ( assuming the lattice spacing @xmath61 thereon ) , with @xmath62 . for more details see appendix [ bulksusceptibility ] . by doing so , we can readily write down the electronic hamiltonian @xmath63 , with @xmath64 where @xmath65 are pauli matrices that act in the nambu ( particle - hole ) space , i.e. on the vectors @xmath66 . the coupling to the cavity , on the other hand , simply reads @xmath67 so that the susceptibility in the time domain can be written as : @xmath68|0\rangle\,,\ ] ] with @xmath69 . utilizing this description , after some lengthy but straightforward calculations , we obtain for the susceptibility ( in the @xmath70 space ) : @xmath71 where @xmath72 is the bogoliubov spectrum of the 1d @xmath0-wave superconductor [ from diagonalizing the bdg hamiltonian in eq . ] @xcite and @xmath73 is a small positive number that accounts for causality . for @xmath74 , the imaginary part @xmath75 acquires a simple analytical form , and it is given by @xmath76 ^ 2}{\displaystyle 4{\rm t}^2\mu^2}}\,,\end{aligned}\ ] ] for @xmath77 and is zero otherwise . the topological phase transition takes place at @xmath78 , with the system being in the topological ( trivial ) phase for @xmath79 ( @xmath80 ) . in fig . [ topo_trans ] we plot @xmath75 ( main plot ) and @xmath81 ( inset ) as a function of the chemical potential @xmath38 for various values of the cavity frequency @xmath70 . we see that this function shows a large peak at the transition point ( @xmath78 ) , which becomes narrower and more pronounced for smaller @xmath70 ( compared to the gap @xmath37 ) . physically , this is due to the fact that the electronic levels close to the zero energy have larger curvatures , i.e. they are more susceptible close to the phase transition point . the real part also serves for detecting the phase transition , although not as directly as the imaginary part , as shown in fig . [ topo_trans ] , where the phase transitions are inferred from the kinks in this function . we have checked that the same peak structure holds for the cases when @xmath82 , too , the only modification being a shift in the scale for @xmath70 , which should be of the order of @xmath83 . ] as a function of @xmath38 for the kitaev model . the topological phase transition takes place at @xmath84 , where this function reaches its maximum , indicating the transition point . inset : the real part of the density - density correlation function [ @xmath81 ] , which also shows features ( kinks ) around the topological phase transition point . the full , dashed , dot - dashed curves correspond to the @xmath86 , @xmath87 , and @xmath88 , respectively . we assumed @xmath89 , @xmath90 and all energies are expressed in terms of @xmath36 . ] the case of a realistic so coupled nanowire is more complicated that the kitaev model showed above , and so is the evaluation of susceptibility . this is so because the so coupled wire has four bands ( because of the spin ) , instead of two , and a more complicated quasiparticle spectrum . nevertheless , writing the electronic operators in the fourier space as @xmath91 , we can write again the electronic hamiltonian as @xmath92 , with @xmath93\tau_z^k+v_z\sigma_x+\delta\tau_x^k\,,\ ] ] and the coupling to the cavity the same as in eq . . however , the expression for @xmath94 becomes rather cumbersome for the general case and to get some analytical insights we need to resort to approximations . for that , the hamiltonian can be put in a different form by the use of a unitary transformation ( see appendix [ nanowire_periodic ] ) : @xmath95\tau_z^k\nonumber\\ & + \frac{\delta\gamma\sin{k}}{\sqrt{(\gamma\sin{k})^2+v_z^2}}\tau_x^k-\frac{\delta v_z}{\sqrt{(\gamma\sin{k})^2+v_z^2}}\sigma_y\tau_y^k\,,\end{aligned}\ ] ] while the @xmath96 stays unchanged . progress can be made if we assume the limit of large magnetic field , @xmath97 , in which case we can neglect the last term in the above hamiltonian . by doing so , we recover two copies of the kitaev chain , for @xmath98 . the susceptibility becomes : @xmath99 where @xmath100 are given by the kitaev spectrum with : @xmath101 all the results from the previous section apply to this case but with the @xmath102-dependent parameters showed above . the system is in the topological nontrivial ( trivial ) regime for @xmath103 ( @xmath104 ) . in fig . [ topo_trans_so ] we plot the imaginary part of the susceptibility as a function of the zeeman splitting @xmath46 for two different values of the cavity frequency @xmath70 . we see a similar behavior as in the case of the kitaev chain : a peak emerges in @xmath75 at the topological phase transition point , which becomes narrower as omega becomes smaller . however , an extra peak emerges at a larger @xmath46 , and it is due to the resonance condition with the gaps around the @xmath105 ( external gaps in the so coupled nanowire spectrum ) . ] as a function of the zeeman splitting @xmath46 for a so coupled nanowire in case of pbcs for parameters @xmath106 , @xmath107 , @xmath108 , @xmath109 , @xmath110 , @xmath111 , and @xmath112 mev ( @xmath86 mev in the inset ) . the topological transition takes place for @xmath113 mev , for which the susceptibility reaches its maximum . the emergence of a second peak is due to the resonance condition around the external gaps ( @xmath105 ) . ] in this section , we consider a finite 1d topological system coupled to the cavity ( therefore with open boundary conditions , or obcs ) , so that there are two majorana fermions emerging in the topological region , each localized at one of the two ends of the chain . taken together , they give rise to a zero - energy fermionic state in the infinite wire limit , which can be either empty or occupied , thus labeling the _ parity _ of a 1d @xmath0-wave superconductor @xcite . the majorana wavefunctions decay exponentially in the wire on the scale of the superconducting correlation length @xmath114 , and for a finite wire it can lead to a finite energy splitting @xmath115 of the initially zero energy fermionic state @xcite . in the following , we will show that both the presence of the majorana fermions and the parity of the majorana fermionic state can be inferred from the susceptibility @xmath9 . in the finite chain case we can not obtain exact results for @xmath9 anymore , therefore we proceed to calculate this quantity numerically ( see appendix [ openboundaries ] ) . we will treat the two models , the kitaev chain and the so coupled wire on equal footing , showing that they give similar results . for starters , the electronic hamiltonian can be casted in the following form : @xmath116 with @xmath117 where @xmath1 counts internal degrees of freedom , such as spin , band index , etc . for the kitaev chain @xmath118 ( and thus we can disregard it ) , while for the so coupled nanowire @xmath119 . here , @xmath120 is a @xmath121 matrix @xcite , and we can write @xmath122 , with @xmath123 @xmath124 is a unitary matrix ( @xmath125 ) whose columns are the eigenvectors of @xmath120 @xcite . also , @xmath126 , with @xmath127 are the eigenenergies of the electronic hamiltonian , including the majoranas ( if present ) . thus , the electronic hamiltonian can be re - written as @xmath128 and @xmath129 , where @xmath130 with @xmath131 ( @xmath132 ) are the creation ( annihilation ) operators for the bogoliubov quasiparticles in the finite wire , with @xmath133 labeling the energy levels . finally , we can write @xmath134 and also define the spinorial wavefunction for the state of energy @xmath135 at position @xmath41 as @xmath136 , where @xmath137 are the electron ( hole ) components of the wavefunction at position @xmath41 in the wire . the electron - cavity coupling hamiltonian can be then written in the new basis as follows : @xmath138\nonumber\\ & \times(a^\dagger+a)\ , , \label{el_cav}\end{aligned}\ ] ] where @xmath139 are coefficients that depend on the transformation from the electronic basis @xmath140 to the bogoliubov basis @xmath141 and read @xcite : @xmath142 here the pseudo - spin @xmath143 acts in the nambu ( or particle - hole ) subspace . in general , all @xmath144 for @xmath145 and @xmath146 , thus there are couplings between all the levels ( and bands ) via the cavity field , and that includes transitions between the majorana and the bulk ( or gaped ) modes . this in turn affects the correlation function in eq . , which can be written as @xmath147 , being the sum of the terms that contain only bulk states ( bulk - bulk , or bb ) , cross terms between majorana and the bulk ( bulk - majorana or bm ) , and majorana contributions only ( majorana - majorana or mm ) , respectively . however , @xmath148 @xcite due to the fact that the cavity can not mix different parities , and in consequence the only contribution from the majorana modes comes through the cross terms @xmath149 . we have found that for @xmath150 the @xmath151 contribution is given by the one obtained from the pbcs in the first part of the paper , i.e. , @xmath152 , while @xmath153 , up to exponentially small terms in @xmath154 . we note in passing that in a real wire , the smallness of the @xmath155 compared to @xmath156 is measured by @xmath157 , with @xmath158 being the fermi wavelength and @xmath159 the length of the wire . on the chemical potential @xmath38 . the blue ( dot - dashed ) , red ( full ) , and the black ( dashed ) lines correspond to the susceptibility for pbcs , obcs for parity @xmath160 , and obcs for parity @xmath161 , respectively . lower inset : the relative strength of the susceptibility @xmath162 as a function of @xmath38 in logarithmic scale , for @xmath163 ( red - full ) , @xmath164 ( green - dashed ) , and @xmath165 ( blue - dot - dashed ) . the size of @xmath166 is exponentially reduced as a function of @xmath37 . we used @xmath167 , @xmath86 , @xmath163 , @xmath168 , and all energies are expressed in terms of @xmath36 . ] in the following , we analyze the cross - terms contribution @xmath149 . for @xmath169 , with @xmath170 , we obtain : @xmath171\ , , \label{pibm}\end{aligned}\ ] ] where @xmath172 and @xmath173 are the occupations of the bulk and majorana states , respectively . this is one of our main results . inspecting the above expression , we see that it is strongly dependent on the majorana state parity @xmath173 . assuming that @xmath174 for @xmath175 and @xmath176 for @xmath177 in the ground state , we obtain that @xmath178 ( @xmath179 ) for @xmath161 ( @xmath160 ) . to get more physical insight into the resulting susceptibility , we write the coefficients @xmath180 in the following way : @xmath181\,.\end{aligned}\ ] ] let us analyze the implication of the above result . when @xmath182 , we also have @xmath183 , and thus @xmath184 , since electron and hole contributions are the same in the majorana state . however , for a finite energy splitting @xmath185 , and thus we have that @xmath186 , which in turn results in @xmath187 . all these suggest that the susceptibility @xmath9 , via @xmath149 should allow us to infer both the parity of the ground state and the zeros in the majorana energy @xmath188 , assuming their spatial overlap is large enough . on the zeeman splitting @xmath46 . above the topological transition ( indicated in the figure ) , the red ( full ) and black ( dashed ) curves correspond to @xmath189 for parity @xmath190 and @xmath161 , respectively . @xmath189 shows oscillations as a function of @xmath46 , that are different in amplitude for the two parities @xmath191 ( but having the same period ) , crossing at points where @xmath182 ( see inset ) . below the topological transition the susceptibility reproduces well the one for pbcs . the parameters we took are @xcite : @xmath192 , @xmath193 , @xmath194 , @xmath109 , @xmath195 , @xmath196 , and @xmath111 . ] in the main plot in fig . [ parities ] , we show the real part @xmath81 for the kitaev chain as a function of the chemical potential @xmath38 for the two parities @xmath197 as well as the bulk value for pbcs . first of all , the values for @xmath9 in case of periodic and obcs are different because of @xmath149 , as this contribution has a different dependence on @xmath38 and @xmath37 from the bulk states . second of all , the open bcs wire susceptibility shows oscillations as a function of @xmath38 on top of the average value , of the form @xmath198 , with @xmath199 corresponding to @xmath161 ( @xmath160 ) , i.e. they are opposite in sign for the two parities . here @xmath200 is the fermi wavevector of the electronic system , and for the range of parameters considered is @xmath201 @xcite . this means that the cavity field can access the parity of the majorana fermions non - invasively and without locally accessing the wire . moreover , the oscillations disappear below the phase transition point @xmath202 , the susceptibility @xmath81 acquires the same value as for the pbcs wire which signals that the majorana fermions exist only above the topological phase transition . in order to get a closer look at the oscillations of @xmath203 , in the lower inset in fig . [ parities ] we show the real part of the relative difference between the two parities , @xmath204 , for different values of @xmath37 . we see that the oscillations have the same periodicity as the majorana energy splitting @xmath205 . notice that the oscillations of the majorana splitting with the chemical potential has been studied in detail @xcite together with the fact that the magnitude of the oscillations becomes exponentially suppressed in @xmath154 @xcite . in the main figure in fig . [ parities - wire ] , we plot the real part of the susceptibility for a 1d topological wire as a function of the zeeman splitting @xmath46 for the two parities @xmath197 . the susceptibility @xmath206 for that figure was computed using realistic parameters that might be appropriate for an insb wire such as in the experiments in ref . . we find similar features as for the kitaev toy model , namely oscillations as a function of the zeeman splitting above the topological transition . these oscillations around the ground state have opposite sign and different amplitudes for each parity . like for the kitaev model , they have the same periodicity as the majorana energy @xmath188 ( see the inset of fig . [ parities - wire ] ) and cross at points where @xmath182 . notice that if the parity is not conserved in the system ( for example , due to the quasi - particle poisoning ) , @xmath207 will follow the ground state and exhibit therefore sharp cusps as a function @xmath208 at the crossing points where @xmath182 ( see also ref . for similar features in a topological josephson junction ) . as for the kitaev chain , we thus find that the cavity phase shift is thus able to detect the majorana fermions and the parity of the ground state of a realistic topological wire . on the cavity frequency @xmath70 for @xmath167 . the full - red ( dashed - black ) line corresponds to the susceptibility in the topological regime for @xmath209 , while the dot - dashed - blue ( dotted - green ) corresponds to the non - topological regime with open ( periodic ) bcs with @xmath210 , so that the effective gap is the same @xmath211 in both regions . inset : a zoom in the region where the majorana peak emerges . for all the plots we used @xmath89 , and all energies are expressed in terms of @xmath36 . ] the imaginary part of @xmath9 gives us also information on the presence of majorana fermions . in fig . [ susc_of_omega ] , we show the dependence of @xmath75 on @xmath70 for the kitaev chain , both in the topological and non - topological regimes , for @xmath74 . we see that the majorana fermions , through @xmath149 , give rise to an extra peak in the susceptibility at half the effective superconducting gap @xmath212 in the topological regime , while such a peak is absent for the same effective gap @xmath213 , but in the non - topological case . for completeness , we also show the result for pbcs , in which case there are no majorana fermions . finally , let us give some estimates for @xmath9 , and in particular for @xmath149 and the resulting phase shift in the exiting photonic signal . we assume typical experimental values for the cavity frequency , @xmath214 ev , and with a quality factor @xmath215 , which results in photon escape rate @xmath216 ev . for an estimate of the capacitive coupling @xmath21 we refer , for example , to the case of carbon nanotubes , which have been under experimental scrutiny in the context of cavity qed @xcite . there , it was found that @xmath217 ev , and we believe similar values should be relevant for semiconductor nanowires too . the phase shift of the radiation exiting the cavity satisfies @xmath218 so that we obtain @xmath219 which is a sizeable value . we studied two paradigmatic examples of 1d topological superconducting systems capacitively coupled to a microwave superconducting stripline cavity : the kitaev chain and a 1d nanowire with strong so interaction in the presence of a magnetic field and in proximity of a superconductor . we analyzed the electronic charge susceptibility of these systems that is revealed in the photonic transport through the microwave cavity via its transmission @xmath220 . we showed that this electronic susceptibility can actually be used to detect the topological phase transition , the occurrence of majorana fermions and the parity of the majorana fermionic state in a _ non - invasive _ fashion . such effects are due to the interplay between the bulk and majorana states , either via virtual or real transitions taking place between the two , and which are mediated by the photonic field . as an outlook , it would be interesting to use the same cavity qed setup to access the physics associated with the fractional josephson effect . _ acknowledgments_| we acknowledge discussions with the lps mesoscopic group in orsay , and in particular insightful suggestions by helene bouchiat , and we thank jukka vayrynen for valuable correspondence . this work is supported by a public grant from the `` laboratoire dexcellence physics atom light matter '' ( labex palm , reference : anr-10-labx-0039 ) and the french agence nationale de la recherche through the anr contract dymesys . in this section , we provide theoretical arguments for the wire hamiltonian utilized in eq . ( [ kitaevhamiltonian ] ) , and the effective electron - cavity hamiltonian used in the main text ( mt ) . in a continuum description , the natural way to account for the interaction between the electrons and the electromagnetic field is via the minimal coupling , i.e. @xmath221 in the electronic hamiltonian , with @xmath222 being the electromagnetic field vector potential and @xmath223 being the momentum of the electrons in the material . in a tight - binding picture instead , one accounts for the coupling between light and matter by performing the peierls substitution to the hopping parameters @xmath224 between neighboring sites @xmath225 and @xmath226 , namely @xmath227 with @xmath228 being the electromagnetic field vector potential at position @xmath229 , and the integration is performed between the sites @xmath225 and @xmath226 . we will focus on the derivation of the effective kitaev model in the tight - binding picture , as the microscopic , continuum model was described in great detail very recently in @xcite . we thus refer the reader to that paper for a detailed calculation of the cavity effects , as well as the derivation of the capacitive coupling starting from the minimal coupling . here we give some details on the derivation of eq . ( [ kitaevhamiltonian ] ) in the mt starting from a non - superconducting nanowire coupled to a bulk @xmath0-wave superconductor with such a coupling being assisted by the cavity field . for simplicity , we assume the bulk to be not @xmath1 , but @xmath0-wave paired , thus the presence of spin - orbit coupling in the wire is not a necessary ingredient . however , the present calculations can be straightforwardly generalized to more realistic system , such as nanowires with soi . the total hamiltonian of the system reads : @xmath230 where @xmath231 with @xmath232 , and @xmath233 ( no intrinsic superconductivity in the wire ) , and @xmath234 the @xmath0-wave pairing in the bulk superconductor . here , @xmath235 ( @xmath236 ) and @xmath237 are the electronic annihilation ( creation ) operator at position @xmath41 and the hopping parameter in system @xmath238 , respectively . the tunneling hamiltonian in the presence of the cavity reads : @xmath239 where @xmath240 , with @xmath241 , @xmath242 , @xmath243 , @xmath7 , and @xmath244 ( @xmath245 ) , being the cavity vector potential , the coupling strength , the cavity frequency , and the cavity photon annihilation ( creation ) operators , respectively . note that we assumed that the cavity field points perpendicularly to the wire , and it has no component along it . if instead such components would exists , we should have modified the wire hamiltonian too in order to account for the cavity induced phase factors . in the following , we will assume that @xmath246 , namely it is constant along the entire wire . finally , the hamiltonian of the cavity reads : @xmath247 with @xmath7 being the ( fundamental ) frequency of the cavity . before deriving an effective wire hamiltonian , it is instructive to switch to the fourier space , for both the bulk and wire hamiltonians . we get : @xmath248 where @xmath249 , with @xmath250 the chemical potential in the @xmath251 system . next we perform the so called lang - firsov transformation on the system hamiltonian , which means @xmath252 with @xmath253 chosen as follows : @xmath254 after some lengthy , but straightforward calculation we obtain the system hamiltonian as follows : @xmath255 which implies we excluded the photonic field from the tunneling term at the expense of adding photon - dependent chemical potential shift in the wire ( third term ) as well as an interaction term ( fourth term ) . note that for @xmath256 , the transformation does not affect the spectrum , as it can be simply undone . however , as will see in the following , in the presence of the tunneling term the photonic field in the form of the capacitive coupling can lead to real effects . in the following , we aim at finding an effective hamiltonian describing the wire only by integrating the bulk superconductor degrees of freedom up to second order in the tunneling @xmath257 . we choose to do so by employing the schrieffer - wolff transformation formalism , which means , as before , that we unitary rotate the system hamiltonian as @xmath258+\dots\,,\end{aligned}\ ] ] and choose @xmath259 = -h_t\ , , \label{condition}\end{aligned}\ ] ] or @xmath260 , with @xmath261 being a superoperator whose action is defined as @xmath262 $ ] , @xmath263 . this is equivalent to the following identity : @xmath264 this term excludes the tunneling hamiltonian @xmath265 in leading order ( assuming there is no diagonal contribution caused by such a term ) . then , we neglect the contributions of the higher order terms on the wire spectrum by averaging over the bulk ground state @xmath266 in order to derive a purely ( renormalized ) wire hamiltonian : @xmath267\nonumber\\ & + \left[s_{sw } , h_{w - c}\right]+\dots\,|0_b\rangle\ , , \label{htransform}\end{aligned}\ ] ] in order to find @xmath268 from eq . explicitly , let us perform bogoliubov transformation for the bulk @xmath0-wave superconductor defined as @xmath269 where @xmath270 and @xmath271 with @xmath272 the phase of the superconducting condensate ( that we choose @xmath273 from now on ) and @xmath274 the spectrum . we can then express the bulk hamiltonian in terms of the @xmath275 and @xmath276 operators : @xmath277 utilizing the fact that : @xmath278 and @xmath279 we can readily find the transformation matrix @xmath268 as follows ( assuming also that @xmath280 , since we are interested in the energies well inside the band gap of the bulk superconductor ) : @xmath281\,.\end{aligned}\ ] ] utilizing this expression for @xmath268 , we can calculate the expectation values for the different commutators in eq . . we obtain : @xmath282|0_b\rangle \approx-\sum_{k}\frac{t_{int}^2}{e_{kb}}\nonumber\\ & \times\big[\left(|u_k|^2 - |v_k|^2\right)c^\dag_{k , w}c_{k , w}-2u_kv_kc^\dag_{-k , w}c^\dag_{k , w}+{\rm h.c.}\big]\,,\end{aligned}\ ] ] which can be interpreted as follows : the first term renormalizes the single particle spectrum in the wire , while the second term is responsible for the induced superconductivity in the wire . the full wire hamiltonian thus becomes : @xmath283 with @xmath284 being the renormalization of the single - particle energies and the @xmath0-wave induced gap ( @xmath285 ) . note that the last term in eq . can be seen as a normalization of the single - particle spectrum in the mean - field , and thus finally we recover the same wire hamiltonian defined in eq . ( [ kitaevhamiltonian ] ) in the mt . and @xmath286 are sent from the right and left mirrors , respectively , towards the cavity , and an output fields @xmath287 and @xmath288 are collected on the same sides . the cavity field , which is quantified by the bosonic operators @xmath244 and @xmath245 , interacts with the electronic system via capacitive coupling , affecting the cavity . the coupling between the cavity field and the external modes is quantified by the decay rate @xmath289 , with @xmath290 the bath density of states and @xmath291 being the average coupling between the cavity and the bath modes on the right ( left ) . in the first part , for simplicity , we assume that @xmath292 , which correspond to a one - side cavity . ] in this section , we present details on the input - output theory for the cavity in the presence of the coupling to a 1d @xmath0-wave sc . we will show that the transmission of the cavity depends on the electronic susceptibility of the electronic system . we concentrate on only single resonance of the cavity with frequency @xmath7 @xcite and , for simplicity , we assume for the moment a single - sided cavity . the total hamiltonian describing the system reads @xmath293 where @xmath294 is the electronic hamiltonian only , and @xmath295 are the cavity , the electronic - cavity , the bath , and the cavity - bath hamiltonians , respectively . above , @xmath244 ( @xmath245 ) is the annihilation ( creation ) operator for the cavity mode with energy @xmath7 , @xmath296 ( @xmath297 ) are the annihilation ( creation ) operators for the bath modes with energy @xmath298 , with @xmath299 labeling their quantum numbers , and the complex coefficients @xmath300 are coupling parameters between the cavity and the external bath . moreover , @xmath21 is the coupling strength between the cavity field and the total number operator in the system @xmath301 , with @xmath302 ( @xmath303 ) being the annihilation ( creation ) operator for the electrons ( fermionic degrees of freedom ) at site @xmath41 in the electronic system . the idea of the input - output theory is to find the output photons ( or field ) in terms of the input ones , as shown schematically in fig . [ sketch ] . following ref . @xcite , we obtain for the cavity equation of motion for a one - side cavity ( @xmath292 in fig . [ sketch ] ) : @xmath304 - \frac{\displaystyle \kappa}{\displaystyle 2}a - \sqrt{\kappa}b_{in}\ , , \label{ain}\end{aligned}\ ] ] for the input field , and @xmath304 + \frac{\displaystyle \kappa}{\displaystyle 2}a - \sqrt{\kappa}b_{out}\ , , \label{aout}\end{aligned}\ ] ] for the output field , where @xmath305 is the cavity decay rate , with @xmath290 the bath density of states and @xmath306 being the average coupling between the cavity and the bath modes . subtracting eq . from eq . we obtain that @xmath307 a result which holds for any general cavity hamiltonian . in the following , we will establish the relationship between @xmath287 and @xmath308 in the presence of the electronic system , as depicted in fig . [ sketch ] . for that , we first evaluate the commutator : @xmath309=-\omega_ca-\alpha n\ , , \label{eq_motion}\end{aligned}\ ] ] where @xmath310 is the time - dependent electronic particle number ( see below ) . in order to utilize this contribution to the equation of motion of the cavity field , we need to evaluate the time - dependent particle number @xmath311 , which itself depends on the coupling to the cavity . at time @xmath312 , we can write : @xmath313 and @xmath314 being the heisenberg and interaction pictures , respectively , with @xmath315 being the initial time and which can be chosen at will . we can represent @xmath316 as follows : @xmath317 where @xmath318 is the evolution operator with @xmath319 the time - ordering operator that puts operators with later times to the left of the ones with earlier times . we can then write eq.([nheisenberg ] ) in the following way @xmath320= n_i(t ) + i\alpha\int_{t_0}^{t}dt'\big[(a + a^\dag)n_i(t ' ) , n_i(t)\big]\ , , \label{noperator}\end{aligned}\ ] ] up to leading order in the coupling constant @xmath21 . thus , the time - evolution of the electronic particle number contains , besides the electronic component , a contribution that arises because of the coupling to the cavity . introducing eq . into eq . we obtain : @xmath321&= -\omega_c a -\alpha n = -\omega_c a -\alpha n_i(t)\nonumber\\ & - i\alpha^2\int_{t_0}^{t}dt'\big[\left(a(t ' ) + a^\dag(t')\right)n_i(t ' ) , n_i(t)\big ] . \label{hsystem2}\end{aligned}\ ] ] let us assume that @xmath322 and write eq . as : @xmath321&= -\omega_c a -\alpha n_i(t)- i\alpha^2\int_{-\infty}^{t}dt'\big[\big(a(t)e^{-i\omega_c(t ' - t)}+ a^\dag(t)e^{i\omega_c(t ' - t)}\big)n_i(t ' ) , n_i(t)\big]\ , , \label{hsystem3}\end{aligned}\ ] ] where @xmath323 and @xmath324 in zeroth order in @xmath21 ( because the expression is already multiplied by @xmath325 we can utilize the bare time dependence in this expression ) . in the following , we switch to the fourier space in order to solve the equation for @xmath326 , and take into account all contributions that affect its time - dependence , namely the external modes too . we obtain : @xmath327\ , . \label{aomega}\end{aligned}\ ] ] before continuing with the derivation , let us describe each term in the above expression . the first term describes the free cavity evolution , the second term the leaking into the continuum of modes ( the external bath ) at rate @xmath328 , the third term is the input field supplied from the right side , the fourth term correspond to another input " contribution to the cavity from the electronic system ( a noise term ) , while the last term leads to both a shift in the cavity frequency as well as to an extra decay channel ( @xmath329-factor modification ) . one can now average over the electronic system , thus neglecting any fluctuation ( i.e. feed - back effects ) . moreover , we can neglect the highly oscillating term @xmath330 , namely we perform the so called rotating wave approximation ( rwa ) . under all these assumptions , the last term in eq . becomes : @xmath331\rangle_0\nonumber\\ & = i\frac{\displaystyle \alpha^2}{\displaystyle \hbar}\int_{-\infty}^{\infty}dt e^{i\omega t}a(t)\int_{-\infty}^{\infty}dt'(-i)\theta(t ' - t)e^{-i\omega_c(t ' - t)}\nonumber\\ & \times\langle[n_i(t ' ) , n_i(t)]\rangle_0 = i\int_{-\infty}^{\infty}dt e^{i\omega t}a(t)\pi(-\omega_c)\nonumber\\ & = ia(\omega)\pi(-\omega_c ) , \label{aomegalastterm2}\end{aligned}\ ] ] where @xmath332\rangle_0\ , , \label{corrfunc}\end{aligned}\ ] ] is the retarded density - density electronic correlation function utilized in the main text , and @xmath333 means the expectation value of the unperturbed electronic system . note that for deriving the above expression we assumed that the electronic system is in equilibrium and thus the time dependence is homogeneous . we are now in position to find the cavity field @xmath326 and the output field @xmath334 in terms of the input field @xmath335 . introducing eq . into eq . we obtain @xmath336 so that @xmath337 we thus have two contributions to the cavity field : the external input and the input from the electronic system . our aim is to relate in fact the output and input fields , which can be done easily via the expression in eq . : @xmath338b_{in}(\omega)-i(\sqrt{\kappa}\alpha/\hbar)\langle n_i(\omega)\rangle_0}{\displaystyle -i(\omega - \omega_c ) + \kappa/2 - i\pi(-\omega_c)}\,.\end{aligned}\ ] ] in the limit of large number of photons in the input beam , we can neglect the contribution from the electronic system so that we obtain : @xmath339 in experiments , one actually encounters a two - sided cavity ( see fig . [ sketch ] for the nomenclature ) , in which case the expression for the cavity equation of motion reads : @xmath304 - \left(\frac{\displaystyle \kappa_1}{\displaystyle 2}+\frac{\displaystyle \kappa_2}{\displaystyle 2}\right)a - \sqrt{\kappa_1}b_{in}-\sqrt{\kappa_2}c_{in}\ , , \label{ain2}\end{aligned}\ ] ] so that for the output fields we get : @xmath340 by following the same reasoning as for the one - sided cavity , we obtain : @xmath341 while if the two mirrors are the same @xmath342 , this becomes @xmath343+i(\alpha/\hbar)\langle n_i(\omega)\rangle_0}{\displaystyle -i(\omega - \omega_c ) + \kappa- i\pi(-\omega_c)}\ , . \label{res2twosided}\end{aligned}\ ] ] assuming again that the input flux is much larger than the electronic contribution , we can write : @xmath344 with @xmath345 being the transmission of the cavity , which is a complex number , and which depends on the electronic susceptibility @xmath9 , as stated in eq.([transmissioncoef ] ) in the mt . in this section we give more details on the derivation of the susceptibility @xmath9 for the case of a ring geometry for which we can apply pbcs . the kitaev chain hamiltonian in real space was defined in eq.([kitaevhamiltonian ] ) as ( for definitions of the parameters please see mt ) : @xmath346 and @xmath347 . in this case , we can switch to the fourier space , which implies we can write the fermionic operators as follows : @xmath348 where @xmath349 is the fermionic annihilation operator with momentum @xmath60 . we can then rewrite the electronic hamiltonian in momentum space @xmath350 with @xmath351 is the bogoliubov de gennes hamiltonian and @xmath352 . the interaction hamiltonian between the electronic system and the cavity can as well be written in the @xmath102-space as @xmath353 one can simply diagonalize the @xmath294 in the @xmath102-space and write : @xmath354 with @xmath355 being the eigenenergies , and we used @xmath356 @xmath357 here , the functions @xmath358 and @xmath359 are given by @xmath360 @xmath361 we can now calculate the susceptibility @xmath9 defined in the previous section , which quantifies the change in the photonic transmission @xmath362 due to the interaction with the electronic system . in the fourier space , we obtain : @xmath363\rangle . \label{correlationfunckspace}\end{aligned}\ ] ] using eq . and eq . with eq . and eq . we find : @xmath364 with the small @xmath365 assuring the convergence of the time - integrals . for large @xmath150 , we can transform the sum into integral , and also write @xmath366 , with : @xmath367\,,\end{aligned}\ ] ] where @xmath368 means the principal value of the function and we used the fact that : @xmath369 we can perform the integral over @xmath102 for the imaginary part @xmath81 to obtain the expression in the mt : @xmath76 ^ 2}{\displaystyle 4{\rm t}^2\mu^2}}\,,\end{aligned}\ ] ] for @xmath77 , and being zero otherwise . for the real part @xmath81 we found no simple solution , and so we chose not to depict it . note that the susceptibility @xmath370 , i. e. it scales linearly with the number of sites . the hamiltonian for a nanowire in the presence of a zeeman field @xmath49 , rashba spin - orbit coupling @xmath371 and induced superconductivity @xmath37 reads @xcite @xmath372 let us diagonalize the hamiltonian @xmath373 in the absence of the induced pairing @xmath37 and then treat the latter perturbatively @xmath374 in order to do so , let us perform the unitary transformation @xcite @xmath375 where @xmath376 if we choose @xmath21 so that @xmath377 @xmath378 takes the from @xmath379\tau_z.\end{aligned}\ ] ] now let us reintroduce the pairing term @xmath380 and apply the transformation @xmath381 to it . then @xmath382 reads @xmath383\tau_z+\frac{\displaystyle \delta up}{\displaystyle \sqrt{u^2p^2+b^2}}\tau_x\nonumber\\ & -\frac{\displaystyle \delta b}{\displaystyle \sqrt{u^2p^2+b^2}}\sigma_y\tau_y . \label{htransnw}\end{aligned}\ ] ] neglecting the last term in eq.([htransnw ] ) the effective hamiltonian reads @xmath384\tau_z+\frac{\displaystyle \delta up}{\displaystyle \sqrt{u^2p^2+b^2}}\tau_x . \label{heff}\end{aligned}\ ] ] the bulk energy spectrum of @xmath385 for @xmath386 is @xmath387 and for @xmath388 is @xmath389 the electronic susceptibility is defined as @xmath390\rangle.\end{aligned}\ ] ] in order to diagonalize the hamiltonian in eq . ( [ heff ] ) , let us perform a transformation @xmath391 where @xmath392 then the electronic susceptibility reads @xmath393\rangle . \label{pit}\end{aligned}\ ] ] @xmath394 introducing eq.([tauxtime ] ) into eq.([pit ] ) , @xmath94 reads @xmath395\rangle\nonumber\\ % = -i\theta(t)\langle - i\sin\left(2\gamma\right)\tau_y+i\sin\left(2\gamma\right)\cos(2\epsilon_p t)\tau_y+i\sin(2\gamma)\sin(2\epsilon_p t)\tau_x+2i\sin^2(\gamma)\sin(2\epsilon_p t)\tau_z\rangle\nonumber\\ = -2\theta(t)\sin^2(\gamma)\sin(2\epsilon_p t ) . \label{pit2}\end{aligned}\ ] ] performing the fourier transform , eq . ( [ pit2 ] ) reads @xmath396 sum can be transformed into integral as @xmath397 then the imaginary part of the susceptibility @xmath366 , eq . ( [ suscept ] ) reads @xmath398 . \label{susceptgeneral}\end{aligned}\ ] ] the low - energy subspace at @xmath399 is formed by the bands for which @xmath386 . @xmath400 when @xmath401 . the effective hamiltonian for @xmath402 , @xmath401 linearized around @xmath403 reads @xmath404 the bulk energy spectrum reads @xmath405 in this limit @xmath406 and the density of states reads @xmath407 introducing eq.([sin1 ] ) and eq.([density1 ] ) into eq.([susceptgeneral ] ) the imaginary part of the electronic susceptibility reads @xmath408 in this case the effective hamiltonian linearized around @xmath409 reads @xmath410\right)\tau_z+\frac{\displaystyle \delta u p}{\displaystyle b}\tau_x = \left(\frac{\displaystyle p^2}{\displaystyle 2m^*}-\mu+b\right)\tau_z+\frac{\displaystyle \delta u p}{\displaystyle b}\tau_x , \label{heffp2}\end{aligned}\ ] ] where an effective mass @xmath411 was introduced as @xmath412 the bulk energy spectrum reads @xmath413 where @xmath414 the density of state reads @xmath415}}. \label{density2}\end{aligned}\ ] ] and @xmath416}{\displaystyle b^2\epsilon^2c}. \label{sin2}\end{aligned}\ ] ] introducing eq.([density2 ] ) and eq.([sin2 ] ) into eq.([susceptgeneral ] ) the imaginary part of the electronic susceptibility reads @xmath417 in this part , we present details on the calculation of the susceptibility for a finite wire with open boundary conditions . in this case , the excitation spectrum changes compared to the previous case , as in the topological region the zero energy majorana fermions emerge . in the mt we found that the susceptibility reads : @xmath418|\rangle \label{suscept2}\ ] ] and which can be written as @xmath419 being the sum of a bulk susceptibility , that can be constructed from only the bulk ( or gaped ) states , and cross terms that involve both bulk and majorana states , respectively . we use the discrete lattice model to numerically diagonalize the hamiltonian for an electronic system with @xmath39 fermionic sites . this can be written in a compact form as follows : @xmath420 with @xmath421 where @xmath120 is a @xmath422 matrix . moreover , assuming all entries in the matrix are real , @xmath120 is also symmetric , we can write it as follows : @xmath423 where @xmath424 is a diagonal matrix with eigenvalues on its diagonal and @xmath124 is a unitary matrix ( @xmath425 ) whose columns are eigenvectors of @xmath120 . the matrix @xmath424 is ordered so that @xmath426 with @xmath427 being the eigenenergies of the bdg hamiltonian and @xmath428 . that pertains to the following diagonal hamiltonian : @xmath429 where @xmath430 ( @xmath431 ) are the annihilation ( creation ) operators for the bogoliubov quasiparticles , which are defined as follows : @xmath432 and @xmath433 it is instructive to introduce the wavefunctions @xmath434 , where @xmath435 , and which are describing the state @xmath436 at position @xmath437 in the lattice and accounts for the electron ( @xmath371 ) and hole ( @xmath438 ) components , respectively . that allows us to write : @xmath439 , \label{ci}\end{aligned}\ ] ] @xmath440 . \label{cidag}\end{aligned}\ ] ] so that we can rewrite eq . in terms of @xmath441 and @xmath442 as follows : @xmath443 next we extract from this expression only the @xmath149 component . this reads : @xmath444 \nonumber\\ & -(n_m - n_k)\big[v_{2k - 1 , i}u_{2 m , i}u_{2k , j}v_{2 m - 1 , j}\nonumber\\ & - v_{2k - 1 , i}u_{2 m , i}v_{2k , j}u_{2 m - 1 , j}+ v_{2 m , i}u_{2k - 1 , i}u_{2 m - 1 , j}v_{2k , j}\nonumber\\ & - v_{2 m , i}u_{2k - 1 , i}v_{2 m - 1 , j}u_{2k , j}\big]\bigg]\ , , \label{suscept4}\end{aligned}\ ] ] and , using that @xmath445 , it can be simplified even further to give : @xmath446 \nonumber\\ & + ( n_m - n_k)\big[-u_{2k , i}u_{2 m , i}u_{2k , j}u_{2 m , j}\nonumber\\ & + u_{2k , i}u_{2 m , i}v_{2k , j}v_{2 m , j } - v_{2 m , i}v_{2k , i}v_{2 m , j}v_{2k , j}\nonumber\\ & + v_{2 m , i}v_{2k , i}u_{2 m , j}u_{2k , j}\big]\bigg].\end{aligned}\ ] ] let us now introduce the coefficients @xmath447 , @xmath448 defined in the mt : @xmath449 which we can utilize to rewrite @xmath149 as follows : @xmath450\,,\end{aligned}\ ] ] and which correspond to the expression eq . ( [ pibm ] ) in the mt . 48ifxundefined [ 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 @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevb.79.094504 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevlett.111.107007 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevb.88.235401 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.035121 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevb.92.075143 [ * * , ( ) ] link:\doibase 10.1103/physrevb.91.085406 [ * * , ( ) ] link:\doibase 10.1103/physrevb.83.121311 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevb.91.045403 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.180503 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.220506 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevb.92.134508 [ * * , ( ) ] @noop * * , ( )

we show under general conditions that the linearized force - based quasicontinuum ( qcf ) operator has a positive spectrum , which is identical to the spectrum of the quasinonlocal quasicontinuum ( qnl ) operator in the case of second - neighbour interactions . moreover , we establish a bound on the condition number of a matrix of eigenvectors that is uniform in the number of atoms and the size of the atomistic region . these results establish the validity of and improve upon recent conjectures ( ( * ? ? ? * conjecture 2 ) and ( * ? ? ? * conjecture 8) ) which were based on numerical experiments . as immediate consequences of our results we obtain rigorous estimates for convergence rates of ( preconditioned ) gmres algorithms , as well as a new stability estimate for the qcf method .

1004.3435

quasicontinuum methods are a prototypical class of multiscale models that directly couple multiple modeling regions to reduce the computational complexity of modelling large atomistic systems . these methods are useful for computing the interaction of localized material defects such as crack tips or dislocations with long - range elastic fields of a crystalline material . the force - based quasicontinuum ( qcf ) method @xcite partitions the material into two disjoint regions , the atomistic region and the continuum region . it assigns forces to the degrees of freedom within each region using only the respective model , be it atomistic or continuum . this simplifies the formulation of the method as no special interaction rules are needed near the atomistic - continuum interface . the simplicity of mixing forces combined with the lack of spurious interface forces ( so - called `` ghost forces '' ) make the force - based method a popular approach , and this technique is widely applied in the multiscale literature @xcite . a potential drawback of the qcf method is that it does not derive from an energy ( as it generally produces a non - conservative field ) . while the practical implications of this fact are still under investigation , it is already clear that the analysis of the qcf method poses formidable challenges . a series of recent articles has been devoted to its study @xcite . for example , it was shown in @xcite that the linearized qcf operator is not positive definite , and that it is not uniformly stable ( in the number of atoms and the size of the atomistic region ) in discrete variants of most sobolev spaces . however , numerical experiments in @xcite showed some unexpected spectral properties . conjecture 2 in @xcite states that the spectrum of @xmath0-eigenvalues of the linearized qcf operator is identical to that of the operator associated with the quasinonlocal qc method @xcite . this is particularly surprising since the quasinonlocal qc method is energy - based , and thus indicates that the qcf operator is diagonalizable and that its spectrum is real . conjecture 8 in @xcite states that the condition number of a matrix of eigenvectors of the qcf operator grows at most logarithmically . this is an important fact for understanding the solution of the qcf equilibrium equations by the gmres method . the purpose of the present paper is to provide rigorous proofs for these numerical observations . we define the qcf method and introduce the necessary notation in section [ sec : prelims ] . in section [ sec : n2 ] we establish all results for the case of next - nearest neighbour interactions as in the numerical experiments in @xcite . then , in section [ sec : fr ] we extend the results to the more technical case of finite - range interactions . in the case of finite - range interactions we can not make the comparison between the spectra of the qcf and qnl operators . instead , we prove that the spectrum of the qcf operator lies between the spectrum of the atomistic operator and the spectrum of the continuum operator ( theorem [ th : fr : evals ] ) . we note , moreover , that we were able to construct a matrix of eigenvectors for the qcf operator whose condition number is bounded uniformly in the number of atoms and the size of the atomistic region . this result is in fact stronger than the conjectures made in @xcite . finally , in section [ sec : prec ] , we analyze variants of preconditioned qcf operators to obtain rigorous convergence rates for preconditioned gmres methods as well as a new stability estimate . for the sake of brevity , we will keep the introduction to the atomistic model and the various flavours of quasicontinuum approximations to an absolute minimum . we refer to @xcite for detailed discussions . note , in particular , that we have left out the usual rescaling factor @xmath1 . this reduces the complexity of the notation and is justified since in this paper we are primarily concerned with algebraic aspects of quasicontinuum operators . in this section , we summarize the notation and certain elementary results for some standard finite difference operators with periodic boundary conditions . we identify @xmath2 with periodic infinite sequences as follows : @xmath3 the @xmath0-inner product on @xmath4 , and its associated norm , are defined as @xmath5 we will frequently use a subspace @xmath6 of mean zero functions , @xmath7 where @xmath8 . the orthogonal projection onto @xmath9 is denoted @xmath10 , @xmath11 or , in matrix notation , @xmath12 the difference operator @xmath13 is defined by @xmath14 we note that @xmath15 and @xmath16 where @xmath17 denotes the range and @xmath18 denotes the kernel of an operator . we also remark that , here and throughout , unless specifically stated otherwise , we will not distinguish between an operator and its associated matrix representation in @xmath19 . the second generic operator that we will encounter is the negative laplace operator @xmath20 , @xmath21 as for the difference operator , @xmath22 and @xmath23 . using summation by parts , we obtain @xmath24 which implies that @xmath25 and hence @xmath26 . since @xmath27 , we have the identities @xmath28 we also note that @xmath29 , and that this bound is attained for even @xmath30 , as well as in the limit @xmath31 . since @xmath32 is singular , we also define the modified negative laplace operator @xmath33 so that @xmath34 if @xmath35 and @xmath36 this operator is invertible and satisfies @xmath37 the translation operator @xmath38 is defined by @xmath39 @xmath40 is an orthogonal operator , i.e. , @xmath41 and its eigenbasis can be written explicitly as @xmath42 the eigenvalues of @xmath40 are located on the unit circle @xmath43 and , in the limit @xmath44 , are dense in @xmath45 . we remark that we can write the difference operator @xmath46 and the negative laplace operator as laurent polynomials in @xmath40 : @xmath47 and @xmath48 where @xmath49 in general , if @xmath50 is a polynomial , then the spectrum of the operator @xmath51 is @xmath52 , and the eigenvectors are the same as for @xmath40 . since @xmath40 is a normal operator , all polynomials @xmath51 are also normal . finally , we note that @xmath53 , which implies that @xmath40 or any polynomial of @xmath40 commutes with @xmath54 . in particular , this implies that all polynomials in @xmath40 ( e.g. , @xmath32 , @xmath46 ) and the operators @xmath55 and @xmath56 commute . we consider an atomistic model problem with periodic boundary conditions . we let @xmath9 be the set of admissible displacements of an @xmath30-periodic chain : the set of all @xmath30-periodic displacements with mean zero . the latter condition is necessary to ensure that the systems of equations that we consider are well posed . if @xmath57 is a fixed _ macroscopic strain _ , then the energy ( per period ) of the atomistic chain subject to a displacement @xmath58 is given by @xmath59 where @xmath60 is a pair interaction potential , for example , a lennard jones or morse potential , and @xmath61 , @xmath62 , can be thought of as a discrete cutoff radius . ( note that , even though we have defined @xmath63 for all @xmath64 , only @xmath65 are admitted in the solution of the minimization problem . ) the cauchy born or local quasicontinuum ( qcl ) approximation of @xmath63 is the functional @xmath66 where @xmath67 is the _ cauchy born stored energy function _ , @xmath68 . our analysis in the present paper concerns properties of the hessians @xmath69 and @xmath70 and the quasicontinuum operators that we derive from them for future reference we write out @xmath71 and @xmath72 explicitly , @xmath73 where the constants @xmath74 and @xmath75 are given by @xmath76 we understand both @xmath71 and @xmath72 as linear operators from @xmath4 to @xmath4 , defined by the above formulas , but are primarily interested in their properties on @xmath9 . for example , we note that if @xmath77 and @xmath78 , then both are positive definite on @xmath9 and in particular invertible ( see ( * ? ? ? * eq . ( 2.2 ) and sec . 2.2 ) and ( * ? ? ? * prop . 1 and prop . 2 ) for the next - nearest neighbour case , and @xcite for finite range ) , however , both operators have a non - trivial kernel that contains @xmath79 . for the continuum operator , @xmath80 the stability condition @xmath81 is sharp , that is , @xmath72 is positive definite if and only if @xmath82 we work to prove our spectral results on @xmath83 up to this sharp stability criterion . the force - based quasicontinuum ( qcf ) method is obtained by mixing the forces from the atomistic and the continuum model . to this end , we define atomistic and continuum regions @xmath84 and @xmath85 that satisfy @xmath86 we define the qcf forces @xmath87 - \frac{\pp \e^\c(u)}{\pp u_\ell } , & \text{if } \ell \in \cs . } \ ] ] linearization of the nonlinear qcf operator @xmath88 at @xmath89 , yields the _ linear qcf operator _ ( or simply , _ qcf operator _ ) , @xmath90 , @xmath91 which is the focus of our studies in the present paper . unfortunately , @xmath83 as defined above , does not map @xmath9 to @xmath9 , hence we will normally consider the projected qcf operator ( see ( * ? ? ? 2.3 ) for more detail ) @xmath92 to conclude this section we represent @xmath83 in a more compact way . by considering the characteristic function of @xmath84 , @xmath93 and the associated diagonal operator @xmath94 , @xmath95 we can write @xmath83 in the form @xmath96 \lc + \chi \la = \lc + \chi [ \la - \lc].\ ] ] the second atomistic / continuum hybrid scheme that will feature prominently in our investigations is the _ quasinonlocal quasicontinuum ( qnl ) method _ we note that the qnl method is only defined for second - neighbour interaction range ( i.e. , @xmath97 ) . extensions to further neighbours exist @xcite , but we only use the version up to second neighbours in this paper . the qnl method is conservative with energy functional @xmath98 the linearized qnl operator is the hessian of @xmath99 at @xmath89 , that is , @xmath100 . the operator @xmath101 , is most easily written in variational form ( * ? ? ? 3.3 ) , @xmath102 based on this representation one can show that , if @xmath78 and @xmath103 then @xmath104 is positive definite on @xmath9 ( see ( * ? ? ? 3 ) for the case when @xmath105 ; the case @xmath106 is trivial ) . in @xcite the invertibility of the qcf operator was investigated analytically and numerically , and several interesting numerical observations were left as conjectures . similar observations were also used in @xcite to study the performance of iterative solution methods for the qcf operator . in the present section we rigorously establish some of the most fundamental of these conjectures in the next - nearest neighbour case . we will then extend the results , to finite range interactions in section [ sec : fr ] . in @xcite it was observed in numerical experiments that the spectra of @xmath107 and @xmath104 coincide . in this section we provide a rigorous proof by explicitly constructing a similarity transformation between @xmath107 and @xmath104 . the main ideas , after which the proof will be straightforward , are the following two new representations of the @xmath83 and @xmath104 operators . let @xmath97 , then @xmath83 and @xmath104 have , respectively , the representations @xmath108 we begin by noting that , for @xmath97 , the operators @xmath71 and @xmath72 may be written as @xmath109 = w_f '' l - \phi_{2f } '' l^2 , \\ \lc = ~ & \phi_f '' l + \phi_{2f } '' [ 4 l ] = w_f '' l.\end{aligned}\ ] ] using these formulas , the operator @xmath83 ( as defined in ) can be written in terms of the atomistic and the continuum operators as @xmath110 \lc = \chi \left [ w_f '' l - \phi_{2f } '' l^2 \right ] + [ i-\chi ] \left [ w_f '' l \right].\ ] ] from this we immediately obtain . to rewrite the qnl operator , we note that we can write as @xmath111 for all @xmath112 , and we therefore obtain . based on and we will deduce the similarity of the qcf and qnl operators . since @xmath32 is not invertible , we introduce the nonsingular operator @xmath113 defined by . the following result confirms conjecture 2 in @xcite ( a related conjecture with different boundary conditions is conjecture 6 in @xcite ) . [ thm : lqcf - lqnl - similarity ] if @xmath97 then the operators @xmath107 and @xmath104 are similar , with similarity transformation @xmath114 defined by : @xmath115 in particular , the spectra of @xmath107 and @xmath104 coincide . using formulas , , and , a straightforward computation yields the desired identity : @xmath116 \lmod = \lqnl \lmod . \qedhere\end{aligned}\ ] ] since @xmath104 is self - adjoint , there exists an orthonormal matrix @xmath117 , and a diagonal matrix @xmath118 containing the eigenvalues of @xmath104 , such that @xmath119 note in particular that @xmath118 also contains the zero eigenvalue . since the operators @xmath104 and @xmath107 are similar , there exists also an invertible operator @xmath120 such that @xmath121 as suggested by theorem [ thm : lqcf - lqnl - similarity ] , a possible choice for the eigenvectors is @xmath122 , since in that case we have @xmath123 however , @xmath124 , which is much worse than the numerical observations in @xcite and suggests a poor scaling of the eigenvectors . to produce a better eigenbasis , it is important to note that the choice of eigenvectors is not unique even after fixing the ordering as we are always free to rescale them . this turns out to be a crucial ingredient in our following construction of an eigenbasis with a uniformly bounded condition number . the following result is inspired by figure 4.2 in @xcite and conjecture 8 in @xcite . it does not precisely confirm these , but establishes a closely related and in fact stronger result . [ th : condell2 ] suppose that @xmath97 , then the operator @xmath125 , @xmath126 \vqnl,\ ] ] diagonalizes @xmath127 that is , @xmath128 , where @xmath118 is the diagonal matrix of eigenvalues associated with @xmath104 . moreover , if @xmath78 and @xmath129 then @xmath130 is bounded above by a constant that depends on @xmath131 , but is independent of @xmath30 and @xmath84 . [ rem : rescale_evecs ] the choice of @xmath132 is motivated by the following calculation . starting with the similarity result of theorem [ thm : lqcf - lqnl - similarity ] , we derive @xmath133 and we scale the eigenvectors by @xmath134 which gives @xmath135^t \vqnl = \lmod^{-1 } \lqnl \vqnl \\ = ~ & \lmod^{-1 } [ w_f '' l - \phi_{2f } '' l \chi l ] \vqnl = [ w_f '' \pu - \phi_{2f } '' \pu \chi l ] \vqnl.\end{aligned}\ ] ] this is equivalent to the choice of @xmath132 in when restricted to @xmath136 _ step 1 : diagonalization . _ in a straightforward computation we obtain @xmath137 \ , \big [ w_f '' i - \phi_{2f } '' \pu \chi l \big ] \vqnl \\ = ~ & \big[w_f '' i - \phi_{2f } '' \pu \chi l \big ] \ , \big [ w_f '' l - \phi_{2f } '' l \chi l \big ] \vqnl \\ = ~ & \big[w_f '' i - \phi_{2f } '' \pu \chi l \big ] \ , \vqnl \lambda \\ = ~ & \vqcf \lambda.\end{aligned}\ ] ] _ step 2 : estimating @xmath130 . we now assume that @xmath78 and @xmath138 to estimate @xmath130 we can ignore the positive constant multiple @xmath75 as well as the orthonormal matrix @xmath139 , that is , we have @xmath140 with constant @xmath141 , and the convention @xmath142 if @xmath143 is not invertible . we note that the condition @xmath144 implies that @xmath145 . elementary estimates give the upper bound @xmath146 we similarly get the lower bound @xmath147 which gives an estimate for @xmath148 whenever @xmath149 in the following we prove a bound for @xmath148 that holds whenever @xmath150 that is , up to the sharp stability limit , which is a more involved result . to estimate @xmath148 we use the facts that ( i ) @xmath151 , and ( ii ) if @xmath152 for some constant @xmath153 , then @xmath154 is invertible and @xmath155 . in lemma [ th : condell2_lowerbd ] we establish precisely this fact , assuming that @xmath145 , with a constant @xmath156 that depends only on @xmath157 but not on @xmath30 or @xmath84 . a generalization of the following technical lemma used in the previous proof will also be required in the finite range interaction case . it follows from lemma [ th : fr : condell2_lowerbd ] by choosing @xmath158 . [ th : condell2_lowerbd ] let @xmath145 , then there exists a constant @xmath153 , which depends on @xmath157 but is independent of @xmath30 and of @xmath84 , such that @xmath159^t u \big\| \geq \gamma_0 \|u\| \qquad \forall u \in \r^{n}.\ ] ] from the proof of lemma [ th : fr : condell2_lowerbd ] we see that , in the case @xmath160 , the constant @xmath156 is explicitly given by @xmath161 where @xmath162 , and the resulting condition number estimate by @xmath163 if @xmath164 is moderate but @xmath165 , then @xmath166 . a brief calculation shows that in this limit @xmath167 behaves asymptotically like @xmath168 in this section we extend the results of section [ sec : n2 ] to the case of finite range interactions ( i.e. , with arbitrary finite @xmath169 ) . we begin by stating a simplified formulation of our main results . we note , however , that our actual results are more general than the following theorem . in particular , we can replace the assumption @xmath170 , @xmath171 , by a more general condition . this theorem will be proved in section [ sec : fr : nonpos ] . [ th : fr : mainthm_simple ] suppose that @xmath170 for @xmath172 and that @xmath173 . \(i ) there exists an operator @xmath174 , which diagonalizes @xmath107 , that is , @xmath175 where @xmath118 is a diagonal real matrix of eigenvalues @xmath176 . \(ii ) if @xmath78 then @xmath132 is invertible and @xmath130 is bounded above by a constant that depends on the coefficients @xmath74 , @xmath177 , but is independent of @xmath30 and @xmath84 . \(iii ) if @xmath78 and if the eigenvalues are ordered , then @xmath178 where @xmath179 and @xmath180 denote the ordered eigenvalues of , respectively , @xmath72 and @xmath71 . in particular , we have the bounds @xmath181 we recall from the definition of the finite range qcf operators , @xmath182 , \quad \text{and } \quad \lqcfp = \lc + \pu \chi [ \la - \lc],\ ] ] where @xmath71 and @xmath72 are , respectively , the atomistic and the continuum operators defined in and . in terms of the translation operator defined in , we can express @xmath71 and @xmath72 as @xmath183 , \quad \text{and } \quad \la = \sum_{r=1}^{r } \phi''_{rf } \big[-t^r + 2i - t^{-r } \big].\ ] ] to show that @xmath107 is diagonalizable , we aim to construct a matrix @xmath184 such that @xmath185 is symmetric . we first note that we can write the difference @xmath186 as a laurent polynomial of the translation operator @xmath40 : @xmath187 \\ = ~ & \sum_{r=2}^{r } \phi''_{rf } \big[(t^r - i)(t^{-r}-i ) - r^2 ( t - i ) ( t^{-1}-i)\big].\end{aligned}\ ] ] thus , if we define the laurent polynomial @xmath188 \quad \text{for } t \in \c \setminus \{0\},\ ] ] then we obtain @xmath189 . the crucial observation is the following : if we can factorize @xmath190 as @xmath191 then we can choose @xmath192 to obtain @xmath193 and we immediately obtain @xmath194 = \big[\lc + y \chi y^t\big ] y.\ ] ] here , we have used the fact that all operators that are polynomials in @xmath40 commute . a similar result holds also for @xmath107 , though this would require further preparation . if @xmath184 were invertible ( this will turn out to be false ) , then this would imply that @xmath83 is similar to a symmetric , hence , normal matrix , and is therefore diagonalizable . the desired polynomial factorization result is essentially the riesz fjer factorization lemma ( * ? ? 53 ) , which we state and prove in a slightly more general form . [ th : symmetric - product ] let @xmath190 be a laurent polynomial with real coefficients such that @xmath195 . then there exists a polynomial @xmath50 such that @xmath191 . if , in addition , @xmath196 for all @xmath197 in lemma [ th : symmetric - product ] , if @xmath198 on @xmath45 , then it can be factorized as @xmath199 , where @xmath50 has real coefficients . let @xmath200 be the leading term in @xmath190 with @xmath201 , then @xmath202 is a polynomial with @xmath203 . moreover , @xmath204 and @xmath205 share the same roots , which we collect into a set @xmath118 , so that @xmath206 where @xmath207 denotes the multiplicity of @xmath208 . next , we define the auxiliary polynomial @xmath209 since @xmath190 has real coefficients it follows that @xmath210 if and only if @xmath211 , and @xmath212 , and hence @xmath213 also has real coefficients . we also note that @xmath214 since @xmath195 , for each root @xmath208 with @xmath215 , @xmath216 is also a root , with the same multiplicity , and therefore we have @xmath217 where @xmath218 for some constants @xmath219 and @xmath220 . by construction , @xmath221 has real coefficients and @xmath222 . we will later assume that @xmath205 has no roots on @xmath45 except for @xmath223 , and for that case , the proof would be complete . to establish the lemma in its full generality , we now distinguish two cases . _ case 1 : @xmath224 on @xmath45 . _ for @xmath225 , we have @xmath226 , and hence , upon reordering the product in @xmath227 , @xmath228 > 0 \quad \text{for all } t \in \ts.\ ] ] thus , if @xmath229 on @xmath45 , then we also have @xmath230 on @xmath45 . this implies that the multiplicity @xmath207 of all roots @xmath231 is even . we can therefore define @xmath232 and argue similarly as above , to obtain that @xmath233 has real coefficients and that @xmath234 for some constants @xmath235 and @xmath236 , and @xmath237 . we can again deduce that @xmath238 and that @xmath239 on @xmath45 , which implies that @xmath240 and @xmath241 . thus , we obtain @xmath50 by defining @xmath242 _ case 2 : @xmath243 on @xmath45 . _ in this case the multiplicity of roots @xmath244 may be odd , and therefore we define @xmath245 thus , @xmath233 contains all roots of @xmath246 with positive imaginary part and @xmath247 contains all roots of @xmath246 with negative imaginary part . we are only left to find any roots at @xmath248 . since @xmath222 it follows that these roots must be even ( possibly @xmath249 ) , and hence we can define @xmath250 to obtain @xmath251 for some constants @xmath235 and @xmath252 . arguing as before we find that @xmath253 and @xmath240 , and hence we obtain the result if we define @xmath254 for any laurent polynomial @xmath190 we call a polynomial @xmath50 that satisfies @xmath191 a grf - factorization ( generalized riesz fjer factorization ) of @xmath190 . since the laurent polynomial @xmath190 defined in clearly satisfies the condition @xmath195 , we have now indeed established and . thus , if @xmath184 were invertible , then we could deduce that @xmath83 ( and similarly @xmath107 ) is diagonalizable . unfortunately , as follows from the next lemma , @xmath184 is always singular . [ th : fr : defn_p1 ] let @xmath50 be any grf - factorization of the laurent polynomial @xmath190 defined in , then there exists a polynomial @xmath213 such that @xmath255 . in particular , the operator @xmath192 can be rewritten as @xmath256 it can be immediately seen from the definition of @xmath190 that @xmath257 and , since @xmath258 , it follows that @xmath259 . next , dividing @xmath190 by @xmath260 we obtain @xmath261,\end{aligned}\ ] ] and hence @xmath262 thus , we see that @xmath263 is also a ( multiple ) root of @xmath264 , and we can conclude that @xmath265 for some polynomial @xmath213 . upon rewriting the representation @xmath255 as @xmath266\big[-t p_1(t)\big ] = \big[- t + 2 - t^{-1 } \big]\big[-t p_1(t)\big]\ ] ] we obtain @xmath267 . thus , as a consequence of lemma [ th : fr : defn_p1 ] , we see that @xmath184 is always singular . nevertheless , in order to construct a similarity transformation to diagonalize @xmath107 , we can make a similar modification as in the next - nearest neighbour case , and simply replace @xmath32 by its invertible variant @xmath114 defined in . of course this still leaves us to verify that @xmath268 is invertible , for which we will introduce conditions in sections [ sec : lr : cond_bounds ] and [ sec : fr : nonpos ] . [ th : fr : diag_lqcf_lqcfp ] let @xmath50 be any grf - factorization of @xmath190 , and let @xmath269 , as in lemma [ th : fr : defn_p1 ] , then @xmath270 \lqcfp = \big [ \lc + y x y^t \big ] [ \lmod y_1].\end{aligned}\ ] ] in particular , if @xmath268 is invertible then @xmath107 is diagonalizable . we use to represent @xmath186 , the fact that polynomials of @xmath40 and @xmath114 commute , and the fact that @xmath271 , to obtain @xmath270 \lqcfp = ~ & [ \lmod y_1 ] \lc + y_1 [ \lmod \pu ] \chi l y_1 y_1^t l \\ = ~ & \lc [ \lmod y_1 ] + y_1 l \chi l y_1^t \pu [ \lmod y_1 ] \\ = ~ & [ \lc + y_1 l \chi l y_1^t ] \ , [ \lmod y_1 ] . \qedhere\end{aligned}\ ] ] to investigate the invertibility of @xmath268 we will study the laurent polynomial @xmath272 ^ 2}\ ] ] on the unit circle @xmath45 . it will quickly become apparent that @xmath246 needs to be bounded away from zero on @xmath45 in order to obtain invertibility of @xmath268 and bounds on the inverse that are uniform in @xmath30 and @xmath84 . consequently , we focus on the case where @xmath190 does not change sign on @xmath45 ( note that @xmath190 is always real on @xmath45 ) . to show that this is a reasonable assumption we will , in section [ sec : fr : nonpos ] , study the case of non - positive coefficients @xmath170 for @xmath273 , which is of particular interest in applications as this condition is satisfied for most practical interaction potentials . we will show that one can obtain bounds of the following type : _ there exist positive constants @xmath274 such that _ @xmath275 clearly , since @xmath190 is real on @xmath45 , it is necessary that @xmath246 does not change sign on @xmath45 , that is , either @xmath276 or @xmath277 . if @xmath276 then the matrix @xmath268 defined in lemma [ th : fr : defn_p1 ] is real , however , if @xmath278 then it is imaginary . for the following analysis we prefer to construct a real coordinate transform . according to remark [ rem : fr : neg_b ] and lemma [ th : fr : defn_p1 ] , we can choose a polynomial @xmath255 with real coefficients such that @xmath279 where @xmath280 is the sign of @xmath246 on @xmath45 . thus , if we ( re-)define @xmath268 accordingly as @xmath281 then proposition [ th : fr : diag_lqcf_lqcfp ] implies that @xmath282 \lqcfp = \big[\lc + \sigma ( l y_1 ) \chi ( l y_1)^t\big ] [ \lmod y_1].\ ] ] next , we show that @xmath283 implies invertibility of @xmath268 , including explicit bounds . beforehand , however , we make a brief remark on the connection of the sign of @xmath190 and a relationship between @xmath71 and @xmath72 . [ rem : sign_b ] the assumption @xmath276 on @xmath45 , or , more generally , @xmath284 on @xmath45 , implies that @xmath285 this follows simply from the fact that @xmath71 and @xmath72 share the same eigenvectors , and that the spectrum of @xmath186 , which is contained in @xmath286 is non - negative . conversely , if @xmath198 on @xmath45 then , by the same argument , @xmath287 these observations are interesting in their own right , and will moreover lead to useful estimates on the eigenvalues of the qcf operator @xmath107 in section [ sec : fr : evals ] . [ th : fr : bounds_y1 ] let @xmath268 be defined as in and suppose that @xmath190 satisfies , then @xmath268 is invertible and we have the bounds @xmath288 let @xmath255 be a grf - factorization of @xmath289 with real coefficients , then @xmath213 is a grf - factorization of @xmath246 with real coefficients . hence , for @xmath225 , we obtain @xmath290 which implies that @xmath291 by assumption , @xmath292 . since @xmath40 is unitary , this implies that @xmath268 has an orthonormal eigenbasis and that the spectrum of @xmath268 is contained in @xmath293 . this immediately gives @xmath294 moreover , we have @xmath295 which gives the second bound . according to , and lemma [ th : fr : bounds_y1 ] , we have @xmath296 \lqcfp [ \lmod y_1]^{-1 } = \lc + \sigma [ ly_1 ] x [ l y_1]^t = : l^{\rm sym},\ ] ] that is , @xmath83 and @xmath297 are similar matrices . since @xmath297 is real and symmetric , there exists an orthogonal operator @xmath298 such that @xmath299^t,\ ] ] where @xmath118 is the diagonal matrix of eigenvalues of @xmath107 , and we obtain @xmath300 \lambda [ y_1^{-1 } \lmod^{-1 } \vsymm]^{-1},\ ] ] that is , the operator @xmath301 diagonalizes @xmath107 . as in the nearest neighbour case , the eigenbasis @xmath302 is poorly scaled and would lead to an @xmath303 condition number . however , the same argument as in remark [ rem : rescale_evecs ] shows how to rescale the basis to obtain the following theorem . [ th : fr : condbnd_abstract ] suppose that the laurent polynomial @xmath190 defined in satisfies , and let @xmath268 be defined by . then the operator @xmath120 , @xmath304 \vsymm\ ] ] diagonalizes @xmath107 , that is , @xmath128 , where @xmath118 is a real diagonal matrix of eigenvalues . moreover , if @xmath78 , and if @xmath305 , then @xmath132 is invertible and @xmath130 is bounded above by a constant that depends only on @xmath306 , and , in particular is independent of @xmath30 and @xmath84 . the proof of this result is very similar to the proof of theorem [ th : condell2 ] , and hence we shall be fairly brief . first , we note that , by lemma [ th : fr : bounds_y1 ] , the matrix @xmath268 can be chosen to be invertible , and hence @xmath132 is well - defined . moreover , we recall that @xmath268 is a polynomial in @xmath40 and hence commutes with all operators that commute with @xmath40 , such as other polynomials in @xmath40 , the modified negative laplace operator @xmath114 , and the projection operator @xmath56 . _ step 1 : diagonalization . _ as in the computation in the proof of theorem [ th : condell2 ] , we obtain @xmath307 \ , \big [ w_f '' y_1^{-1 } + \sigma \pu \chi y_1^t l \big ] \vsymm \\ = ~ & \big [ w_f '' y_1^{-1 } + \sigma \pu \chi y_1^t l \big ] \ , \big [ w_f '' l + \sigma l y_1 \pu \chi y_1^t l \big ] \vsymm \\ = ~ & \big [ w_f '' y_1^{-1 } + \sigma \pu \chi y_1^t l \big ] \ , \vsymm \lambda \\ = ~ & \vqcf \lambda.\end{aligned}\ ] ] _ step 2 : estimating @xmath130 . _ suppose now that @xmath308 . as before , estimating @xmath309 is straightforward . using lemma [ th : fr : bounds_y1 ] , we obtain @xmath310 where @xmath311 and @xmath312 are the constants from . to estimate @xmath132 from below , we will use again the fact that @xmath313^{-1 } \| = \|[\vqcf]^{-t}\| \leq 1/(w_f '' \tilde{\gamma}_0)$ ] where @xmath314 satisfies @xmath315^t u \| \geq \tilde{\gamma}_0 \|u\| \qquad \forall u \in \rper.\ ] ] writing out @xmath316 in full , we get @xmath317^t = ~ & w_f '' y_1^{-t } [ \vsymm]^t \big[i + { { \textstyle \frac{\sigma}{w_f '' } } } y_1^t l y_1 \chi \pu \big ] \\ = ~ & w_f '' [ \vsymm]^t y_1^{-t } \big[i - \alpha y_1^t l y_1 \chi \pu \big],\end{aligned}\ ] ] where @xmath318 . if @xmath319 , then we can use lemma [ th : fr : condell2_lowerbd ] below to deduce that @xmath320 v \big\| \geq \gamma_0 \| v \| \qquad \forall v \in \rper,\ ] ] where @xmath156 depends only on @xmath321 , but is independent of @xmath30 and @xmath84 . this implies with @xmath322 . combining and gives the stated result . [ th : fr : condell2_lowerbd ] let @xmath323 , where @xmath324 commutes with @xmath56 , and where @xmath325 satisfies @xmath326 then there exists a constant @xmath153 , depending only on @xmath327 , such that @xmath328 we decompose @xmath154 into @xmath329 + \pu\big [ i - \alpha z^t l z \chi \pu\big],\end{aligned}\ ] ] where we have used the fact that @xmath330 and that @xmath56 commutes with @xmath331 . since @xmath56 is an orthogonal projection we obtain , again using @xmath330 , @xmath332v\|^2 + \| [ i - \alpha z^t l z \chi]\pu v\|^2 . \end{split}\ ] ] we will show next that @xmath333 w \|^2 \geq ( 1-\epsilon ) \| w \|^2 \qquad \forall w \in \us,\ ] ] where @xmath334 is defined in and depends only on @xmath335 , but not on @xmath30 or @xmath84 . hence , combined with , gives the desired result @xmath336v\|^2 + \| \pu v\|^2 ) = \gamma_0 \| v \|^2,\ ] ] with @xmath337 . _ proof of . _ we begin by splitting the operator , using @xmath338 , into @xmath339 = ~ & \chi [ i-\alpha z^t l z \chi ] + [ i-\chi ] [ i-\alpha z^t l z\chi ] \\ = ~ & \chi[i - \alpha z^t l z]\chi + [ i-\chi][i - \alpha z^t l z\chi ] \\ = : ~ & s_1 + s_2.\end{aligned}\ ] ] since @xmath340 is an orthogonal projection , we have , for any @xmath341 , @xmath342 w \big\|^2 = \| s_1 w\|^2 + \|s_2 w\|^2.\ ] ] _ estimating @xmath343 . _ since @xmath344\chi$ ] is a symmetric operator , the following variational bound is found using the fact that @xmath345 we obtain @xmath346 \chi w , w \big\ > = ~ & \|\chi w \|^2 - \alpha \big\ < l z\chi w , z \chi w \big\ > \\ \geq~ & \min\big ( 1 , 1 - 4 \alpha \|z\|^2 \big ) \| \chi w \|,\end{aligned}\ ] ] where the last equality follows by distinguishing the cases @xmath347 and @xmath348 . in summary , if @xmath349 , then we have the @xmath350-independent bound @xmath351 ^ 2 \|x w\|^2 \qquad \forall w \in \r^{n}.\ ] ] _ estimating @xmath352 . _ due to the good estimate on @xmath343 we only need fairly rough estimates on the term @xmath353 . application of the cauchy schwartz inequality and a weighted cauchy inequality provides the estimate @xmath354 w \|^2 + ( 1-\epsilon^{-1 } ) \alpha^2 \| [ i-\chi ] [ z^t l z ] \chi w \|^2,\end{aligned}\ ] ] for any @xmath334 . using the fact that @xmath355 is an orthogonal projection , @xmath356 , we can further estimate @xmath357 w \|^2 + ( 1-\epsilon^{-1 } ) 16 \|z\|^4 \alpha^2 \|\chi w \|^2.\ ] ] _ combining the estimates . _ if we define @xmath358 , and insert and into , we obtain , for all @xmath359 , @xmath360 w \|^2 \geq~ & \big\ { \min(1 , ( 1-\tilde\alpha)^2 ) + ( 1-\epsilon^{-1})\tilde\alpha^2 \big\ } \| xw\|^2 + ( 1-\epsilon ) \|[i - x]w\|^2 \\ \geq~ & \min\big\ { \min(1 , ( 1-\tilde\alpha)^2 ) + ( 1-\epsilon^{-1})\tilde{\alpha}^2 , 1 - \epsilon \big\ } \| w \|^2,\end{aligned}\ ] ] for any @xmath334 . since @xmath361 it is clear that choosing @xmath362 sufficiently close to @xmath263 gives a positive lower bound . to optimize this constant with respect to @xmath362 , we have to choose @xmath362 to equalize the two competitors in the outer @xmath363 formula . the resulting choice is @xmath364 \tilde\alpha - \tilde{\alpha}^2 + \sqrt{2 ( \tilde{\alpha}^2 - \tilde{\alpha}^3 ) + \tilde{\alpha}^4 } , & 0 < \tilde{\alpha } < 1 , } \ ] ] which concludes the proof of . ( as a matter of interest , @xmath365 as @xmath366 , @xmath367 for @xmath368 , and @xmath365 as @xmath369 . ) using the similarity to a symmetric matrix that we have established in the previous section we can now give sharp bounds on the spectrum of @xmath107 . [ th : fr : evals ] suppose that holds , then @xmath107 has a real , ordered spectrum @xmath176 . if we denote the ordered eigenvalues of @xmath370 and @xmath72 , respectively , by @xmath371 and @xmath372 then @xmath373 we know from theorem [ th : fr : condbnd_abstract ] that @xmath107 is diagonalizable and that it is similar to the self - adjoint operator @xmath297 defined in , which has a real spectrum that is identical to the spectrum of @xmath107 . we will next show that , for all @xmath58 , @xmath374 from these inequalities , the min - max characterisation of eigenvalues ( * ? ? ? xiii.1 ) immediately gives the stated result . to prove we will take the following starting point : @xmath375 from here on , we treat the cases @xmath376 and @xmath377 separately . _ case 1 : @xmath376 . _ using and the fact that @xmath378 , we have @xmath379 u , u \ > = \ < \la u , u \>.\ ] ] for the lower bound we use and the fact that @xmath380 is non - negative to obtain @xmath381 _ case 2 : @xmath377 . _ the idea for @xmath377 is essentially that the roles of @xmath71 and @xmath72 are reversed . note that , now , @xmath382 . hence , for the upper bound , we get @xmath383 whereas , for the lower bound , we get @xmath384 u , u \ > = \ < \la u , u \>. \qedhere\ ] ] for many of the common interaction potentials , such as the lennard jones potential , @xmath385 , or the morse potential , @xmath386 , there exists a minimal strain @xmath387 such that , for all @xmath388 , we have @xmath389 in most cases , it is reasonable to assume that the macroscopic strain @xmath390 remains in this region , as it would require extreme compressive forces to violate it . hence , for the remainder of the section , we will assume that is satisfied . the following two lemmas reduce this case to the one studied earlier in this section . [ th : p - has - real - coeffs ] suppose that holds , then @xmath229 in @xmath45 . similarly as in the proof of lemma [ th : fr : defn_p1 ] we rewrite @xmath190 in the form @xmath391 \\ = ~ & ( t-1)(t^{-1}-1 ) \sum_{r=2}^r \phi_{rf } '' \big [ ( t^{r-1}+t^{r-2}+\ldots+1 ) ( t^{-r+1}+t^{-r+2}+\ldots+1 ) - r^2 \big ] . \label{eq : p - has - real - coeffs : intermediate_1}\end{aligned}\ ] ] it is easy to see that @xmath392 is non - negative on @xmath45 , and moreover , for @xmath225 , @xmath393 hence we obtain the stated result . [ th : fr : nonpos_implicit_bound ] suppose that holds and that @xmath394 . then is satisfied with constants @xmath311 and @xmath312 that are independent of @xmath30 and @xmath84 . the upper bound @xmath312 in obviously exists since @xmath246 is a continuous function on a compact set @xmath45 . to show the existence of the lower bound it is sufficient to show that @xmath395 suppose that @xmath396 at some point @xmath397 . then @xmath398 and hence at least one term in the definition of @xmath190 is non - positive . to be precise , there exists @xmath399 such that @xmath400 and @xmath401 \le 0\ ] ] it follows from and that this may only happen at @xmath402 . however , a striaghtforward computation shows that @xmath403 ^ 2 } = \frac{r^4 - r^2}{12 } > 0.\ ] ] hence no point @xmath397 such that @xmath396 exists . we are now in a position to complete the proof of theorem [ th : fr : mainthm_simple ] . item ( i ) is a special case of proposition [ th : fr : diag_lqcf_lqcfp ] , taking into account that , for non - positive coefficients , @xmath190 defined in is non - negative and @xmath268 can therefore be chosen to be real . item ( ii ) follows from theorem [ th : fr : condbnd_abstract ] and proposition [ th : fr : nonpos_implicit_bound ] . item ( iii ) is established in theorem [ th : fr : evals ] . we conclude this section with a result that gives the sharp bounds on @xmath311 , @xmath312 for the case of non - positive coefficients . [ th : cor_beta_nonpos ] let @xmath246 be defined by and suppose that holds ; then holds with constants @xmath404 the lower bound is attained at @xmath405 and the upper bound is attained at @xmath406 . the proof of this proposition is based on the following technical lemma . [ lem : sin - r - beta - estimate ] for @xmath407 , @xmath171 , let @xmath408 \to \r$ ] be defined as @xmath409 then @xmath410 _ proof of . _ first , notice that @xmath411 which proves that the left - hand side of is not larger than the right - hand side . to prove the that @xmath412 for all @xmath413 , transform @xmath414 = & \displaystyle \frac{1}{\sin^2\beta } \left(r^2\cos^2\beta - \frac{\sin^2 r\beta}{\sin^2\beta}\right ) + r^2 = \frac{1}{\sin^2\beta } \left(\frac{r^2 \sin^2 2\beta}{4\sin^2\beta } - \frac{\sin^2 r\beta}{\sin^2\beta}\right ) + r^2 \\[1em ] & \displaystyle\hfill = \frac{r^2}{\sin^4\beta } \left(\frac{\sin^2 2\beta}{4 } - \frac{\sin^2 r\beta}{r^2}\right ) + r^2 \end{array}\ ] ] and consider the three cases : @xmath415 , @xmath416 , and @xmath417 . _ case 1 . _ ( @xmath418 ) further transform the function @xmath419 in : @xmath420 the expression in the brackets is positive since @xmath421 is a decreasing function for @xmath422 . this proves @xmath423 . _ case 2 . _ ( @xmath424 ) in this case @xmath425 , hence @xmath426 _ case 3 . _ ( @xmath417 ) denote @xmath427 ( @xmath428 ) and consider the two different subcases : @xmath429 is even / odd . _ case 3.1 . _ ( @xmath429 is even ) in this case @xmath430 , @xmath431 , and @xmath432 . hence @xmath433 as expressed in takes the form @xmath434 _ case 3.2 . _ ( @xmath429 is odd ) in this case @xmath430 , @xmath435 , and @xmath436 . hence transforms into @xmath437 here we used the fact that @xmath438 ( @xmath428 ) . _ proof of . _ first compute @xmath439 to prove the inequality @xmath440 consider the two cases : @xmath441 and @xmath442 . _ case 1 . _ ( @xmath441 ) in this case follows directly from the following computation : @xmath443^{-1 } \ge~ & \frac{1}{3 } r^2 ( r^2 - 1 ) \left[\frac{r^2}{\sin^2{\beta}}\right]^{-1 } = \frac{1}{3 } ( r^2 - 1)\,\sin^2{\beta } \\ \ge~ & \frac{1}{3 } ( r^2 - 1)\,\sin^2 \frac{\pi}{r } \ge \frac{1}{3 } ( r^2 - 1)\ , \left(\frac{2}{r}\right)^2 = \frac{4 r^2 - 4}{3 r^2 } \ge \frac{4}{3 } - \frac{1}{3 } = 1.\end{aligned}\ ] ] _ case 2 . _ ( @xmath442 ) we need to prove @xmath444 for integer @xmath429 , but instead we prove that it is valid for all real values of @xmath445 . first , notice that the following calculation @xmath446 proves for @xmath447 . as is shown below , @xmath448 which concludes the proof of . _ proof of . _ we use the standard inequalities @xmath449\end{aligned}\ ] ] and obtain @xmath450 and hence @xmath451 to reduce the problem to the statement of lemma [ lem : sin - r - beta - estimate ] we make the substitution @xmath452 , @xmath453 , which gives @xmath454 ^ 2 } \\ = ~ & \sum\limits_{r=2}^r ( -\phi_{rf}'')\ , \frac{r^2 ( e^{2i\beta}-1 ) ( e^{-2i\beta}-1 ) - ( e^{2ir\beta}-1 ) ( e^{-2ir\beta}-1)}{\big[(e^{2i\beta}-1 ) ( e^{-2i\beta}-1)\big]^2 } \\ = ~ & \sum\limits_{r=2}^r ( -\phi_{rf}'')\ , \frac{r^2\ , 4 \sin^2{\beta } - 4 \sin^2{r\beta}}{16 \sin^4{\beta } } = \sum\limits_{r=2}^r ( -\phi_{rf}'')\ , \frac{f_r(\beta)}{4 } .\end{aligned}\ ] ] application of lemma [ lem : sin - r - beta - estimate ] gives proposition [ th : cor_beta_nonpos ] . in this final section we present two further interesting applications of our foregoing analysis . first , we discuss the gmres solution of a linearized qcf system . we rigorously establish some conjectures used in @xcite and briefly discuss their consequences . second , we prove a new stability result for the linearized qcf operator in a discrete sobolev norm , which we hope will become a useful tool for future analyses of the qcf method . we assume throughout this section that holds and that @xmath268 is defined by . moreover , we recall the definition of @xmath297 from . since the results are fairly straightforward corollaries from our analysis in section [ sec : fr ] we will derive them in a less formal manner . we consider the linearized qcf system @xmath455 where @xmath456 , which is to be solved for @xmath35 . if this system is solved using the gmres algorithm ( see @xcite for a general introduction and @xcite for a detailed discussion of using gmres for solving the qcf system ) , then standard estimates on gmres convergence @xcite , along with the analysis of the previous sections , show that the residual of the @xmath457-th iterate , @xmath458 , satisfies the bound @xmath459 where @xmath460 ( see also ( * ? ? ? * prop . the fraction @xmath461 is used instead of @xmath462 since we are solving the system in @xmath9 . this convergence is rather slow and hence two variants of preconditioned gmres ( p - gmres ) algorithms were suggested in @xcite . the preconditioner used was the negative laplacian @xmath32 . we will use the preconditioner @xmath114 instead of @xmath32 , however , this is purely for the sake of a consistent notation since @xmath463 ( note that @xmath464 and that @xmath465 is well - defined on @xmath9 ) . the first variant of p - gmres that was considered in @xcite was the standard left - precon- ditioned gmres algorithm where gmres is applied to the preconditioned system @xmath466 to obtain convergence rates , we require bounds on the eigenvalues and eigenbasis of @xmath467 ( see ( * ? ? ? 6.2 ) ) . the second variant considers again the left - preconditioned system but this time the residual is minimized in the norm induced by the operator @xmath114 . the convergence rates of the resulting method are governed by the spectrum and eigenbasis of the operator @xmath468 ( see ( * ? ? ? 6.3 ) ) . we consider @xmath468 first . using , and the fact that @xmath469 commutes with @xmath470 , we obtain @xmath471 [ \lmod^{-1/2 } \lqcfp \lmod^{-1/2 } ] [ \lmod y_1]^{-1 } = \lmod^{-1/2 } \lsymm \lmod^{-1/2 } = \wsymm \lamtil ( \wsymm)^t,\ ] ] where @xmath472 is the real diagonal matrix of eigenvalues and @xmath473 an orthonormal matrix of eigenvectors of @xmath474 . hence , we conclude that @xmath475 is diagonalizable with real spectrum @xmath472 : @xmath476 [ y_1^{-1 } \lmod^{-1 } \wsymm ] = [ y_1^{-1 } \lmod^{-1 } \wsymm ] \lamtil.\ ] ] multiplying the equation by @xmath469 , we obtain @xmath477 [ y_1 \lmod^{-3/2 } \wsymm ] = [ y_1^{-1 } \lmod^{-3/2 } \wsymm ] \lamtil,\ ] ] thus showing that also @xmath467 is diagonalizable with the same real spectrum @xmath472 . we note that this rigorously establishes a variant of ( * ? ? ? * conjecture 10 ) . using the fact that @xmath473 is orthogonal , and lemma [ th : fr : bounds_y1 ] to bound @xmath478 , we can obtain the following upper bounds on the condition number of the matrices of eigenvectors : @xmath479 this rigorously establishes ( * ? ? ? * conjectures 12 and 13 ) . since these bounds are not uniform in @xmath30 the question arises whether we can define a better scaling for the eigenvectors to improve them . note , however , that ( * ? ? ? 4.3 ) implies that @xmath480 , and hence no choice of eigenbasis can achieve an upper bound in that is uniform in @xmath30 . moreover , numerical experiments in ( * ? ? ? 3 , figs . 2 and 3 ) indicate that our bounds may be optimal . to establish convergence rates for the p - gmres solution of the qcf system , we still need to obtain bounds on the eigenvalues contained in @xmath472 . let @xmath481 denote the ordered eigenvalues of @xmath472 , and let @xmath482 and @xmath483 denote , respectively , the ordered eigenvalues of @xmath71 and @xmath72 . since @xmath484 , we know that @xmath485 in view of remark [ rem : sign_b ] , replacing @xmath486 by @xmath487 in the formulas , we obtain that @xmath488 u , u \big\ > \geq \big\ < [ \lmod^{-1/2 } \lc \lmod^{-1/2 } ] u , u \big\ > \qquad \forall u \in \us , \\ \text{or } & \qquad \big\ < [ \lmod^{-1/2 } \la \lmod^{-1/2 } ] u , u \big\ > \leq \big\ < [ \lmod^{-1/2 } \lc \lmod^{-1/2 } ] u , u \big\ > \qquad \forall u \in \us.\end{aligned}\ ] ] hence , we can repeat the proof of theorem [ th : fr : evals ] verbatim to show that @xmath489 at this point we need to make an assumption on the stability of the atomistic system . we assume that the macroscopic strain @xmath390 is chosen so that @xmath490 the upper bound can be obtained in a straightforward computation that gives a constant @xmath491 depending only on the coefficients @xmath74 , @xmath177 . the lower bound means that the homogeneous deformation @xmath492 lies in the region of stability of the atomistic energy ( see @xcite for a detailed discussion of this point , in particular , that @xmath493 is indeed independent of @xmath30 ) . for example , in the case of non - positive coefficients remark [ rem : sign_b ] shows that this bound holds with @xmath494 , and that in the case @xmath97 one can choose @xmath495 . upon noting that the stability assumption is equivalent to the statement that @xmath496 and @xmath497 , we can now deduce that @xmath498 which are bounds that are independent of @xmath30 and @xmath84 . from the foregoing discussion we obtain the following convergence rates for the p - gmres solution of ( see ( * ? ? ? 6.2 and sec . 6.3 ) for details of these derivations ) : for the standard left - preconditioned gmres algorithm we obtain bounds on the preconditioned residual , @xmath499 where @xmath500 and @xmath501 are independent of @xmath30 and @xmath84 . for the left - preconditioned p - gmres algorithm , which minimizes the preconditioned residual in the norm induced by @xmath114 , we obtain @xmath502 where @xmath503 and @xmath501 are independent of @xmath30 and @xmath84 . we also note that a finer analysis ( see ( * ? ? ? 6.2 and 6.3 ) ) shows that both variants of p - gmres reduce the residual to zero in at most @xmath504 iterations . we define the discrete sobolev - type norm @xmath506 which is a norm on the space @xmath9 of mean - zero functions and denote the space @xmath9 equipped with @xmath507 by @xmath505 . moreover , we denote the space @xmath9 equipped with the norm @xmath508 by @xmath509 . we are interested in the question whether @xmath510 is stable , uniformly in @xmath30 and @xmath84 . to begin with , we note that @xmath511 thus , the question reduces to the analysis of the operator @xmath512 . using the representation @xmath513 where @xmath268 is defined in , we obtain @xmath514 we now argue similar as in the proof of theorem [ th : fr : bounds_y1 ] . on the space @xmath9 the operator @xmath515 can be replaced by @xmath516 for this modified operator lemma [ th : fr : condell2_lowerbd ] shows that it is invertible and provides uniform bounds on the inverse . restricting the argument back to @xmath9 we obtain the following theorem . [ th : prec : u22stab ] suppose that @xmath78 , that holds , and that @xmath517 ; then @xmath107 is invertible and @xmath518 is bounded above by a constant that depends only on @xmath306 ( that is , on the coefficients @xmath519 ) but is independent of @xmath30 and of the choice of @xmath84 . we have established a comprehensive @xmath0-theory of a linearized force - based quasicontinuum ( qcf ) operator making several conjectures from previous work @xcite regarding its spectrum and eigenbasis rigorous ( at least up to a modification of the boundary conditions ) . we have given elementary derivations in the case of next - nearest neighbour interactions but have also provided proofs for arbitrary finite range interactions . finally , as an immediate corollary of our analysis we have also obtained a new stability estimate in the space @xmath505 . our results heavily use the fact that the nonlinear qcf operator is linearized at a homogeneous deformation and a question of immediate relevance is whether our results can be generalized , at least partially , to linearizations around non - uniform states . even in small neighbourhoods of homogeneous deformations it is unclear whether this can be done . s. kohlhoff and s. schmauder . a new method for coupled elastic - atomistic modelling . in v. vitek and d. j. srolovitz , editors , _ atomistic simulation of materials : beyond pair potentials _ , pages 411418 . plenum press , new york , 1989 .

we study theoretically dynamics in a josephson junction coupled to a mechanical resonator looking at the signatures of the resonance in d.c . electrical response of the junction . such a system can be realized experimentally as a suspended ultra - clean carbon nanotube brought in contact with two superconducting leads . a nearby gate electrode can be used to tune the junction parameters and to excite mechanical motion . we augment theoretical estimations with the values of setup parameters measured in the samples fabricated . we show that charging effects in the junction give rise to a mechanical force that depends on the superconducting phase difference . the force can excite the resonant mode provided the superconducting current in the junction has oscillating components with a frequency matching the resonant frequency of the mechanical resonator . we develop a model that encompasses the coupling of electrical and mechanical dynamics . we compute the mechanical response ( the effect of mechanical motion ) in the regime of phase bias and d.c . voltage bias . we thoroughly investigate the regime of combined a.c . and d.c . bias where shapiro steps are developed and reveal several distinct regimes characteristic for this effect . our results can be immediately applied in the context of experimental detection of the mechanical motion in realistic superconducting nano - mechanical devices .

1112.5807

nanoscale electromechanical systems ( nems ) convert small amplitude mechanical motion into measurable electrical currents @xcite . devices based on nems have found applications as sensitive detectors of mass @xcite , force @xcite and electrical charge @xcite . considerable research efforts have been dedicated to improving detection sensitivity by fabricating devices with higher resonance frequencies , lower damping rates ( high quality factors ) and larger coupling between electrical and mechanical degrees of freedom . the problem of detecting the quantum state of a macroscopic mechanical resonator gave rise to several measuring schemes , proposed @xcite as well as realized @xcite . continuous improvements in device fabrication in combination with new techniques for cooling mechanical motion @xcite have pushed the sensitivity threshold to the quantum limit @xcite . the use of superconducting devices , in particular , superconducting qubits to detect and control the mechanical motion is in focus of modern research @xcite . it gives rise to a growing interest in techniques of coupling nems to superconducting circuits . superconducting nano - devices frequently use coulomb blockade that makes their transport properties sensitive to the gate voltages . the same gate voltage can be used to excite the mechanical motion which is detected from the change of d.c . transport properties of the device @xcite . without superconductors , this scheme has been successfully realized for a metallic single - electron transistor @xcite and for a coulomb - blockaded quantum dot in an ultra - clean carbon nanotube ( cnt ) @xcite . the results revealed high resonance frequencies , reaching gigahertzs , and unprecedented quality factors of the order of @xmath0 . these devices can be made superconducting by connecting them to superconducting leads and providing sufficiently large coupling between the states of the lead and device . we have successfully realized josephson junctions based on the ultra - clean cnt . the supercurrent observed demonstrates a pronounced gate - voltage sensitivity that indicates a well - developed coulomb blockade @xcite . a very interesting proposal that combines josephson dynamics and mechanical resonator has been recently put forward by gothenburg collaboration @xcite . the authors consider an ideal ballistic cnt between two superconducting leads biased at voltage @xmath1 . owing to josephson relation , the current in the nanotube oscillates at frequency @xmath2 . the authors notice that in external magnetic field this gives rise to an oscillating lorentz force . if the frequency matches the frequency of the mechanical resonator , the force excites mechanical motion which rectifies the josephson current enabling the observation of the effect in d.c . electric response of the junction . one would observe a narrow current peak in @xmath3 characteristics of the device . the same mechanism is responsible for fiske steps @xcite : the difference is that in fiske experiments the resonance is electrical rather than mechanical . this provides us motivation for the present theoretical study where we address superconducting nems where a mechanical resonator is integrated with a superconducting circuit element , a josephson junction . the focus of this work is to describe the techniques of driving and detecting mechanical motion using superconducting current and to explore the coupled dynamics of the oscillator displacement and superconducting phase difference . while the study is general for any type of oscillating nanowire , for the purpose of illustration and for concrete estimations we refer to one of the more successful nems devices : a suspended metallic cnt resonator connected to the metallic leads . we have fabricated and studied such devices . studies of cnt josephson junctions have shown that their josephson energy can be modulated by the gate - induced charge on the cnt . in this paper , we have shown that this gives rise to a different and generally more important mechanism of mechanical driving than that considered in @xcite . we consider all possible non - linearities in coupled josephson - mechanical dynamics . we have shown that in our situation the most important one is the mechanical non - linearity . we provide detailed estimations of the displacement , force and electrical current scales involved . it is our conclusion that the mechanical response in our devices should be small modifying the current on the scale of @xmath4 of the critical current ( in this case , the critical current is estimated from the product of the junction conductance and superconducting energy gap and can exceed the experimentally measured switching current by two orders of magnitude ) . in the present paper , we have investigated in detail the mechanical response under conditions of phase bias and d.c . voltage bias . we dedicated special attention to the dynamics in the presence of external a.c . drive , in the regime where josephson junction exhibits well - developed shapiro steps @xcite . one of the motivations of the research on shapiro steps was the better synchronization conditions in comparison with d.c . voltage bias . the reason for this is that the big quality factor @xmath5 of the nanomechanical resonance results in a narrow fiske - like current peak . its width in voltage can be estimated as @xmath6 . this imposes a severe limitation on voltage noise @xmath7 : to resolve the peak one must achieve @xmath8 . this may be challenging under realistic experimental circumstances . there is a way out : the voltage can be synchronized with the frequency of external irradiation . this effectively reduces the noise . from the other hand , the external irradiation can excite the mechanical motion by itself . the resulting complicated dynamics should be augmented with non - linear effects . in this paper we present our theoretical results concerning the mechanical response manifested in d.c . electric response of the junction , in particular , the modification of the width and position of shapiro steps , and analyze a variety of distinct regimes that differ in relative and absolute magnitudes of driving forces and resonant conditions , and are manifested in distinct lineshapes of the resonant response . our preliminary experimental results show corresponding features . they will be presented elsewhere @xcite upon completion of detailed analysis and comparison with our theoretical findings . the paper is organized as follows . in section [ prop ] we describe the setup concentrating separately on the electrical and mechanical properties in the corresponding subsections and introduce the notations . in section [ fe ] we analyze in detail the mechanisms of coupling between mechanical and electrical degrees of freedom . we consider electrostatic energy , derive and explicate the concept of _ josephson mechanical force_. we present a rather involved analysis of competing non - linearities and conclude that for our devices the mechanical non - linearities dominate and the josephson force can be strong enough to induce the non - linear mechanical response . we give the workflow we use to compute the mechanical response . in the end of the section , we specify a set of concrete parameters based on experimentally measured values . in section [ sec : phasebias ] we address the phase bias conditions . we reveal that the _ phase - dependent shift _ of the resonant frequency can be quite noticeable and discuss lorentz - like and fano - like frequency dependences of the mechanical response . in section [ sec : dcvoltage ] we discuss the fiske - type mechanical response at josephson frequency matching the resonance frequency of the mechanical resonator , or an integer fraction of this frequency by higher harmonics , and give the estimations of the effect . we shortly discuss the parametric excitation . in section [ sec : shapiro - res ] we study the mechanical response at the shapiro steps in the regime where the a.c . driving frequency matches the resonant frequency and present the mechanisms and peculiarities of this response . in section [ sec : shapiro - non - res ] we consider the non - resonant driving that appears to efficiently excite the mechanical motion in the regime of shapiro steps . our concluding remarks are presented in section [ concl ] . the setup under consideration is sketched in figure [ cnt ] where we concentrate on a case where both josephson junction and mechanical resonator are realized with the same single cnt . even in this case , the coupling between mechanical and electrical degrees of freedom is relatively weak . this permits us to describe the electrical and mechanical aspects of the setup separately . we provide such description in this section , while in the next section we concentrate on the coupling . we consider a conducting link between two superconducting leads ( the cnt in fig . 1 ) . in general , the current flowing in this junction is a complicated non - linear and time - delayed function of superconducting phase difference between the leads . however we assume that in the relevant frequency range the current response of the junction is superconducting and instant . the junction is included into an external electric circuit and the voltage drop at the junction is related to the time derivative of josephson phase , @xmath9 . in general , a circuit that connects the leads can be described by a complex frequency - dependent impedance @xmath10 in series with a voltage source @xmath11 . we typically assume that @xmath12 by far exceeds the typical junction impedance at low frequencies while at frequency scale of josephson generation frequency @xmath13 , @xmath10 is negligible in comparison with the junction response . in this case , the junction is current - biased at low frequencies with @xmath14 and voltage - biased at josephson frequencies . while this scheme looks different from the traditional rsj model where the external impedance is connected in parallel and the junction is current - biased , it is equivalent to a generalized rsj upon transforming the impedance and the voltage source . for instance , the linear part of possible quasiparticle response of the junction can be incorporated into @xmath10 . in addition , the junction is affected by the gate electrode biased by voltage source @xmath15 . the bias and gate circuits are disconnected at zero frequency . at finite frequency , there is a cross - talk between the circuits which is difficult to eliminate or even characterize in realistic experimental circumstances . we account for that by correlating a.c . parts of the voltage sources @xmath16 . for instance , if the gate voltage consists of a d.c . part and a harmonic signal at frequency @xmath17 , @xmath18 , the bias voltage source should also oscillate at the same frequency , @xmath19 . the ratio of two a.c . amplitudes and their mutual phase shift @xmath20 is determined by details of the crosstalk . we will show below that the interference of these two a.c . signals may strongly affect the d.c . currents in the junction . the superconducting current is determined by the instant phase difference , @xmath21 . in this case , it is related to the josephson energy @xmath22 of the junction , @xmath23 it is essential for us that the josephson energy is not only a function of phase difference but also depends on the gate voltage through the charge @xmath24 induced in the resonator , @xmath25 . for a nanotube device , the origin of this charge sensitivity is ( weak ) coulomb blockade of electrons in the middle of the nanotube . the nanotube can be viewed as two junctions in series , those being formed at contact with metallic leads . if the conductance of the junctions is smaller or comparable with the conductance quantum @xmath26 coulomb interactions become important and set a quasiperiodic dependence of josephson energy on @xmath27 with a period @xmath28 . this corresponds to charge quantization in the middle of the nanotube . we routinely observe the quasi - periodic modulation of superconducting currents in fabricated nanotube devices . the modulation can be tuned by changing the gate voltage at scale of @xmath29 from values of the order of @xmath30 to several per cent . big modulation and well - developed coulomb blockade require big junction resistances , this strongly suppresses the superconducting current . it is therefore advantageous to have intermediate resistances @xmath31 . at @xmath32 we typically observe 30% modulation . the superconducting current is a periodic function of the phase @xmath33 and therefore can be expanded in harmonics as @xcite : @xmath34 if one neglects all harmonics except the first one , @xmath35 gives the critical current of the josephson junction . we will typically assume this , and will mention the effect of higher harmonics only if it is crucial . -direction by an electrostatic force produced by the gate voltage . the superconducting leads are parts of an electrical circuit characterized by an impedance @xmath12 . the setup can be biased by either voltage or current source . ] mechanical resonators can be realized in a variety of ways @xcite . in many cases the adequate description of the resonator can be achieved with a minimum model that accounts for excitations of a single resonator mode , neglecting coupling to any other modes . the minimum model is given by the following equation of motion for a displacement variable @xmath36 : @xmath37 here @xmath38 is the time - dependent driving force , @xmath39 is the effective mass corresponding to the mode , @xmath40 stands for the resonant frequency , @xmath41 is the damping rate , and @xmath42 is the parameter describing the leading cubic non - linearity @xcite . the cubic non - linearity provides the important restriction on the magnitude of the displacement at resonant frequency as a reaction on resonant force . we also keep the second - order non - linearity @xmath43 . although it is not important in analysis of the reaction at resonant force , it describes the shift of the resonant frequency due to constant force . our preferable realization of mechanical resonator is a suspended ultra - clean cnt @xcite that demonstrates best quality factors observed so far(@xmath44 ) . in this subsection we review the parameters of the minimal model for this realization . in the setup shown in figure [ cnt ] , the nanotube displacement from equilibrium position and the driving force are in the @xmath36-direction towards the gate , that is , perpendicular to the nanotube axis . the mechanical variable @xmath45 is the displacement of the mid - point of the nanotube . in the case of a cnt , the adequate model of mechanical properties involves a suspended thin cylindrical rod clamped at both ends where the nanotube touches the metal leads . the parameters are the rod length @xmath46 , the cylinder radius @xmath47 , and the tube cross section area @xmath48 . in our experiments , @xmath49 , @xmath50 , and @xmath51 for a single - wall nanotube , @xmath52 being the layer spacing in graphite . the relevant elastomechanical material constants , carbon young s modulus @xmath53 and graphite density @xmath54 are estimated as @xmath55 @xcite and @xmath56 . the bending modes of the rod and their complete dynamics are described by the euler - bernoulli equation of motion @xcite . we concentrate on the lowest frequency bending mode that has no nodes in the rod and therefore is easy to excite . the resonant frequency can be tuned by `` tightening '' the tube , that is , changing the elastic tension . this is achieved by applying a sufficiently big d.c . gate voltage @xmath57 . the resulting electrostatic force strives to elongate the nanotube , thus producing the tension . in such a way , the resonant frequency can be increased by a factor of three in comparison with that of the `` loose '' nanotube . for estimations , we concentrate on the case of untightened rod . in this case , the resonance frequency corresponding to the lowest cnt bending mode can be estimated in terms of the bending spring constant and the carbon mass density @xmath58 @xcite , where @xmath59 is the principal moment of inertia . in our devices of length @xmath60 @xmath61 , the frequency is @xmath62 ghz , similar to frequencies reported in @xcite . the effective force is evaluated using the eigenfunction of the mode @xmath63 , @xmath64 , @xmath65 being the force per unit length . for electrostatic forces , an ad - hoc assumption is that the force distribution is uniform , so the total force is @xmath66 . in this case , the effective mass is given by @xmath68 , @xmath69 kg . the cubic non - linearity originates from the tension produced by the nanotube displacement , the corresponding parameter can be estimated as @xmath70 , assuming uniform distribution of force along the length of the rod . the second - order non - linearity @xmath43 vanishes for untightened straight rod for symmetry reasons . it however appears if the rod is tightened such that its frequency change with respect to the untightened value @xmath40 is of the order of @xmath40 . in this case , the non - linearity is obtained as @xmath71 , @xmath72 being the equilibrium displacement induced by the tightening . if @xmath38 oscillates at frequency @xmath73 close to the resonant frequency , eq . ( [ yt ] ) can be solved in resonant approximation for the complex amplitude @xmath74 : @xmath75 with @xmath76 being the complex force amplitude . here we introduce the detuning @xmath77 implying that @xmath78 . we also introduce the duffing parameter @xmath79 corresponding to the amplitude - dependent frequency shift . in our experiments , we estimate @xmath80 . we will re - write eq . in dimensionless form introducing a critical amplitude @xmath81 , @xmath82 . at this amplitude scale , the response of the resonator becomes a two - valued function of detuning ( see fig . [ fig : real - imaginary ] ) for a cnt , it can be estimated as @xmath83 which corresponds for our experiments to @xmath84 . the driving force corresponding to @xmath81 is @xmath85 . we estimate it for a cnt @xmath86 n. the dimensionless form of eq . ( [ y ] ) is : @xmath87 here we have introduced a dimensionless complex response function @xmath88 . in the linear regime @xmath89 its dependence on @xmath76 can be neglected : @xmath90 . figure [ fig : real - imaginary ] shows the real and imaginary parts of @xmath74 as a function of detuning for three values of the driving force amplitude that correspond to quasi - linear , critical and bistable regimes . versus detuning . right panel : imaginary part of the complex displacement amplitude @xmath91 versus detuning . the three curves in each panel correspond to @xmath92 . ] in this section , we analyze the coupling between mechanical and electrical degrees of freedom , with emphasis on phase - dependent mechanical force . since we prove that this force emerges from charging effect , we will start with a detailed discussion of electrostatic energy in the setup , and express the forces and superconducting current in terms of the induced charge . by doing so , we assume that the typical time of charge equilibration is much shorter than the typical time scale @xmath93 of the mechanical motion . we compare the electrostatic phase - dependent force with lorentz force proposed in @xcite . we derive the coupled equations of motion governing the josephson and mechanical dynamics and identify the dominant source of non - linear behavior . the junctions connecting the middle of the nanotube to the leads have intermediate resistance so that the middle of the nanotube forms a coulomb island that is neither isolated from nor ideally connected to the leads . while this situation is difficult to quantify from a microscopic calculation , it can be completely analyzed at phenomenological level . the consideration below is just a case of elementary non - linear electrostatics and is similar to the discussion in @xcite . however , it contains some important and less obvious details so we choose to present it at a comprehensive level . to start with , let us assume that the capacitance to the gate is vanishingly small while @xmath15 is diverging such that the charge induced to the coulomb island by the gate @xmath94 tends to a constant limit . a part of the ground - state energy of the setup , @xmath95 does depend on @xmath27 . general coulomb - blockade considerations @xcite imply that this part is a ( quasi ) periodic function of @xmath27 with a period of @xmath28 . in the limit of full isolation , for instance , this energy is piecewise parabolic , @xmath96 , @xmath97 being an integer number of extra cooper pairs stored in the island . in general , it is a smooth function of @xmath27 and may depend on the superconducting phase difference @xmath98 and , in principle , on mechanical displacement @xmath36 . this energy results in a non - zero electrostatic potential difference between the island and the leads , @xmath99 . let us now turn to finite @xmath100 and therefore finite @xmath15 that is the potential difference between the leads and the gate electrode . since this is not the potential difference between the island and the gate anymore , the induced charge @xmath27 is not equal to @xmath101 . rather , it is determined from the voltage division in a nonlinear capacitance network , or , equivalently , from the minimization of the total electrostatic energy with respect to @xmath27 , @xmath102 the charge is then found from the condition of the minimum , @xmath103 there are two implicit dependences in this equation that distinguish it from pure electrostatics , and that we make explicit now . first of all , the electrostatic energy depends on the mechanical displacement of the nanotube . geometric considerations suggest that this dependence can be ascribed to @xmath100 : indeed , the modification of capacitance to the gate is linear in @xmath36 , @xmath104 while the modification of @xmath105 is expected to be @xmath106 . second , the electrostatic energy depends on the superconducting phase difference : indeed , the josephson energy is just the phase - dependent part of @xmath105 , @xmath107 . the electrostatic charge @xmath27 depends both on displacement @xmath36 and on superconducting phase @xmath98 . to single out these contributions , we assume that i. the voltage between the middle of the nanotube is smaller than the gate voltage , @xmath108 , this is fulfilled if the induced charge @xmath109 , i.e. , in any practical setup ; ii . the mechanical displacement is small in comparison with the distance to the gate , @xmath110 ( in our experiments @xmath111 ) . with this , we linearize eq . ( [ eq : for_charge ] ) with respect to the josephson energy and mechanical displacement to arrive at ( @xmath112 ) @xmath113 the first term is the common expression for the gate - induced charge while the second and the third are the corrections of interest . at the moment , we keep @xmath27 in the argument of @xmath22 , although @xmath114 . the point is that the josephson energy is sensitive to variations of @xmath27 of the order of @xmath115 , and ( @xmath116 ) can in principle be of this order . since @xmath117 , we may disregard the possible @xmath36 dependence of @xmath118 . the mechanical resonator is affected by the electrostatic force @xmath119 : @xmath120 it is convenient to distinguish three separate contributions to the total force : the static force , the gate driving force and the phase - dependent josephson force . the static force is produced by the d.c . gate voltage . its magnitude is given by @xmath121 , corresponding to the first dominating term in .the effect of the static force is to pull on the resonator , thereby tuning its resonance frequency@xcite . since it is stationary , it does not excite the oscillations . the a.c . gate driving force arises due to the a.c . modulation of the gate voltage and is given by @xmath122 . the phase - dependent josephson force , not discussed in previous literature , comes about the product of the first and third term in ( [ linset1 ] ) , @xmath123 in fact it is similar to the gate driving force , with @xmath124 replaced by the phase - dependent voltage arising in the capacitive network , @xmath125 . in contrast to the gate driving force , the time dependence of the josephson force is determined by the phase dynamics rather than the external modulation of the gate voltage . the scale of the josephson force is @xmath126 , where we assume @xmath127 , which is true for intermediate contact conductances @xmath128 . it can be compared to the scale of a.c . gate driving force , @xmath129 . for sufficiently low a.c . driving amplitude @xmath130 the josephson force dominates @xmath131 . the josephson force explained above arises from the combined effect of charge sensitivity of the josephson coupling and the capacitive coupling to the gate electrode . the alternative mechanism of generating a @xmath98-dependent force was recently proposed by g. sonne _ _ e__t al @xcite . this force is of lorentz type arising from the interaction of the @xmath98-dependent superconducting current with an external magnetic field @xmath132 applied in the perpendicular direction ( for the setup of figure [ cnt ] , along the @xmath133-axis . ) this mechanism does not require the presence of a gate . let us compare the lorentz force and the electrostatic josephson force . the lorentz force is in @xmath36-direction , that is , perpendicular to both @xmath132 and the supercurrent , @xmath134 . it is natural to express @xmath135 in terms of the electric field @xmath136 produced by the gate electrode . for estimates , we assume @xmath137 and @xmath138 . this yields @xmath139 where @xmath140 @xmath141 is the speed of light and @xmath142 is the fine structure constant . typical magnetic fields used in experiments are @xmath143 @xmath144 . they are limited from above by the critical fields of the superconducting leads . the typical electric fields are @xmath145 @xmath146 . this corresponds to a potential drop of @xmath147 @xmath1 over a distance of @xmath148 @xmath61 . for these values @xmath149 suggesting that the josephson force dominates . therefore in the rest of the paper we will disregard the lorentz force . if one imagines a ballistic nanotube , the josephson coupling is not affected by the induced charge . in this case the lorentz force would be the only superconducting phase - dependent driving mechanism . however , the ideally ballistic nanotubes have not been realized experimentally . let us bring together three coupled equations governing the dynamics of the setup @xmath150 the first equation describes the dynamics of superconducting phase difference @xmath151 and is obtained by applying kirchhoff s laws to the circuit . the second equation is for the mechanical displacement @xmath45 where we substitute the electrostatic force ( [ eq : for_force ] ) . the induced charge @xmath27 enters both equations , and at the same time is defined by the third equation , that is , its value depends both on @xmath98 and @xmath36 . therefore the equations are coupled . we wish to simplify these equations under experimentally relevant assumptions . for this , we shall analyze the relative importance of different non - linearities in the coupling . there are natural non - linearity scales for all three variables , @xmath152 , @xmath153 , @xmath154 . this could change if the coupling is sufficiently strong . for instance , the displacement may cause the variation of phase that is subject to josephson non - linearity . the resulting variation of phase would produce the non - linear variation of @xmath27 , this will result in non - linear feedback on @xmath36 . this could in principle cause a non - linear scale of @xmath36 to be smaller than @xmath81 . so first of all we shall quantify the coupling between electrical and mechanical variables by comparing the non - linear terms in the mechanical force resulting from the coupling with those coming from the intrinsic non - linearities characterized by @xmath43 and @xmath42 . the conclusion of this subsection is that the mechanical non - linearity is the dominant one . we prove this with a rather involved reasoning given below . for the estimations , it is convenient to introduce the following dimensionless parameters : @xmath155 for estimations , we assume that @xmath156 are either small or of the order of @xmath30 . this assumption is valid for @xmath157 ; it compares the josephson energy to the charging energy under conditions of well - developed coulomb blockade . the parameter @xmath158 is a coefficient of josephson feedback at high frequency and depends on the details of the external circuit via the impedance @xmath12 . unless a special effort is made to increase the circuit impedance at high frequency , @xmath158 will not be big . given a variation of displacement @xmath159 we estimate the linear responses of the charge @xmath160 and the superconducting phase @xmath161 on @xmath159 using eqs . ( [ set1 ] ) and ( [ set2 ] ) as @xmath162 assuming @xmath163 the linear responses can be estimated as @xmath164 we use this to find a scale of @xmath165 for which the responses of charge and superconducting phase may become comparable with the scales of their intrinsic non - linearities @xmath166 and @xmath167 . we conclude that @xmath168 . comparison with the scale @xmath169 of the mechanical non - linearity yields @xmath170 two last factors in this expression are big , while the first one can be small . we estimate the biggest @xmath171 from the condition that @xmath40 is changed significantly by applying the gate voltage , that is , the stationary displacement @xmath172 . this yields @xmath173 @xmath174 being atomic scale , @xmath175 for our devices . the fact that the intrinsic scale of the non - linearity is smaller signals that the intrinsic non - linearity of the resonator dominates the non - linearity arising due to coupling . with this , we estimate the first two factors as @xmath176 . this is @xmath177 for our geometries and we conclude that @xmath178 for any @xmath179 . this proves that the dynamics of charge and phase is linear in @xmath36 provided our estimations of mechanical non - linearities @xmath180 hold . we still need to show that the coupling to josephson junction does not change these nonlinearities significantly . so we estimate the quadratic and cubic non - linearities of the mechanical force due to coupling . first we find the quadratic and cubic variations of charge with respect to displacement using eq . ( [ set1 ] ) . @xmath181 we can now estimate the terms in the mechanical force that are quadratic and cubic in @xmath159 . @xmath182 we compare these terms with the intrinsic non - linearities . the second order term @xmath183 needs to be compared with the mechanical quadratic non - linearity @xmath184 . assuming the static displacement of the order of cnt radius , @xmath185 , we estimate @xmath186 . @xmath187 here we use the estimation @xmath188 , @xmath189 for typical cnt geometries . ( see eq . ) . the third order term @xmath190 needs be compared with the third - order non - linearity @xmath191 . this yields @xmath192 to summarize , we proved that the non - linear scales correspond to @xmath152 , @xmath153 , @xmath154 and that for a cnt resonator the intrinsic mechanical non - linearities dominate the non - linearities arising from coupling . this permits a simplification of the dynamical equations . we may linearize the terms describing the coupling of mechanical displacement and electricity , thus separating josephson and mechanical non - linearities . this sets the following workflow . * at given a.c . and bias and gate voltages we solve for josephson dynamics neglecting the mechanical coupling and setting @xmath193 . we find @xmath194 and @xmath151 . using these , we compute the josephson force @xmath135 given by eq . . * we solve the non - linear mechanical equation @xmath195 to find @xmath45 . we are mostly interested in a part that oscillates with frequency @xmath196 . this may be excited by both @xmath135 and @xmath197 . * in most cases , we are interested in a ( d.c . ) current response on the mechanical motion , the mechanical response . it arises due to direct modulation of charge by the mechanical displacement , @xmath198 in the first order in @xmath45 . we will mostly concentrate on the situation when the displacement oscillates at the resonant frequency while a d.c . component of @xmath199 is of interest . the d.c . signal them arises from the rectification of @xmath45 by an oscillating part of @xmath200 , that we call the detecting current . * if we can neglect the feedback in josephson dynamics , we are done , since the response is given directly by @xmath199 . otherwise , we linearize the josephson dynamics to determine the response of the superconducting phase on the mechanical displacement found , @xmath201 @xmath202 here the kernel @xmath203 represents the combined linear impedance of the external circuit and the junction . * taking this into account , we obtain the current response sought : @xmath204 + the first term is the direct modulation of the current by the charge induced by the mechanical displacement while the second one is a feedback of josephson junction by means of @xmath205 . the mechanical response is thus typically a small correction to the maximum superconducting current . we can estimate it at maximum taking the @xmath206 ( we remind that @xmath207 , @xmath208 for our devices ) . perhaps unexpectedly , the typical response becomes smaller upon increasing the quality factor . the reason for this is clear : the maximum displacement becomes smaller . however , the large @xmath5 results in sharp frequency dependence of the response making it easier to identify . we thus concentrate on this dependence . let us specify the values of parameters employed . we choose these values such that they closely match those of a typical experimental realization of our setup . @xmath209 let us start our considerations with the junction biased with a time - independent phase @xmath98 : such bias condition can be achieved by embedding the junction into a superconducting loop . unfortunately , our present experimental setup does not allow measurements under these bias conditions . we present the theoretical results in hope that they will be useful for future experiments . the simplest experimental signature of josephson force under phase bias conditions is the _ phase - dependent shift _ of the resonant frequency . the mechanism of this shift in our situation is the mechanical non - linearity : the static josephson force tightens or looses the nanotube resulting in the frequency change . the frequency response on the static force in our model reads @xmath210 so the phase - dependent frequency shift reads @xmath211 the shift is clearly observable provided it exceeds the broadening @xmath212 . the estimation gives @xmath213 for the parameter set ( [ eq : parameter - set ] ) , maximum value of the shift @xmath214 this is , the shift is clearly observable . let us consider an example of a mechanically - induced response under conditions of the phase bias . to excite mechanical oscillations , we apply an additional a.c . voltage to the gate that oscillates at the frequency @xmath17 close to the resonant frequency @xmath40 . @xmath215 assuming @xmath124 to be sufficiently small to provide a linear response of the displacement , we obtain the following expression for the resonant part of the displacement : @xmath216 here , @xmath217 is the detuning that includes the phase dependent shift of the resonance frequency discussed above . owing to the mechanical non - linearity , the oscillating displacement produces a stationary displacement @xmath218 . this induces a stationary charge that affects the d.c . superconducting current at constant phase bias . rather remarkably , this effect is related to the phase - dependent shift discussed above . indeed , both are proportional to charge - dependent part of the josephson energy and to the non - linearity coefficient @xmath43 . the resulting current response reads @xmath219 the contribution to the current being of the order of @xmath220 if we measure quantum fluctuations of displacement @xmath221 . the dependence on frequency of the a.c . modulation is a lorentzian one , as it is frequently expected ( fig . [ fig : shift ] ) , the lorentzian center being shifted with changing the phase . as a function of frequency detuning for a set of phase bias values : from the lowermost to the uppermost curve the phase changes from @xmath222 to @xmath223 , with interval @xmath224 . the curves are offset for clarity . dashed lines give the positions of zero . [ fig : shift ] ] the above mechanism of response exploits the dominating mechanical non - linearity . it is proportional to @xmath225 . upon increase of the a.c . amplitude @xmath124 the oscillating displacement saturates owing to the non - linearities . in this case , the dominating d.c . current signal can arise from the electrical non - linearity as a result of a mixing of the oscillating displacement @xmath74 and the oscillating charge @xmath226 . the resulting current is thus proportional to @xmath227 and may exceed the contribution @xmath228 provided the latter saturates . the expression for this contribution reads @xmath229 interestingly , it exemplifies a fano - type dependence on the detuning that is quite different from a lorentzian . in the linear regime , @xmath230 so that the signal changes sign at the resonance point . fig . [ fig : asymmetric ] illustrates the fano - type dependence in the non - linear regime . comparing expressions ( [ eq : iem - phasebias ] ) and ( [ eq : im - phasebias ] ) we conclude that the fano - shaped @xmath231 dominates provided @xmath232 , this is , deep in non - linear regime . let us turn to d.c . voltage bias . in this case , the superconducting phase is in first approximation a linear function of time , @xmath233 , @xmath2 being the josephson frequency which corresponds to the voltage @xmath1 across the junction . in the same approximation , the current is a purely oscillatory function of time . the time - dependent current can be expanded into harmonics of the josephson frequency , @xmath234 a d.c . current emerges from the feedback on josephson dynamics : oscillatory currents produce oscillatory corrections to the phase proportional to the impedances at frequencies @xmath235 . taking this into account in the first approximation in @xmath236 , we arrive at @xmath237 the above relation holds provided @xmath238 . non - perturbative treatment of josephson dynamics is required otherwise . in the upper panel and @xmath239 in the lower panel . the current is in units of @xmath240 that amounts to @xmath241 for the parameter set chosen . [ fig : asymmetric ] ] let us consider the mechanical effects . it is important to note that under the d.c . the josephson force also oscillates in time , @xmath242 @xmath243 being the harmonics of josephson energy . therefore , the force can efficiently excite the mechanical resonator provided @xmath244 . let us first concentrate on the case where the resonance frequency is matched by the first harmonics , @xmath245 . the detuning is defined as @xmath246 . to start with , let us assume that the josephson force is sufficiently weak so that the mechanical response is linear and thus given by eq . . the direct mechanical contribution to the d.c . josephson current is obtained by averaging eq . [ eq : workflow - tilde - im ] , and reads @xmath247 this can be cast to the form similar to ( [ eq : current - impedance ] ) , @xmath248 where the current is replaced with detecting current @xmath249 and the `` mechanical impedance '' @xmath250 is defined as @xmath251 ( here , @xmath252 ) . this form of the presentation of the mechanical response makes evident an analogy with fiske steps @xcite that are observed at voltages corresponding to resonant frequencies of an electrical impedance . this may be either an impedance of external circuit or an effective impedance that is essentially contributed josephson inductance . to comprehend the scale of the response , we note first that for a sufficiently well - developed coulomb blockade @xmath253 . therefore , to compare the current([eq : current - impedance ] ) and the mechanical response , we need to compare @xmath254 and a typical environmental impedance . the latter can be estimated as the impedance of free space @xmath255 . the typical mechanical impedance far from the resonance , @xmath256 , should be much smaller than that . indeed , if we substitute the parameter set ( [ eq : parameter - set ] ) into eq . [ eq : z0 m ] we end up with @xmath257 . however , the `` mechanical '' impedance is enhanced by a factor of @xmath5 at the resonant frequency . with this , @xmath258 and the current peak produced by the mechanical response should exceed the background current given and be clearly observable . the voltage dependence of d.c . current response in linear regime is determined by @xmath259 and thus takes a lorentzian shape with the half - width @xmath260 . this assumes a noiseless voltage source . it is known @xcite that the voltage noise suppresses the coherence of josephson generation . for white noise spectrum of intensity @xmath7 , the resulting line - width reads @xmath261 . comparing the two , we conclude that the mechanical response will be broadened by the noise and essentially suppressed provided @xmath262 , this is , @xmath263 . this gives a condition on the possibility of detection of the mechanical response that may be challenging to meet in practical circumstances . . the crosses indicate the values of @xmath27 that correspond to the values of josephson force used in other panels . top right panel : the mechanical response as function of detuning for relatively low values of the force @xmath264 , at which the response increases with increasing the force . bottom panels : frequency dependence at the force values @xmath265 ( from the left to the right panel ) where the response decreases with increasing the force . [ fig : lorentzian ] ] as mentioned , the josephson force can be big enough to exceed @xmath266 , this makes it relevant to address the non - linear response as well . we illustrate the non - linear response in fig . [ fig : lorentzian ] . to produce the figure , we took the charge - dependent part of the josephson energy to be of the form @xmath267 . tuning @xmath27 with the d.c . gate voltage tunes the josephson force from @xmath268 to some maximum value . we chose @xmath53 such that the maximum force equals @xmath269 ( @xmath270 for the parameter set in use ) and compute the response using eq . and eq . at a set of the values of @xmath27 , or , equivalently , @xmath135 . the response is lorentzian at small forces , increases and develops a jump characteristic for bistability . it is interesting to note that the response slowly decreases upon increasing @xmath135 at @xmath271 . this is because the response is proportional to @xmath272 that quickly decreases at big driving forces . in this limit , @xmath273 . if we take higher harmonics of current - phase characteristic into account , we note that josephson force emerges at a set of frequencies that are integer multipliers of @xmath274 ( eq . ) . this implies that the resonant mechanical response can be also observed in the vicinities of a set of discrete voltage values satisfying @xmath275 , this is , at lower voltages than the resonance described above . the response is computed along the same lines with replacing @xmath276 by @xmath243 . in linear regime , the response reads @xmath277 ( cf . [ eq : current - mech - impedance ] , the factor @xmath278 in the present expression is canceled by lower voltage @xmath279 ) the response scales with the relative values of the harmonics and is in principle of the same order of magnitude for several low harmonics . its dependence on voltage in the vicinity of the resonance is similar to that discussed above and does not have to be illustrated separately . for the sake of completeness , let us mention the possibility of the resonant mechanical response at _ higher _ voltages by means of parametric excitation @xcite . generally , parametric resonance in a non - linear oscillator is achieved by applying an a.c . driving force with frequency about a double of the resonant frequency , @xmath280 @xcite . in our case , this is achieved by applying a d.c . bias voltage with @xmath281 , so that the josephson force oscillates at @xmath282 and integer multiples of this frequency and thus provides the parametric driving required . the response of at resonant frequency emerges provided the parametric driving force exceeds a certain threshold value , and , as in case of direct resonance , achieves values @xmath283 . the point is that this threshold driving force is parametrically bigger than @xmath266 , @xmath284 . for our devices at @xmath285 , the parametric excitation requires @xmath286 that by a factor of 30 exceed the value from the parameter set and are not practical . this is why we do not explore the regime of parametric excitation in detail . besides , the manifestation of the oscillating amplitude is not as straightforward as in the case of direct resonance . the contribution of displacement at @xmath40 to the mechanical current response ( [ eq : workflow - tilde - im ] ) oscillates at the same frequency and is not readily rectified to a d.c . current . under our assumptions , the d.c . mechanical response is dominated by the displacement oscillating at @xmath287 and is by a factor of @xmath288 smaller than the typical responses studied in this paper . from now on , we turn to the situation where the setup is a.c . driven at frequency @xmath17 . as discussed in section [ sec : ele - setup ] , in our setup this gives rise to two a.c . signals @xmath289 , @xmath290 . the effect of @xmath291 is a formation of shapiro steps @xcite . a common approach to shapiro steps takes into account only the first harmonics of the current - phase relation and starts with the assumption that the time - dependent superconducting phase difference can be presented as a sum of three terms @xmath292 here , the first term describes the a.c . driving ( @xmath293 , @xmath294 , the second term corresponds to a d.c . voltage @xmath295 , and the third term is a lock - in phase important for further consideration . with this , @xmath296 can be presented as a sum over harmonics @xmath297 with @xmath298 . here , @xmath299 denotes the @xmath300-th bessel function of the first kind . shapiro steps are formed at discrete values of d.c . voltage @xmath301 . in this case , the time - dependent current @xmath302 has a d.c . component @xmath303 simplest assumption is ideal current bias at zero frequency and ideal voltage bias at frequencies @xmath304 . in this case , the i - v curve of a.c . driven junction consists of a series of separate pieces . at each piece ( shapiro step ) the voltage is locked to one of the discrete values . the current within each piece may vary from minimum values @xmath305 to the maximum value @xmath306 provided the bias current fits this interval . in this case , the actual value of the lock - in phase @xmath307 is set by the bias current . the extremal values @xmath308 are achieved at the lock - in phases given by @xmath309 we in main follow this approach while admitting extreme simplifications it brings . the higher harmonics of current - phase relation and/or non - ideal voltage bias not only modify the relation between the current and lock - in phase : they also provide phase - locking at fractional ratios of @xmath310 @xcite and formally at all rational values of this ratio . these fractional shapiro steps are however more sensitive to noise than the integer ones and more likely to vanish . the i - v curves of our devices do show well - developed steps at integer values of @xmath274 and only traces of phase - locking at intermediate values . for this reason , we do not consider fractional shapiro steps in this paper and concentrate on integer ones where @xmath311 . it is advantageous to look at the mechanical response at shapiro steps rather than at d.c . bias conditions . the external a.c . driving synchronizes josephson oscillations . the inductive response present at shapiro steps also reduces significantly the voltage noise at low frequencies so that it does not broaden the resonant lines . in this section , we will consider the mechanical response in the simplest situation of resonant driving where the driving frequency matches the resonant frequency , @xmath312 . let us first concentrate on the first shapiro step , the one at voltage @xmath313 , that is the biggest in the limit of small driving voltages @xmath314 , and determine the d.c . part of the response at the oscillating displacement @xmath74 . to represent the results , we normalize @xmath74 to the non - linearity scale @xmath81 and introduce a convenient current scale @xmath315 for the values of our parameter set , @xmath316 making use of eqs . and , we express the mechanical response in terms of the amplitudes @xmath74 and @xmath317 at the resonant frequency , @xmath318 this displacement is a response on the force at resonant frequency which is a sum of josephson force and gate force . the time - dependent josephson force is expanded in harmonics in the form @xmath319 its amplitude at resonant frequency is contributed by the terms @xmath320 and reads @xmath321 let us discuss first the relative scale of the gate force in comparison with the josephson force . it may seem there is none , and varying the a.c . gate voltage @xmath124 one can achieve any ratio @xmath322 between the forces . however , we should take into account the fact that in our setups a.c . driving also induces an appreciable bias voltage @xmath291 . if the oscillating phase @xmath323 produced by this voltage becomes large shapiro steps can hardly be observed . it is in general reasonable to expect @xmath324 . in this case , @xmath325 corresponds to @xmath326 . the latter ratio is typically @xmath327 in our setups ( for our parameter set it is @xmath328 ) . this implies that typically we can disregard a.c . gate force in comparison to the josephson force . we will analyze this case first and consider the effect of the gate force in the end of the subsection . with this , the mechanical response is given by @xmath329 here , @xmath330 defined by eq . gives the non - linear mechanical response . the expression is naturally separated onto two terms . the first term is proportional to @xmath331 and therefore exhibits a lorentz - like dependence on frequency . it does not depend on the lock - in phase and can be regarded as a _ shift _ in the current . owing to the shift , the maximum and minimum currents @xmath332 at the step are no more opposite : the mechanical effect breaks the symmetry of the shapiro step . the shift is however opposite for opposite voltages . the origin of the shift may be traced to the fiske response ( eq . [ eq : current - mech - impedance ] ) formed at @xmath245 in the absence of the a.c . driving . indeed , in the limit of vanishing @xmath323 the first term in the mechanical response does not vanish : rather , it approaches the expression ( [ eq : current - mech - impedance ] ) . so it looks like the fiske response persists also for well - developed shapiro steps and contributes to the current at the step . this suggest perhaps the easiest way to observe and identify the mechanical response : measure maximal and minimal currents at a step as function of the a.c . frequency . in the rest of the paper we thus mainly concentrate on the modification of extremum currents . the second term in eq . can not however be observed in this way . in ideal current bias conditions at low frequency , the second term in the current response fact amounts to a shift of the lock - in phase . indeed , since the current at the step as function of @xmath307 reads as @xmath333 , the second term can be seen as a modification of the lock - in phase at constant bias current which does not depend on this current , @xmath334 this response is of fano - type . since such shift of the phase does not modify the values of the current extrema , the effect can not be observed in the course of two - terminal electrical measurement in our setup . the shift of the lock - in phase can be however revealed if the josephson junction under consideration is a part of a squid , or with the aid of lock - in measurement at non - resonant a.c . frequency . with respect to this , we ought to mention yet another effect of josephson force manifesting itself in the mechanical response considered . in fact , the situation at a shapiro step is similar to the phase bias conditions considered in section [ sec : phasebias ] , with lock - in phase playing the role of @xmath98 . we thus expect @xmath307-dependent shift of the resonance frequency . the static josephson force at a shapiro step is given by @xmath335 the frequency shift caused by this force thus reads @xmath336 here @xmath337 is a maximum frequency shift in the absence of the a.c . driving , given by eq . . the frequency shift vanishes at extremum points @xmath338 and therefore can not be observed by measuring the extrema of the current . and at the first shapiro step @xmath339 versus the oscillating phase @xmath323 . the first ( upper ) plot gives the maximum current at the step . the second and fourth plots give the mechanical response defined as the extremum of the modification of this maximum current over the frequencies near the resonance , at maximum josephson forces @xmath340 and @xmath341 , respectively . the actual amplitudes of the resonant josephson forces are given at the lower plots , respectively third and fifth . ] we illustrate the mechanical response in fig . [ fig : firststep - fj ] . in this figure as well in all subsequent figures except fig . [ fig : shapirowithgate ] , we concentrate on the modification of the maximum current on the step . instead of presenting the ( rather trivial lorentz - like ) frequency dependence of this modification , we give the extremum of this modification over the frequency range and plot it versus @xmath323 . the extremum is proportional to the maximum of @xmath331 over the frequency . we shall note that the dependence of this maximum on the force is rather specific one : it is a constant until the bistability threshold @xmath342 , has a cusp at this value of force , and decreases monotonously at higher forces . this accounts for rather strange appearance of the response curves . if the josephson force is smaller than the bistability threshold , they coincide with the linear response given by the dotted curves . otherwise , the response is smaller than linear one and exhibits kinks . the so - defined maximum response is plotted in fig . [ fig : firststep - fj ] for two values of @xmath343 , those correspond to slightly and strongly non - linear regime , respectively . in both cases , the response vanishes when the width of shapiro step reaches maximum , or becomes zero ( except @xmath344 ) . in slightly non - linear regime , the response reaches maximum value at @xmath307 . upon increasing @xmath323 , it exhibits bessel - like oscillations . the envelop of these oscillations shrinks with increasing @xmath323 . this shrinking is much faster than that for either step width or josephson force . in strongly non - linear regime , the amplitude of the response is determined by competition of two factors : it is increased by the bigger value of the detecting current , and decreased owing to smaller imaginary part of oscillating displacement at higher josephson forces . . in the plots from top to bottom the maximum josephson force @xmath343 assumes the values @xmath345 . we choose @xmath346 and @xmath347 . dashed lines in the plots for @xmath348 give values opposite to the corresponding @xmath349 , to stress the symmetry or asymmetry of the response . ] let us turn to the effect of the gate force . the josephson force of the kind considered can change sign and therefore be tuned to zero by tuning @xmath27 . in the vicinity of this particular @xmath27 , the gate force should compete with the josephson one and eventually dominate . the full amplitude of the force at the resonant frequency then reads @xmath350 the frequency shift @xmath20 between the bias and the gate voltage being a relevant parameter . the mechanical response is given by eq . where @xmath351 depends on the full force plus an addition proportional to @xmath197 , @xmath352 the last equation holds for @xmath353 . the expression for @xmath354 is obtained by interchanging @xmath355 and @xmath356 . evaluating this at the extremum points of lock - in phase , @xmath338 , we obtain @xmath357 therefore , the contribution of the gate force to extremum currents is not like a shifts : rather , it modifies the width of the step . these terms are even in voltage and display a mixture of fano - type and lorentzian - type response as function of frequency , this being tuned by the phase @xmath20 . to illustrate a rather complex interplay of josephson and we plot in fig . the frequency dependence of the mechanical response for a constant @xmath15 and a set of values of @xmath135 that pass zero . the plots show the modifications of extremum currents @xmath332 . these modifications are the same for the josephson force contribution and opposite for the gate force contribution . besides , the frequency dependence is fano - like for the gate force contribution and lorenz - like for the josephson force contribution . in the central plot , the josephson force contribution is absent , the modifications of @xmath332 are opposite , and the frequency dependence is fano - like . upon increasing the josephson force , these features are transformed into the opposite ones . the plots are symmetric upon simultaneous change of signs of the current and the josephson force . at the same conditions of the resonant driving , we analyze the mechanical response at other shapiro steps @xmath358 , those correspond to higher voltages @xmath359 . both the amplitudes of the detecting current and the josephson force display a complex dependence on the step number @xmath300 and the oscillating phase @xmath323 . they are given by @xmath360 where the dependence on the sign of the voltage is incorporated into @xmath361 . we consider only the situation when josephson force dominates . with this , we obtain a relation similar to eq . : @xmath362 as in the first step , the response consists of two terms . the first one gives a shift in the current , and gives a modification of the maximum and minimum currents at the step , this is to be measured . as in the previous case , the shift is odd in voltage . however , its @xmath323-dependence is quite rather distinct . the measuring of the mechanical response at higher steps is important to check the consistency of results and thereby unambiguously identify the mechanism of the response . the characteristic dependences on @xmath323 make the identification easy . we illustrate the response for higher steps in fig . [ fig : higher - steps - second ] ( second step ) and fig . [ fig : higher - steps - fifth ] ( fifth step ) . in both cases , the response correlates with the shapiro step width given in the upper plots : it vanishes when the width achieves a maximum or becomes zero . in distinction from the first step , the responses vanish at vanishing @xmath323 . their typical values are of the same order . however , the envelopes of the responses decrease rather slow with increasing @xmath323 . for @xmath363 and @xmath364 at the second shapiro step @xmath365 . ] at the fifth shapiro step @xmath366.,title="fig : " ] [ fig : higher - steps - fifth ] in the previous section , we concentrate on the case when the driving frequency @xmath17 matches the resonant frequency of the mechanical oscillator . it is not a necessary condition for an efficient excitation of the resonant mode . the josephson dynamics at shapiro steps are essentially non - linear . as a consequence , the spectrum of current oscillations contain all higher harmonics @xmath367 of the driving frequency @xmath17 . the same pertains the josephson force . therefore , the resonator can be efficiently excited for @xmath368 , @xmath369 being an integer number . at any given @xmath97 , the resonant conditions are achieved for any shapiro step number @xmath300 , and thus for voltages @xmath370 . these non - resonant driving conditions are advantageous for observation of the josephson force since the a.c . gate voltage force is not in the resonance , does not cause any appreciable displacement and therefore does not mask the effect of the josephson force . in this short section , we will thus concentrate on the case of the non - resonant driving @xmath371 . the amplitudes of the detecting current and the josephson force depended not only on the step number @xmath300 and the oscillating phase @xmath323 , but also on @xmath97 . they are given by @xmath372 where the dependence on the sign of the voltage is again incorporated into @xmath361 . since the gate force is absent , the response is given by a relation similar to eq . [ eq : firststep - jo ] that contains the josephson force only : @xmath373 it is again separated onto two terms discussed above , only the first term being responsible for the modification of the extremum currents of the shapiro steps . we illustrate the dependences on @xmath323 in fig . [ fig : offresonantresponse1 ] ( for @xmath374 ) and fig . [ fig : offresonantresponse2 ] ( for @xmath375 ) for the first three steps with @xmath376 . the vertical arrangement of the plots is the same as in the previous figures except we chose a single value of the maximum josephson force @xmath364 that brings us deep into the non - linear regime . in this regime , the response is of the same order of magnitude for all steps and ratios @xmath97 , while retaining unique @xmath377 specific dependence on @xmath98 that can be used for identification of the effect and the characterization of the josephson force . in conclusion , we have studied josephson junction dynamics affected by excitation of a mechanical resonator . we have demonstrated that the mechanical oscillations can be rectified giving rise to an additional d.c . current that can be used for detection . the mechanical response is proportional to the oscillation amplitude , and is estimated as @xmath378 . the resonator can be driven by the a.c . voltage applied to the gate electrode as well as an additional mechanical force , termed the josephson force , that depends on the superconducting phase difference at the junction . we estimate the josephson force as @xmath379 and show that it is sufficiently strong to drive the mechanical resonator into the non - linear regime . we also show that it is typically larger than magneto - induced force proposed in @xcite . we have presented a general and detailed analysis of the coupling between electrical and mechanical degrees of freedom , discussing the competing non - linearities . this analysis is applied to a josephson device with a suspended cnt resonator , where we show that the intrinsic non - linearity scales dominate those arising from the coupling . we have provided analytical formulas for the response of the device to mechanical excitations in a wide interval of the excitation strengths and for various biasing schemes . we discuss distinct frequency dependencies , lorentz and fano - like , of the mechanical response both for linear and non - linear regimes and show how these arise based on the nature of the resonant mechanical force . in the case of a phase biased junction we show that the resonant frequency of the mechanical mode acquires a measurable phase - dependent shift ( see fig . [ fig : shift ] ) . we have discussed conditions of detecting the enhanced mechanical response arising when the josephson frequency matches the resonance frequency of the mechanical mode . we reasoned that the regime of shapiro steps is advantageous , since the fluctuations of the voltage drop over the junction are suppressed . we provided expressions for the mechanical response in the regime of shapiro steps and demonstrated that it manifests as modifications of the extrema of the steps . we show that the mechanical mode can be efficiently excited not only by resonant a.c . signals , but also by a.c . signals with frequencies close to an integer fraction of the mechanical resonance frequency . our preliminary experimental results confirm this behavior ; these will be reported elsewhere . the authors would like to thank g. a. steele , s. m. frolov , l. p. kouwenhoven , and h. s. j. van der zant for many useful discussions . this research was supported by the dutch science foundation nwo / fom . a. n. cleland , _ foundations of nanomechanics _ ( springer - verlag , berlin , 2003 ) . k. l. ekinci , x. m. h. huang , and m. l. roukes , appl . lett . * 84 * , 4469 ( 2004 ) . m. p. blencowe and e. buks , phys . b * 76 * , 014511 ( 2007 ) . a. n. cleland and m. l. roukes , nature ( london ) * 392 * , 160 ( 1998 ) . a. d. armour , m. p. blencowe , and k. c. schwab , phys . lett . * 88 * , 148301 ( 2002 ) . t. l. schmidt , k. brkje , c. bruder , and b. trauzettel , phys . * 104 * , 177205 ( 2010 ) . oconnell , m. hofheinz , m. ansmann , r.c . bialczak , m. lenander , e. lucero , m. neeley , d. sank , h. wang , m. weides , j. wenner , j.m . martinis , a.n . cleland , nature ( london ) * 464 * , 697 ( 2010 ) . f. marquardt , j. p. chen , a. a. clerk , and s. m. girvin , phys . lett . * 99 * , 093902 ( 2007 ) . g. sonne , m. e. pea - aza , l. y. gorelik , r. i. shekhter , and m. jonson , phys . lett . * 104 * , 226802 ( 2010 ) . m. d. lahaye , j. suh , p. m. echternach , k. c. schwab , and m. l. roukes , nature ( london ) * 459 * , 960 ( 2009 ) . h. g. craighead , science * 290 * , 1532 ( 2000 ) ; m. blencowe , phys . rep . * 395 * , 159 , ( 2004 ) . ; k. l. ekinci , and m. l. roukes , rev . * 76 * , 061101 ( 2005 ) . j. guckenheimer and p. holmes , _ nonlinear oscillations , dynamical systems , and bifurcations of vector fields _ ( springer - verlag , new york , 1983 ) . v. sazonova , y. yaish , h. stnel , d. roundy , t.a . arias , and p. l. mceuen , nature ( london ) * 431 * , 284 ( 2004 ) . b. witkamp , m. poot , and h. s. j. van der zant , nano . lett . * 6 * , 2904 ( 2006 ) . g. a. steele , a. k. httel , b. witkamp , m. poot , h. b. meerwaldt , l. p. kouwenhoven , and h. s. j. van der zant , science * 325 * , 1103 ( 2009 ) . a. krishnan , e. dujardin , t.w . ebbesen , p.n . yianilos , and m.m.j . treacy , phys . b * 58 * , 14013 ( 1998 ) . a. naik , o. buu , m. d. lahaye , a. d. armour , a. a. clerk , m. p. blencowe , and k. c. schwab , nature ( london ) * 443 * , 193 ( 2006 ) . l. d. landau and e. m. lifshitz , _ theory of elasticity _ ( pergamon , new york , 1986 ) . s. sapmaz , y. m. blanter , l. gurevich , and h. s. j. van der zant , phys . b * 67 * , 235414 ( 2003 ) . g. sonne , r. i. shekhter , l. y. gorelik , s. i. kulinich , m. jonson , phys . b * 78 * , 144501 ( 2008 ) . k. k. likharev , _ dynamics of josephson junctions and circuits _ ( gordon and breach , new york , 1986 ) . y. v. nazarov and y. m. blanter , _ quantum transport : introduction to nanoscience _ ( cambridge university press , cambridge , 2009 ) . m. d. fiske , rev . mod . phys . 36 , 221 ( 1964 ) ; d. d. coon and m. d. fiske , phys . 138 , a744 ( 1965 ) . s. shapiro , phys . 11 , 80 ( 1963 ) . c. j. h. keijzers , l. p. kouwenhoven , unpublished . p. w. anderson and a.h . dayem , phys . 13 , 195 ( 1964 ) .

indirect inference ( ii ) is a methodology for estimating the parameters of an intractable ( generative ) model on the basis of an alternative parametric ( auxiliary ) model that is both analytically and computationally easier to deal with . such an approach has been well explored in the classical literature but has received substantially less attention in the bayesian paradigm . the purpose of this paper is to compare and contrast a collection of what we call parametric bayesian indirect inference ( pbii ) methods . one class of pbii methods uses approximate bayesian computation ( referred to here as abc ii ) where the summary statistic is formed on the basis of the auxiliary model , using ideas from ii . another approach proposed in the literature , referred to here as parametric bayesian indirect likelihood ( pbil ) , uses the auxiliary likelihood as a replacement to the intractable likelihood . we show that pbil is a fundamentally different approach to abc ii . we devise new theoretical results for pbil to give extra insights into its behaviour and also its differences with abc ii . furthermore , we examine in more detail the assumptions required to use each pbii method . the results , insights and comparisons developed in this paper are illustrated on simple examples and two other substantive applications . the first of the substantive examples involves performing inference for complex quantile distributions based on simulated data while the second is for estimating the parameters of a trivariate stochastic process describing the evolution of macroparasites within a host based on real data . we create a novel framework called bayesian indirect likelihood ( bil ) that encompasses pbii as well as general abc methods so that the connections between the methods can be established . ,

1505.03372

approximate bayesian computation ( abc ) now plays an important role in performing ( approximate ) bayesian inference for the parameter of a proposed statistical model ( called the generative model here ) that has an intractable likelihood . despite the intense attention abc has recently received , the approach still suffers from several drawbacks . an obvious disadvantage is the usual necessity to reduce the data to a low dimensional summary statistic . this leads to a loss of information that is difficult to quantify . the second , often less severe but sometimes related , drawback is the computational challenge of achieving stringent matching between the observed and simulated summary statistics . in situations where an alternative parametric model ( referred to as an auxiliary model ) can be formulated that has a tractable likelihood , the methodology known as indirect inference ( ii ) ( see , e.g. , gourieroux , monfort and renault , @xcite and @xcite ) is applicable . ii has been thoroughly examined in the classical framework . most methods differ in the way that observed and simulated data are compared via the auxiliary model . we expand on this later in the article . for the moment , we note that some key references are @xcite , @xcite and @xcite . however , ii has been far less studied in the bayesian paradigm . @xcite developed an abc approach that uses ii to obtain summary statistics . in particular , the estimated parameter of the auxiliary model fitted to the data becomes the observed summary statistic . we adopt a similar naming convention to @xcite and refer to this method as abc ip where `` i '' stands for `` indirect '' and `` p '' stands for `` parameter '' . @xcite also list another method , abc il ( where `` l '' stands for `` likelihood '' ) , which is essentially an abc version of @xcite . this approach follows abc ip in the sense that the parameter estimate of the auxiliary model is again the summary statistic . however , the abc discrepancy is based on the auxiliary likelihood , rather than a direct comparison of the auxiliary parameters . @xcite advocate a slightly different approach to abc with ii , which is effectively an abc version of the classical approach in @xcite . here , @xcite use the score vector based on the auxiliary model as the summary statistic , which is referred to as abc is ( here `` s '' stands for `` score '' ) . the parameter value used in the score is given by the mle of the auxiliary model fitted to the observed data . this approach can be far cheaper from a computational point of view since it avoids an expensive fitting of the auxiliary model to each dataset simulated from the generative model required in abc ip and abc il . throughout the paper , the collection of approaches that use the parametric auxiliary model to form summary statistics is referred to as abc ii methods . an advantage of this approach over more traditional summary statistics is that some insight can be gained on the potential utility of the ii summary statistic prior to the abc analysis . additionally , if the auxiliary model is parsimonious , then the summary statistic can be low - dimensional . @xcite ( see also @xcite ) suggest an alternative approach for combining ii with bayesian inference . this method has similar steps to abc ip and abc il but essentially uses the likelihood of the auxiliary model as a replacement to the intractable generative model likelihood . we note here that this is a fundamentally different approach as it is not a standard abc method . in particular , there is no comparison of summary statistics and no need to choose an abc tolerance . here , we refer to this method as parametric bayesian indirect likelihood ( pbil ) . the focus of this paper is the application of a parametric auxiliary model for the full data , which we refer to as pdbil ( where `` d '' stands for `` data '' ) . however , the ideas in this paper are equally applicable if a parametric model is applied at the summary statistic ( not necessarily obtained using abc ii techniques ) level ( i.e. , some data reduction technique has been applied ; see @xcite , for a review ) . this is referred to as psbil ( where `` s '' stands for `` summary statistic '' ) . we show that the bayesian version of the synthetic likelihood method of @xcite is a psbil method . in the paper , we refer to the collection of abc ii and pbil approaches as pbii methods ( `` bayesian '' `` indirect '' `` inference '' using a `` parametric '' auxiliary model ) . in the process of reviewing these pbii methods , we create a novel framework called bayesian indirect likelihood ( bil ) which encompasses pbii as well as abc methods generally . in particular , if a specific nonparametric auxiliary model is selected ( npbil ) instead of a parametric one ( pbil ) , then the general abc method is recovered . a nonparametric kernel can be applied either at the full data ( npdbil ) or summary statistic ( npsbil ) level . the abc ii approaches are thus a special type of npsbil method where the summary statistic is formed on the basis of a parametric auxiliary model . this framework is shown in figure [ fig : bil ] , which also highlights the methods that this paper addresses . this article does not develop any new algorithms for pbii . however , this paper does make several interesting and useful contributions . firstly , we explore the pdbil method in more detail theoretically , and recognise that it is fundamentally different to abc ii . the behaviour of this method is also substantially different . a technique sometimes applied with classical ii methods is to increase the simulated dataset size beyond that of the observed data in order to reduce the variability of the estimated quantities under the auxiliary model ( see , e.g. , @xcite ; @xcite ; @xcite ) . we demonstrate that pdbil and abc ii behave differently for increasing size of the simulated data . our second contribution is to compare the assumptions required for each pbii approach . our theoretical and empirical results that the pbil method will provide good approximations if the auxiliary model is sufficiently flexible to estimate the likelihood of the true model well based on parameter values within regions of posterior probability . abc ii methods rely on the parameter estimate or the score of the auxiliary model to provide a near - sufficient summary statistic . finally , our creation of the general bil framework provides a clear way to see the connections between pbii and other methods . the paper is organised as follows . in section [ sec : notation ] , the notation used throughout the paper is defined . the abc ii methods are reviewed in section [ sec : abc ii ] . the pbil approach is presented in section [ sec : bil ] . the theoretical developments in this section , which offer additional into the pbil approximation , are new . in addition , this section demonstrates how the synthetic likelihood approach of @xcite is a pbil method on the summary statistic level . section [ sec : abcasbil ] shows how abc can be recovered as a bil method via a nonparametric choice of the auxiliary model . section [ sec : comparebii ] provides a comparison between abc ii and pdbil . the contributions of this article are demonstrated on examples with varying complexity in section [ sec : results ] . the highlights of this section include improved approximate inferences for quantile distributions and a multivariate markov jump process explaining the evolution of parasites within a host . the article concludes with a discussion in section [ sec : discussion ] . consider an observed dataset @xmath0 taking values in @xmath1 of dimension @xmath2 assumed to have arisen from a generative model with an intractable likelihood @xmath3 , where @xmath4 is the parameter of this model . intractability here refers to the inability to compute @xmath3 pointwise as a function of @xmath5 . we assume that there is a second statistical model that has a tractable likelihood function . we denote the likelihood of this auxiliary model by @xmath6 , where @xmath7 denotes the parameter of this auxiliary model . there does not necessarily need to be any obvious connection between @xmath5 and @xmath8 . the auxiliary model could be purely a data analytic model that does not offer any mechanistic explanation of how the observed data arose . the parameter estimate of the auxiliary model when fitted to the observed data is given by @xmath9 . assuming a prior distribution on the parameters of the generative model , @xmath10 , our interest is in sampling from the posterior distribution , @xmath11 , or some approximation thereof . we denote data simulated from the model as @xmath12 . in this paper , we also consider the effect of using @xmath13 independent replicates of data simulated from the model , which we denote @xmath14 and define @xmath15 . therefore the total size of @xmath16 is @xmath17 . note that this could also relate to a stationary time series simulation of length @xmath17 ( see , e.g. , @xcite ) . in the case of a stationary time series or independent and identically distributed ( i.i.d . ) data it is not a requirement for the simulated dataset size to be a multiple of the observed data size . however , for the sake of simplicity we restrict @xmath13 to be a positive integer . since the likelihood of the auxiliary model is tractable , we can potentially consider richly parameterised statistical models to capture the essential features of the data . we assume throughout the paper that the dimensionality of the auxiliary model parameter is at least as large as the dimensionality of the generative model parameter , that is , @xmath18 . @xcite note that it still may be possible to obtain useful estimates of a subset of the parameters when this assumption does not hold . we do not consider this here . abc is widely becoming a standard tool for performing ( approximate ) bayesian inference on statistical models with computationally intractable likelihood evaluations but where simulation is straightforward . abc analyses set @xmath19 so that the simulated dataset is the same size as the observed . however , for the purposes of this paper , we relax this commonly applied restriction for the moment . set @xmath20 simulate @xmath21 draw @xmath22 simulate @xmath23 compute @xmath24 @xmath25 and @xmath26 @xmath27 and @xmath28 in abc , a summary statistic is defined by a collection of functions @xmath29 , for each @xmath30 . henceforth , the subscript @xmath13 is omitted to ease presentation . proposed parameter values that produce a simulated summary statistic , @xmath31 , `` close '' to the observed summary statistic , @xmath32 , are given more weight . here , we define `` close '' by the discrepancy function @xmath33 and a kernel weighting function , @xmath34 , where @xmath35 is a bandwidth referred to as the abc tolerance . the abc target distribution is given by @xmath36 where @xmath37 is referred to here as the abc likelihood . it can be shown that if @xmath38 is a sufficient statistic and @xmath19 then @xmath39 as @xmath40 @xcite . unfortunately , abc can not be trusted when the value of @xmath13 is increased in the sense that the target @xmath41 can move further away from @xmath11 . in the simple example in appendix a of the supplemental article @xcite , the posterior distribution for a univariate @xmath42 converges to a point mass centred on the single observation @xmath43 as @xmath44 and @xmath40 ( see also @xcite , pages 2829 , for a similar example ) . we also verify this behaviour empirically on a toy example . this suggests that whilst it is tempting to increase the simulated dataset size to reduce the variability of the simulated summary statistic , such an approach is fraught with danger . standard abc procedures can make use of @xmath13 simulated datasets but in a different way . here the abc likelihood is estimated via @xmath45 . however , we note that this is an unbiased estimate ( regardless of @xmath13 ) of a more standard abc likelihood , and thus this process does not alter the abc approximation @xcite . therefore , for the remaining presentation in this section we set @xmath19 . however , increasing @xmath13 may improve the performance of abc algorithms and allow a smaller value of @xmath35 for the same computational cost . we do not investigate this here . how the parameter value is proposed depends on the chosen abc algorithm . here , we use markov chain monte carlo ( mcmc abc , @xcite ) with the proposal distribution @xmath46 chosen carefully , and sometimes based on previous abc analyses . the approach is shown in algorithm [ alg : mcmc ] for completeness . in the algorithm , @xmath47 is the number of iterations . we choose @xmath20 , the initial value of the chain , so that it is well supported by the target distribution . this initial value may come from a previous analysis or by performing some preliminary runs . this algorithm is mainly used for simplicity , although it is important to note that other abc algorithms could be applied . one difficult aspect in abc is that often some form of data reduction is required , to avoid the curse of dimensionality associated with comparing all the data at once @xcite . this choice of summary statistic is therefore crucial for a good approximation , even when @xmath35 can be reduced to practically 0 . in some applications , it is possible to propose another model that provides a good description of the data . the summary statistic used in abc can be formulated based on this auxiliary model . these approaches are summarised below . @xcite suggest using the parameter estimate of the auxiliary model as the summary statistic to use in abc . data are simulated from the generative model based on a proposed parameter value , then the auxiliary model parameter is estimated based on this simulated data . the way the auxiliary parameter is estimated provides a mapping between the generative model and the auxiliary model parameters . the abc algorithm uses a noisy mapping between @xmath5 and @xmath8 through the simulated data , @xmath48 , generated on the basis of @xmath5 , @xmath49 . for this purpose ( explained below ) , we use the maximum likelihood estimate ( mle ) of the auxiliary model @xmath50 where @xmath51 . abc ip relies on the following assumption : [ ass : abcip ] the estimator of the auxiliary parameter , @xmath49 , is unique for all @xmath5 with positive prior support . it is important to note that abc ip ( as well as other abc approaches below ) use @xmath19 so the approximation quality of the method can depend on the statistical efficiency of the estimator @xmath49 based on this finite sample . additionally , the mle is asymptotically sufficient ( @xcite , page 307 ) . for this reason , we advocate the use of the mle in general as it is typically more efficient than other estimators like sample moments ( for the auxiliary model ) . sample moments may be computationally easier to obtain , but are likely to result in a poorer abc approximation if the statistical efficiency is lower than the mle . we note that the optimal choice of auxiliary estimator ( trading off between computational effort and statistical efficiency ) may be problem dependent . an additional complication is that the auxiliary model is fitted based on data generated from a different model , @xmath51 . therefore , the efficiency of @xmath52 should be based on @xmath53 not @xmath54 ( see , e.g. , @xcite ) . this can be investigated by simulation . in section [ subsec : quantile_example ] , we provide an example where the auxiliary model does not satisfy assumption [ ass : abcip ] , creating difficulties for abc ip . the abc ii methods ( and the pdbil method ) that follow do not necessarily require unique auxiliary parameter estimates . an advantage of the ii approach to obtaining summary statistics is that the summary statistic will likely be useful if the auxiliary model fits the observed data ( see section [ subsec : abciisummary ] for more discussion ) . in the case of abc ip , the ( approximate ) covariance matrix of the auxiliary parameter estimate based on the observed data can be estimated by the inverse of the observed information matrix [ we denote this information matrix by @xmath55 . intuitively , we expect this discrepancy function to be more efficient than a euclidean distance , as it can take into account the variability of summary statistics and the correlations between summary statistics . denoting the observed summary statistic as @xmath9 and the simulated summary statistic as @xmath56 ( dropping @xmath5 for notational convenience ) , we use the following discrepancy for abc ip : @xmath57 it is important to note that this is essentially an abc version of the classical approach in @xcite . a more appropriate weighting matrix may involve considering the variance of @xmath52 when the data are generated under an alternative model , @xmath51 ( @xcite provides a result , the so - called sandwich estimator ) . @xcite also describe an approach that uses the auxiliary likelihood to set up an abc discrepancy . here , the abc discrepancy is @xmath58 this is effectively an abc version of the classical approach of @xcite . we note that @xmath59 will remain unchanged throughout the algorithm and provides an upperbound for values of @xmath60 obtained for every simulated dataset . abc il uses the same summary statistic as abc ip but uses a discrepancy based on the likelihood rather than the mahalanobis distance . note that the discrepancy function for abc ip appears in the second - order taylor series approximation of @xmath61 about @xmath62 ( @xcite , page 126 ) assuming standard regularity conditions for @xmath59 and @xmath9 . the abc tolerance could be viewed here as a certain cut - off value of the auxiliary log - likelihood . the abc il approach relies on the following assumption . [ ass : abcip_likelihood ] the auxiliary likelihood evaluated at the auxiliary estimate , @xmath63 , is unique for all @xmath5 with positive prior support . we note that this assumption can still be satisfied even when the auxiliary model does not have a unique mle ( see section [ subsec : quantile_example ] for an example ) . the abc ip and abc il methods use parameter estimates of the auxiliary model as summary statistics and can thus be expensive as it can involve a numerical optimisation every time data is simulated from the generative model . the next approach to obtaining summary statistics from ii avoids this optimisation step . @xcite advocate the use of the score vector of the auxiliary model evaluated at the auxiliary mle , @xmath9 , as the summary statistic . we denote the score vector of the auxiliary model as @xmath64 where @xmath65 . each component of the summary statistic involving the observed data and the mle , @xmath66 , is assumed to be numerically 0 under standard regularity assumptions ( see below , assumption [ ass : abcis ] ) . thus , the search is for parameter values of the generative model that lead to simulated data , @xmath48 , that produces a score close to @xmath67 . noting that the approximate covariance matrix of the observed score is given by the observed information matrix @xmath68 , the following abc discrepancy is obtained for abc is @xmath69 this is essentially an abc version of @xcite . this approach is fast relative to abc ip when the mle of the auxiliary model is not analytic whilst the score is analytic since no numerical optimisation is required every time data are simulated from the generative model . of course , it may be necessary to estimate the score numerically , which would add another layer of approximation and may be slower . in the examples of this paper , we are able to obtain the score analytically . abc is relies on the following assumptions . [ ass : abcis ] the mle of the auxiliary model fitted to the observed data , @xmath9 , is an interior point of the parameter space of @xmath8 and @xmath68 is positive definite . the log - likelihood of the auxiliary model , @xmath70 , is differentiable and the score , @xmath71 , is unique for any @xmath48 that may be drawn according to any @xmath5 that has positive prior support . we note that assumption [ ass : abcis ] is generally weaker than assumption [ ass : abcip ] ( abc ip ) , since it may still hold even if the mle of the auxiliary model is not unique . only models in the exponential family possess a minimal sufficient statistic with dimension equal to that of @xmath72 . for other models , under suitable conditions , the pitman koopman darmois theorem states that the dimension of any sufficient statistic increases with the sample size . for many complex models , such as those considered in the abc setting , the minimal sufficient statistic will be the full dataset ( or the full set of order statistics if the data are i.i.d . ) . the summary statistic produced by abc ii will always have dimension @xmath73 , and thus will not produce sufficient statistics in general ( this argument of course carries over to any abc method that uses some data reduction technique ; see @xcite , for a review ) . intuitively , our suggestion is that the summary statistic produced by abc ii should carry most of the information contained in the observed data provided that the auxiliary model provides a good description of the data . unfortunately , this is difficult to verify since it is usually not possible to quantify the amount of information lost in data reduction . despite this , by conducting goodness - of - fit tests and/or residual analysis on the auxiliary model fit to the data will at least provide some guidance on the usefulness of the summary statistic produced by abc ii . this is in contrast to the more traditional approach of summarising based on simple functions of the data ( e.g. , @xcite ) , whose utility is difficult to assess prior to running an abc analysis without performing an expensive simulation study . furthermore , abc ii methods provide natural discrepancy functions between summary statistics as shown above . selecting the discrepancy function and determining appropriate weighting of the summary statistics in traditional abc can be problematic . it is well known that the choice of summary statistic in abc involves a compromise between and dimensionality @xcite . a low - dimensional and near - sufficient summary statistic represents an optimal trade - off . another advantage of abc ii over usual abc is the dimensionality of the abc ii summary statistic can be controlled by selecting parsimonious auxiliary models and using standard model choice techniques to choose between a set of possible auxiliary models [ e.g. , the akaike information criterion ( aic ) and the bayesian information criterion ( bic ) ] . @xcite and @xcite propose a method that has similar steps to abc ip and abc il but is theoretically quite different , as we show below . after data are simulated from the generative model , the auxiliary parameter is estimated . this auxiliary estimate is then passed into the auxiliary likelihood of the observed data . this likelihood is then treated in the usual way and fed into a bayesian algorithm , for example , mcmc . one first defines a collection of functions @xmath74 . the artificial likelihood is then defined as follows : @xmath75 and the target distribution of this approach is given by @xmath76 where the subscripts @xmath77 and @xmath13 denote the dependence of the target on the auxiliary model choice and the number of replicate simulated datasets , respectively . this approach is effectively a bayesian version of the simulated quasi - maximum likelihood approach of @xcite . @xcite proposes to maximise @xmath78 with respect to @xmath5 . instead of applying this as an abc discrepancy as in the abc il method above , @xcite and @xcite treat this auxiliary likelihood as a replacement to the likelihood of the generative model in the same way that @xcite does . it is important to note that this approach does not perform a comparison of summary statistics , and hence there is no need to choose an abc tolerance . thus , it is not a standard abc algorithm . we refer to this approach simply as pdbil , since we apply a `` parametric '' auxiliary model for the full `` data '' . when the full data have been summarised as a summary statistic , @xmath32 , an alternative approach is to apply a parametric auxiliary model for the summary statistic likelihood , @xmath79 ( see section [ subsec : psbil ] for more details ) . in figure [ fig : bil ] , these methods fall under the pbil class . within this class , the focus of this paper is on the pdbil method . the theoretical aspects of this approach are yet to be investigated in the literature . some clues are offered in @xcite and @xcite , but we formalise and extend the theory here . the subscript @xmath13 is used to denote that this target remains an approximation to the true posterior distribution and that the approximate target may change with @xmath13 ( we show below that in general the target does depend on @xmath13 ) . however , because it is not an abc algorithm , it is unclear how pdbil behaves as @xmath13 increases . @xcite use a very large simulation size ( @xmath80 ) , without a theoretical investigation . with @xmath0 fixed , we consider a potential limiting likelihood @xmath81 with associated posterior @xmath82 note that the parameter of @xmath78 is @xmath83 but the parameter of @xmath81 is @xmath84 . we emphasise that the results below all assume that @xmath0 is a fixed value in @xmath1 . to ease presentation , we define the random variable @xmath85 where @xmath86 is a sequence of i.i.d . random variables distributed according to @xmath87 , and we can write @xmath88,\ ] ] where @xmath89 is expectation with respect to the distribution of @xmath90 . the results below provide sufficient conditions under which , as @xmath91 , @xmath92 where @xmath93 is some function of @xmath5 whose posterior expectation is of interest . this does not assume that @xmath94 , which in general will not be the case . a useful tool to allow us to answer both questions is provided by @xcite . [ thm : billingsley ] if @xmath95 is a sequence of uniformly integrable random variables and @xmath95 converges in distribution to @xmath96 then @xmath96 is integrable and @xmath97 . a simple sufficient condition for uniform integrability is that for some @xmath98 , @xmath99 [ res : conv_posterior_and_expectation ] assume that @xmath100 as @xmath101 for all @xmath5 with positive prior support , @xmath102 and @xmath103 . then @xmath104 furthermore , if @xmath105 is a continuous function satisfying @xmath106 for some @xmath98 then @xmath107 the first part follows from the fact that the numerator of @xmath108 converges pointwise and the denominator is positive and converges by the bounded convergence theorem . for the second part , if for each @xmath109 , @xmath110 is distributed according to @xmath111 and @xmath5 is distributed according to @xmath112 then @xmath110 converges to @xmath5 in distribution as @xmath101 by scheff s lemma @xcite . since @xmath93 is continuous , @xmath113 converges in distribution to @xmath114 as @xmath101 by the continuous mapping theorem and we conclude by application of theorem [ thm : billingsley ] . a simple condition for @xmath115 as @xmath101 to hold is provided by the following result . [ res : conv_likelihood ] assume that @xmath116 converges in probability to @xmath117 as @xmath101 . if @xmath118<\infty\ ] ] for some @xmath98 then @xmath115 as @xmath119 . the result follows by application of theorem [ thm : billingsley ] . although the results above hold under conditions on the fixed , observed data @xmath0 , they will often hold for a range of possible values of @xmath0 . the function @xmath120 is often referred to as the mapping or binding function in the ii literature . in @xcite , it is assumed that this function is 11 but the results above demonstrate that this is not a necessary condition for result [ res : conv_likelihood ] to hold . the following example where the auxiliary model is a mixture model demonstrates this principle . assume that the true model is @xmath121 while the auxiliary model is a mixture of normals , @xmath122 so that @xmath123 . assuming an infinite sample from the true model , there are an infinite number of mles of the auxiliary model ; @xmath124 where @xmath125 , @xmath126 where @xmath127 or @xmath128 where @xmath129 . all of these possible mappings produce the same value of the auxiliary likelihood , which coincides with the value of the generative likelihood . it is straightforward under the assumptions of results [ res : conv_posterior_and_expectation][res : conv_likelihood ] to show that pdbil will target the true posterior as @xmath44 if the true model is contained within the auxiliary model . when the generative model is a special case of the auxiliary model , using the notation of @xcite , the auxiliary and generative parameter can be written as @xmath130 and @xmath131 where @xmath132 and @xmath133 denote `` extended '' and `` reduced '' respectively and @xmath134 is fixed . the proof of this result is straightforward . it involves demonstrating that @xmath135 is consistent also for the parameter of the generative ( reduced ) model . therefore , when @xmath91 the generative and auxiliary likelihoods will coincide . this theoretical result can not typically be realised in practice since a model which incorporates an intractable model as a special case is likely to also be intractable . however , it does suggest that the auxiliary model be chosen to be adequately flexible to give a good approximation of the generative likelihood for @xmath5 values with positive prior support . in practice , our empirical evidence indicates that it is only necessary for the auxiliary likelihood to mimic the behaviour of the generative likelihood for the values of @xmath5 with nonnegligible posterior support and for the auxiliary likelihood to be negligible in regions of the parameter space with little posterior support . if this is not the case , it is likely that the pdbil method will lead to poor approximations ( as we demonstrate in section [ subsec : toy_example ] ) . if the binding function , @xmath120 , were known , the pdbil method would proceed straightforwardly . since it will not be available in practice , it can be estimated indirectly through the simulated data @xmath16 . from the above , it is desirable for the pdbil method if the auxiliary likelihood is as close as possible to the true likelihood . @xcite show , for a particular choice of the auxiliary model , that choosing the mle for @xmath135 minimises the kullback leibler divergence between the generative and auxiliary likelihoods . furthermore , this choice will often lead to results [ res : conv_posterior_and_expectation][res : conv_likelihood ] holding . therefore , we advocate the use of the mle with this method . when the mle of the auxiliary model is used , @xcite provides an expression which defines @xmath120 [ see equation ( 25 ) of @xcite ] . set @xmath20 simulate @xmath136 compute @xmath137 draw @xmath22 simulate @xmath138 compute @xmath139 compute @xmath140 @xmath25 @xmath141 @xmath27 @xmath142 it remains to be seen what the target of @xmath143 is relative to @xmath144 . from an intuitive perspective , increasing @xmath13 leads to a more precise determination of the mapping @xmath120 , and thus should lead to a better approximation . a more theoretical argument is as follows . the approximations will coincide with each other provided @xmath145 = p_a({\mathbf{y}}|{\boldsymbol{\phi}}({\boldsymbol{\theta}}))$ ] @xcite . unfortunately , even if one uses an unbiased estimator for the auxiliary parameter , that is @xmath146 = { \boldsymbol{\phi}}({\boldsymbol{\theta}})$ ] for any value of @xmath13 , this result will still rarely hold in general . the likelihood can be viewed here as a nonlinear function of the auxiliary parameter estimate , and so @xmath147 \neq p_a({\mathbf{y}}|{\boldsymbol{\phi}}({\boldsymbol{\theta}}))$ ] in general . our empirical evidence ( see section [ sec : results ] ) suggests that @xmath143 typically becomes less precise relative to @xmath144 as the likelihood estimate becomes more noisy , that is , when @xmath13 is reduced . the message of results [ res : conv_posterior_and_expectation][res : conv_likelihood ] is that , provided the auxiliary model is suitably chosen , a better approximation can be anticipated by taking @xmath13 as large as possible , which has the effect of reducing the bias of @xmath148 . as an example , mcmc can be used to sample from the pdbil target ( referred to here as mcmc pdbil , see @xcite ) . this approach is presented in algorithm [ alg : mcmc - bil ] . as results [ res : conv_posterior_and_expectation][res : conv_likelihood ] suggest , the aim with pdbil is to select @xmath13 as large as possible . we demonstrate in the examples that it is desirable to consider values of @xmath13 greater than one due to the improved statistical efficiency of mcmc pdbil ( and potentially other algorithms that implement pdbil ) when increasing @xmath13 . of course , the method will become computationally infeasible for very large @xmath13 . @xcite proposes a method called synthetic likelihood when it is convenient to perform inference on the basis of a set of summary statistics rather than the full data . considering bayesian inference , the target distribution when the data have been reduced to a summary statistic is given by @xmath149 the major issue with this construction is that there is no analytical form for the likelihood function of the summary statistic , @xmath79 . @xcite overcomes this by applying , based on the terminology in this paper , a parametric auxiliary model for this probability distribution , @xmath150 . in our framework , an approach that applies a parametric auxiliary model to form the likelihood of the summary statistic rather than the likelihood of the full data ( as is presented in section [ subsec : bil_target ] ) is referred to here as psbil ( where `` s '' denotes `` summary statistic '' ) . therefore the bayesian version of @xcite is a psbil approach . for the auxiliary likelihood , @xmath150 , @xcite considers using the likelihood of a multivariate normal distribution , @xmath151 , where @xmath152 is the auxiliary parameter . as is the case with pdbil , the binding function @xmath120 is rarely known but can be estimated via simulated data from the generative model for a particular value of @xmath5 . using our notation , we obtain the following dataset from the true model of the summary statistic , @xmath153 . this represents @xmath13 i.i.d . observations from @xmath154 . an advantage of selecting such a simple auxiliary model is that the mle has the analytic form @xmath155 where the superscript @xmath47 denotes transpose . the auxiliary likelihood used is then based on @xmath156 . our results indicate that the target distribution of this method will depend on @xmath13 , and , if the auxiliary model for the summary statistic likelihood is reasonable , better approximations of @xmath157 are likely to be obtained for large @xmath13 . an alternative and perhaps natural candidate for @xmath158 is to use a kernel density estimate based on the samples @xmath16 . this corresponds to choosing @xmath159 and we define @xmath160 @xcite , where @xmath35 is the bandwidth parameter . we have then @xmath161 and this is exactly the form of the standard abc likelihood . in addition , @xmath13 does not affect the likelihood ( although it may help computationally in some algorithms ) and @xmath35 controls the level of approximation . here , we see that this is an estimate of the abc likelihood where the comparison is made between the full datasets . here , we obtain the npdbil approach as presented in figure [ fig : bil ] ( where `` d '' corresponds to full `` data '' ) . alternatively , a nonparametric density estimate of the auxiliary model of the summary statistic likelihood , @xmath162 , could be applied . using a similar procedure to above , we obtain @xmath163 . we refer to this in figure [ fig : bil ] as npsbil ( npbil based on a `` summary statistic '' ) . this is the approach adopted by @xcite , however , their focus is on point estimation ( posterior mean ) . unfortunately , @xcite refer to their bayesian estimator as bil , however , under our framework bil is a more general class of methods . of course , if the summary statistic is derived from some parametric auxiliary model , then the abc ii class of method is recovered as an npsbil method . the reader is again referred to figure [ fig : bil ] to see the connection between these methods . by selecting a parametric model for the auxiliary likelihood ( pbil ) , we can potentially overcome the curse of dimensionality associated with the nonparametric aspect of abc . this requires further research . of course , finding a suitable parametric auxiliary model may be challenging in practice . there are a few remarks to be made about the above results in relation to theoretical comparisons between abc ii and pdbil . [ rem : bil_increasing_n ] under suitable conditions better approximations with pdbil are obtained by increasing @xmath13 . this is in stark contrast with abc ii , which can not be trusted for @xmath164 . [ rem : abc_never_correct ] in the case where the true model is a special case of the auxiliary model , the pdbil method will be exact in the limit as @xmath165 . in contrast , in this ideal situation , abc ii still does not produce sufficient statistics ( see the dimensionality argument in section [ subsec : abciisummary ] ) and will not target the true posterior in the limit as @xmath40 . an example is where the true model is a t - distribution with location , scale and degrees of freedom of @xmath166 , @xmath167 and 1 , respectively . the auxiliary model is a more general t - distribution with degrees of freedom @xmath168 . in this case , the pdbil method is exact in the limit as @xmath44 as the true model is incorporated within the auxiliary model . unfortunately , abc ii does not produce a sufficient statistic as the summary statistic will be of dimension three whilst it is known for this model the minimal sufficient statistic consists of all the order statistics . of course , finding an auxiliary model that satisfies this condition in practice will rarely be feasible . [ rem : bil_never_correct ] even if the auxiliary parameter estimate or score happen to be a sufficient statistic for the generative model , pdbil still will not generally target the true posterior , as the auxiliary and generative likelihoods will still not match up . in this situation , the abc ii approaches will enjoy convergence to the true posterior as @xmath40 whilst pdbil will not converge to the true posterior as @xmath44 . however , sufficient statistics are rarely achieved in practice . remarks [ rem : abc_never_correct ] and [ rem : bil_never_correct ] demonstrate that pbii methods generally will not ( and rarely will ) target the true posterior distribution asymptotically . this is generally the case for other techniques in the literature for dealing with models that have intractable likelihood functions . there are some exceptions to this . for example , exact techniques are available for so - called doubly intractable models when perfect simulation from the generative model is possible ( e.g. , @xcite ; @xcite ) . furthermore , so - called pseudo - marginal methods @xcite are applicable when a positive and unbiased estimator of the likelihood is available and is a current area of research . despite not being exact , we demonstrate that pbii methods can produce quite good approximations in some applications . the characteristics of a good auxiliary model differ between the abc ii and pdbil methods . in the context of abc ii , we simply require a good summarisation of the data , that is , a low - dimensional summary statistic that hopefully carries most of the information in the observed data . therefore , we feel that it is useful if the auxiliary model in this context provides a good fit to the data and is parsimonious , so that the essential features of the data are described well and as succinctly as possible . this is independent of the process for selecting a generative model . therefore , the same auxiliary model should be used regardless of which generative model is fitted to the data . for pdbil , we require a flexible auxiliary model that can mimic the behaviour of the generative model for different values of @xmath5 within the posterior support . here , it is not necessary for the auxiliary model to provide a good fit to the data considering the fact that the generative model being proposed might be mis - specified . the auxiliary model chosen for pdbil may alter depending on the generative model being proposed . in our examples , the generative model is either known or provides a good fit to the data . in such cases , it would not be uncommon to choose the same auxiliary model for the abc ii and pdbil methods . the conditions required for pdbil to produce exact results are very strong and finding an auxiliary model that is sufficiently flexible so that the auxiliary likelihood can mimic the generative likelihood could be difficult in practice . in some applications , an auxiliary model that is a simplified version of the generative model may be specified where the parameter of each model has the same interpretation . for example , the auxiliary model for a continuous time markov jump process may be its corresponding linear noise approximation . in such situations , the pdbil method is unlikely to perform well whilst it remains possible that such an approximate model could produce summary statistics for abc even though the auxiliary model would not fit the data well . @xcite show that ii can work well in the classical framework when the auxiliary model is a simplified version of the generative model . further research is required in the bayesian setting . an additional advantage of the abc ii approach over pdbil is the extra flexibility of being able to accommodate additional summary statistics that do not involve an auxiliary model , since this method belongs in the more general npsbil class ( see @xcite , for an example where the summary statistic is a combination of auxiliary parameter estimates and other summary statistics ) . @xcite and @xcite consider ii applications in a classical framework where the comparison of observed and simulated data is made on the basis of both an auxiliary model and supplementary summary statistics . in this example , we consider a simple model so that exact bayesian inferences are trivially obtained . our intention here is to investigate the theoretical considerations in section [ sec : bil ] . in particular , we show that when the auxiliary model is reasonable , pdbil produces better approximations as the size of simulated datasets goes beyond that of the observed data and as a useful by - product increases the acceptance probability of the mcmc moves . we also demonstrate empirically that unfortunately abc approaches ( including those using ii to obtain summary statistics ) do not possess this same desirable property as @xmath13 is increased . additionally , we investigate the output of pdbil when the auxiliary model is poorly chosen . [ fig : toy ] here , the data are @xmath169 independent draws from a poisson distribution with a mean of @xmath170 , @xmath171 . the prior is @xmath172 ( where @xmath173 and @xmath174 ) , which results in a @xmath175 posterior . for such a relatively large value for the mean of the poisson distribution , a normal approximation with mean , @xmath166 , and variance , @xmath176 , will be reasonable . we use this normal distribution as the auxiliary model . here , the auxiliary likelihood , mle and score are trivial to compute . the anderson darling test for normality produced a @xmath177-value of about 0.576 , which indicates no evidence against the assumption that the normal auxiliary model provides a good description of the data . the summary statistic based on this auxiliary model includes the sample mean , @xmath178 , which is a sufficient statistic for the generative model . thus , the abc ii approaches can be expected to produce essentially exact inferences ( excluding monte carlo error ) as long as the abc tolerance is low enough . as demonstrated in figure [ fig : toy ] , this is the case . such sufficiency is not usually achieved in practice . however , it can be seen that the abc posterior is grossly over - precise when the size of the simulated datasets is increased to 1000 ( i.e. , @xmath179 ) . in the limit as @xmath44 , the pdbil method boils down to a @xmath180 distribution approximating a @xmath181 distribution . the central limit theorem states that the normal approximation improves as @xmath182 increases . since @xmath170 , pdbil can never target the true posterior . the pdbil target distribution ( for @xmath44 ) is proportional to @xmath183 while the true posterior is proportional to @xmath184 . figure [ fig : toy](d ) demonstrates a small amount of bias for the pdbil method ( this is an illustration of remark [ rem : bil_never_correct ] ) . figure [ fig : toy](d ) presents the results for the pdbil method based on simulated dataset sizes of @xmath19 and @xmath179 ( results for @xmath185 and @xmath186 are even closer to the true posterior but are not shown on the figure ) . it is evident from the figure that a more precise posterior is achieved when using larger simulated datasets , without necessarily over - shooting the true posterior . additionally , there was an increase in the mcmc acceptance rate as @xmath13 increased . for the @xmath13 values investigated here , the acceptance rates were roughly 46% , 67% , 72% and 73% for increasing @xmath13 . these acceptance rates are very high , especially relative to abc algorithms which generally suffer from quite low acceptance probabilities . this example demonstrates that better inferences using pdbil can be obtained by increasing the size of the simulated dataset beyond that of the observed . unfortunately , abc inferences that use a simulated data size larger than that of the observed data can not be trusted in the same way ( see remark [ rem : bil_increasing_n ] ) . the reason for improved inferences from pdbil as @xmath13 is increased is apparent from figure [ fig : toy_loglike ] . here , it can be seen from increasing @xmath13 the log - likelihoods of the generative and auxiliary models are becoming more correlated with the slope of the relationship becoming approximately one . figure [ fig : compare_true_aux](a ) shows the true @xmath187 and auxiliary @xmath180 log - likelihood values for @xmath188 within the 99% highest prior density region . the vertical lines indicate the bounds of a 99% credible interval based on the true posterior . it can be seen that the auxiliary log - likelihood is a poor approximation to the true log - likelihood in regions with negligible posterior support . this is even the case for larger @xmath182 values where it would be expected that a normal approximation would be more appropriate . however , the normal approximation will perform relatively poorly in the tails of the distribution . it is evident that the auxiliary likelihood acts as a useful replacement likelihood in the region of high posterior support [ see figure [ fig : compare_true_aux](b ) ] , and this is enough to result in a good approximation of the true posterior for large @xmath13 . finally , we investigate the output from pdbil when the auxiliary model is chosen poorly . figure [ fig : compare_true_aux](c ) shows the results for when the auxiliary model is @xmath189 , where @xmath190 is fixed . here , the pdbil posterior as @xmath44 is proportional to @xmath191 . here , we consider @xmath192 ( over - dispersed ) and @xmath193 ( under - dispersed ) . the results show over - precise and conservative results in the under - dispersed and over - dispersed case , respectively . the under - dispersed and over - dispersed auxiliary models have thinner and fatter tails , respectively , than the likelihood of the generative model in the parameter space well supported by the posterior distribution ( see figure [ fig : compare_true_aux ] ) . in these cases , the auxiliary model is not providing a useful replacement likelihood . just by chance abc ii is still exact here as @xmath40 since the parameter estimate for @xmath166 is a sufficient statistic for @xmath182 . quantile distributions ( or functions ) represent a class of distributions that are defined in terms of their quantile function . such functions can be formulated to create more flexible distributions than other standard distributions . in this example , the focus is on the @xmath194-and-@xmath195 distribution described in , for example , @xcite ( the reader is also referred to the references therein ) . this quantile function , which can also be interpreted as a transformation of a standard normal random variate , has the following form : @xmath196 \\[-8pt ] & & \phantom{a + } { } \cdot\bigl(1 + z(p)^2\bigr)^kz(p).\nonumber\end{aligned}\ ] ] here , @xmath177 denotes the quantile of interest while @xmath197 represents the quantile function of the standard normal distribution . the model parameter is @xmath198 , though common practice is to fix @xmath199 at 0.8 , which we do here ( see @xcite , for a justification ) . the likelihood function can be computed numerically , although this is more expensive than model simulation which is cheaply implemented for quantile distributions via the inversion method . full likelihood - based inference is more expensive than the simulation - based approaches for the relatively large dataset considered here . the observed dataset consists of 10,000 independent draws from the @xmath194-and-@xmath195 distribution with @xmath200 , @xmath201 , @xmath202 , @xmath203 and @xmath204 ( same as considered in @xcite ) . a nonparametric estimate of the probability density function based on these samples is shown in figure [ fig : gandk_data ] . the data exhibit significant skewness and kurtosis . we use a three component normal mixture model with 8 parameters as the auxiliary model . a mixture model is a suitable choice for an auxiliary distribution since it can be made arbitrarily flexible whilst maintaining a tractable likelihood function . therefore , auxiliary mles are computationally easy to obtain [ here we use the expectation - maximisation ( em ) algorithm ] and the subsequent likelihood can be evaluated cheaply . on the other hand , mixture models can he highly irregular and the mle is not consistent in general . the invariance of the likelihood to a re - labelling of the components causes an immediate issue for abc ip , which requires a unique auxiliary parameter estimate . in an attempt to overcome this , we post - process the mixture model parameter estimates generated throughout the abc ip algorithm by ordering them based on the component means . since pdbil and abc il use the likelihood of the auxiliary model , they more naturally overcome the label switching issue . however , the mixture model can give other numerical issues such as those resulting from infinite likelihoods . this would create serious issues for methods that use the auxiliary likelihood ( the auxiliary likelihood would not be unique ) . from investigations on the dataset here , it appears that the likelihood is well behaved and that the modes in the likelihood correspond only to re - labelling of components . therefore , we proceed with abc il and pdbil with caution . the abc is method , based on the score vector , appeared to not have any difficulties accommodating the auxiliary mixture model . from figures [ fig : gandk_data](a ) and [ fig : gandk_data](b ) , it can be seen that there is a correspondence between both the densities and the cumulative distribution functions of the mixture model and the data . however , we performed a hypothesis test to assess the goodness - of - fit of the three component mixture model with a parameter given by the mle . the test - statistic was the kolmogorov smirnov statistic that computes the maximum absolute difference between the theoretical and empirical c.d.f.s . to avoid any distributional assumption about this test - statistic , we simulated 10,000 values of this statistic under the assumption that the mixture model is correct . we found that the observed test - statistic was exceeded 0.25% of the time , indicating strong evidence against the mixture providing a good fit to the data . figure [ fig : gandk_data](b ) shows the differences between the empirical and theoretical c.d.f.s . however , from figure [ fig : gandk_data](a ) it is evident that the mixture model can explain several features of the true model , and since the dataset size is large there is a high probability of detecting a difference . our results below show that we are able to obtain quite accurate posterior distributions with the pbii methods despite the lack of fit suggested by the hypothesis test . in appendix b of the supplemental article @xcite , we present results from using a four component mixture model . unfortunately we found this was substantially more expensive to apply and resulted in some numerical problems . the proposal distribution in the mcmc for the pdbil algorithm was guided using the results in @xcite , who analysed the same data via a traditional abc approach that used robust measures of location , scale , skewness and kurtosis as the summary statistics . pdbil was run using @xmath13 values of 1 , 2 , 4 , 10 , 20 and 50 for a number of iterations given by 1 million , 500,000 , 500,000 , 200,000 , 100,000 and 75,000 , respectively . the mcmc acceptance probabilities obtained were about 2.8% , 5% , 7.1% , 13.1% , 18.5% and 20.8% , respectively . the average effective sample size ( ess , averaged over the four parameters ) divided by the computing time ( in hours ) were roughly 63 , 127 , 106 , 124 , 70 and 41 , respectively . this demonstrates how pdbil is still feasible as @xmath13 increases to a certain point . however , for very large @xmath13 the computation becomes unmanageable . figure [ fig : gandk_large ] shows the results for @xmath19 , @xmath179 and @xmath205 ( the results for @xmath206 and @xmath205 were quite similar ) . a very time consuming exact mcmc algorithm was run for 10,000 iterations to obtain a gold - standard ( producing an average ess per hour of 6 ) . the results show an increase in precision of the pdbil posteriors as @xmath13 increases . the results for @xmath207 and @xmath208 are very accurate , while the pdbil posteriors for @xmath194 and @xmath195 show some bias ( also with a loss of precision for @xmath194 ) . abc ip and abc il were run for 1 million iterations with a tolerance tuned to achieve an acceptance rate of about 1% . due to the abc is method being so much faster than the other pbii approaches , we aimed for a relatively lower abc tolerance and ran the algorithm for more iterations . more specifically , 7 million iterations were used and the abc tolerance chosen resulted in an acceptance rate of 0.3% . we also applied a regression adjustment to the ( appropriately thinned ) abc ii samples using the approach of @xcite in an attempt to eliminate the effect of the abc tolerance . in order to apply regression adjustment for abc il , the same post - processing procedure used for abc ip was required . without regression adjustment [ see figure 2 of the supplemental @xcite ] , the abc is method gave slightly better results than other abc ii methods , which could be due to the ability of abc is getting to a lower abc tolerance . the unadjusted abc il results were also slightly better than the unadjusted abc ip results . abc is produced an average ess per hour of 90 while the corresponding number was 50 and 30 for abc ip and abc il , respectively , showing that the abc is method required less time to produce a better approximation . regression adjustment offered improvement to all the abc ii methods . we compared the pbii approaches with the abc results of @xcite . it should be noted that we applied a local regression adjustment to the abc results here as we found some improvement for the parameters @xmath207 and @xmath194 ( results were very similar for @xmath208 and @xmath195 relative to those obtained in @xcite ) . the results are shown in figure [ fig : gandk_compare ] . overall , the pbii results present a marked improvement over the abc analysis of @xcite , with @xmath194 seemingly the most difficult parameter to estimate . the abc ii methods with regression adjustment produced very similar results . taking into account accuracy and computational efficiency , abc is with regression adjustment is probably the preferred method . when a four component auxiliary model was used [ appendix b of the supplemental article @xcite ] , the abc ii methods with regression adjustment produced similar results and outperformed pdbil in terms of accuracy . further , the abc is approach was able to avoid the heavy computation associated with fitting the four component mixture model at every iteration , and thus also avoided other numerical issues such as the em algorithm converging to potentially local optima . the abc ii regression adjustment results showed some improvement for @xmath194 and @xmath195 when going from the three to four component mixture model . drovandi , pettitt andfaddy ( @xcite ) developed an abc ip approach to estimate the parameters of a stochastic model of macroparasite population evolution developed by @xcite ( see also @xcite ) . data was collected independently on 212 cats , who were initially injected with a certain number of juvenile _ brugia pahangi _ parasites . some time after each cat was sacrificed , the number of parasites that had reached maturity were counted and recorded ( see @xcite ) . @xcite discovered that a beta - binomial model provided a good description of the data . below provides a brief review of the generative and auxiliary models , with the reader referred to @xcite for more details . at time @xmath209 , any host is described by three random variables @xmath210 , where @xmath211 is the number of mature parasites , @xmath212 is the number of larvae and @xmath213 is a discrete version of the host s immunity . it is assumed that each larva matures at a constant rate of @xmath214 per day . larvae die at a rate @xmath215 per larva where @xmath216 represents the rate at which natural death of larvae occurs and @xmath217 is a rate parameter that describes additional death of larvae due to the immune response of the host . the acquisition of immunity is assumed to be dependent only on the number of larvae and occurs at rate @xmath218 , and a host loses immunity at a rate @xmath219 per unit of immunity . mature parasites die at a rate of @xmath220 adults per day . parameters @xmath214 and @xmath220 have been previously estimated at 0.04 @xcite and 0.0015 @xcite , respectively . the data were modelled via a continuous time discrete trivariate markov process . given current values of the states at time @xmath209 , @xmath221 , @xmath222 , @xmath223 , and a small time increment @xmath224 the transition probabilities at time @xmath225 are given by @xmath226 and the probability of remaining in the same state is one minus the sum of the above probabilities . only the final mature count is observed whilst the immunity and larvae counts are unobserved throughout the process . moreover , the immune response variable @xmath213 is unbounded . data generative likelihood - based approaches appear infeasible due to computational issues ( see @xcite ) . simulation is straightforward via the algorithm of @xcite . the prior distributions are : @xmath227 , @xmath228 , @xmath229 and @xmath230 . here , we denote the observed data as @xmath231 where @xmath232 is the mature count for the @xmath233th host . covariates for the @xmath233th host are given by @xmath234 ( initial larvae count ) and @xmath235 ( sacrifice time ) . for the auxiliary model , @xcite capture the overdispersion via a beta - binomial regression model and take into account the effect that @xmath235 and @xmath234 have on @xmath232 . denote @xmath236 and @xmath237 as the beta - binomial parameters for the @xmath233th host . more specifically , the @xmath233th observation has the following probability distribution : @xmath238 where @xmath239 denotes the beta function . consider a re - parameterisation in terms of a proportion , @xmath240 , and over - dispersion , @xmath241 , parameter . the auxiliary model relates these parameters to the covariates via @xmath242 where @xmath243 and @xmath244 hence , the auxiliary model has the parameter @xmath245 while the generative model has the parameter @xmath246 . using the approach outlined in appendix c of the supplemental article ( @xcite,@xcite ) , we obtained goodness - of - fit @xmath177-values of 0.37 and 0.47 , indicating no evidence against the beta - binomial model providing a good description of the data . @xcite use the aic to select this auxiliary model over competing auxiliary models . for validation of the pbii methods for this example , data was simulated using the same experimental design as the observed data based on the parameter configuration estimated by @xcite ; @xmath247 , @xmath248 , @xmath249 and @xmath250 . we found that the pbii methods were able to recover the parameters @xmath168 and @xmath216 well , @xmath219 was determined less precisely and @xmath217 was not recovered . the data are not particularly informative about @xmath219 and @xmath217 ( see @xcite , for more discussion ) . the abc is gave the most precise posterior distributions for @xmath168 and @xmath251 out of the pbii methods . for full details on the analysis of this simulated data , see appendix c of the supplemental article @xcite . here , we used the abc ip results of @xcite to form an mcmc proposal distribution . the pdbil method with @xmath19 , @xmath206 and @xmath205 was run for 1 million , 100,000 and 50,000 iterations , respectively . acceptance probabilities of roughly 1.4% , 23.5% and 28.2% , respectively , were obtained . the average ess per hour was 37 , 79 and 58 , respectively . the substantial increase in acceptance probability allowed us to use fewer iterations . the results are shown in figure [ fig : cat_bil ] . the figures suggest that we are not able to gain any additional information from the data for the parameter @xmath252 from the pdbil approach by increasing @xmath13 . however , an increase in precision is obtained for @xmath251 as @xmath13 is increased . the posteriors are shifted slightly for the other two parameters , however , they are still largely uninformative , although the posterior for @xmath219 for large @xmath13 may indicate some preference for smaller values of @xmath219 . we now compare the results of pdbil with abc . abc ip and abc il mcmc algorithms were all run for 1 million iterations . the abc ip and abc il tolerances were chosen so that the acceptance rate was about 1.5% . due to the increased computational efficiency of abc is , we ran this algorithm for 20 million iterations and tuned the tolerance to obtain an acceptance rate of about 0.1% . abc ip and abc il used about 15 hours of computing time while abc is only required 11 hours even though 20 times more iterations were run . [ figsub : cat_ii_compare_nu_reg][figsub : cat_ii_compare_mul_reg ] the estimated posterior densities ( after appropriate thinning ) for the different approaches are presented in figure [ fig : cat_ii_compare ] . in general , the data are not informative about the @xmath219 and @xmath217 parameters , so we turn our focus to the parameters @xmath168 and @xmath251 . we note that it is difficult to compare the approximations without having available a gold standard . it can be seen that pdbil produces the most precise inferences for the parameters @xmath168 and @xmath251 . despite being able to reduce the abc tolerance , the abc is method appears to be the least precise . this is in contrast to the results for the simulated data in appendix c of the supplemental article @xcite , where abc is produced the most precise results . regression adjustment was also applied to the abc ii methods in an attempt to reduce the effect of the abc tolerance . these adjustments were applied individually to @xmath253 and @xmath254 [ see appendix c of the supplemental article ( drovandi , pettitt and lee , @xcite ) ] and the results are shown in figure [ fig : cat_ii_compare_reg ] . the regression adjustment does increase the precision of the abc ii posteriors . the regression adjustment appears to shift the modes of the abc ii results slightly for @xmath168 . for @xmath251 , the regression adjustment brings the abc ii results closer to that obtained by pdbil for @xmath205 . this paper has provided an extensive comparison of pbii methods , from theoretical , practical and empirical perspectives . we discovered that the pdbil method of @xcite is fundamentally different to abc ii approaches developed in the literature . more specifically , we showed that pdbil can produce better approximations by increasing the size of the simulated datasets as long as the auxiliary model provides a useful replacement likelihood for the generative likelihood for a variety of @xmath5 values . in contrast , abc methods ( including those that use ii to form the summary statistic ) should simulate datasets the same size as the observed . the pdbil method has the additional advantage of not having to determine an appropriate abc tolerance . furthermore , we found that increasing the size of the simulated dataset beyond that of the observed does not necessarily make computation infeasible due to the increase in statistical efficiency . however , it is of interest to determine the size of the simulated dataset upon which negligible improvement will be obtained . this requires further research . we have also established that bil is a rather flexible framework since the synthetic likelihood approach of @xcite is a pbil method that applies a parametric auxiliary likelihood to the summary statistic likelihood while abc can be recovered by selecting a specific nonparametric auxiliary model . our focus in this paper has been on the pbil method where a parametric auxiliary model is proposed for the full data likelihood . however , the ideas in this paper may carry over to when the auxiliary model is applied to a summary statistic likelihood as in @xcite . for the pbil method to have some chance of a good approximation to the true posterior for the specified generative model , it is important that the auxiliary model is able to well fit data simulated from the generative model for parameter values within nonnegligible posterior regions , at least in the majority of simulations . it would be possible to perform a goodness - of - fit test on the auxiliary model for every dataset generated from the proposed model during the mcmc pbil algorithm in order to assess the usefulness of the auxiliary model in the context of the pbil method . this is the subject of further research . in this paper , we have not addressed the issue of which abc ii method provides the best approximation . abc is is much faster ( when the auxiliary score vector is analytic ) and requires only weak assumptions , but did not always outperform the other abc ii methods in the examples considered in this paper . the abc ip and abc il methods differ only in their discrepancy function and it is not clear if one discrepancy function dominates the other across applications . furthermore , it remains unknown if the auxiliary parameter estimate or auxiliary score carries the most information in the observed data . it could be that the optimal choice of abc ii approach is problem dependent . until further research is conducted , we suggest trying all three methods ( assuming that abc ip and abc il are computationally feasible ) . one approach to speed up abc ip and abc il might be to start with a computationally simple but consistent estimator ( e.g. the method of moments ) and apply one iteration of a newton raphson method to produce an asymptotically efficient estimator ( @xcite , page 308 ) in a timely manner . from a practical perspective , these methods have led to improved approximate analyses for two substantive problems compared with that obtained in @xcite and @xcite . across applications considered in this paper , abc is was the most computationally efficient and led to good posterior approximations . overall , pdbil avoids having to choose an abc discrepancy function and the abc tolerance . if an auxiliary model can be proposed that satisfies a rather strong condition , more precise inferences can be obtained by taking @xmath13 large , which we showed is still computationally feasible with mcmc pdbil up to a point . however , abc ii appears to provide a more general framework for pbii problems , due to the extra flexibility of being able to incorporate additional summary statistics outside the set formed from the auxiliary model and potentially providing better approximations when the auxiliary model is a simplified version of the generative model . it is this extra flexibility that may see abc ii as the method of choice as ever - increasingly complex applications are encountered . the first two authors were supported by the australian research council discovery projectdp110100159 . the authors would like to express their gratitude to the referees and editorial team associated with this paper . the comments and suggestions provided have led to a substantially improved paper . the authors would like to thank edwin michael and david denham for access to the macroparasite data . the first author is grateful to ewan cameron , richard everitt , dennis prangle , andy wood and christian robert for useful discussions about this work . the first author is also grateful to warwick university ( where this work was partially completed ) for providing some funding for a three month visit .

in this work , we show that the zero field excitation spectra in the quantum spin ice candidate pyrochlore compound ybtio is a continuum characterized by a very broad and almost flat dynamical response which extends up to mev , coexisting or not with a quasi - elastic response depending on the wave - vector . the spectra do not evolve between 50 mk and 2 k , indicating that the spin dynamics is only little affected by the temperature in both the short - range correlated and ordered regimes . although classical spin dynamics simulations qualitatively capture some of the experimental observations , we show that they fail to reproduce this broad continuum . in particular , the simulations predict an energy scale twice smaller than the experimental observations . this analysis is based on a careful determination of the exchange couplings , able to reproduce both the zero field diffuse scattering and the spin wave spectrum rising in the field polarized state . according to this analysis , ybtio lies at the border between a ferro and an antiferromagnetic phase . these results suggest that the unconventional ground state of ybtio is governed by strong quantum fluctuations arising from the competition between those phases . the observed spectra may correspond to a continuum of deconfined spinons as expected in quantum spin liquids .

1506.01729

understanding , characterizing and classifying novel states of matter is one of the main goal of the research in solid state physics . in particular , systems where the thermal or quantum fluctuations are able to melt long range order , the so - called spin liquids , draw a lot of attention since they generally go beyond nel paradigm @xcite . the pyrochlore lattice , made of corner - sharing tetrahedra is the archetype of a three - dimensional frustrated lattice and has proven during the last years to be a rich playground for studying such spin liquid states @xcite . among the variety of possible pyrochlore systems , the compound yb@xmath0ti@xmath0o@xmath1 has been presented as one of the possible realizations of a quantum spin liquid ground - state . more precisely , it has been proposed as a candidate for the quantum variant of the spin ice state observed in ho@xmath0ti@xmath0o@xmath1 and dy@xmath0ti@xmath0o@xmath1 . in these classical spin ices , the ising - like anisotropy of the magnetic moments along the local @xmath3 directions , combined with an effective ferromagnetic interaction induce a macroscopically degenerated ground state characterized by the local ice - rule @xcite ( in each tetrahedron , two spins point in and two spins point out ) . in yb@xmath0ti@xmath0o@xmath1 , the yb@xmath4 magnetic moment shows a weak xxz planar anisotropy perpendicular to the local @xmath3 directions @xcite . however , the strongly anisotropic interactions tensor @xcite , whose main component is ferromagnetic , induces a local constraint analog to the ice - rule . the static and dynamical magnetic correlations in yb@xmath0ti@xmath0o@xmath1 have been investigated by means of neutron scattering experiments . short range correlations settle around @xmath5 k , giving rise to rods of diffuse scattering along high symmetry directions @xcite . at lower temperature , a phase transition is observed in specific heat measurements around 0.2 k @xcite , but the critical temperature depends on the nature of the samples ( polycrystal or single crystal ) and their quality @xcite . this transition was shown to be first order @xcite , in both single crystals and powder samples but its nature remains debated . several studies ( including neutron scattering and muon spin relaxation measurements ) evidence the stabilization of an ordered ferromagnetic state , with a reduced static magnetic moment @xcite , while others do not @xcite . to understand the underlying microscopic mechanism leading to such a non - conventional ground - state , several studies attempted to determine the exchange couplings , trying to reproduce the experimental results . different sets of interactions were obtained , depending on the fitted quantity : diffuse scattering @xcite or field induced spin wave excitations @xcite . the latter parameters allow reproducing specific heat @xcite , and were further refined using terahertz spectroscopy measurements @xcite . these parameters place yb@xmath0ti@xmath0o@xmath1 in a `` splayed ferromagnetic '' phase , quite far from the `` quantum spin liquid '' phase of the theoretical phase diagram @xcite . since the precise nature of the elementary excitations generally depends on the ground - state itself , we propose in this article to address the unconventional magnetic properties of yb@xmath0ti@xmath0o@xmath1 through a detailed study of the spin dynamics on a single crystal in the zero field correlated phase below 2 k , which has been little explored up to now @xcite . in section [ sec : exp ] , we first present macroscopic measurements performed on a piece of our single crystal , which show evidence for a transition at 175 mk . by means of inelastic neutron scattering , we show that the excitations spectrum is a continuum characterized by a very broad and almost flat dynamical response extending up to @xmath2 mev , and coexisting or not with a quasi - elastic response depending on the wave - vector . these excitations do not depend on the temperature below the stabilization of short - range correlations ( i.e. 2 k ) , and , in particular , entering the ordered phase does not affect the spectra . then , in section [ sec : num ] , we compare these inelastic data with classical calculations combining monte carlo and spin dynamics simulations . we show that , although the qualitative features of the spectra are reproduced , calculations predict an energy scale twice smaller than the experimental observations . these calculations are based on our determination of a reliable and robust set of exchange couplings , able to reproduce both the spin - wave spectrum of the field polarized phase as well as the diffuse scattering in zero field . the obtained couplings do place yb@xmath0ti@xmath0o@xmath1 in a ferromagnetic phase , yet very close to the boundary with an antiferromagnetic phase . then , although yb@xmath0ti@xmath0o@xmath1 is not a canonical quantum spin liquid as initially proposed , we propose that the proximity of competing ferromagnetic and antiferromagnetic correlations restore strong quantum fluctuations@xcite and is at the origin of the anomalous static and dynamical behaviors observed in our experiments . our measurements were performed on a single crystal . first , a polycristalline sample of yb@xmath0ti@xmath0o@xmath1 stoichiometry was synthesized from yb@xmath0o@xmath6 and tio@xmath0 starting powders by solid state reaction . to obtain the pyrochlore phase several thermal treatments at 1400 c with intermediate grindings were necessary . the progress of the synthesis reaction was followed by powder x - ray diffraction . for the last thermal treatment the powder was shaped as a cylinder of 5 mm diameter and 90 mm length and the obtained road was used as feed road for the single crystal synthesis . crystal growth was performed using the floating zone technique in a four - mirror optical image furnace nec sc - n15hd . the obtained crystal was then annealed under o@xmath0 gas flow for two days . to characterize the macroscopic properties of our yb@xmath0ti@xmath0o@xmath1 sample , we have performed thermodynamic measurements , magnetization and specific heat , on a flat disk sample ( mass 244 mg ) , cut from the single crystal used for neutron scattering experiments . these measurements were performed at the institut nel in purpose built experiments equipped with @xmath7he-@xmath8he dilution refrigerators . magnetization was measured in a squid magnetometer as a function of field and temperature down to 90 mk , together with ac susceptibility measurements @xcite . specific heat was measured by the relaxation method down to 70 mk in several applied fields up to 100 mt . the high sensitivity and fast response of the thermometers used in this experimental set - up as well as their high stability allow to measure both short and long time heat relaxation @xcite . the magnetic field was applied in an arbitrary direction , parallel to the plane of the disk . ( red squares ) and magnetization @xmath9 ( blue dots ) vs temperature @xmath10 . the specific heat is the long time specific heat ( see text ) and the magnetization was measured in a field of 3.83 mt , with a step of 10 mk every 1350 s.,width=321 ] the magnetization exhibits a curie - weiss temperature of 0.5 k , and reaches 1.65 @xmath11/yb at 90 mk and 3 t which is consistent with previous measurements @xcite . the specific heat presents a broad maximum around 2.5 k characteristic of short - range order correlations in agreement with previous results @xcite . when decreasing the temperature , these measurements show a transition at @xmath12 mk , in both specific heat and magnetization , which is evidenced by a peak in the specific heat , an upturn in the magnetization ( see figure [ fig_cmvst ] ) , and the onset of the out - of - phase ac susceptibility ( not shown ) . as previously observed @xcite , the magnetization shows a small hysteresis between the cooling and warming curves below the transition , indicative of a first order transition . nevertheless , although the peak in specific heat is rather narrow , the increase of the magnetization when decreasing the temperature below the transition is smooth , and fails to saturate , even at 90 mk . this result suggests that the magnetic transition expands on a broad temperature range and that no net spontaneous ferromagnetic moment is stabilized in this sample down to 90 mk . vs temperature @xmath10 for several magnetic fields . open symbols correspond to the long - time specific heat ( @xmath13 s ) and full symbols to the short time specific heat ( @xmath14 s ) . , width=283 ] when a small field is applied ( @xmath15 mt ) , the transition persists and is shifted to higher temperatures , confirming the ferromagnetic nature of the transition ( see figure [ fig_cvsh ] for specific heat ) . it is worth noting that below 1 k , the heat relaxation is not exponential , resulting in two contributions for the specific heat : the short time one ( characteristic time of about 9 s ) and the long time one ( characteristic time of about 600 s ) . these contributions show qualitatively the same temperature dependence in zero magnetic field , which evidences that dynamics with different time scales exist in the system above and below the transition . when increasing the magnetic field , the peak in the short time response is suppressed more drastically than the long one ( see figure [ fig_cvsh ] ) . the neutron scattering measurements were performed on a large yb@xmath0ti@xmath0o@xmath1 single crystal grown with the floating zone technique , as detailed above . the crystal was aligned in the @xmath16 scattering plane and cooled down to 50 mk in a dilution fridge . note that due to the large size of the crystal , its poor thermal conductivity at very low temperature , and the heat provided by the neutron beam , the true temperature of the crystal might be slightly larger . special care to the thermalisation of the sample was taken during the experiments : a waiting time of a few hours was used between the measurements to ensure a stable temperature inside the sample . polarized and unpolarized experiments were respectively conducted on the 4f1 and 4f2 triple - axis spectrometers ( llb , france ) , with final neutron wave - vectors @xmath171.3 @xmath18 , resp . @xmath18 , yielding an energy resolution of @xmath19 150 @xmath20ev , resp . @xmath19 85 @xmath20ev . a cooled beryllium filter was placed in the incident beam to remove high order contaminations . the polarization analysis allows measuring the spin - spin correlation functions @xmath21 and @xmath22 where the subscripts @xmath23 and @xmath24 indicate that the spin components are perpendicular to @xmath25 within the scattering plane and along the vertical axis respectively . both are measured in the spin flip channel , with polarization ( direction of the incident neutron s spin ) @xmath26 applied along @xmath24 and @xmath23 respectively . the flipping ratio was about fr=42 at high temperature but slightly decreased below @xmath27 k , down to fr=33 at @xmath28 k. because of technical difficulties during the polarized neutron experiment , the lowest reachable temperature was @xmath29 mk . note that the magnetic intensity measured in unpolarized experiments is the usual spin - spin correlation function @xmath30 . scattering plane at @xmath31 k. white lines correspond to the brillouin zones boundaries . black dashed lines denote the directions probed in the inelastic neutron scattering experiments ( see section [ sec : exp_ine ] ) . ( b ) elastic intensity ( full dots ) along the @xmath32 direction obtained by integrating figure ( a ) over @xmath33 rlu in the perpendicular direction . the lines with open symbols show calculated elastic scattering functions obtained from classical spin dynamics simulations for different parameter sets @xmath34 mev ( gray ) , @xmath35 mev ( blue ) and @xmath36 k ( see section [ ssec : simus ] ) . ( c ) spin flip elastic intensities ( full symbols ) along the @xmath32 direction at @xmath37 k with polarization applied along @xmath23 ( blue dots ) and @xmath24 ( red squares ) . the blue and red lines with open symbols correspond to calculated elastic scattering functions @xmath38 and @xmath39 obtained from classical spin dynamics simulations for @xmath40 mev ( see section [ ssec : simus]).,width=321 ] elastic data collected at @xmath31 k are shown in figure [ fig1:elastic ] , revealing the same qualitative features as the energy integrated diffuse scattering described at length in prior works above the critical temperature @xcite . it is characterized by a line of scattering along @xmath41 accompanied by another branch from @xmath42 to @xmath43 ending with a large spot at @xmath44 . the presence of spectral weight around both @xmath45 and @xmath43 is better evidenced in figure [ fig1:elastic ] ( b ) , where the elastic intensity is plotted versus wave - vector along the @xmath32 direction . while these rods of diffuse scattering are undoubtedly related to the presence of strongly anisotropic exchange interactions @xcite , their origin is still unclear . first ascribed to a dimensionality reduction of the static correlations @xcite , the ground - state has more recently been shown to be fully 3d in the region of parameter space where yb@xmath0ti@xmath0o@xmath1 is supposed to lie @xcite . the spectral weight along the rods is particularly intense around the @xmath46 brillouin zone center , as expected in the presence of ferromagnetic correlations : in the limit of a long range collinear ferromagnetic order , the strongest intensity is obtained for @xmath47 bragg peaks , while the @xmath48 ones are around three times weaker and the @xmath49 peaks are extinct . a finite spectral weight at @xmath50 is characteristic of antiferromagnetic spin arrangements , which points out the coexistence of both ferro- and antiferromagnetic correlations in the system . these results reveal that the short range correlations still remain at very low temperature and coexist with the long range ferromagnetic order . this might be related to the fact that , in our sample , the magnetization increases smoothly below the transition temperature . finally figure [ fig1:elastic ] ( c ) displays data in the same direction measured using polarized neutrons at higher temperature ( @xmath37 k ) . it points out that the signal around @xmath44 is polarized along @xmath24 , while the one around @xmath45 has similar intensity in both the @xmath23 and @xmath24 channels . the present data are thus consistent with the one obtained in ref . . , @xmath51 , and @xmath32 at @xmath31 k. the blue area stands for the background . dashed red and blue plain lines are the result of the fit considering a dual response consisting in quasi - elastic and inelastic contributions . each contribution is represented in red while the total is represented in blue.,width=321 ] series of inelastic data were then collected along several high symmetry directions @xmath52 , @xmath53 , @xmath54 , @xmath32 and @xmath55 at the base temperature of @xmath31 k. for all measured wave - vectors @xmath25 ( see figure [ fig1:elastic](a ) , black dots ) , the spectra show above the elastic line a very broad signal extending up to about @xmath2 mev ( see figure [ fig1:inelastic ] ) . despite the rather good energy resolution @xmath56=85 @xmath20ev , _ no well - defined collective excitations _ could be observed . this is in sharp contrast with the resolution - limited spin - waves observed in the field polarized phase of yb@xmath0ti@xmath0o@xmath1 @xcite . two qualitatively different behaviors can be observed depending on the wave - vector position relative to the rods of scattering . the spectra taken at @xmath25 vectors located on these rods are dominated by quasi - elastic fluctuations , whereas away from them , the signal appears inelastic . this feature is evidenced on the raw data presented in figure [ fig1:inelastic ] , displaying directions @xmath52 , @xmath32 and @xmath51 on the left , middle and right panels respectively . on the left panel , while `` flat - toped '' at @xmath57 , the signal becomes quasi - elastic on approaching the rod position at @xmath46 . the same feature is observed for the scans taken along @xmath32 : the line - shape of the spectra becomes quasi - elastic - like while going through the rod position at @xmath42 and at @xmath43 . finally , along @xmath58 , i.e. away from the rods , an inelastic line - shape is observed . a phenomenological fit of the data can be performed considering a dual response consisting in quasi - elastic and inelastic contributions multiplied by the detailed balance factor ( see figure [ fig1:inelastic ] , plain and dashed lines ) . since the experimental resolution is one order of magnitude smaller than the observed excitations , the fitting function was not convolved with the experimental resolution function . the fitting procedure is as follows . first , the width @xmath59 mev of the quasi - elastic response is evaluated using a lorentzian profile at different wave - vectors along or close to the rods of scattering . @xmath60 is then considered as wave - vector independent . in addition , a broad gaussian peak is used to model the inelastic response . the fit is able to converge only if the width of this inelastic peak is fixed to a given value . the free parameters of the fit are thus the intensity of both quasi - elastic and inelastic contributions as well as the peak position . this modeling shows qualitatively the different behaviors close to and away from the rods of scattering , respectively dominated by quasi - elastic and inelastic response . note that , in such an analysis , the inelastic contribution is peaked at about 0.5 mev and seems not to disperse . these inelastic data , and especially the flat energy dependence , are for the most part consistent with earlier inelastic results obtained below @xmath61 mev around two positions only , @xmath62 and @xmath63 ( see figure 4 in reference ) . the present results demonstrate that the on- and off - rods behaviors can be generalized throughout the brillouin zone . furthermore they show that the dynamical response extends up to @xmath2 mev ( see appendix [ annexe : neutron ] for a further detailed analysis ) . interestingly , the spectra recorded at higher temperature do not show any change compared to the low temperature data , indicating that the spin dynamics is only little affected by the increase of temperature up to 850 mk . this remains valid whether @xmath25 lies close to the rods of diffuse scattering or not ( see figure [ fig5a : tk ] ( a , b , d ) and ( c , e ) respectively ) . although the lack of temperature dependence of the `` off - rod '' scattering is consistent with ref . , the reported `` on - rod '' depletion at energies less than 0.2 mev is not observed in our data . the spectra remain quasi - elastic . as a result we can not conclude on a slowing down of the spin fluctuations on entering the ordered phase . this difference may be explained by the low transition temperature observed in our sample , as denoted in section [ sec : exp_mag ] . inelastic polarized neutron measurements have also been performed at higher temperature ( @xmath64 and 4.5 k ) at @xmath65 . the data , shown in figure [ fig5b : tk ] , first confirm the magnetic nature of the inelastic signal . at @xmath66 k , the full polarization analysis points out that the @xmath67 contribution is a little bit more intense than the @xmath68 . surprisingly , it also appears that @xmath68 ( red curve ) vanishes above @xmath69 mev while @xmath67 ( blue curve ) extends up to @xmath70 mev . a depletion of the spectrum is also observed at low energies , making the generalized susceptibility @xmath71 quite narrow , with a maximum around @xmath72 mev ( see figure [ fig5b : tk ] ( c ) ) . interestingly , this approximately well defined excitation observed at @xmath66 k spreads out and becomes less defined with decreasing temperature ( cf . figure [ fig5b : tk](b ) ) . at @xmath31 and 2 k , @xmath73 is superimposed , showing that the spectrum does not evolve anymore below 2 k ( see panel ( c ) ) . to 850 mk.,width=302 ] in summary , this study of yb@xmath0ti@xmath0o@xmath1 points out very unconventional spin dynamics . starting from the high temperature , the spin excitations spectrum broadens below @xmath74 k. this is largely unexpected , since , in general , the width of inelastic spectra , related to the rate of the spin fluctuations , tends to increase with increasing temperature . the crossover temperature @xmath75 between the two regimes coincides with the specific heat bump observed around @xmath76 k @xcite , below which short range spin ice like correlations establish @xcite and the rods of diffuse scattering rise @xcite . at low temperature , no well defined excitations are observed but rather a continuum of excitations . finally , entering the ordered phase does not affect the spectra . taken at @xmath65 and @xmath66 k. red squares , blue circles and green triangles respectively denote the spin flip contribution in the @xmath24 , @xmath23 and @xmath77 polarization channels . the inset shows the calculated spectra , see section [ ssec : simus ] at @xmath37 k. ( b ) spin flip intensity with polarization along @xmath77 at 4.5 ( red squares ) and 2 k ( blue circles ) . ( c ) shows the imaginary part of the magnetic susceptibility @xmath73 at @xmath65 for 4.5 ( red squares ) , 2 ( blue circles ) and 0.05 k ( green triangles).,width=321 ] these experimental results are now confronted to numerical calculations . first , a model of exchange interactions at play in yb@xmath0ti@xmath0o@xmath1 is proposed . based on rpa numerical simulations @xcite , we show that a large range of parameters accounts for the available neutron data obtained in the conventional polarized phase induced by applying magnetic fields of 2 and 5 t . reproducing the strong quasi - elastic scattering around @xmath44 with monte carlo spin dynamics simulations @xcite allows however to constrain further the set of parameters and determine an optimal set of coupling constants . in zero field , these simulations show that yb@xmath0ti@xmath0o@xmath1 is a canted ferromagnet , which yet lies very close to an antiferromagnetic phase . finally , rpa and classical spin dynamics calculations are presented . although the latter reproduce some of the experimental features , they both fail to explain the very large energy range of the experimental data . this points out the role of quantum fluctuations that may be amplified by the proximity to a competing antiferromagnetic phase , and which are not captured using a classical approach . the starting point is the widely accepted hamiltonian for pyrochlore systems @xcite : @xmath78 @xmath79 , is the zeeman term , with @xmath80 the applied magnetic field and @xmath81 the magnetic moment at site @xmath82 . @xmath83 is the crystal electric field ( cef ) hamiltonian , with @xmath84 the stevens operators and @xmath85 scalar parameters @xcite . several estimations of the @xmath86 parameters may be found in the literature @xcite . in the present paper , we use the cef parameters @xmath87 k obtained in ref . . these parameters lead to a doublet cef ground - state , separated from the first excited level by an energy gap @xmath88 k. the magneto - crystalline anisotropy of the ground - state doublet is xxz - like , with an easy plane perpendicular to the local @xmath89 directions . the @xmath90 ratio denotes a weak in - plane anisotropy with land factors @xmath91 and @xmath92 . @xmath93 is a bilinear coupling hamiltonian , where the interaction tensor @xmath94 couples next neighbors magnetic moments @xmath95 at sites @xmath82 and @xmath96 . note that @xmath97 gathers all possible physical ( e.g. exchange and dipolar coupling restricted to nearest neighbors ) , whether they are isotropic or not . by symmetry arguments , the 9 coupling constants of the @xmath98 tensor @xmath94 are reduced to only four@xcite . here , we assume an exchange tensor which is diagonal in the ( @xmath99 frame linked with a r - r bond @xcite : @xmath100 since the energy gap between the cef ground - state and the first excited levels are order of magnitudes larger than the exchange interactions and the zeeman term , it is possible to define effective spin @xmath101 operators , denoted by @xmath102 , by projecting the full moment @xmath103 onto the cef ground - state doublet . as a result , an effective hamiltonian can be defined in terms of new anisotropic couplings between the @xmath102 spin components : @xmath104 a popular convention used in ref consists in using @xmath105 defined as : @xmath106 \end{aligned}\ ] ] where @xmath107 are c - numbers @xcite . note that `` sanserif '' notations refer to local bases . @xmath105 are related to @xmath108 by the following relations @xcite : @xmath109 with @xmath110 . in the following , we will follow the alternative convention of ref , using the set of effective parameters @xmath111 ( correspondence between the different conventions are given in appendix [ annexe : conversion ] ) : @xmath112 the zero - temperature phase diagram for the nearest - neighbor pseudo spin-1/2 hamiltonian given by eq . ( [ eq : effective ] ) has been studied in ref . . owing to the xxz nature of the yb@xmath4 magnetic anisotropy , and depending on the sign and amplitude of the exchange parameters , four different magnetic phases are obtained in the phase diagram : a canted ( or splayed ) ferromagnetic phase ( labeled fm ) , the antiferromagnetic palmer - chalker phase which results from the so - called @xmath113 representation @xcite , and the basis states @xmath114 and @xmath115 of the @xmath116 representation @xcite , whose combination form the classically continuously degenerated manifold . figure [ fig7:phasediag ] shows a sketch of the @xmath114 , @xmath115 and fm spin configurations on one tetrahedron . as quoted previously , the parameters of the hamiltonian determined for yb@xmath0ti@xmath0o@xmath1 in prior studies place it in the ferromagnetic phase . the aim of the next paragraphs is to test the unicity of this set of parameters , and further explore the consequences of the location of yb@xmath0ti@xmath0o@xmath1 in the phase diagram . as already reported in ref . , a polarized state sets in when a magnetic field @xmath117 t is applied along @xmath118 $ ] . the spin dynamics consists in spin - wave excitations propagating in different directions of reciprocal space , as measured by inelastic neutron scattering for @xmath119 and 5 t @xcite . the data collected in these studies along directions @xmath41 , @xmath54 , @xmath120 , @xmath121 and @xmath122 are shown in figure [ fig : swfield](a ) . four modes can be distinguished in the experimental spectra , one of them being almost flat throughout the brillouin zone into the @xmath123 plane . to determine the exchange couplings , we solve the general hamiltonian ( [ gen_ham ] ) in the rpa approximation @xcite , and perform exhaustive calculations of the spin - spin correlation function @xmath124 . the quality of the fit relies on a good matching of the location of the maximum neutron intensity in @xmath41 , @xmath54 , @xmath120 , @xmath121 and @xmath122 directions , as well as of the overall spectral weight @xcite . the calculated mean - field magnetic moment is checked to reproduce the experimental one @xcite at @xmath31 k for @xmath119 and 5 t. finally @xmath108 are transformed into @xmath111 . in contrast with prior reports , we find that this procedure does not lead to a unique solution . representative spectra are shown in figure [ fig : swfield ] ( b , c , d ) for different relevant sets . they all reproduce most of the experimental features , yet slight differences , mainly in the spectral weight distribution remain . for instance , while the spectral weight of the two higher branches in the @xmath54 direction is minimum at @xmath125 , the inverse is observed in the simulations . more generally , the spectral weight distribution of the modes is never fully reproduced , although the agreement seems slightly better in the intermediate parameter range ( e.g. figure [ fig : swfield ] ( c ) ) . the sets of parameters giving such good agreement are reported in figure [ fig : sets ] . clearly , there is a linear relation between the @xmath126 s , so that the 4 parameters space is now reduced to a one dimensional space . colored areas delimit the uncertainties . note also that , since both magnetic fields stabilize the same ferromagnetic ground - state , fitting the @xmath119 and 5 t data provides similar constraints in the parameter space , resulting in the same linear relation between coupling constants . for this reason , the parameters obtained from the @xmath119 and @xmath127 t spectra are not distinguished and represented together in figure [ fig : sets ] . it must be emphasized that the set of parameters reported in ref . is consistent with our determination . it should be stressed also that a number of parameters do not lead , in zero field , to the expected ferromagnetic phase ( labeled `` fm '' in figure [ fig : sets ] ) but to the @xmath115 antiferromagnetic ground - state . the phase boundary separating the @xmath115 and the fm phases at @xmath128 , while exploring the one dimensional parameter space described above , is depicted in figure [ fig : sets ] by a vertical dashed line and corresponds to @xmath129 mev . and 5 t. exchange interactions reported in ref . are indicated . those reported in ref . are out of the parameter range of the figure , and are instead indicated in the fig . [ corres ] of appendix [ annexe : conversion].,width=359 ] in the @xmath130 plane obtained from spin dynamics simulations for @xmath131 mev at @xmath36 k.,width=321 ] to further constrain the dimension of parameter space , other experimental results have to be taken into account . to this end , @xmath132 was calculated using a combination of monte carlo and spin dynamics simulations at finite temperature , aiming at reproducing the rods of diffuse scattering . this numerical method is briefly described in appendix [ annexe : spindynamics ] and more details may be found in ref . . calculations are based on the effective hamiltonian ( [ eq : effective ] ) and were performed in the short - range correlated regime at @xmath36 k , just above the critical temperature , for different coupling sets within the one dimensional parameter space determined above . those numerical simulations ( see figure [ fig : diffus ] ) show that the rods of scattering around @xmath46 and @xmath43 are simultaneously observed when approaching the fm / afm phase boundary ( in agreement with ref . ) . this confirms the observation made in section [ sec : exp_ela ] of coexisting fm and afm short range correlations . the cuts in direction @xmath32 shown in figure [ fig1:elastic ] ( b ) and ( c ) further show that a good qualitative agreement between the experimental data ( filled circles ) and the spin dynamics simulations ( empty circles ) may only be achieved while lying very close to the fm / afm phase boundary . combining the different results , the best agreement is obtained for @xmath133 mev ( grey area in figure [ fig : sets ] ( d ) ) . the associated curie - weiss temperature is @xmath134 k , in excellent agreement with the experimental value @xmath135 k determined in section [ sec : exp_mag ] ( see also ref . therein ) . it is worth noting that this set of values locates yb@xmath0ti@xmath0o@xmath1 closer to the afm / fm phase boundary than the one obtained in ref . . it is also quite different from that obtained by thompson _ et al _ @xcite as well as that reported by chang _ et al _ @xcite . actually , the latter studies yield spin wave spectra in strong disagreement with experiment , while our set reconciles both the diffuse scattering and inelastic neutron data . , ( b ) @xmath115 and ( c ) fm phases . ( d ) phase diagram calculated as a function of temperature and of exchange coupling varying along the optimal one dimensional parameter space determined from the spin - wave fitting procedure . the black line roughly indicates the range of parameters for which the transition toward the ferromagnetic order is first order.,width=321 ] we now examine in more detail the consequences of the above results on the zero field ground state . to this end , the phase diagram within the 1d parameter space discussed in the previous section is calculated based on monte carlo simulations , for different lattice size up to @xmath136 , @xmath137 ( see figure [ fig7:phasediag ] ) . since the critical temperatures does not seem to evolve much with the lattice size for @xmath138 ( not shown ) , the phase boundaries have been roughly determined by locating the maximum of the specific heat for @xmath139@xcite . for the optimal values @xmath140 close to the boundary located at @xmath141 , the specific heat and magnetic susceptibility show , upon cooling , a succession of several transitions . two second order transitions first occurs at about @xmath142 and @xmath143 k towards the @xmath114 and @xmath115 afm states respectively . this is followed by an abrupt transition at @xmath144 k towards a fm state ( sketches of the magnetic configuration are shown in figure [ fig7:phasediag ] ( a , b , c ) ) . the first order character of this second transition is demonstrated by a finite latent energy at the critical temperature . the latent energy decreases while getting deeper into the fm state , such that the first order transition smoothly transforms into a second order one . this behavior while approaching the phase boundary is consistent with the study by yan et al . @xcite . both the reentrant behavior of the fm phase at low temperatures and the evolution of the order of the transition on approaching the phase boundary might result from thermal order by disorder : although the ferromagnetic state has a lower internal energy for @xmath145 , the @xmath146 phases may be favored in a small temperature range because of their strong antiferromagnetic fluctuations ; the long - range ferromagnetic order is then stabilized at lower temperatures where thermal fluctuations are reduced . we finally turn to the calculations of the classical spin dynamics aiming at a final discussion of the experimental results . @xmath147 was first calculated in the rpa approximation . spectra have been averaged over equi - populated domains . they are presented in figure [ fig10:sdmaps](c ) for different high - symmetry directions in reciprocal space . for comparison , experimental data in the same directions are shown in panel ( a ) . calculations have also been performed for different exchange interactions around the fm / afm phase boundary , along the 1d - parameter space determined in the previous section . while the excitations are `` gapless '' in the afm phase ( we did not consider order by disorder or crystal field effects at this point , which could however be at the origin of a small spin gap ) , an anisotropy gap opens in the fm region because of a sizeable tilt of the spins from their easy - plane ( see figure [ fig10:sdmaps](c ) ) . obviously , at this level of approximation , the model fails to reproduce the experimental data . actually , the calculated spectra consist in conventional spin wave excitations , which can not capture the flat line shape of the experimental spectra . a more striking discrepancy especially affects the energy scale , which is about twice larger experimentally than in the simulations at most wave - vectors . in the simulations , the highest spin - wave branch is observed at @xmath72 mev , a value which does not evolve much by varying the exchange parameters . experimentally , the spectral weight reaches energies @xmath148 mev depending on the wave - vector ( see figure [ fig1:inelastic ] ) . note that such a difference can not be explained by the experimental resolution , which is around @xmath149 mev for @xmath150 and an energy transfer @xmath151 mev . along high symmetry directions : ( a ) neutron data ( this work ) taken at @xmath31 k ; ( b ) calculated spectra obtained from classical spin dynamics simulations in the short - range correlated regime ( @xmath36 k ) ; ( c ) rpa spectra obtained in the fm phase.,width=321 ] to go a step further , the classical spin dynamics was solved numerically by taking into account non - linear effects associated with thermal fluctuations ( see appendix [ annexe : spindynamics ] and ref . for technical details about the method ) . such spin dynamics simulations of @xmath147 have been performed at different temperatures from @xmath152 to @xmath153 k , on the basis of the effective hamiltonian ( [ eq : effective ] ) and optimal exchange parameters @xmath140 proposed in the previous section . intensity maps of @xmath147 are shown figure [ fig10:sdmaps ] ( b ) for @xmath36 k in high symmetry directions . the simulations succeed in reproducing the overall shape of the excitation spectrum observed experimentally : @xmath154 in direction @xmath52 , the excitation spectrum goes from quasi - elastic approaching @xmath155 to flat - toped excitations at @xmath156 ; @xmath157 the excitations are mainly flat or weakly inelastic along direction @xmath51 ; @xmath158 in direction @xmath32 , a strong quasi - elastic signal overwhelm the spectra between @xmath42 and @xmath43 . more generally , the computed @xmath147 displays a quasi - elastic ( resp . flat - toped / weakly inelastic ) line - shape close to ( resp . away from ) the rods of scattering , as observed experimentally . this is illustrated figure [ fig11:sdvst ] , displaying the calculated scattering function for different temperatures @xmath159 and @xmath153 k , and wave - vectors @xmath42 , @xmath43 , @xmath160 , and @xmath156 . while the two first positions are dominated by a strong quasi - elastic contribution , the latter are mostly inelastic . the experimental data measured at the same wave - vectors are shown in the insets . moreover , the calculated spectra do not evolve much with temperature below @xmath161 k. experimentally , this behavior is even more impressive since the spectra do not depend on temperature between @xmath31 and 2 k. however , despite this success , the spin dynamics simulations fail to reproduce the large energy range , which does not exceed @xmath162 mev in the calculations , as also observed in rpa simulations . obtained from classical spin dynamics simulations at different temperatures @xmath159 and @xmath153 k , and for relevant wave - vectors @xmath42 , @xmath43 , @xmath160 , and @xmath156 . insets show the neutron data at the same wave - vectors.,width=321 ] in summary , our results show that for the optimal values of coupling constants determined to reproduce the field induced spin wave spectra , classical calculations captures some of the magnetic properties of yb@xmath0ti@xmath0o@xmath1 . it reproduces different experimental features , such as the non - trivial structure of the diffuse scattering which points out coexisting fm and afm correlations above @xmath163 , as well as the transition towards a long - range ordered canted ferromagnetic state at @xmath163 . at lower temperature , thermal afm fluctuations have then been shown to affect the stability of the fm order : approaching the boundary from the fm side , the order of the transition smoothly goes from second to first order ; sufficiently close to the boundary , a long ranged antiferromagnetic order may even be favored in a significant temperature range above @xmath163 . this succession of phase transitions observed close to the phase boundary could clarify the peculiar shape of the magnetization curve as a function of temperature on a single crystal recently reported ( see figure 2(a ) of ref . ): with decreasing temperature , it shows first a reversible bump around @xmath164 k , followed by the irreversible step at @xmath165 k associated with the long range ferromagnetic order . however , the classical approach fails to explain the strongly reduced ordered moment observed experimentally below @xmath163 as well as the unconventional nature of the spin dynamics , not only at very low temperatures , but up to @xmath166 k. both features are obviously connected , since the part of the moment which is not condensed in the static component remains , by essence , fluctuating . to explain this unconventional spin dynamics , the coupling to other degrees of freedom can not be ruled out . this phenomena has been shown to be of utmost importance in tb@xmath0ti@xmath0o@xmath1 for instance @xcite , but there is , to our knowledge , no experimental evidence in favor of such a coupling in yb@xmath0ti@xmath0o@xmath1 . moreover , the unconventional dynamics appears around 2 k which coincides with the apparition of short range correlations , which is a clue for a purely magnetic origin . it is thus tempting to invoke quantum effects as the missing ingredient in the proposed model . indeed , the proximity with the afm magnetic states is likely a source of enhanced quantum fluctuations . to account for the unconventional spin dynamics , two scenarii come to mind : @xmath154 the fractionalization of magnons into deconfined new spin 1/2 particles and @xmath167 ) the scattering by longitudinally polarized two - magnons excitations . the former case is very well known in one dimensional systems @xcite . here the so - called spinons , which can be seen as domain walls , propagate freely and separate two `` nel '' configurations @xcite . in two and three dimensions , the situation is more complex as the interactions create an attractive potential between spinons that bind them to form conventional magnons @xcite . the quasi - degeneracy of the fm and afm phases in the relevant part of the phase diagram phase for yb@xmath0ti@xmath0o@xmath1 could cancel these interactions and release the spinons . in this view , the continuum would be characterized by the absence of edge singularity at its lower threshold , with the exception of the `` on rod '' q positions , as revealed by the quasi - elastic like shape . actually , classical examples of continuum demonstrate that both situations are possible : the classical de cloizaux - person spectrum in the case of the spin 1/2 heisenberg chain , shows an edge singularity at the lower threshold @xcite , the spectra decreasing roughly as @xmath168 ( the exact exponent depends on the anisotropy @xcite ) . however , for `` xxz '' chains , close to the ising limit of a strong coupling along the z - axis , have a spectrum determined by gapped solitons , with no singularity at the edges @xcite . theoretical studies have shown that the essence of the singularity relies on the fact that many low - lying modes add up @xcite . in this picture , recent theoretical work argue that the spontaneous magnetization along @xmath169 below the critical temperature should confines the spinons by creating a string tension between them@xcite . this should result in the discretization of the two - spinon continuum into multiple spin - wave branches , a feature which is however not observed in the data , that especially show no significant evolution below @xmath163 . alternatively , since a strongly reduced ordered momentum is the sign for significant non - linear longitudinal excitations , the scattering by a multi - magnons continuum should be considered . in this case , the ground state is long range ordered , with propagating magnons as the elementary ( transverse ) quasi - particles . in the longitudinal channel , however , fluctuations along the length of the ordered moment may develop @xcite . experimental realizations are however quite scarce and usually give rise to a kind of tail above the spin wave dispersion which is still very well defined . for this reason , we believe that this second scenario is less relevant . it should be stressed finally that the measurements performed at @xmath66 k ( see figure [ fig5b : tk ] ) show that the spectra tend to become more conventional when increasing the temperature . further inelastic neutron scattering measurements should be performed as a function of temperature , to look for an eventual transition or cross - over between quantum and classical regimes , with spin dynamics respectively governed by quantum and thermal fluctuations . in this work , we show that the short range correlations that develop below 2 k in yb@xmath0ti@xmath0o@xmath1 come along with non - conventional excitations : no well defined spin waves are observed in both the short range correlated and long range ordered regimes ; rather , excitations spectra are characterized by a very broad and almost flat dynamical response which extends up to @xmath2 mev and does not evolve with temperature below 2 k , sometimes coexisting with a quasi - elastic response depending on the wave - vector . based on monte carlo spin dynamics and rpa calculations , we determine a new set exchange couplings that allows to reproduce both the diffuse scattering in the short range correlated regime and the spin wave spectra observed in the field polarized phase . this careful determination of the exchange tensors places yb@xmath0ti@xmath0o@xmath1 in the close vicinity to a fm / afm phase boundary , and , in any case , far from the canonical quantum spin liquid state . this location points out the possible role of quantum fluctuations arising from these two competing phases . we show further that conventional rpa fails to reproduce the experimental excitation spectra at low temperature . spin dynamics simulations performed in the short range correlated regime reproduce however some features of the excitation spectra but lead to an energy bandwidth which is twice smaller than the experimental observations . we speculate that quantum fluctuations between fm and afm phases govern the spin dynamics in yb@xmath0ti@xmath0o@xmath1 and especially that the observed spectra rather correspond to a continuum of deconfined spinons as expected in quantum spin liquids , than simple paramagnons . we acknowledge m. gingras and l. jaubert for fruitful discussions . we thank c. paulsen for allowing us to use his squid dilution magnetometers . we also thank ph . boutrouille , ( cryogeny group at llb ) for his technical help while setting up he dilution fridge . to further analyze the continuum of excitations observed in neutron experiments , the data can be modeled with the following function : @xmath170 \sum_{i=0, .. ,6 } r_i(\omega)-r_i(-\omega)\ ] ] @xmath171 is a constant background , @xmath172 a standard gaussian profile of width @xmath56 to model the incoherent elastic scattering at @xmath173 , @xmath174 is the detailed balance factor and @xmath175 a series of flat spectral bands with a fixed width of @xmath176 @xmath20ev , and centered at @xmath177 and @xmath178 mev ( see top of figure [ band ] for an example ) : @xmath179 the different spectra were analyzed through this model and the relative weights @xmath180 are reported in figure [ band ] as a function of @xmath25 for the high symmetry directions . those weights sequence in decreasing order as @xmath25 approaches the rods positions , close to @xmath46 , @xmath45 , @xmath181 and @xmath43 ( see the orange bars ) , while the weights @xmath182 become roughly equal away from the rod . the spectra then appear `` flat toped '' . along @xmath183 a weak maximum can be detected for the @xmath184 band ( @xmath185 mev ) . these inelastic data , and especially the flat energy dependence demonstrates that on- and off - rods is a general feature of the spin dynamics throughout the brillouin zone . furthermore , the measurements show that the dynamical response extends up to @xmath2 mev . . bottom : relative weights @xmath180 of the different bands for the data recorded at base temperature along @xmath52 , @xmath51 , @xmath54 , @xmath32 and @xmath55 . @xmath186 is defined via the spectral function @xmath187 \sum_{i=0, .. ,6 } r_i(\omega)-r_i(-\omega)$ ] , where @xmath175 is constant , equal to @xmath186 in a given energy range centered at @xmath188 and @xmath189 mev , with a fixed width of @xmath176 @xmath20ev.,width=321 ] the widely accepted starting point is given by the following hamiltonian @xmath190 @xmath79 , is the zeeman term , with @xmath80 the applied magnetic field and @xmath81 the magnetic moment at site @xmath82 . @xmath191 is the crystal electric field ( cef ) hamiltonian and @xmath192 is a bilinear coupling hamiltonian , where the interaction tensor @xmath193 couples next neighbors magnetic moments @xmath95 at sites @xmath82 and @xmath96 . by symmetry arguments , the 9 coupling constants of the @xmath98 tensor @xmath193 are reduced to only 4 @xcite . here , we assume an exchange tensor which is diagonal in the ( @xmath99 frame linked with a r - r bond @xcite : @xmath100 alternatively , we can use @xmath194 defined by the simple relations : @xmath195 since the energy gap between the cef ground - state and the first excited levels are order of magnitudes larger than the exchange interactions and the zeeman term , it is possible to define effective spin @xmath101 operators , denoted by @xmath102 , by projecting the full moment @xmath103 onto the cef ground - state doublet . the effective spin 1/2 hamiltonian reads : @xmath196 a popular convention consists in using @xmath105 defined as : @xmath106 \end{aligned}\ ] ] where @xmath107 are numbers defined in ref . note that `` sanserif '' notations refer to local bases . the connection between @xmath197 and @xmath198 is realized using the anisotropic land factor @xmath199 @xcite . defining @xmath110 : @xmath109 alternatively , a set of effective parameters @xmath111 can be used @xcite : @xmath112 figure [ corres ] provides the different sets of parameters and correspondence between these 4 conventions able to capture the spin wave spectra measured in the field polarized phase of yb@xmath0ti@xmath0o@xmath1 . the correspondence between the optimal sets , obtained by taking into account the fit of the zero field diffuse scattering , is finally given in table [ param ] . .optimal sets of exchange in yb@xmath0ti@xmath0o@xmath1 [ cols="^,^,^,^ " , ] ) .,width=321 ] we consider the heisenberg model @xmath200 , where the summation is limited to nearest neighbors , @xmath201 is the interaction tensor , and @xmath202 are classical pseudo - spins located at the pyrochlore sites . our interest lies in the time evolution of the spin - pair correlations emerging in such a model . it is convenient to probe such dynamical correlations by calculating the dynamical scattering function @xmath203 , which can be done by combining monte carlo and spin dynamics methods . the classical dynamics of the pseudo - spins is described by the non - linear bloch equations @xmath204 where sites @xmath96 are the nearest neighbors of @xmath82 . these equations of motions were numerically integrated using an 8th - order explicit runge - kutta ( rk ) method with an adaptive step - size control , offering an excellent compromise between accuracy and computation time . the initial spin configurations used for the numerical integration are generated at each temperature by a hybrid monte carlo method using a single - spin - flip metropolis algorithm @xcite . a thousand spin configurations are used at each temperature to evaluate the ensemble average in @xmath203 while the number of monte carlo steps needed for decorrelation is adapted in such a way that the stochastic correlation between spin configurations is lower than @xmath205 . the numerical results have been obtained for different lattice sizes ranging from @xmath206 ( @xmath207 ) to @xmath136 ( @xmath208 ) with periodic boundary conditions . the classical monte carlo approach described above was also used to derive thermodynamics quantities such as the specific heat and the magnetic susceptibility , which allowed us to determine the phase diagram presented in figure [ fig7:phasediag ] . , edited by c. lacroix , p. mendels , and f. mila ( springer - verlag , berlin , 2011 ) . j. s. gardner , m. j. p. gingras , and j. e. greedan , rev . mod . phys . * 82 * , 53 ( 2010 ) . m. j. harris , s. t. bramwell , d. f. mcmorrow , t. zeiske , and k. w. godfrey , phys . lett . * 79 * , 2554 ( 1997 ) . h. cao , a. gukasov , i. mirebeau , p. bonville , c. decorse , and g. dhalenne , phys . letters , * 103 * , 056402 ( 2009 ) ; j. a. hodges , p. bonville , a. forget , m. rams , k. krlas , and g. dhalenne , j. phys . : condens . matter * 13 * , 9301 ( 2001 ) . b. z. malkin , a. r. zakirov , m. n. popova , s. a. klimin , e. p. chukalina , e. antic - fidancev , ph . goldner , p. aschehoug , and g. dhalenne , phys . b * 70 * , 075112 ( 2004 ) . h. b. cao , a. gukasov , i. mirebeau , and p. bonville , j. phys . matter * 21 * , 492202 ( 2009 ) . k. a. ross , l. savary , b. d. gaulin , and l. balents , phys rev x * 1 * , 021002 ( 2011 ) . j. d. thompson , p. a. mcclarty , h. m. rnnow , l. p. regnault , a. sorge , and m. j. p. gingras , phys . lett . * 106 * , 187202 ( 2011 ) . k. a. ross , j. p. c. ruff , c. p. adams , j. s. gardner , h. a. dabkowska , y. qiu , j. r. d. copley , and b. d gaulin , phys . lett . * 103 * , 227202 ( 2009 ) . k. a. ross , l. r. yaraskavitch , m. laver , j. s. gardner , j. a. quilliam , s. meng , j. b. kycia , d. k. singh , th . proffen , h. a. dabkowska and b. d. gaulin , phys . b * 84 * , 174442 ( 2011 ) . h. w. blte , r. f. wielinga , and w. j. huiskamp , physica ( amsterdam ) * 43 * , 549 ( 1969 ) . a. yaouanc , p. dalmas de r ' eotier , c. marin , and v. glazkov , phys rev b * 84 * 172408 ( 2011 ) . r. m. dortenzio , h. a. dabkowska , s. r. dunsiger , b. d. gaulin , m. j. p. gingras , t. goko , j. b. kycia , l. liu , t. medina , t. j. munsie , d. pomaranski , k. a. ross , y. j. uemura , t. j. williams , and g. m. luke , phys rev b * 88 * 134428 ( 2013 ) . e. lhotel , s. r. giblin , m. r. lees , g. balakrishnan , l. j. chang and y. yasui , phys . b * 89 * , 224419 ( 2014 ) . chang , s. onoda , y. su , y .- j . tsuei , y. yasui , k. kakurai , and m. r. lees , nat . * 3 * , 992 ( 2012 ) . j. a. hodges , p. bonville , a. forget , a. yaouanc , p. dalmas de rotier , g. andr , m. rams , k. krlas , c. ritter , p. c. m. gubbens , c. t. kaiser , p. j. c. king , and c. baines , phys . rev . * 88 * , 077204 ( 2002 ) . j. s. gardner , g. ehlers , n. rosov , r. w. erwin and c. petrovic , phys . b * 70 * , 180404(r ) ( 2004 ) . p. bonville , j. a. hodges , e. bertin , j .- ph . bouchaud , p. dalmas de rotier , l .- p . regnault , h. m. rnnow , j .- sanchez , s. sosin , and a. yaouanc , hyperfine interact . * 156 - 157 * , 103 ( 2004 ) . r. applegate , n. r. hayre , r. r. p. singh , t. lin , a. g. r. day , and m. j. p. gingras , phys . * 109 * 097205 ( 2012 ) . l. pan , s. k. kim , a. ghosh , c. m. morris , k. a. ross , e. kermarrec , b. d. gaulin , s. m. koohpayeh , o. tchernyshyov , and n.p . armitage , nat . commun . * 5 * , 4970 ( 2014 ) . l. savary , k. a. ross , b. d. gaulin , j. p. c. ruff , and l. balents , phys rev letters * 109 * , 167201 ( 2012 ) . a. w. c. wong , z. hao , and m. j. p. gingras , phys rev b * 88 * , 144402 ( 2013 ) . h. yan , o. benton , l. d. c. jaubert , and n. shannon , arxiv:1311.3501 ( 2013 ) . c. paulsen , in _ introduction to physical techniques in molecular magnetism : structural and macroscopic techniques - yesa 1999 _ , edited by f. palacio , e. ressouche , and j. schweizer , servicio de publicaciones de la universidad de zaragoza , zaragoza , 2001 , p. 1 . f. zougmor , j. c. lasjaunias and o. bthoux , j. phys . france * 50 * , 1241 ( 1989 ) ; j. c. lasjaunias , s. sahling , k. biljakovic , and p. monceau , j. low temp . 130 * , 25 ( 2003 ) . s. t. bramwell , m. n. field , m. j. harris and i. p. parkin , j. phys . : condens . matter * 12 * , 483 ( 2000 ) . j. d. thompson , p. a. mcclarty , and m. j. p. gingras , j. phys . : condens . matter * 23 * , 164219 ( 2011 ) . s. guitteny , s. petit , e. lhotel , j. robert , p. bonville , a. forget , and i. mirebeau , phys . b * 88 * , 134408 ( 2013 ) . s. petit , j. robert , s. guitteny , p. bonville , c. decorse , j. ollivier , h. mutka , m. j. p. gingras and i. mirebeau , phys rev b * 90 * , 060410(r ) ( 2014 ) . j. robert , b. canals , v. simonet , and r. ballou , phys . * 101 * , 117207 ( 2008 ) . m. taillefumier , j. robert , c. l. henley , r. moessner , and b. canals , phys . rev . b * 90 * , 064419 ( 2014 ) . x. tao , d. p. landau , t. c. schulthess , and g. m. stocks , phys . * 95 * , 087207 ( 2005 ) . p. h. conlon and j. t. chalker , phys . lett . * 102 * , 237206 ( 2009 ) . g. mller , phys . rev . lett . * 60 * , 2785 ( 1988 ) . malkin , t. t. a. lummen , p. .h m van loosdrecht , g dhalenne and a r zakirov , j. phys . : condens . matter * 22 * , 276003 ( 2010 ) . a. bertin , y. chapuis , p. dalmas de rotier and a. yaouanc , j. phys . : condens . matter * 24 * , 256003 ( 2012 ) . s. h. curnoe , phys . b * 78 * , 094418 ( 2008 ) . j. d. m. champion , m. j. harris , p. c. w. holdsworth , a. s. wills , g. balakrishnan , s. t. bramwell , e. cizmar , t. fennell , j. s. gardner , j. lago , d. f. mcmorrow , m. orendac , a. orendacova , d. mck.paul , r. i. smith , m. t. f. telling , and a. wildes , phys . b. * 68 * , 020401 , ( 2003 ) . although a finite size scaling could allow to determine more precisely the critical temperatures , our goal here is only to qualitative discuss the possible consequences of lying close to the fm / afm phase boundary . s. sachdev , _ quantum phase transitions _ , cambridge university press , 2011 . d. a. tennant , r. a. cowley , s. e. nagler and a. m. tsvelik , phys . b * 52 * , 13368 ( 1995 ) . b. lake , d. a. tennant , c. d. frost and s. e. nagler , nat . mater . * 4 * , 329 ( 2005 ) . b. dalla piazza , m. mourigal , n. b. christensen , g. j. nilsen , p. tregenna - piggott , t. g. perring , m. enderle , d. f. mcmorrow , d. a. ivanov and h. m. rnnow , nat . physics * 11 * , 62 ( 2015 ) . j. des cloizaux , j.j . pearson , phys . rev . * 128 * , 2131 ( 1962 ) . h. j. schulz , phys . b * 34 * , 6372 ( 1986 ) . n. ishimura , h. shiba , progr . . phys . * 63 * , 743 ( 1980 ) . mikeska , phys . b * 12 * , 2794 ( 1975 ) ; h .- j . mikeska , w. pesch : j. phys . c * 12 * , l37 ( 1979 ) ; h .- j . mikeska , e. patzak : z. phys . b * 26 * , 253 ( 1977 ) . n. b. christensen , h. m. rnnow , d. f. mcmorrow , a. harrison , t. g. perring , m. enderle , r. coldea , l. p. regnault and g. aeppli , pnas * 104 * , 15264 ( 2007 ) .

a family of operators on a complete metric space is called contractive if there exists such that for any in we have for some . stein conjectured that for any contractive family there is some composition of the operators that has a fixed point . austin gave a counterexample to this , and asked if stein s conjecture is true if we restrict to compact spaces . our aim in this paper is to show that , even for compact spaces , stein s conjecture is false .

1312.0587

let @xmath7 be a ( non - empty ) complete metric space . given @xmath8 continuous functions @xmath9 , and a real number @xmath10 , we call @xmath11 a _ @xmath12-contractive _ family if for every pair of points @xmath13 in @xmath1 there is @xmath5 such that @xmath14 . further , we say that @xmath11 is a _ contractive _ family if it is a @xmath12-contractive family for some @xmath10 . in particular , when @xmath15 is a function on @xmath1 and @xmath16 is a contractive family we say that @xmath15 is a _ contraction_. recall the well - known theorem of banach @xcite which says that any contraction on a complete metric space has a unique fixed point . + in @xcite , stein conjectured the following generalisation of the theorem of banach : + let @xmath11 be a @xmath12-contractive family on a complete metric space . then some composition of @xmath17 ( i.e. some word in @xmath18 ) has a fixed point . + in @xcite , austin constructed a counterexample to this statement . there is a complete metric space @xmath7 with a contractive family of operators @xmath19 , such that no word in @xmath20 has a fixed point . furthermore , austin asked if this is possible in a compact space . does every contractive family of operators on a compact space have a composition with a fixed point ? the purpose of this paper is to show that even with the additional assumption of compactness , there still need not be a fixed point . + [ thm1 ] there is a compact metric space @xmath7 with a contractive family of operators @xmath19 , such that no word in @xmath20 has a fixed point . we remark that our construction provides the counterexample for any given @xmath10 . since our construction of the counterexample which proves the theorem [ thm1 ] is fairly complicated , we provide some motivation for our steps . suppose that we are in the setting of theorem [ thm1 ] . given any set @xmath21 of points in our space , since @xmath1 is compact , we know that @xmath21 is bounded . let @xmath22 be its diameter . then , it is not hard to show that the diameter of one of @xmath23 or @xmath24 is at most @xmath25 . to see this , pick any point @xmath26 in @xmath21 and consider set @xmath27 and let @xmath28 . if @xmath29 , we are done , so suppose that @xmath30 . if we can find @xmath31 such that @xmath32 for all @xmath33 , then , by looking at distance form @xmath34 , the diameter of @xmath23 does not exceed @xmath25 . on the other hand , if there is no such @xmath35 , then each point of @xmath36 is on distance at most @xmath37 from some point in @xmath38 which are on distance at most @xmath37 from @xmath39 , so @xmath24 has diameter at most @xmath25 . + this simple observation leads us to the idea that instead of considering distances between each pair of points in the wanted space , thinking about diameters of sets should be much more convenient in our problem . with this in mind , we develop the notion of ` diameter spaces ' , which will play a key role throughout our construction . due to their importance , we include the proper definition . + let @xmath1 be a non - empty set . given a collection @xmath40 of subsets of @xmath1 , we call @xmath41 a _ diametrisable space _ provided the following conditions are met , 1 . given @xmath42 there is @xmath43 with @xmath44 2 . if @xmath45 and @xmath46 then @xmath47 we refer to elements of @xmath1 as points and elements of @xmath40 as _ diametrisable sets_. further , if a function @xmath48 is such that given @xmath45 with a nonempty intersection , we have that @xmath49 holds , we call @xmath50 the _ diameter _ and @xmath51 a _ diameter space_. we refer to this inequality as the _ triangle inequality for the diameter spaces_. + in a very natural way , one can use a diameter space to induce a pseudometric on the underlying space , by simply finding the infimum of diameters of all the diametrisable sets containing any two given distinct points . furthermore , by imposing suitable conditions on the diametrisable sets , one can get nice properties to hold for the pseudometric space . + in order to continue further , we must first specify the underlying set . hence , let us look for the space that should , in some vague sense , be the minimal counterexample . one of the possible ways to approach this issue is to fix a point @xmath52 , and then look what other points can we obtain . it is not hard to see that completion of the set of all images that one can get by applying @xmath15 and @xmath53 to @xmath52 is itself a compact metric space , and that @xmath54 form a @xmath12-contractive map on this subspace of @xmath1 as well . now , starting from @xmath52 we must include all the points described , and ( provided we can not get to the same point in different ways ) we can actually biject all these points that are obtained using @xmath54 from @xmath52 with finite words over a two letter alphabet . therefore , our construction will start from an underlying set @xmath1 of all finite words over @xmath55 , with obvious functions @xmath54 each of which adds one of the characters to the beginning of the word given as input . then , provided we have a metric on @xmath1 , we will take its completion , and hope that the metric space that we get , along with @xmath54 satisfies all the properties of the theorem [ thm1 ] . this is where diameter spaces come into play . we will describe the properties of a collection of diametrisable sets @xmath40 on @xmath1 , that guarantee the completion of the induced pseudometric space , along with @xmath54 , ( being the concatenation functions described ) , to be a counterexample . being a relatively long list , we refer the reader to the corollary [ sumcor ] to get the idea of what conditions we impose on @xmath40 , which will in fact be a sequence of subsets of @xmath1 , denoted by @xmath56 . in this subsection we show how one can obtain a pseudometric space from diameter spaces . the following proposition tells us how to induce pseudometric on the underlying set . let @xmath57 ) be a diameter space . define a function @xmath58 by @xmath59 when @xmath26 and @xmath35 are distinct , where infimum is taken over all diametrisable set that contain points @xmath26 and @xmath35 , and @xmath60 otherwise . then @xmath61 is well - defined and @xmath62 is a pseudometric space . firstly , suppose we are given two distinct points @xmath26 and @xmath35 . then @xmath21 , the set of all values that diameter of a diametrisable set containing @xmath13 can take , is non - empty , and bounded from below , so @xmath63 exists , and @xmath61 is well - defined . + to prove that @xmath61 is a pseudometric , we just need to show that the triangle inequality holds , since @xmath64 holds for all points @xmath26 and @xmath61 is symmetric by construction . therefore let @xmath65 be three points . if any of these are equal , we are done . otherwise , given @xmath66 we can find sets @xmath45 such that @xmath44 , @xmath67 , @xmath68 . but @xmath69 and @xmath70 intersect hence their union forms a diametrisable set , and further @xmath71 , so @xmath72 . but this holds for any positive @xmath73 proving the triangle inequality , and therefore the proposition . the pseudometric that we constructed from the diameter space will be referred to as the _ induced pseudometric _ by @xmath50 . + remark that given any metric space @xmath62 we can construct a diameter space @xmath51 by taking diametrisable sets to be the finite subsets of @xmath1 , and the diameter function @xmath50 to have the usual meaning , that is for any @xmath43 , we set @xmath74 to be @xmath75 where maximum is taken over all pairs of points in @xmath69 . then , the metric @xmath61 coincides with the pseudometric induced by @xmath50 . thus in this way , we can consider every metric space to be a diameter space at the same time . + in addition we note that any finite measure space is a diameter space as well . in the previous subsection we saw how to obtain a pseudometric space from a diameter space . to construct a counterexample to the conjecture , we will use this procedure , but as the proposition only guaranties that we get a pseudometric space , we need to add additional properties of a diameter space to ensure that we reach our goal . first of all , we work with specific underlying set @xmath1 and the diameter function which are consistent with the nature ( or more precisely the geometry ) of the problem . as we are considering essentially compositions of functions applied to an element of the given metric space satisfying the assumptions of the conjecture , we will work with points of @xmath1 being finite words over the alphabet consisting of two letters @xmath76 , including the empty word . before we proceed further , let us introduce some notation . + * notation . * if @xmath77 are two words of @xmath1 , we write @xmath78 for the word obtained by writing first @xmath79 then @xmath80 . say that @xmath79 is a _ prefix _ or an _ initial segment _ of @xmath80 , if there is another word @xmath81 with @xmath82 , and if this holds write @xmath83 . the _ length _ of a word @xmath79 , denoted by @xmath84 is the number of letters in @xmath79 . characters are considered to be words of length one simultaneously as being characters . when @xmath21 is a subset of @xmath1 we write @xmath85 and @xmath86 , when @xmath87 . given a positive integer @xmath8 and word @xmath79 , write @xmath88 for @xmath89 , where @xmath79 appears @xmath8 times . the empty word is denoted by @xmath90 . finally , we allow infinite words occasionally ( although these are not elements of @xmath1 ) , and write @xmath91 to stand for the infinite word obtained by writing consecutive copies of @xmath79 infinitely many times . + we will take our functions @xmath92 to be given by @xmath93 and @xmath94 for all words @xmath87 . then every word is actually the same as the corresponding composition of functions applied to the empty word . on the other hand , having in mind the contraction property of the family of the functions that we want to hold , we take @xmath40 to be a sequence of subsets of @xmath1 , namely @xmath95 and set @xmath96 for some fixed @xmath10 . now , we just need to specify what needs to hold for @xmath40 so that we get a counterexample . + let us start with ensuring that @xmath51 is a diameter space . we accomplish this by requiring the following property : + * a1 . * for any nonnegative integers @xmath97 if @xmath98 then @xmath99 and @xmath100 . + then we see that @xmath41 is a diametrisable space and that given any two diametrisable sets @xmath101 that intersect , we have @xmath102 or @xmath103 so the triangle inequality holds , thus this is indeed a diameter space . + consider now @xmath62 where @xmath61 is the induced pseudometric by @xmath50 . to make @xmath61 non - degenerate , we introduce another property : + * a2 . * each point belongs to only finitely many diametrisable sets . + since @xmath50 is always positive , and the infimum defining @xmath104 for @xmath105 is actually minimum taken over finitely many positive values , we get that @xmath61 is nondegenerate , and thus a metric . + as far as the compactness is concerned , the fact that we can obtain a compact space from a totally bounded one by taking its completion is what motivates our following step . hence , another condition is introduced : + * a3 . * for any positive integer @xmath106 , there are integers @xmath107 greater than @xmath106 such that @xmath108 is finite . + the completion @xmath109 of the metric space induced by @xmath51 is compact , provided @xmath40 satisfies * a1 * -*a3*. by the comments from the above , we just need to show that @xmath62 is totally bounded . let @xmath66 be given , and choose @xmath106 for which @xmath110 holds . by * a3 * , there are @xmath111 for which the union of @xmath112 covers all but finitely many points , denoted by @xmath113 . then , each @xmath114 is contained in @xmath115 for some @xmath116 and @xmath117 , so @xmath1 is covered by finitely many open balls of radius @xmath73 , and so the metric space is totally - bounded . say that a cauchy sequence @xmath118 is proper if there is no @xmath106 with @xmath119 . the three conditions described so far give us a nice characterisation of the proper cauchy sequences in @xmath109 , whose elements lie in @xmath1 . [ cauchychar]suppose that @xmath40 satisfies properties * a1 * -*a3*. then a sequence of points in @xmath1 is proper cauchy with respect to the induced metric if and only if for any given positive integer @xmath120 there is @xmath121 such that @xmath122 contains all but finitely many points of the sequence . _ only if direction . _ let @xmath123 be a proper cauchy sequence in @xmath1 and let a positive integer @xmath120 be given . take a positive @xmath124 . then , as the given sequence is cauchy , we have @xmath106 such that @xmath125 implies @xmath126 . + now fix any @xmath127 and let @xmath128 . by the definition of @xmath61 , we know that this set is nonempty , therefore has a minimal element @xmath129 . if @xmath130 then @xmath131 , so there is @xmath132 with @xmath133 both belonging to @xmath134 . but @xmath135 so @xmath136 intersect and by the choice of @xmath129 we have @xmath137 , so almost all points of the sequence are contained in @xmath138 . + _ if direction . _ given @xmath66 take @xmath120 such that @xmath139 . then there is @xmath121 with @xmath122 containing almost all points in the sequence , and the distance between two points in @xmath122 is at most @xmath140 , so the sequence is cauchy . if it was not a proper one , the point which is equal to almost all of its members would belong to infinitely many of the sets @xmath141 which is impossible . under the same assumptions as in the previous proposition , no proper cauchy sequence in @xmath109 converges to a point in @xmath1 . suppose that a point @xmath142 is a limit of a proper cauchy sequence @xmath143 in @xmath144 . first of all , if @xmath145 , substitute @xmath146 by a point @xmath147 such that @xmath148 . the newly obtained sequence now lies in @xmath1 , and is still proper cauchy , with unchanged limit . as @xmath142 belongs to @xmath141 for only finitely many @xmath5 , we have @xmath149 , for sufficiently large @xmath8 . this is however a contradiction , due to the proposition [ cauchychar ] . let @xmath109 be the metric space considered so far . suppose @xmath150 is a function on @xmath1 that preserves cauchy sequences , that is given cauchy sequence @xmath123 , the sequence of images @xmath151 is cauchy as well . then , extension of @xmath150 to the completion of the space given by @xmath152 , where @xmath143 is any cauchy sequence in @xmath1 tending to @xmath26 in @xmath153 , is continuous . firstly , we should prove that such extension of @xmath150 is well - defined . take arbitrary @xmath154 , and thus since @xmath144 is completion of @xmath26 , there must be cauchy sequence in @xmath1 whose limit is @xmath26 . but image of this sequence under @xmath150 is cauchy as well , so it has limit in @xmath144 , so we just need to show its uniqueness . therefore , suppose that @xmath155 are two sequences in @xmath1 tending to @xmath26 . merging these two sequence into @xmath156 , where @xmath157 , implies that @xmath156 is cauchy , hence @xmath158 is also cauchy , so @xmath151 and @xmath159 have the same limit , as required . + secondly , we should prove that @xmath150 is continuous in @xmath144 . let @xmath123 be sequence tending to some @xmath160 . if @xmath142 , then sequence is eventually constant and equal to @xmath26 , hence @xmath151 trivially tends to @xmath161 . otherwise , @xmath162 , so consider new sequence @xmath156 given as follows . if @xmath163 , set @xmath164 , and if this does not hold , there is cauchy sequence @xmath165 in @xmath1 whose limit is @xmath146 . by assumption , @xmath166 is cauchy in @xmath1 , and as we have shown previously , it tends to @xmath167 . hence , for sufficiently large @xmath168 , we have that @xmath169 , so set @xmath170 . thus , as for all @xmath8 we have @xmath171 , we have that @xmath156 is cauchy in @xmath1 and tends to @xmath26 , so its image under @xmath150 is cauchy sequence with limit @xmath161 , but @xmath172 holds for all @xmath8 , thus @xmath173 , as required , implying continuity of @xmath150 . this lemma suggests the fourth property of the diametrisable sets : + * a4 . * if @xmath174 are indices such that @xmath175 then , given any @xmath106 , we can find @xmath176 for which @xmath177 contains all but finitely many elements of @xmath178 for some @xmath168 , and @xmath179 contains all but finitely many elements of @xmath180 for some @xmath181 . + if @xmath40 satisfies * a1 * -*a4 * , then @xmath182 , defined before and then extended as in the previous lemma are continuous with respect to the induced metric . we will show the claim for @xmath15 , proof for @xmath53 follows the same lines . we only need to show that @xmath15 preserves cauchy sequences in @xmath1 , in fact it is sufficient to prove that if @xmath123 is proper cauchy , then @xmath183 is cauchy . thus , suppose we are given a proper cauchy sequence @xmath123 in @xmath1 , so there indices @xmath184 ( without loss of generality @xmath174 ) such that @xmath114 covers all but finitely many elements of the sequence for every @xmath185 . due to intersections and * a1 * we have @xmath175 . let @xmath66 be given and take @xmath106 which satisfies @xmath186 . further , by * a4 * there is @xmath130 and some @xmath168 for which @xmath187 contains all but finitely many elements of @xmath178 . exploiting the fact that almost all elements of @xmath183 are contained in @xmath178 yields that there is @xmath120 such that @xmath188 implies @xmath189 hence @xmath190 , as required . to make @xmath191 a contractive family , we define the property : + * a5 . * for any @xmath192 , there is @xmath193 such that @xmath194 for some character @xmath195 . + if @xmath40 satisfies * a1 * -*a5 * , then @xmath54 form a @xmath12-contractive family in @xmath196 . let us consider first @xmath42 . taking largest possible @xmath8 for which @xmath197 , we have @xmath198 by * a5 * such that without loss of generality @xmath199 , thus @xmath200 , as wanted . ( if we had character @xmath201 instead , we would get contraction when applying @xmath53 . ) + in general case , @xmath202 , we can find sequences in @xmath1 , @xmath118 tending to @xmath26 , and @xmath203 tending to @xmath35 ( if one of these is already in @xmath1 , then take trivial sequence ) . then , by previous case , one of @xmath54 contracts infinitely many pairs @xmath204 , @xmath15 say . let indices of those pairs be @xmath174 . then , @xmath205 , since @xmath15 is continuous , as required . suppose that a function @xmath150 , which is a word in @xmath54 , has a fixed point . let @xmath81 be the nonempty word which corresponds to @xmath150 , i.e. @xmath206 for all @xmath26 . then there are infinitely many @xmath5 such that there is @xmath207 in @xmath141 with @xmath208 also being a member of @xmath141 . suppose @xmath150 fixes @xmath79 . then @xmath209 , so take @xmath123 in @xmath1 converging to @xmath79 . hence @xmath210 converges to @xmath211 as well , so merging these two sequences together , we get a cauchy sequence , and the result follows from the proposition [ cauchychar ] . therefore , let us introduce the final property of the diametrisable sets that we need to hold . the following corollary does this and sums up our work so far . [ sumcor]let @xmath10 . consider a collection @xmath212 of diametrisable subsets of @xmath1 that obeys : * a1 * : : the set @xmath213 is the whole of @xmath1 and given any nonnegative integers @xmath97 , @xmath98 implies @xmath99 . * a2 * : : any point in @xmath1 belongs to only finitely many diametrisable sets . * a3 * : : for any positive integer @xmath106 , there are integers @xmath107 greater than @xmath106 such that @xmath108 is finite . * a4 * : : if @xmath174 are indices such that @xmath175 then , given any @xmath106 , we can find @xmath176 for which @xmath177 contains all but finitely many elements of @xmath178 for some @xmath168 , and @xmath179 contains all but finitely many elements of @xmath180 for some @xmath181 . * a5 * : : for any @xmath192 , there is @xmath193 such that @xmath194 for some character @xmath214 . * a6 * : : given a nonempty word @xmath81 in @xmath1 there are only finitely many diametrisable sets @xmath141 with @xmath215 for which @xmath216 . then , with constructions described above , @xmath109 is a compact metric space with continuous functions @xmath217 that form a @xmath12-contractive family , but no word in @xmath54 has a fixed point . our main task now is to find a collection of diametrisable sets @xmath40 which has the properties * a1 * -*a6*. the most natural subsets of @xmath1 to be examined are @xmath218 , where @xmath81 is any word in @xmath1 . immediately , we observe that these , ordered by length of @xmath81 , then by alphabetical order ( so that @xmath219 ) satisfy all the properties except for * a6*. hence , we will use these as the pillar of our construction , however , to make * a6 * hold , we need to modify these slightly . the issue with sets described is that given any nonempty word @xmath220 , we allow @xmath221 to contain all the initial segments of @xmath222 . with this in mind , we say that a nonempty word @xmath81 in @xmath1 is forbidden if there is another finite word @xmath79 such that @xmath81 is a prefix of @xmath91 and @xmath223 . otherwise , say that @xmath81 is available . for example , @xmath224 are available , while @xmath225 is forbidden . [ addavailprop]given a word @xmath220 either @xmath226 or @xmath227 is available . if @xmath228 , @xmath226 is available . suppose @xmath229 and that the claim is false , so @xmath226 is an initial segment of @xmath230 for some non - empty @xmath231 , @xmath227 is an initial segment of @xmath232 for some non - empty @xmath233 and @xmath234 . we can permute @xmath231 cyclically to @xmath235 so that @xmath236 holds , and we can correspondingly transform @xmath233 to @xmath237 . observe that the last character of @xmath235 is @xmath238 and of @xmath237 is @xmath201 . this way , @xmath81 becomes a prefix of @xmath239 for @xmath240 . but @xmath241 , hence @xmath242 and so @xmath243 , by comparing them as initial segments of @xmath81 . however , this implies equality of the last characters of @xmath235 and @xmath237 , which is impossible . the claim now follows . in the same spirit , we prove the following statement : given a word @xmath220 either @xmath244 or @xmath245 is available . suppose contrary , both are forbidden . let @xmath246 be words such that @xmath247 , @xmath248 , @xmath249 . observe that both @xmath250 and @xmath251 are prefixes of @xmath81 , so being of the same length , they coincide . but , then @xmath252 , which is a contradiction , as otherwise @xmath253 would be two prefixes of the same length of this infinite word . [ availcor]let @xmath81 be a non - empty word . let @xmath79 be an initial segment of @xmath222 , and suppose @xmath254 . take character @xmath255 such that @xmath256 is not a prefix of @xmath222 . then @xmath80 is available . say that a word @xmath81 is minimal if it is non - empty and given non - empty @xmath79 for which @xmath257 we have @xmath258 . a non - empty word @xmath81 is not minimal iff there is word @xmath79 such that @xmath259 , some @xmath260 . if @xmath259 with @xmath261 then @xmath262 and @xmath263 , therefore @xmath81 is not minimal suppose now that we have non - empty @xmath81 , for which there exists @xmath79 such that @xmath257 , but @xmath264 . write @xmath265 , so @xmath266 , for some positive integers @xmath267 , in particular @xmath268 . further @xmath269 , where @xmath270 are of length @xmath61 . considering successive copies of @xmath79 we have that @xmath271 , when @xmath272 and similarly by looking at copies of @xmath81 we have @xmath273 when @xmath274 . so @xmath275 for @xmath276 $ ] , ( where for a positive integer @xmath106 , @xmath277 $ ] denotes the set @xmath278 ) . observe that @xmath279 and as @xmath181 and @xmath280 are coprime @xmath281 take all possible values modulo @xmath181 hence @xmath282 , allowing us to conclude @xmath283 , @xmath284 . having established these results about the words , we are ready to choose the diametrisable subsets @xmath40 of @xmath1 . consider the following subsets . + for all available words @xmath81 , including @xmath90 , we include @xmath285 in @xmath40 ( we refer to these as _ w - type sets _ , i.e say that @xmath141 is of @xmath286-type if @xmath287 , for some available @xmath81 ) . + for all minimal words @xmath81 , all integers @xmath288 such that @xmath289 and @xmath290 , we set @xmath291 initial segment of @xmath292 ( call these _ @xmath293-type _ sets ) . + finally , for each minimal word @xmath81 and @xmath294 , we define @xmath295 , which we refer to as _ @xmath296-type _ sets . for these , we need additional notation . + first of all , for each @xmath294 we define an infinite arithmetic progression @xmath297 . we set @xmath298 , @xmath299 . for each integer @xmath300 , if @xmath185 is an integer such that @xmath301 , we set @xmath302 where @xmath303 , when @xmath185 is odd ; if @xmath185 is even , then put @xmath304 where @xmath305 . on the other hand , if @xmath185 is an integer such that @xmath306 , then we set @xmath307 . + note that , given a minimal word @xmath81 and @xmath308 , by the corollary [ availcor ] we get a unique word @xmath309 such that : it is of the form @xmath310 where @xmath311 and @xmath312 is a character , @xmath309 is available , and @xmath313 . at last , we define @xmath314 . + thus , we set @xmath315 is available@xmath316 is minimal , @xmath317 is minimal@xmath318 . + to illustrate the definition of @xmath296-type sets , we list a few examples : @xmath319 . let us make a few easy remarks about @xmath40 . fix a minimal word @xmath81 . then we have @xmath320 . also , if @xmath300 is an integer , and @xmath185 is an odd integer that satisfies @xmath321 , then @xmath322 . furthermore , if @xmath323 then @xmath324 or vice - versa . + for any @xmath43 , observe that there is a unique word in @xmath69 of the shortest length , which we will denote by @xmath325 . + let us now establish a few claims about the structure of @xmath40 , which will be exploited in the rest of the proof . [ containprop]if @xmath326 are two diametrisable sets and they intersect , then one is contained in the other . furthermore , if they are not identical , then @xmath327 is infinite . we are going through possible types of @xmath69 and @xmath70 . 1 . case 1 : @xmath69 and @xmath70 are @xmath286-type . + suppose @xmath328 without loss of generality @xmath329 . if @xmath231 is not initial segment of @xmath233 then @xmath69 and @xmath70 do not intersect , so we must have @xmath330 , that is @xmath331 . if @xmath332 , then @xmath333 , and hence @xmath334 , where @xmath255 is character for which @xmath335 is not prefix of @xmath233 . we will use this case for showing the other ones . 2 . case 2 : @xmath69 and @xmath70 are @xmath293-type . + suppose @xmath336 , where @xmath337 are some minimal words , and @xmath338 are suitable integers . there is word @xmath339 , so @xmath81 is initial segment of @xmath340 and @xmath232 , while @xmath234 , from which we deduce that @xmath341 , being the initial segment of @xmath81 of length @xmath342 . therefore @xmath343 and due to minimality @xmath344 . now , due to definition of @xmath293-type sets for fixed minimal word , we get @xmath345 or vice versa and infinite symmetric difference . 3 . case 3 : @xmath69 is @xmath293-type , @xmath70 is @xmath296-type . + let @xmath346 . this makes @xmath81 a prefix of some @xmath347 , where @xmath348 is minimal and @xmath349 , as @xmath350 , @xmath181 , @xmath351 being suitable integers . also @xmath352 , some available @xmath80 , so as @xmath80 is an initial segment of @xmath81 , hence @xmath347 , we have @xmath353 , since @xmath80 is available . but then @xmath354 , which proves the first claim . + on the other hand @xmath69 has no available words , but @xmath70 has infinitely many of these , which gives the second part . 4 . case 4 : @xmath69 is @xmath293-type , @xmath70 is @xmath286-type . + same proof as in the case 3 . case 5 : @xmath69 and @xmath70 are @xmath296-type . + suppose @xmath355 , for some minimal words @xmath337 and positive integers @xmath356 . let @xmath339 , thus @xmath357 and @xmath358 . by case 1 , without loss of generality , @xmath359 holds . + if @xmath360 , letting @xmath207 be @xmath361 without the last character , gives us prefix of @xmath362 , @xmath363 $ ] , and @xmath364 , so as before @xmath343 and due to minimality @xmath344 . so @xmath345 or @xmath365 , due to construction of @xmath296-type sets for a fixed minimal word , and also the second part of the claim follows . + on the other hand , if @xmath366 , then @xmath367 , @xmath368 being @xmath235 after omitting the last character , as before . further we have that @xmath368 is a prefix of @xmath340 , so @xmath237 is too , but @xmath237 is available , hence @xmath369 , hence @xmath370 . + for the second part of the claim , consider any other @xmath286-type set contained in @xmath70 , distinct form @xmath371 . case 6 : @xmath69 is @xmath296-type , @xmath70 is @xmath286-type . + let @xmath372 , some available word @xmath80 , and @xmath339 , so @xmath373 . if @xmath374 , then @xmath365 , so we are done , and second part follows as in previous case . otherwise , by case 1 , @xmath375 , so @xmath376 and @xmath377 . we have @xmath378 , some minimal @xmath348 and integer @xmath185 , so @xmath379 and available so @xmath353 , hence @xmath380 . also , @xmath381 , which is infinite . if @xmath69 is any of the diametrisable sets , then there are unique @xmath382 ( up to ordering ) proper subsets of @xmath69 in @xmath40 , such that @xmath383 has almost all elements of @xmath69 . firstly , let us show the existence of such sets . of course , we go through the possible types of set @xmath69 . the only non - trivial case , that is the one that does not directly follow from construction of @xmath40 is the third one . 1 . case 1 : @xmath69 is of @xmath293-type . + say @xmath384 , some minimal @xmath81 and integers @xmath288 . then , we can take @xmath385 2 . case 2 : @xmath69 is of @xmath296-type . + say @xmath386 , some minimal @xmath81 and integer @xmath185 . then , by construction , we have either @xmath387 , some two integers @xmath356 , or @xmath388 , some available @xmath80 , giving us the needed sets @xmath382 . 3 . case 3 : @xmath69 is of @xmath286-type . + suppose @xmath389 , for some available @xmath81 . then , if @xmath244 and @xmath245 are both available , we can take @xmath390 and @xmath391 . + otherwise , suppose this is not true , and @xmath392 is not available for some character @xmath255 . therefore , there is a word @xmath348 such that @xmath392 is prefix of @xmath347 , and @xmath393 . but @xmath81 is also a prefix of @xmath347 and available so @xmath394 so @xmath395 , i. e. @xmath396 . + the next thing to do is to establish minimality of @xmath348 . let @xmath79 be any word such that @xmath397 and @xmath398 . then , as before @xmath392 is initial segment of @xmath91 and @xmath399 , so by same arguments we get @xmath400 , thus @xmath401 , as wanted . but then @xmath402 , proving the existence part of the claim . + suppose now that @xmath403 is any set strictly contained in @xmath69 . then , by proposition [ containprop ] , @xmath404 is infinite , so @xmath403 can not contain both @xmath382 . also , from the same proposition we get that @xmath405 and @xmath406 are disjoint , as otherwise , one must contain the other and thus be equal to @xmath69 . so , @xmath403 intersects at precisely one of @xmath382 , since all diametrisable sets are infinite . w.l.o.g . @xmath403 intersects @xmath405 . if @xmath407 , then @xmath408 is infinite , and thus intersects @xmath406 , which is impossible . hence , by proposition [ containprop ] , @xmath409 . so , if we had any other @xmath410 with the property in assumption , we would have @xmath411 and @xmath412 , reordering if necessary , ( can not have both sets included in the same @xmath413 ) , so unless these are both equalities @xmath414 would be infinite , yielding a contradiction , and concluding the proof . [ structcor ] if @xmath69 is a diametrisable set with proper diametrisable subsets @xmath415 , such that @xmath416 is finite , then @xmath382 are disjoint and one of them contains @xmath403 . [ finprop ] given word @xmath81 , there are only finitely many diametrisable sets containing it . if @xmath417 , some @xmath79 , then @xmath79 is prefix of @xmath81 , hence there are only finitely many such sets containing @xmath81 . + if @xmath418 , then , @xmath419 , so there are only finitely many choices for @xmath420 . + finally , suppose @xmath421 . length of @xmath81 must be greater than @xmath422 , which gives us finitely many choices for minimal word @xmath348 . fix @xmath348 . recalling the definition of @xmath296-type sets , we have @xmath423 , where @xmath424 . but if @xmath168 is a nonnegative integer , then @xmath425 for @xmath426 . thus , for such @xmath185 , we get @xmath427 , so @xmath421 for only finitely many @xmath185 , as desired . these claims serve us to understand better the structure of @xmath40 . in particular , we can view @xmath40 as a binary tree whose nodes are the diametrisable sets , root is @xmath428 and given a set @xmath429 , its sons @xmath382 are given by the corollary [ structcor ] . what is not clear , however , is to see that the tree so defined actually contains all the diametrisable sets . but , given any such a set @xmath429 , we have either @xmath430 or @xmath69 is contained in one of the sons of the root , by corollary [ structcor ] . proceeding further in this fashion , either we reach @xmath69 , or we get an infinite collection of diametrisable sets whose subset @xmath69 is . but , that implies that if do not reach @xmath69 , its elements belong to infinitely many members of @xmath40 , which contradicts proposition [ finprop ] . hence this binary tree has precisely @xmath40 for its set nodes . moreover , to say that a diametrisable set @xmath431 is a subset of another such set @xmath432 is equivalent to having @xmath431 as an ancestor of @xmath432 in this binary tree . to depict what has just been discussed , we include the figure [ treefig ] which shows the first few layers of tree . we refer to this tree as @xmath433 . in order to finish constructing the counterexample , we must make @xmath40 well - ordered , so that then we know which diametrisable set is actually which @xmath187 . defining such an order , and proving that it is in fact what we need , is the purpose of this section . + consider the relation @xmath434 on @xmath40 , given as follows : if @xmath332 , we say @xmath435 if any of these holds : * @xmath436 * @xmath437 and @xmath331 * @xmath437 , none is contained in the other and we have that either @xmath69 is of @xmath293-type , but v is not , or @xmath69 is of @xmath296-type and @xmath70 is of @xmath286-type . * @xmath437 , none is contained in the other , they are of the same type and @xmath325 is alphabetically before than @xmath438 . [ typeprop]if @xmath435 by * o2 * , then either : 1 . both @xmath69 and @xmath70 are @xmath293-type , or 2 . both @xmath69 and @xmath70 are @xmath296-type , or 3 . @xmath69 is @xmath296-type , @xmath70 is @xmath286-type . regard @xmath40 as a binary tree @xmath433 that was described in the concluding remarks of the previous section . then , @xmath345 tells us that @xmath69 is an ancestor of @xmath70 , and as @xmath435 by * o2 * , we have @xmath439 . thus , the shortest word must same for all sets on the path from @xmath69 to @xmath70 in @xmath433 . now , we analyze the splits , i.e. given a node , what its sons are . returning back to the choice of the diametrisable sets , we see that @xmath293-type sets always split into two @xmath293-type sets , and @xmath296-type sets split into @xmath296 and @xmath286-type sets . we conclude that the claim will be true if when @xmath286-type set splits into an @xmath293-type and a @xmath296-type set , the shortest word of the father is not in any of the sons . but , suppose that we have such a situation , a set @xmath440 , whose children are @xmath441 and @xmath442 . from the proof the proposition , recall that @xmath396 . but then @xmath443 and @xmath444 , as desired . relation @xmath434 on the chosen sets is strict total order , and further collection of chosen sets is well ordered under @xmath434 . if @xmath445 are two distinct sets among the chosen ones , we want that precisely one of @xmath446 , @xmath447 is true . + suppose neither of these holds . hence @xmath439 . but then their intersection is non - empty , hence one is contained in the other , so @xmath435 or vice versa by * o2*. + now , assume that both hold . therefore , @xmath437 , none is contained in the other ( otherwise we have @xmath345 and @xmath448 , thus @xmath449 ) , they are of same type , and further @xmath439 . but as before , these must intersect , leading us to a contradiction . + having proved the trichotomy , we now move to establishing transitivity of the relation . suppose we have three chosen sets @xmath450 such that @xmath451 , from which @xmath332 , @xmath452 follows . further , we can not have @xmath453 as this would imply @xmath454 . hence , we can assume that all three sets are distinct . if @xmath455 holds for @xmath435 or @xmath456 , then it also holds for @xmath457 . so , assume this is not true . then we have the following cases : case 1 : @xmath435 by * o2 * , @xmath456 by * o2*. + then @xmath458 , so @xmath459 by * o2 * as well . case 2 : @xmath435 by * o2 * , @xmath456 by * o3*. + by proposition [ typeprop ] , we see that ( as @xmath70 can not be of @xmath286-type ) , @xmath69 and @xmath70 are of same type . as @xmath365 and @xmath460 , then @xmath461 , so either @xmath462 and @xmath459 by * o2 * , or @xmath459 by * o3*. case 3 : @xmath435 by * o2 * , @xmath456 by * o4*. + as above , we get @xmath461 , so either @xmath462 implying @xmath463 or if @xmath69 is @xmath296-type and @xmath70 is @xmath286-type , so @xmath459 by * o3 * , or by proposition [ typeprop ] @xmath69 and @xmath70 have the same type , hence so does @xmath464 , and @xmath439 ( because of inclusion ) so @xmath459 by * o4*. case 4 : @xmath435 by * o3 * , @xmath456 by * o2*. + if @xmath465 , then @xmath345 , which is impossible , so @xmath466 . if @xmath462 , we re done , otherwise from proposition [ typeprop ] we obtain @xmath459 by * o3*. case 5 : @xmath435 by * o3 * , @xmath456 by * o3*. + thus @xmath69 is of @xmath293-type , @xmath70 is of @xmath296-type , @xmath464 is of @xmath286-type . by proposition [ typeprop ] , we conclude that @xmath467 , @xmath468 , so @xmath459 by * o3*. case 6 : @xmath435 by * o3 * , @xmath456 by * o4*. + by proposition [ typeprop ] , @xmath461 , so either @xmath462 giving @xmath459 by * o2 * , or none is contained in the other and @xmath459 by * o3*. case 7 : @xmath435 by * o4 * , @xmath456 by * o2*. + having @xmath69 as a subset of @xmath464 implies @xmath345 which is impossible . so @xmath461 and if @xmath462 , then @xmath459 by * o2*. otherwise , unless @xmath69 and @xmath464 are of same type , we have @xmath459 by * o3 * , due to proposition [ typeprop ] . finally , from inclusion we deduce @xmath469 and thus @xmath459 by * o4*. case 8 : @xmath435 by * o4 * , @xmath456 by * o3*. + if @xmath470 , by proposition [ typeprop ] , we reach contradiction . hence @xmath471 , so @xmath459 by * o2 * , or @xmath459 by * o3 * otherwise . case 9 : @xmath435 by * o4 * , @xmath456 by * o4*. + if @xmath472 or vice versa , we have @xmath469 which is impossible . as all three sets are of same types , we get @xmath459 by * o4*. finally , given a subset @xmath473 of @xmath40 , consider its subset @xmath474 of those sets @xmath475 such that @xmath476 . this is finite by the proposition [ finprop ] , hence we can find the minimal element of @xmath474 with respect to @xmath434 , which is smaller than any member of @xmath477 by * o1 * , making @xmath40 well - ordered under @xmath434 . hence , as @xmath40 is countable , we can take @xmath478 and for @xmath479 , @xmath480 . observe that for any given @xmath43 there are only finitely many words of length at most @xmath481 . the proposition [ finprop ] then tells us that there are only finitely many @xmath482 such that @xmath483 . hence , there are only finitely many @xmath482 such that @xmath447 , so @xmath484 for some @xmath294 . therefore , @xmath485 . note also that @xmath486 being the only set with shortest word whose length is non - positive . all that is left is to show that @xmath487 satisfy * a1 * - * a6*. having done the most of the work already , the proof of the following claims will either be the rather short or the obvious case - examination . @xmath488 satisfy * a1*. let @xmath101 be such that @xmath97 and they intersect . thus , @xmath489 , @xmath490 and also @xmath491 or @xmath492 , by proposition [ containprop ] . but if @xmath491 , then we must have @xmath493 and hence @xmath494 by * o2 * , which is contradiction . also @xmath100 . @xmath488 satisfy * a2*. note that this is the proposition [ finprop ] , but we include this as another claim for completeness . @xmath488 satisfy * a3*. let @xmath106 be given . by the corollary [ structcor ] , given @xmath141 we can find @xmath495 , disjoint subsets of @xmath141 , which cover all but finitely many elements of @xmath141 , therefore @xmath496 by * o1 * or * o2*. so , we can start with @xmath213 , and perform such splits until we are left with sets @xmath497 , with @xmath498 , which cover almost all elements of @xmath1 , as in each split we lose only finitely many elements . @xmath488 satisfy * a4*. suppose that @xmath175 for some @xmath174 . as usual , we consider different cases . case 1 : there is set of @xmath293-type among these . + let @xmath114 be such a set . as @xmath293-type set splits into @xmath293-type sets , we have that when @xmath499 , @xmath500 , some minimal word @xmath81 , and integers @xmath501 , and @xmath502 , so in fact @xmath503 , as these are powers of 2 . in particular , we deduce that for sufficiently large @xmath8 , @xmath504 and thus @xmath505 is prefix of all words in @xmath506 implying @xmath507 . by the proposition [ addavailprop ] , one of @xmath508 is available , w.l.o.g . the former is true , hence @xmath509 , some @xmath510 . + consider cyclic permutation @xmath79 of @xmath81 , such that @xmath511 , @xmath255 being the last character of @xmath81 . let @xmath80 be a word such that @xmath512 and @xmath513 . but then for cyclic permutation @xmath348 of @xmath80 such that for character @xmath514 we have @xmath515 , we have @xmath516 , so @xmath517 , proving the minimality of @xmath80 . if @xmath518 , we would have that @xmath519 is not available as it would be initial segment of @xmath91 , but @xmath520 . hence @xmath521 , and @xmath522 , where @xmath523 , unless @xmath524 and then @xmath525 , @xmath510 suitable index . as @xmath510 tends to infinity as @xmath8 does , @xmath526 holds in this case . case 2 : there are no @xmath293-type sets , but there infinitely many of @xmath286-type . + denote @xmath286-type sets among these by @xmath527 . as @xmath528 is prefix of @xmath529 , for all @xmath5 , these define an infinite word @xmath81 whose initial segments the words @xmath530 are . by the proposition [ addavailprop ] , w.l.o.g . @xmath531 is available infinitely often , so we can take @xmath532 for suitable @xmath174 , to establish @xmath238 part of the claim . + similarly , if there are infinitely many initial segments of @xmath227 available , choosing these and their corresponding @xmath286-type sets establishes the claim . suppose contrary , i. e. there is @xmath168 such that prefixes of @xmath227 of length greater or equal to @xmath168 are all forbidden . denote by @xmath533 initial segment of @xmath227 of length @xmath8 , and let @xmath534 be the shortest word such that @xmath535 and @xmath533 is initial segment of @xmath536 . now , suppose there is no @xmath537 such that @xmath538 , hence @xmath539 , so @xmath540 , some @xmath8 . but these means that @xmath530 are forbidden from some point , resulting in contradiction . so , there must be such an @xmath537 , so pick the smallest possible . hence @xmath541 is not the initial segment of @xmath536 , but @xmath533 is both initial segment of @xmath536 and @xmath542 . further , @xmath543 so as before , @xmath544 , but @xmath541 is prefix of the later , but not former , giving us contradiction . case 3 : almost all sets are of @xmath296-type . given @xmath545 , let @xmath255 be the last character of @xmath81 , so @xmath546 is forbidden . let @xmath79 be the cyclic permutation of @xmath81 such that @xmath511 , and so @xmath79 is minimal as well , by arguments in case 1 of this proof . then , @xmath547 or @xmath548 for suitable index @xmath549 and word @xmath80 , so take @xmath550 . let @xmath514 be the character not equal to @xmath255 and @xmath79 be @xmath551 without the last character . then as @xmath552 is forbidden , so @xmath553 is available and @xmath554 . as @xmath510 s tend to infinity as @xmath8 does , we re done . @xmath488 satisfy * a5*. let @xmath187 be given . if it is of @xmath293 or @xmath286-type , @xmath555 , so choose character @xmath255 such that @xmath556 is available and hence @xmath557 and @xmath558 , since @xmath559 . if @xmath187 is of @xmath296-type , then let @xmath79 be the shortest word in @xmath187 after erasing the last character , so @xmath560 and once again choose @xmath561 for which @xmath562 is available , hence @xmath563 . if @xmath564 intersects @xmath565 , it must be its subset and @xmath79 prefix of @xmath562 , but then @xmath566 , so @xmath562 is forbidden , as @xmath567 . so @xmath564 and @xmath565 are disjoint , hence @xmath564 and @xmath187 are also . combining this with @xmath568 and comparing the types gives @xmath569 by * o3*. @xmath488 satisfy * a6*. let @xmath570 be given . we will divide the proof into three sections , each showing the claim for particular set type . * suppose for some word @xmath79 we have @xmath571 , where @xmath348 is minimal . hence @xmath572 . suppose further @xmath573 . so @xmath81 is initial segment of @xmath79 , as @xmath79 is initial segment of @xmath574 . but then , @xmath575 is also initial segment of @xmath79 , etc . up to @xmath576 as @xmath577 . hence @xmath578 , but @xmath348 is minimal so @xmath579 . hence we have only finitely many choices for @xmath293-type @xmath187 , as @xmath579 and @xmath580 , must hold . * if @xmath187 is of @xmath286-type , say @xmath581 , we have @xmath582 , @xmath80 available . as there are only finitely many @xmath80 such that @xmath583 , w.l.o.g . we have @xmath585 , some word @xmath586 . hence @xmath587 so @xmath588 . but then @xmath589 , some @xmath590 , so @xmath591 , for some @xmath235 , as @xmath592 , and @xmath593 implies @xmath594 . we can iterate this until @xmath595 . but @xmath596 , some @xmath597 , making @xmath80 initial segment of @xmath222 , but it is available hence @xmath598 , therefore we have finitely many choices for @xmath187 of @xmath286-type . * finally , suppose @xmath187 is @xmath296-type , and @xmath599 . then we have two available words @xmath600 , where @xmath601 are characters , @xmath337 prefixes of @xmath347 , some minimal word @xmath348 , @xmath602 . so @xmath603 , @xmath604 holds for some words @xmath605 . + suppose @xmath606 . let @xmath607 be the initial segment of @xmath574 of length @xmath422 thus it is prefix of @xmath233 hence of @xmath347 . thus @xmath608 is prefix of @xmath609 . hence , @xmath610 is prefix of @xmath609 for all @xmath611 $ ] . in particular , @xmath576 is prefix of @xmath612 , thus @xmath613 , implying @xmath614 , but @xmath348 is minimal , so @xmath615 , some @xmath616 , so @xmath617 as @xmath606 . now we have @xmath618 , which differ from @xmath347 for the first time at @xmath619 and @xmath620 , hence @xmath621 . but for only finitely many @xmath185 can we have @xmath295 having such @xmath622 as subsets due to construction of @xmath296-type sets , and the fact that the common difference of arithmetic progression @xmath297 is at least @xmath623 . + now , suppose @xmath624 . if the claim is to be false , we can assume without loss of generality , that for infinitely many @xmath185 we have some @xmath625 and @xmath626 too . hence we can assume that there are @xmath337 prefixes of @xmath347 , with some characters @xmath601 , such that @xmath600 are available , @xmath627 , @xmath337 arbitrarily long , thus say @xmath628 . so @xmath603 , @xmath604 for some words @xmath605 , thus @xmath629 . hence @xmath630 is prefix of @xmath233 as @xmath631 , and similarly @xmath632 is initial segment of @xmath633 so both are prefixes of @xmath574 hence @xmath630 is prefix of @xmath632 , which is then prefix of @xmath634 , etc . and prefix of @xmath635 , so @xmath614 and due to minimality @xmath636 , some @xmath637 . hence @xmath638 so the character where @xmath574 first differs from @xmath347 is at the same time at @xmath619 and @xmath620 , hence @xmath639 , implying @xmath640 , but as previously explained , this can occur just for finitely many @xmath641 , which proves the claim . having showed the required statements we are ready to conclude : + * theorem . * given @xmath10 , there is a compact ( pseudo-)metric space @xmath62 on which we have continuous functions @xmath642 such that given @xmath42 either @xmath643 or @xmath644 holds , but no word in @xmath20 has a fixed point . even though we have given an answer to austin s question and stein s conjecture , some natural questions remain to be answered . indeed , what kind of additional assumptions are necessary to ensure that stein s conjecture is true . here we mention two theorems related to our problem that could indicate what approach should be taken . + the theorem [ cthmmerrystein ] was proved by merryfield and stein in @xcite , and later merryfield , rothschild and stein removed the assumption of continuity in @xcite ; this statement is also known as the generalized banach contraction theorem . in @xcite , austin proved the theorem [ cthmaustin ] and conjectured that the result holds for arbitrarily many commuting operators . + theorem [ thm1 ] tells us that one should not look for a topological condition , and that algebraic condition as in the theorem of austin or in the generalized banach contraction theorem is more convenient . the later type of the condition directly defines the geometry of the problem , and in general the assumption should , vaguely speaking , provide us with a finite - dimensional problem , and forbid compositions of functions behaving very badly , i.e. prevent the tree structure used in the counterexample from occurring . an appropriate assumption could be that @xmath646 is of subexponential or polynomial growth in @xmath168 . + to conclude , we raise the following question and conjecture which are extensions of a conjecture of austin in @xcite . what algebraic conditions should be imposed on a finite contractive family @xmath11 of operators on a complete metric space to ensure the existence of a composition of the functions @xmath6 with a fixed point ?

we present the results of a detailed study of energy correlations at steady state for a 1-d model of coupled energy and matter transport . our aim is to discover via theoretical arguments , conjectures , and numerical simulations how spatial covariances scale with system size , their relations to local thermodynamic quantities , and the randomizing effects of heat baths . among our findings are that short - range covariances respond quadratically to local temperature gradients , and long - range covariances decay linearly with macroscopic distance . these findings are consistent with exact results for the simple exclusion and kmp models . .5 in

cond-mat0610021

transport processes , such as heat flow through a conducting medium in contact with unequal heat reservoirs , are intrinsically nonequilibrium phenomena because of the presence of nonzero currents @xcite . a central problem in nonequilibrium statistical physics is to explain how such large - scale , macroscopic processes arise from complex microscopic interactions . the most basic questions are perhaps those of mean profiles of quantities of physical interest and their responses to external forces . nonequilibrium steady states , on the other hand , are well known to be characterized by large fluctuations , and among the simplest measures of fluctuations are temporal and spatial correlations . in this paper , we present a systematic study of spatial correlations at steady state for a class of 1-d stochastic models called random halves models . a detailed description of this class of models is given in sect . [ sec : ey - desc ] . very briefly , a linear chain of open cells is connected to two unequal heat baths , which inject into the chain tracer particles at characteristic rates and energies . energy storage devices are systematically placed throughout the chain to mark local energy levels . their contents are redistributed by the tracer particles as they move through the chain . these models were introduced in @xcite as stochastic idealizations of certain mechanical models @xcite . our study is based on a combination of analytic arguments and numerical simulations . we believe random halves models are excellent candidates for this method of investigation for two reasons . first , their dynamics are richer and more complex than rigorously - understood models such as the simple exclusion @xcite and kmp models @xcite , as random halves models have two transported quantities ( energy and matter ) and highly nonlinear interactions involving two different time scales . these features make a purely analytical study more difficult . second , all of the forces acting on this system are clearly identified . this is seldom the case in more realistic physical models . our main results can be summarized as follows . given boundary conditions , _ i.e. _ , the temperature and injection rate of each bath , we let @xmath0 denote the stored energy at the @xmath1th site , and @xmath2 the covariance of @xmath0 and @xmath3 in a chain of length @xmath4 . our first finding is @xmath5 for @xmath6 , which leads us to consider the following two functions describing the covariances of stored energies at microscopic and macroscopic distances : @xmath7},s_{[xn]+1})\ , \qquad x \in ( 0,1);\\ { \cal c}_2(x , y ) & = & \lim_{n \to \infty } n \cdot { { \rm cov}}_n(s_{[xn]},s_{[yn]})\ , \qquad x , y \in ( 0,1),\ \ x\neq y.\end{aligned}\ ] ] we show that @xmath8 has the form @xmath9 where the right side is to be interpreted as follows : @xmath10 is a measure of effective distance to the baths , a rough approximation of which is @xmath11 ; @xmath12 is local temperature , and @xmath13 is temperature gradient ; @xmath14 is local particle density , and @xmath15 as @xmath16 . note the quadratic dependence on the local temperature gradient . for long - range covariances , we find that for fixed @xmath17 , the function @xmath18 is continuous but not differentiable at @xmath19 , decreasing roughly linearly as @xmath20 increases . our numerical results are consistent with exact analytic results on the simple exclusion and kmp models @xcite . we extend the existing picture to a setting of coupled energy and matter transport , with features not present in these two previously studied models . for example , we show that energy covariances at microscopic distances respond quadratically to local temperature gradients ( which vary along the chain ; see @xcite ) . there is a second transported quantity , namely matter in the form of particles , and energy covariances are shown to be inversely related to local particle density . throughout the paper , we provide cross - checking numerical evidence for the phenomena we identify , and venture to give a physical interpretation whenever we can . see @xcite for other relevant works on this topic . while most of the results presented here are specific to 1-d , they also serve as a basis for direct comparisons with higher dimensions . a corresponding analysis of random halves models in 2 and 3-d is underway . the results will be reported in a separate paper . [ [ note - on - simulations . ] ] note on simulations . + + + + + + + + + + + + + + + + + + + + all of our numerical results are obtained via direct simulation . that is , our computer programs faithfully implement the dynamics of the random halves model described in sect . [ sec : ey - desc ] , and expectation values with respect to the invariant measures are computed as time averages over long trajectories . some relevant numerical issues are discussed in appendix a. a class of models of nonequilibrium phenomena is introduced in @xcite . in each model , there is a homogeneous conducting medium represented by a linear chain of @xmath4 identical `` cells '' with stochastic heat baths coupled to the ends of the chain . within each cell , there is a _ stored energy _ which characterizes the `` temperature '' at that location . matter ( in the form of tracer particles ) and energy ( in the form of tracer kinetic energy ) are injected into the system by the heat baths ; they are eventually absorbed by the heat baths . tracer particles interact with the local system at each site , redistributing the stored energies as they move about in the chain ; they do not interact directly with each other . this general framework was introduced as an abstraction of the mechanical models in @xcite . it encompasses both `` hamiltonian '' models with conservative deterministic dynamics ( the heat baths being the only sources of randomness ) , and stochastic models , in which the microscopic dynamics are defined by conservative stochastic rules . for concreteness , we begin by describing a slightly simplified version of the mechanical models studied in @xcite . in this _ rotating disc model _ , each cell contains a disc nailed down at its center , about which it rotates freely ( see fig . [ fig - model ] ) . whenever a tracer collides with a disc , it exchanges kinetic energy with the disc via a deterministic rule , _ e.g. _ the angular momentum of the disc may be interchanged with the tangential component of the tracer s momentum . when a tracer collides with a cell wall , it reflects elastically . here , the stored energy at each site is the rotational energy of the disc . [ fig - model ] the random halves model studied in the present paper are stochastic idealizations of mechanical models like the rotating disc model described above . nonequilibrium energy and particle density profiles of random halves models have been analyzed in @xcite . the rest of this section reviews relevant parts of that paper . [ [ precise - description - of - the - random - halves - model . ] ] precise description of the random halves model . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + there are @xmath4 sites , labeled @xmath21 , with tracers moving through the chain . at each site there is an abstract _ energy storage tank_. we let @xmath22 denote the amount of energy in the tank at site @xmath1 at time @xmath23 . the microscopic dynamics are defined by the following rules . fix @xmath24 . each tracer is equipped with two independent exponential clocks . clock 1 , which signals the times of energy exchanges with tanks , rings at rate @xmath25 , where @xmath26 is the kinetic energy of the tracer . clock 2 , which signals the times of site - to - site movements , rings at rate @xmath27 . suppose a tracer is at site @xmath1 and one of its clocks rings . then instantaneously : 1 . if clock 1 rings , the tracer energy @xmath26 and the stored energy @xmath0 are pooled together and split randomly . that is , the tracer gets @xmath28 units of energy and the tank gets @xmath29 , where @xmath30 $ ] is uniformly distributed and independent of all other random variables . 2 . if clock 2 rings , the tracer leaves site @xmath1 . it jumps with equal probability to sites @xmath31 . a tracer entering site @xmath32 or site @xmath33 exits the system forever . all tracers originate from and eventually exit to one of the heat baths . each heat bath injects tracers with independent , exponentially - distributed energies into the system . the left bath injects tracers with mean energy @xmath34 into site 1 at an exponential rate of @xmath35 , and the right bath injects tracers with mean energy @xmath36 into site @xmath4 at a rate of @xmath37 . tracers in the system are indistinguishable . [ [ physical - interpretation - and - remarks . ] ] physical interpretation and remarks . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + it is natural to think of the parameters @xmath38 and @xmath39 as temperatures . this leads us to define the temperature @xmath40 at site @xmath1 in an @xmath4-chain to be @xmath41 , where the expectation is taken with respect to the invariant measure of the @xmath4-chain . the injection rates @xmath42 and @xmath43 can be rewritten in terms of chemical potentials ; we do not pursue this analogy further . the hamiltonian model contains a small length parameter which does not appear in the stochastic model , namely a length @xmath44 which measures the size of the cell . in the stochastic model , we set @xmath45 . it is useful to keep this in mind in dimensional analysis . there are two time scales in our system , one associated with the local dynamics at each site and the other the movement of tracers along the chain . the ratio of these two time scales is @xmath46 . for example , @xmath47 means that on average , tracer - tank energy exchanges occur much more frequently than site - to - site movements of tracers . steady - state macroscopic profiles such as temperature and tracer density do not depend on @xmath46 , but @xmath46 can have a significant impact on the numerical values of spatial and temporal correlations . in this paper , we have chosen to simplify matters by _ fixing _ @xmath46 , which throughout the paper is set equal to @xmath48 . at any moment in time , the number of tracers at each site can vary from @xmath32 to @xmath49 . observe also that the interaction in this simple model is highly nonlinear , even though there are no direct tracer - tracer or tank - tank interactions : tracers at the same site exchange energy via interacting with the tank , and all actions including energy exchanges and site - to - site jumps take place at rates proportional to the `` speeds '' ( the square roots of the kinetic energies ) of the tracers at that moment in time . this is essential if these stochastic models are to mimic the behaviors of their hamiltonian counterparts . at equilibrium , _ i.e. _ , when the left and right baths have equal temperatures and injection rates , the invariant measure of the random halves model is known explicitly . to give its density , we need a little bit of notation : the state of a single cell in which there are exactly @xmath50 tracers is specified by @xmath51 where @xmath52 is the stored energy , _ i.e. _ , the energy of the tank , and @xmath53 is an unordered set of @xmath50 positive numbers representing the @xmath50 tracer energies . let @xmath54 denote the set of all possible states of a single cell with exactly @xmath50 tracers present . the state space for a single cell is then the disjoint union @xmath55 . @xcite [ prop2.1 ] let @xmath56 , and @xmath57 . then the unique invariant probability measure of the @xmath4-chain is the @xmath4-fold product @xmath58 where @xmath59 is the measure giving the statistics within each cell . the measure @xmath59 is defined by : 1 . the number of tracers present is a poisson random variable with mean @xmath60 , _ i.e. _ , @xmath61 2 . the conditional density of @xmath59 on @xmath54 is @xmath62 where @xmath63 here @xmath64 , and @xmath65 is the normalizing constant . proposition [ prop2.1 ] shows that at equilibrium , the stored energies at different sites are entirely uncorrelated . a generalization of this result is given in proposition 5.1 . we now fix arbitrary boundary conditions @xmath34 , @xmath36 , @xmath42 , and @xmath43 , and give the equations of temperature and tracer profiles in nonequilibrium steady states . we assume that for each @xmath4 , there is a unique invariant probability measure @xmath66 to which all initial data converge . because the energies are unbounded , a tightness argument is needed to guarantee existence . uniqueness should be straightforward . in what follows , the word `` mean '' refers to expectation with respect to @xmath66 . we identify each site @xmath1 with the point @xmath67 in the unit interval , and think of @xmath68 and @xmath69 as the locations of the baths . for @xmath70 , define @xmath71 that is to say , the mean number of jumps out of site @xmath1 per unit time is @xmath72 ; half go to the right and half to the left , and so on . the quantities @xmath73 and @xmath74 are well - defined in steady state and have very simple behavior : [ lemma1.1 ] @xcite with @xmath73 and @xmath74 defined as above , we have , for @xmath75 , 1 . @xmath76 ; 2 . @xmath77 . as @xmath78 , these functions converge ( trivially ) to linear or , more accurately , affine functions on @xmath79 . note that @xmath80 and @xmath81 are the steady - state tracer and energy currents , respectively . these currents by themselves , however , do not determine steady state profiles such as those for temperature and tracer density . the following assumption is used to deduce such information : * ( a version of lte ) * for @xmath82 , let @xmath83 denote the marginal of @xmath66 at the site @xmath1 . we assume that for every @xmath84 , @xmath85}$ ] converges as @xmath78 to @xmath86 for some @xmath87 . a tightness condition for all the @xmath88}$ ] is also assumed . let @xmath89 denote the number of tracers present at site @xmath1 . recall that @xmath0 is the stored energy . [ thm1 ] @xcite let arbitrary boundary conditions @xmath90 be given . under assumptions 1 and 2 , @xmath91})\qquad\mbox { and } \qquad \kappa(x ) = \lim_{n\to\infty}\mathbb{e}_{\mu_n}(k_{[xn]})\ ] ] are well - defined and are given by @xmath92 see @xcite for details . this paper concerns the spatial energy covariances at steady state of the random halves model described in sect . [ sec : ey - desc ] . more precisely , let boundary conditions @xmath93 and @xmath37 be specified , and consider an @xmath4-chain . the quantities of interest are @xmath94 we begin by examining how these quantities scale with @xmath4 . for an @xmath4-chain with boundary conditions @xmath95 and @xmath43 , we consider the total energy variance @xmath96 keeping the boundary conditions fixed , we treat @xmath97 as a function of @xmath4 , and observe that @xmath98 where @xmath99 by theorem 1 , @xmath100 that is , @xmath101 for @xmath102 with @xmath103 . when @xmath104 , @xmath105 ( see sect . [ sec : ey - equilibrium ] ) , so that @xmath106 . we expect this to be true when @xmath107 on physical grounds , which leads to the question : is it true that @xmath108 , and if so , what is the sign of @xmath109 ? two sets of numerical simulations are performed to resolve this . in one , we compute @xmath110 directly as a function of @xmath4 . in the other , we start from equilibrium , and hope to observe a quadratic response as a temperature gradient is introduced ; the sign of the quadratic term is then the sign of @xmath109 . the reason we expect a quadratic ( as opposed to linear ) response is symmetry : the temperature gradients @xmath111 and @xmath112 clearly lead to the same energy variances and covariances . [ fig - total ] [ exp1 ] fix bath temperatures @xmath113 and injection rates @xmath114 and compute @xmath115 and @xmath116 for increasing @xmath4 . the results , plotted in fig . [ fig - total ] , show that @xmath117 _ i.e. _ @xmath118 for some @xmath119 . [ exp2 ] fix @xmath4 , @xmath120 , and a number @xmath121 . compute @xmath122 and @xmath123 for various pairs @xmath124 chosen so that @xmath125 , and investigate @xmath126 as a function of @xmath127 . the results show that @xmath128 depends quadratically on @xmath129 with a strictly positive coefficient . taken together , these two simulations suggest that for @xmath120 and fixed mean temperature , @xmath130 at least when @xmath131 is not too large compared to the mean temperature in the chain . we will see later that both the linear scaling of @xmath128 with system size and its quadratic response to temperature gradient are consistent with the scaling of @xmath2 . note that one advantage of probing spatial correlations via global quantities such as @xmath128 is that it can be computed more reliably than small , local quantities such as @xmath132 . we begin now to investigate the individual terms in the sum @xmath133 . the two most basic characteristics of @xmath2 are its _ sign _ and _ order of magnitude_. the fact that @xmath134 suggests that many of the covariances are positive . indeed , all of our numerical evidence points to @xmath135 , as does the theoretical reasoning in the sections to follow . there is reason to remain cautious , however : since @xmath136 when the system is in equilibrium ( proposition 1.1 ) , and can be zero even when there is a tracer flux ( proposition 5.1 ) , one can not conclude definitively via numerics alone that there are no strictly negative correlations . see the remark at the end of this subsection on why the nonnegativity of covariances in this model may be a delicate question . we proceed to an analysis of the order of magnitude of @xmath2 , assuming in the heuristic discussion below that @xmath135 for all @xmath137 . we propose to decompose @xmath128 into @xmath138 and reason as follows : 1 . for @xmath1 away from the two ends of the chain , @xmath139 since these terms sum to @xmath140 and it is unlikely that they scale differently with @xmath4 . ( for @xmath1 close to the boundary , @xmath141 may be smaller due to the entrance of tracers with _ i.i.d . _ energies . ) 2 . for each fixed @xmath1 , the function @xmath142 should decrease monotonically as @xmath143 increases : sites farther apart are expected to be less correlated because they `` communicate '' via longer and noisier `` channels . '' these considerations imply that @xmath144 , with @xmath1 away from @xmath32 and @xmath4 , are among the larger of the @xmath145 terms in @xmath128 . since @xmath144 is one of @xmath4 terms in @xmath141 , ( i ) above implies @xmath146 we point out that the order of magnitude of @xmath144 contains a fair amount of information about the shape of the function @xmath147 : if @xmath148 , then @xmath149 for a definite fraction of @xmath150 ( because these terms have to add up to @xmath151 ) . this points to an extremely slow decay of @xmath2 with increasing distance between sites @xmath1 and @xmath150 . if , on the other hand , @xmath152 , then the function @xmath147 , @xmath153 , would have a very sharp peak at @xmath154 . the question is : which scenario is the case here ? to resolve this issue , we perform the following numerical simulation : [ fig - middle - n ] [ exp3 ] for fixed boundary conditions , compute @xmath155},s_{[n/2]+1})$ ] for a range of @xmath4 . two sets of results are plotted in fig . [ fig - middle - n ] . they ( and other data sets not shown here ) show that @xmath156},s_{[n/2]+1})\sim\frac{1}{n}.\ ] ] as explained above , one may infer from this that all pair covariances @xmath157 , a fact confirmed by many subsequent simulations ( see _ e.g. _ fig . [ fig - renorm ] ) . [ [ remarks - on - sign - of - rm - cov_ns_is_j . ] ] remarks on sign of @xmath2 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as far as we know , existing analytic techniques for proving or disproving positivity of correlations do not apply to the random halves model . this , in part , is due to the fact that in our model , mechanisms conducive to both positive and negative correlations are at work , and the actual covariance reflects a balance between these two tendencies . for instance , a large , upward surge in the energies of incoming tracers has the effect of raising some of the stored energies above their expected values . it is believed that such excursions lead to long - wavelength fluctuations with slow relaxation times , resulting in positive correlations @xcite . however , upon interacting with one of the tanks , a tracer that acquires a higher - than - normal fraction of the energy is likely to move more quickly to a neighboring site and to interact with the tank there , possibly transferring its energy to the tank in its new location . such a phenomenon creates negative correlations between neighboring sites . in equilibrium , these two tendencies balance perfectly , giving zero covariance ( see proposition [ prop2.1 ] ) . further numerical evidence in support of nonnegative covariances is given in sect . [ sec : local - grad ] . the purpose of this short section is to make explicit the objects to be studied in the rest of this paper and the assumptions under which we plan to operate . the discussion in the sect . [ sec : pair - covs ] points to the following functions : given boundary conditions @xmath159 and @xmath37 , and @xmath160 , @xmath161 , define @xmath8 and @xmath162 by @xmath163},s_{[xn]+1}),\\ { \cal c}_2(x , y ) & = & { \cal c}_2(x , y ; { t_l},{t_r},{\rho_l},{\rho_r } ) \ = \ \lim_{n \to \infty } \ n \cdot { { \rm cov}}_n(s_{[xn]},s_{[yn]}),\end{aligned}\ ] ] if these limits exist . supposing these functions are well defined , it is natural to ask ( i ) how @xmath8 depends on local thermodynamic quantities such as @xmath12 , @xmath14 and their gradients , and ( ii ) if these local quantities completely determine @xmath8 . we present below an answer to the second question ; it is one of a number of possible answers . question ( i ) is addressed in the next two sections . given @xmath84 and @xmath164 , theorem 1 gives a quadruple @xmath165 where @xmath166 , @xmath167 , @xmath168 and @xmath169 . conversely , given @xmath84 and a quadruple @xmath165 , it is easy to check that one can solve the equations in theorem 1 backwards and find the corresponding boundary conditions . this one - to - one correspondence between boundary and local conditions defines a function @xmath170 with @xmath171 since we work only with finite chains , it is necessary to have a version of @xmath172 before the infinite - volume limit . given @xmath164 , @xmath17 and @xmath4 , we let @xmath173 $ ] , and let @xmath174 and @xmath175 be as follows : @xmath40 is the mean temperature at site @xmath1 , @xmath176 is the `` injection rate '' ( see sect . [ sec : ey - eqs ] for precise definition ) , @xmath177 and @xmath178 . let @xmath179 where @xmath180 . the next lemma shows that the function @xmath181 is well defined : [ lemma3.1 ] given @xmath182 and @xmath175 , there is a unique set of boundary conditions @xmath183 for the @xmath4-chain that leads to these values at site @xmath1 provided @xmath4 is sufficiently large ( or @xmath184 and @xmath175 are sufficiently small ) . first , @xmath176 and @xmath175 determine @xmath42 and @xmath43 by linearity of @xmath73 ( see lemma [ lemma1.1 ] , [ sec : ey - eqs ] ) , provided @xmath175 is small enough that both of these numbers are nonnegative . for definiteness , assume @xmath185 . we will find suitable @xmath34 and @xmath36 by varying both until correct values are attained at site @xmath1 : fix the left bath temperature temporarily at @xmath186 , and vary the right bath temperature from @xmath40 to @xmath49 . since the temperature at site @xmath1 increases strictly monotonically with the right bath temperature , there exists a unique @xmath187 giving the correct value of @xmath40 at site @xmath1 . now look at the chain with bath temperatures @xmath188 and @xmath189 . as @xmath188 increases to @xmath40 , @xmath184 decreases to @xmath32 , so for @xmath184 sufficiently small , there is a unique @xmath34 for which @xmath190 is equal to @xmath184 . with @xmath172 and @xmath181 as above , and for fixed boundary conditions , the limit in the definition of @xmath8 is equivalent to @xmath191 such a limit involves difficult issues beyond the scope of this paper . for example , while the convergence of @xmath40 to @xmath192 as @xmath193 follows from lte ( assumption 2 ) , the convergence of @xmath194 to @xmath195 can not be deduced from previous assumptions . in addition to assumptions 1 and 2 in sect . [ sec : ey - eqs ] , we now introduce two other sets of assumptions on which the rest of our analysis relies . the functions @xmath196 and @xmath197 are well defined and finite - valued . 1 . in the context of ( [ phi - phin ] ) , @xmath198 as @xmath78 the function @xmath172 defined by ( [ c - phi ] ) is differentiable . assumption 3 asserts the presence of a well - defined structure at `` @xmath199 '' that governs the covariance relationships of the invariant measures @xmath66 for large @xmath4 . such a structure goes beyond the idea of lte to treat information of the next order , namely how @xmath66 deviates from local equilibrium at microscopic length scales , and how it deviates from products of gibbs measures ( at varying local temperatures ) globally . assumption 4 identifies some technical issues which we take for granted . this section and the next concern nearest - neighbor covariances in long chains , _ i.e. _ @xmath200 for large @xmath4 . we treat in this section the simpler case where there is no tracer flux in the system ; the equality @xmath201 is assumed throughout . our aim is to discover how the functions @xmath202 , equivalently @xmath170 , depend on the various quantities . in this subsection , we fix @xmath203 , and study the function @xmath204 where @xmath205 is as defined in sect . [ sec : cov - struct ] . here @xmath206 and @xmath207 are quantities associated with the midpoint of the chain ; in particular , @xmath207 is the temperature gradient at @xmath208 . we begin by examining the dependence of @xmath172 on @xmath207 . fix @xmath209 and @xmath73 , and consider @xmath210 that is to say , @xmath211},s_{[n/2]+1})$ ] for large @xmath4 with boundary conditions @xmath212 and @xmath201 . clearly , @xmath213 , and @xmath214 is an even function due to the left - right symmetry . this leads one to expect a quadratic response when the system is taken out of equilibrium . the following simulation confirms that the coefficient of the quadratic term is indeed nonzero . [ exp4 ] for various pairs of @xmath209 and @xmath73 , compute @xmath155},s_{[n/2]+1})$ ] for a sample of @xmath207 . a set of results are plotted in fig . [ fig - middle - dt ] . the data show that ( i ) @xmath215 , and in fact , ( ii ) @xmath216 is fairly well approximated by a quadratic function over the entire range @xmath217 of @xmath207 . ( the minimum temperature tends to @xmath32 as @xmath218 . ) note that if @xmath216 is smooth , then the only other possibility is @xmath219 for some @xmath220 , which is clearly not the case here . other data sets ( not shown ) with different values of @xmath221 confirm these conclusions . [ fig - middle - dt ] we investigate next the dependence of @xmath222 on @xmath209 and @xmath73 . first , we establish a simple result ( which is expected from dimensional analysis ) : [ lemma3.4 ] fix @xmath209 and @xmath73 , and view @xmath223 as a parameter . define @xmath224 and @xmath225 as in ( [ f ] ) . then @xmath226 given @xmath227 and @xmath207 , let @xmath228 and @xmath229 be the boundary conditions that give rise to these values at the two middle sites , and consider a second system with boundary conditions @xmath230 , @xmath231 and @xmath232 and initial tank energies @xmath233 times those of the first system . via a standard coupling argument , the sample paths for these two systems are easily matched . corresponding sample paths give rise to time evolutions with identical tracer counts in both systems but with the second system at energies @xmath233 times that of the first and running at speeds @xmath234 times that of the first . this leads to ( [ lambda - scaling ] ) . let @xmath235 . it follows from lemma [ lemma3.4 ] that @xmath236 thus @xmath237 is independent of @xmath234 , and @xmath238 can be written as @xmath239 where @xmath240 is the mean tracer count with respect to @xmath59 ( see sect . [ sec : ey - equilibrium ] ) . the preceding argument tells us that for any @xmath241 related as above , the function @xmath242 is well - defined , and @xmath243 a natural question is how @xmath244 depends on @xmath245 . other things being equal , one would think intuitively that local thermal fluctuations are smaller when more tracers are present , because stored energy is affected less by the entrance of a tracer with very large or very small energy . to confirm this , and to collect data for later use , we compute the function @xmath246 . practical reasons dictate that very short chains be used due to the large number of data points ( each one of which requiring a separate run ) and the time it takes for covariances to converge in long chains ; see appendix a for a discussion of relevant numerical issues . however , some understanding of the errors introduced by the use of very short chains is necessary if these data are to be useful in later predictions . [ fig - a - curves ] [ exp5 ] compute @xmath244 systematically for a range of @xmath245 using chains of 8 cells . compute also some values of @xmath244 using 16-chains for comparison . the 8-chain results are shown in fig . [ fig - a - curves ] . note that the function decreases monotonically as expected , and @xmath247 as @xmath248 . let @xmath249 and @xmath250 denote the values of @xmath244 computed using 8 cells and 16 cells respectively . we computed @xmath250 for some values of @xmath245 , and find these values to be somewhat larger than the corresponding values for @xmath251 . this is consistent with fig . [ fig - middle - n ] , and the true @xmath246-values ( defined at `` @xmath252 '' ) are likely to be larger still . we find , however , that the _ ratios _ of the two sets of values remain fairly constant as @xmath245 varies . for example , @xmath253.\ ] ] we will use the @xmath254-chain @xmath246-curve data in our study of long - chain covariances . all of our long - chain simulations involve @xmath245 in subintervals of the range shown above , and we will only assume that @xmath255 . we demonstrated in sect . [ sec : middle - sites ] that at @xmath256 , the response to the local temperature gradient @xmath207 is quadratic . we now examine the situation at @xmath257 . recall that our argument for showing that the leading term has order @xmath258 at @xmath208 uses the left - right symmetry of the chain , a property not present at @xmath257 . reasoning physically , however , it is hard to imagine that the response to local temperature gradient is sometimes quadratic , sometimes linear , or that the coefficient of the linear term would vanish at exactly @xmath208 independent of boundary conditions . the following simulation confirms this thinking . [ fig - offcenter ] [ exp6 ] for fixed @xmath84 , @xmath259 and @xmath73 , compute @xmath260 } , s_{[xn]+1})$ ] for various values of @xmath261 . repeat for various @xmath262 . [ fig - offcenter ] shows the results for some values of @xmath17 . the function @xmath263 for fixed @xmath17 , @xmath209 , and @xmath73 is clearly quadratic to leading order and contains a nonzero third - order term ( which is not present when @xmath264 ) . other sets of data , computed using different values of @xmath209 and @xmath73 , lead to the same conclusion . notice that a quadratic response to @xmath207 is consistent with the observation in sect . [ sec : total - var ] that @xmath265 . in sects . 4.1 and 4.2 , we showed that ( i ) for each fixed @xmath17 , @xmath209 , and @xmath73 , @xmath266 ( at least for @xmath267 not too large ) ; and ( ii ) at @xmath208 , the coefficient in front of @xmath268 is a function of @xmath269 , that is to say , @xmath270 , the function @xmath244 being _ defined _ by this relation at @xmath208 . with @xmath271 so defined , we now ask if the same relation holds at every @xmath84 , _ i.e. _ , if it is true that @xmath272 . consider , for the moment , a different situation in which the system is infinite in length , _ i.e. _ , @xmath273 , and a temperature gradient is maintained by , say , a constant external field . one would expect such a system to be translation - invariant , which implies that @xmath260 } , s_{[xn]+1 } ) = { { \rm cov}}_n(s_{[yn ] } , s_{[yn]+1})$ ] if the quantities @xmath274 at @xmath17 and @xmath275 are identical , _ i.e. _ , we would expect the answer to the question above to be affirmative . our system has no translation invariance to speak of and , at least for finite @xmath4 , the randomizing effects of the baths are apparent : nearest - neighbor covariances are significantly smaller near the boundaries because the baths inject tracers with _ i.i.d . _ sequences of energies into the system . the answer to the question posed above depends on how fast these randomizing effects dissipate . do they decrease by a certain amount per lattice site , or do they decrease with macroscopic distance ? the aim of the following simulation is to shed light on these questions . [ fig - fixed - local - conds ] [ exp7 ] fix a triplet @xmath276 and compute @xmath260},s_{[xn]+1})$ ] for various @xmath17 , using @xmath276 as the local conditions at @xmath17 . that is , for each @xmath17 , we first compute boundary conditions which give @xmath277 , @xmath278 , and @xmath279 , and then use these boundary conditions to compute the pair covariance at @xmath17 . repeat for various choices of @xmath276 . the results , shown in fig . [ fig - fixed - local - conds ] , show clearly that the randomizing effects of baths dissipate in a manner more like a diffusion process than by a definite amount per lattice site . note also the asymmetry of the data points : identical local conditions at @xmath17 and @xmath280 do not produce equal covariance relations . this suggests that from the conduction point of view , effective distance from the boundary may not equal physical distance . to capture the phenomenon observed , we introduce , for given boundary conditions @xmath281 and @xmath201 , a function @xmath282 defined by the expression @xmath283 we add the superscript `` bc '' to stress the fact that this function depends on boundary conditions . by definition , @xmath284 as @xmath285 . since we do not know of any other forces acting on the system , we think of @xmath286 as a measure of the randomizing effects of the baths at location @xmath17 , and continue our investigation based on this thinking . we now seek to identify the function @xmath286 . to capture the idea of `` effective distance to boundary , '' we introduce the following _ time - to - boundary _ functions : in an @xmath4-chain , let @xmath287 denote the expected time for a tracer at site @xmath1 to reach one of the baths . we then have the relation @xmath288 where @xmath40 is the temperature profile and @xmath289 the expected time a tracer spends at site @xmath1 . identifying the @xmath1th site with @xmath290 and letting @xmath291 , we obtain , as @xmath78 , the differential equation @xmath292 we define @xmath293 where @xmath294 is the solution of ( [ tau1 ] ) and the constant is chosen so that @xmath295 , and conjecture that @xmath296 is approximately equal to @xmath297 . note that @xmath297 also depends on boundary conditions . we now test this conjecture numerically : [ fig - dist - fn ] [ exp8 ] fix boundary conditions and @xmath4 . for @xmath298 , estimate @xmath282 by computing @xmath260},s_{[xn]+1})$ ] and dividing the result by @xmath299 . compare the resulting function to @xmath300 . the results are displayed in fig . [ fig - dist - fn ] . they show that to the degree that we are able to estimate these functions accurately , the two graphs are in near - perfect agreement . these graphs are not far from a perfect parabola , but the definitive presence of asymmetry in both fig . 7 and fig . 8 provides convincing evidence that at comparable distances , it takes longer to reach the bath at the lower - temperature end of the chain . [ [ remark . ] ] remark . + + + + + + + when the system is in equilibrium , _ i.e. _ , when @xmath301 , the solution of ( [ tau1 ] ) is easily computed to be @xmath302 . recall that the expression @xmath303 also appears in the formulas for spatial covariances in the simple exclusion @xcite and kmp models @xcite . an important difference between these models and ours is that there , the clocks signaling state changes ring at rate 1 , whereas in ours , they ring at energy - dependent rates . this property is responsible for , among other things , the slight asymmetry in @xmath294 . from the limited data available , we clearly can not conclude the validity of our conjecture , but details aside , the general ideas seem to point in the right direction . we return to the general case where both energy and tracer fluxes may be present . the situation here is more complex , and we are less able to separate the contributions of the various factors . we will focus on the dominant effects , and make some observations along the way . as before , we consider first @xmath208 . for fixed @xmath209 and @xmath73 , we consider the function @xmath305 and begin with a few observations on which terms may be absent in its taylor series . [ lemma4.2 ] @xmath306 for all @xmath80 . using the notation in sect . [ sec : ey ] , we claim that when @xmath307 , the invariant measure @xmath66 of the chain is the product @xmath308 where @xmath309 and @xmath310 for @xmath70 , so that stored energies at distinct sites are uncorrelated . this is proved by direct verification ( along the lines of the proof of proposition 3.7 in @xcite ) . see appendix b for details . interchanging @xmath38 with @xmath39 and @xmath35 with @xmath37 , we see that @xmath311 , so that there are no odd order terms in the polynomial approximation of @xmath312 . proposition [ lemma4.2 ] says there are no terms that are purely powers of @xmath80 , and we have shown in sect . [ sec : middle - sites ] that among the terms that are pure powers of @xmath207 , the leading one is @xmath268 . perturbing from equilibrium , we have shown that @xmath313 . consider the hessian @xmath314 . since @xmath315 ( proposition [ lemma4.2 ] ) , it follows that for @xmath316 , @xmath317 recall our brief discussion of the sign of @xmath318 in sect . [ sec : pair - covs ] . our next lemma says that the degeneracy of the hessian @xmath314 is equivalent to all covariances at @xmath208 having the same sign , namely the sign of @xmath319 . @xmath320 in a neighborhood of @xmath321 if and only if @xmath322 and @xmath323 . as a quadratic form , the hessian @xmath314 has maxima and minima in perpendicular directions . we know already that it is degenerate along the @xmath80-axis . thus either the minimum is negative , that is @xmath324 for some @xmath325 , or the hessian is positive semi - definite , _ i.e. _ @xmath322 and @xmath323 . ( 5,2.5)(0,0 ) ( -0.2,0 ) ( 2.2,1.1 ) term & coeff . ( @xmath326 ) & coeff . ( @xmath327 ) + @xmath328 & 0.094 & 0.067 + @xmath329 & @xmath330 & @xmath331 + @xmath332 & @xmath333 & @xmath334 + @xmath335 & @xmath336 & @xmath337 + @xmath338 & @xmath339 & @xmath340 + @xmath341 & @xmath342 & @xmath343 + [ fig - interp2d ] to determine the leading - order terms of the taylor expansion of @xmath214 near @xmath321 , in particular to see if @xmath344 ( equivalently @xmath345 ) , we perform the following simulation : [ exp9 ] fix @xmath209 and @xmath73 , and set @xmath346 , @xmath347 . compute @xmath348 for a grid of @xmath349 values and use the resulting data to find the best fitting 4th degree polynomial of the form @xmath350 two sets of results is shown in fig . [ fig - interp2d ] . we find that in these simulations ( and in others not shown here ) , the leading - order term in the taylor polynomial of @xmath312 is @xmath351 with @xmath352 . in all cases , @xmath353 , which is consistent with all observed covariances being @xmath354 . as expected , we can not conclude definitively that @xmath344 . from here on we take as a working assumption that the quadratic form associated with @xmath314 above is degenerate , _ i.e. _ , at @xmath208 , the only second order term is @xmath355 as in the @xmath120 case . for similar reasons as before , namely that the qualitative picture elsewhere in the chain should not differ from that at @xmath208 , we assume further the absence of @xmath356-terms in the taylor expansion of @xmath214 at all @xmath84 . this completes our discussion of how covariance depends on local quantities when the system is not far from equilibrium . we have shown that at @xmath208 , @xmath357 is a good approximation of the response to the local temperature and injection gradients _ provided @xmath267 and @xmath358 are sufficiently small_. in the @xmath359 case , we noted in sect . [ sec : middle - sites ] that the function @xmath360 is close but not equal to a perfect quadratic for large @xmath207 . we now investigate the effect of larger @xmath358 on @xmath361 . [ fig - far - from - eq ] [ exp10 ] for fixed values of @xmath221 , compute @xmath361 at @xmath208 as a function of @xmath207 and @xmath80 , and study the deviations from the function @xmath355 . a set of results is shown in fig . [ fig - far - from - eq ] . the results for other choices of @xmath221 ( not shown ) are similar . observe that ( i ) the dominant factor is @xmath274 , but a moderate correction ( roughly 20% for @xmath362 ) to the leading coefficient is sometimes needed ; and ( ii ) a larger @xmath358 tends to increase covariances . along the lines of the conjectural thinking that positive covariances are caused by waves of abnormally high ( or low ) energy tracers with long wavelengths , the presence of a tracer flux appears to amplify the propagation of these waves . we have seen that the presence of a tracer flux complicates the behavior of covariances at @xmath208 when the system is far from equlibrium . we do not know how it affects the propagation of boundary effects , or how to separate these contributions ( or if they can be separated at all ) . we will demonstrate in this subsection , however , that the _ main ingredients _ in @xmath144 are those already identified in sect . [ sec : nn - equal - rate ] . more precisely , let @xmath296 and @xmath297 be as defined in sect . [ sec : bdry - effects ] . that is to say , @xmath296 is as defined in ( [ defphi ] ) , @xmath294 the solution of ( [ tau1 ] ) , and @xmath363 where @xmath364 is chosen to ensure @xmath365 . we test the validity of @xmath366 } , s_{[xn]+1 } ) \approx \frac{1}{n } \cdot { \hat\varphi^{bc}}(x ) \cdot a(\kappa ( x ) ) \cdot t'(x)^2 \label{nn}\ ] ] as an approximate relation . [ exp11 ] for fixed boundary conditions with @xmath367 , and @xmath4 taken as large as possible , compare empirically computed values of @xmath144 to their predicted values given by ( [ nn ] ) . the results for two sets of boundary conditions are plotted in fig . [ fig - long - chains ] . notice that only the shapes of @xmath368 are being predicted because of unknown normalization factors . + [ fig - long - chains ] fig . [ fig - long - chains ] confirms many features of our predictions . first , the shapes of @xmath8 confirm that the randomizing effects of the heat baths are diffusive in nature ; these effects result in an effective - distance factor whose graph takes the form of a distorted parabola . second , the locations of the peaks of @xmath8 show the influence of @xmath207 ; see the left column of fig.11 . and @xmath246 increase from left to right ; in ( b ) @xmath268 increases from @xmath369 to @xmath370 while @xmath246 decreases from @xmath371 to @xmath372 . ] finally , comparison of the data for different @xmath4 in each set provide another confirmation of the @xmath373 scaling . [ fig - long - chains ] also shows the need for corrective factors on the order of @xmath374 away from the boundaries . while we do not know the precise nature of these corrections , we point out that in the presence of a tracer flux , the functions @xmath296 ( as computed from empirical data ) are a little more asymmetric than in the case @xmath120 ( shown in fig . [ fig - dist - fn ] ) . in sect . [ sec : cov - struct ] , we introduced the idea of a pair - covariance function @xmath162 . the existence of the limit in the definition of @xmath162 implies the following : for all sufficiently large @xmath375 , @xmath376},s_{[yn ] } ) \approx n'\cdot{{\rm cov}}_{n'}(s_{[xn']},s_{[yn']})\ .\ ] ] this can be seen as a statement about the _ renormalizability _ of pair covariances : consider an @xmath4-chain with @xmath377 for some integers @xmath378 and @xmath379 , and subdivide the chain into @xmath378 groups of @xmath379 consecutive sites . for convenience , we take @xmath379 to be odd . let @xmath380 , _ i.e. _ , @xmath381 is the index of the middle of the @xmath379 sites in the @xmath1th group . for any @xmath137 with @xmath382 and @xmath383 , we compare the covariance at sites @xmath1 and @xmath150 in an @xmath378-chain to that at sites @xmath381 and @xmath384 in an @xmath4-chain with the same boundary conditions . ( [ eqn : limit equiv ] ) tells us that @xmath385 if @xmath378 is sufficiently large , and that as @xmath386 , @xmath387 converges to a constant . we test this renormalizability to confirm that @xmath162 is well defined . [ fig - renorm ] [ exp12 ] fix @xmath38 , @xmath39 , @xmath35 , @xmath37 , and @xmath388 . for various pairs @xmath389 in the @xmath378-chain , compute @xmath390 for @xmath377 , @xmath391 . a subset of the data is shown in fig . [ fig - renorm ] . because the value of @xmath378 used is relatively small ( @xmath392 ) , one can expect the plotted values @xmath393 to converge or stabilize to constants only as @xmath379 increases . the results show that they , in fact , stabilize fairly quickly . that @xmath162 is well defined implies that the function @xmath394 } , s_{[yn]})$ ] , when normalized , settles down to a fixed shape for large @xmath4 . we now investigate the shapes of these functions . while carrying out simulations [ exp8 ] and [ exp12 ] , we also collected data for @xmath2 for various pairs of @xmath137 . graphs of @xmath395 with @xmath396 $ ] are shown in fig . [ fig - long - dist - covs ] for three sets of boundary conditions . as predicted in sect . [ sec : pair - covs ] , these functions are bounded , and they decrease monotonically to @xmath32 as @xmath143 increases . this decay rate is roughly linear in macroscopic distance , and extraordinarily slow per lattice site . for example , if @xmath397 and @xmath398 $ ] , then @xmath399 . we note that these findings are consistent with fluctuating hydrodynamics ; see _ e.g. _ , @xcite . [ cols="^,^,^ " , ] [ fig - long - dist - covs ] a natural generalization of the nearest - neighbor covariance formula ( [ nn ] ) to pair covariances separated by macroscopic distances is @xmath366 } , s_{[yn ] } ) \approx \frac{1}{n } \cdot \varphi^{bc}_2(x , y ) \cdot \bar a(x , y ) \cdot \bar t'(x , y)^2\ . \label{c2}\ ] ] we think of this as an approximate formula that holds for @xmath400 with @xmath401 : @xmath402 is a notion of effective distance of _ the pair _ @xmath403 to the boundary , and @xmath404 and @xmath405 are generalizations of corresponding quantities in ( [ nn ] ) . as a rough approximation , one may take @xmath406 [ fig-2pt - dist - fn ] to obtain information on @xmath402 , we estimate it using ( [ c2 ] ) , taking @xmath404 and @xmath405 as above . slices of the graphs of these functions of 2 variables , with @xmath17 fixed and @xmath275 varying , are shown in fig . [ fig-2pt - dist - fn ] . piecing together these slices , we deduce the following geometric facts about our 2-point distance function @xmath407 defined for @xmath408 : it is continuous and piecewise smooth , with a `` ridge '' along the line @xmath409 . along that line , it is equal to @xmath410 , the value of the 1-point distance function introduced in sect . [ sec : bdry - effects ] . for fixed @xmath17 , the function @xmath411 is roughly piecewise linear , peaking at @xmath19 . observe that if this function were _ exactly _ piecewise linear , then by exchanging @xmath17 and @xmath275 , we would arrive at the relation @xmath412 this relation implies that @xmath413 for some constant @xmath364 . since @xmath296 is roughly parabolic ( see sects . [ sec : bdry - effects ] and [ sec : long - dist - covs ] ) , the approximate piecewise linearity of @xmath407 is consistent with our results from previous sections . ( exact formulas for the simple exclusion @xcite and kmp models @xcite also contain piecewise linear functions . ) on the other hand , we know from sections 4 and 5 that for the random halves model , @xmath414 . thus @xmath407 can not be exactly piecewise linear either . in summary , our results for long - range covariances are rough and are obtained by extrapolating from what we know about covariances at microscopic distances , together with a numerical determination of the distance function @xmath407 . an approximate formula is @xmath366 } , s_{[yn ] } ) \ \approx \ \begin{cases } \ \frac{1}{n } \cdot \left(\frac{1-y}{1-x}\right ) \cdot { \hat\varphi^{bc}}(x ) \cdot \bar a(x , y ) \left(\frac{t(y)-t(x)}{y - x}\right)^2 \ \ \ \ { \rm for } \ \ \ \ 0 \leq x \leq y \leq 1\ , \\ \ \frac{1}{n } \cdot \left(\frac{y}{x}\right ) \cdot { \hat\varphi^{bc}}(x ) \cdot \bar a(x , y ) \left ( \frac{t(y)-t(x)}{y - x } \right)^2 \ \ \ \ \quad { \rm for } \ \ \ \ 0 \leq y \leq x \leq 1 \ , \end{cases}\ ] ] where @xmath297 is our approximate one - point distance function and @xmath404 is as in ( [ avg ] ) . both this formula and fig . [ fig - long - dist - covs ] show clearly the following : 1 . covariance decays essentially linearly with macroscopic distance , _ i.e. _ , @xmath415 2 . for fixed @xmath17 , the function @xmath18 is continuous but not differentiable at @xmath19 ; it has opposite concavity on the two sides of @xmath17 if the temperature profile @xmath416 is nonlinear . that the curvature changes sign at @xmath17 is clearly visible in fig . [ fig - long - dist - covs ] ; it is also evident from the second derivative of @xmath417 ( see fig . [ fig - long - chains ] , left column , for the temperature profiles @xmath416 ) . via a series of theoretical arguments and numerical simulations , we have developed a coherent picture for the spatial covariances at steady state of the 1-d random halves model . we have established firmly that stored - energy covariances have order of magnitude @xmath418 away from the boundaries of @xmath4-chains . this in itself points to the presence of long - range covariances which decay very slowly . subsequent analysis shows that covariances decay _ linearly _ with macroscopic distance . for sites separated by microscopic distances , we have a simple formula that encapsulates the main ingredients on which energy covariances depend , including ( i ) a quadratic response to local temperature gradient , ( ii ) diffusive nature of the randomizing effect of the heat baths , and ( iii ) stabilizing effects of large numbers of tracers . since the random halves models are stochastic idealizations of certain mechanical models , we hope our results will also shed light on these and similar hamiltonian systems . there are , however , important differences , such as mixing and memory issues ( see _ e.g. _ @xcite ) . the extent to which the picture established here carries over to the hamiltonian setting remains to be seen . finally , it is well known that nonequilibrium phenomena are quite different in higher dimensions . our detailed study here provides a baseline for explorations in 2-d and 3-d , a project currently being carried out by the authors . the results , which are indeed different as predicted by fluctuating hydrodynamics , will be reported in a forthcoming paper . .2 in _ acknowledgements . _ we are grateful to jonathan goodman , leo kadanoff , oscar lanford , joel lebowitz , charles newman , luc rey - bellet and david ruelle for helpful conversations . the simulations carried out in this paper implement directly the dynamics described in sect . [ sec : ey - desc ] ; expectation values with respect to the invariant measures are computed via time averages over long trajectories . the numerical issues are quite similar to those of markov chain monte carlo computations . we calculate empirical error bars using a standard `` batch means '' estimator . these error bars measure only statistical errors that arise from the fact that expectation values are estimated by time - averaging over finite intervals of time , _ i.e. _ finite - length trajectories . the error bars do not account for _ finite - size effects _ , _ i.e. _ bias due to the fact that we can only simulate @xmath4-chains with finite @xmath4 . to improve clarity and readability , we have suppressed the error bars in most of the figures , displaying only those that are directly relevant to the issues being discussed . a variety of variance reduction techniques have been invented to speed up the convergence of monte carlo calculations , ranging from multigrid monte carlo @xcite to large deviations - based importance sampling @xcite . most of these techniques require additional information , such as an explicit expression for the invariant measure or a large deviations functional . such methods do not work in our setting . there is on - going work on a class of algorithms which do not require detailed knowledge of the invariant measure @xcite . however , it is not known whether such techniques can be applied to the random halves model . the quantities of interest in most of our simulations are the @xmath419 . in general , these are relatively small numbers that are differences of two much larger numbers and can be rather costly to compute . we estimate the cost of computing @xmath420 as follows : for temperatures which are @xmath421 , both @xmath422 and @xmath423 are @xmath424 , while @xmath425 , where @xmath426 is the typical tracer density per site . ( for the present discussion , we focus on @xmath137 away from the boundaries so that @xmath427 . ) in order to compute @xmath2 with a relative error of @xmath428 , we need @xmath429 with a relative error of @xmath430 ; the same is true for @xmath423 . since the statistical error in the time average @xmath431 is @xmath432 for some @xmath433 , this means we need to integrate the system for a time @xmath294 which is proportional to @xmath434 . now , the computational cost of simulating the system up to time @xmath294 , as measured by the total number of `` events , '' is proportional to the number of tracers @xmath435 in the system and the mean rate of activity of each tracer ( which is @xmath436 ) . thus , we have @xmath437 for our simulations involving long chains , we have found @xmath433 to be typically @xmath438 . as an example , for @xmath439 , @xmath440 , @xmath441 , @xmath442 , @xmath443 , we have @xmath444 , and @xmath445 , so that @xmath446 events are needed . on our computer system , this requires @xmath447 days . ( [ eqn : runtime ] ) tells us that the computational cost grows rapidly as @xmath448 decreases . in simulations [ exp4 ] , [ exp6 ] , and [ exp9 ] , where we study the small-@xmath207 behavior of the covariance function , this rapid growth prevents us from taking @xmath207 too small . however , we do not always need small @xmath448 , and a large temperature gradient not only reduces the computational cost , it also takes the system farther out of equilibrium so that some effects are made more transparent . the cost also grows rapidly as @xmath4 increases . this is why we use short chains wherever possible , for example in computing the @xmath246-curve . in simulation [ exp5 ] , we use @xmath449 because we feel that the amount of accuracy gained from using longer chains is perhaps not worth the additional cost . however , it is not always possible or advisable to use short chains , particularly where infinite - volume limits are involved . in such cases , one can sometimes obtain better results by increasing @xmath448 . _ proof of proposition [ lemma4.2 ] : _ the proof of invariance of the measure @xmath453 follows closely that of proposition 3.7 in @xcite . we refer the reader to @xcite for some of the background notation . fix a phase point @xmath454 we assume @xmath455 are distinct , and let @xmath456 be arbitrarily small numbers . fix arbitrary @xmath457 with @xmath458 . we will compare the 3 probabilities , @xmath459 , @xmath460 and @xmath461 defined below : let @xmath462 be the probability that at time @xmath463 , in every site @xmath150 , there are exactly @xmath464 tracers the energies of which lie in disjoint intervals @xmath465 , \qquad \dots , \qquad [ \bar x^{(j)}_{k_{j } } , \bar x^{(j)}_{k_{j}}+\varepsilon]\ , \ ] ] _ and _ before @xmath466 , the tracer in site @xmath457 with energy in @xmath467 $ ] exits the site . the number @xmath460 is the probability that at time @xmath32 , the tracer configuration is as above except that the tracer with energy in @xmath467 $ ] is in site @xmath468 instead of site @xmath457 ( _ i.e. _ , there are @xmath469 tracers in site @xmath457 and @xmath470 tracers in site @xmath468 ) , _ and _ before time @xmath466 , this tracer jumps from site @xmath468 to site @xmath457 . the number @xmath461 is defined analogously with site @xmath471 playing the role of site @xmath468 . to prove the invariance of @xmath472 , three sets of balancing conditions have to be met . the interaction with tanks is as in [ ey ] . we verify below the equation involving interaction between neighbors , namely @xmath473 , and leave the one involving interaction with a bath to the reader . let @xmath474 and @xmath475 be as in proposition [ prop2.1 ] ( sect . [ sec : ey - equilibrium ] ) . we will use the shorthand @xmath476 , and write @xmath477 to remind ourselves that @xmath478 at site @xmath1 ( @xmath479 and @xmath480 do not depend on @xmath73 ) . first , @xmath481 where @xmath482 is defined by the equality above . next , @xmath483 where @xmath484 this product can be written as @xmath485 the equality above follows from the relation @xmath486 which can be derived from the characterization of @xmath59 ( proposition [ prop2.1 ] ) . an analogous argument holds for @xmath461 , and the desired equality follows from @xmath487 @xmath488

a recent time - of - flight ( tof ) expansion experiment with polarized fermionic erbium atoms measured a fermi surface deformation from a sphere to an ellipsoid due to dipole - dipole interaction , thus confirming previous theoretical predictions . here we perform a systematic study of the ground - state properties and tof dynamics for trapped dipolar fermi gases from the collisionless to the hydrodynamic regime at zero temperature . to this end we solve analytically the underlying boltzmann - vlasov equation within the relaxation - time approximation in the vicinity of equilibrium by using a suitable rescaling of the equilibrium distribution . the resulting ordinary differential equations for the respective scaling parameters are then solved numerically for experimentally realistic parameters and relaxation times that correspond to the collisionless , collisional , and hydrodynamic regime . the equations for the collisional regime are first solved in the approximation of a fixed relaxation time , and then this approach is extended to include a self - consistent determination of the relaxation time . the presented analytical and numerical results are relevant for a detailed quantitative understanding of ongoing experiments and the design of future experiments with ultracold fermionic dipolar atoms and molecules . in particular , the obtained results are relevant for systems with strong dipole - dipole interaction , which turn out to affect significantly the aspect ratios during the tof expansion . _ keywords _ : ultracold fermi gas , dipolar interaction , boltzmann - vlasov equation

1608.06448

atomic and molecular ultracold gases offer many advantages for studying quantum phenomena , especially within the realm of many - body physics , due to the high degree of tunability of inter - atomic interactions @xcite . in particular , dipolar quantum gases of atoms and molecules have received much attention in recent years , as the anisotropic and long - range nature of the magnetic or electric dipole - dipole interaction ( ddi ) gives rise to a rich spectrum of novel properties in such systems @xcite . such systems include those made up of ultracold atoms , as well as those consisting of heteronuclear molecules with large dipole - dipole interactions . furthermore , in the recent theoretical and experimental research @xcite a novel kind of strongly dipolar quantum gas was introduced . these are weakly bound polar molecules produced from atoms with large magnetic dipole moments , such as erbium and other lanthanides . these molecules can have a very large magnetic moment , which amounts to twice that of its individual atoms @xcite . in 2005 an anisotropic deformation of the expanding dipolar bosonic chromium condensate due to ddi was observed @xcite . in the recent experiment @xcite , also the rosensweig instability was detected in a @xmath0dy bose - einstein condensate , which represents a quantum ferrofluid due to the large atomic magnetic dipole moments . namely , after a sudden decrease of the scattering length , the dipolar quantum gas creates self - ordered surface structures in form of droplet crystals , which can only be understood by taking into account ddi @xcite and the corresponding quantum fluctuations @xcite . for dipolar fermi gases it was predicted that the long - range and anisotropic ddi leads in equilibrium to an anisotropic deformation of the fermi surface from a sphere to an ellipsoid @xcite . a recent time - of - flight ( tof ) expansion experiment has now unambiguously detected such an ellipsoidal fermi surface ( fs ) deformation in a dipolar quantum gas of fermionic erbium atoms , which turns out to be of the order of a few percent @xcite . within the hartree - fock mean - field theory for a many - body system , first - order contributions of ddi to the total energy of the system taken into account are in terms of both the hartree direct interaction and the fock exchange interaction @xcite . in the case of a fermi gas with isotropic interaction , the hartree and the fock interactions cancel out @xcite , thus leading to a spherically symmetric fs . but in the case of a fermi gas with anisotropic ddi the hartree term gives rise to a distortion in real space @xcite , whereas the fock term gives rise to a distortion in momentum space , i.e. , to an ellipsoidal deformation of the fermi sphere . note in this context that the fock exchange term in dipolar fermi gases is the consequence of a combined effect of the ddi and the pauli exclusion principle . in the current experimentally relevant range of dipolar interactions the theory beyond hartree - fock , where the total energy is determined up to second - order in the ddi , yields only small differences , which can not yet be resolved experimentally . therefore , the hartree - fock mean - field approximation yields already quantitatively accurate results for present - day experiments @xcite . the investigations of collective oscillations and tof dynamics of dipolar fermi gases have so far focused on either the collisionless ( cl ) regime @xcite , where collisions can be neglected , or on the hydrodynamic ( hd ) regime @xcite , where collisions occur so often that local equilibrium can be assumed . the recent paper of wchtler _ et al . _ @xcite studied even the behavior of collective oscillations when the system undergoes a crossover from one regime to the other . motivated by the experimental observation of the ellipsoidal fs deformation in the tof experiment @xcite , we continue here the analytical analysis along the lines of reference @xcite and investigate in detail the expansion dynamics for the collisional regime , which represents the transition zone between the limiting cl and hd regimes . we also extend previous approaches based on the relaxation - time approximation by introducing a self - consistently determined relaxation time , and study how this quantitatively affects the tof dynamics . the paper is structured as follows . in section [ sec : amdfg ] we introduce our notation and summarize recent experiments on atomic and molecular dipolar fermi gases . in section [ sec : ge ] we analyze the global equilibrium of the system by minimizing the hartree - fock total energy in order to obtain the thomas - fermi radii and momenta . afterwards , in order to study the dynamics , in section [ sec : bve ] we follow reference @xcite and introduce the boltzmann - vlasov equation for dipolar fermi gases as well as an approximative solution , which is based on a suitable rescaling ansatz for the equilibrium wigner function . in section [ sec : tof ] we study in detail the tof expansion dynamics of an initially trapped fermi gas . to this end we present our analytical and numerical results of the tof analysis all the way from the collisionless to the hydrodynamic regime and reveal how the expanding cloud bears the signature of the underlying ddi . finally , section [ sec : con ] gathers our concluding remarks and gives an outlook for future research . we consider a trapped ultracold quantum degenerate dipolar gas of single - component fermions of mass @xmath1 and magnetic dipole moment @xmath2 or electric dipole moment @xmath3 at zero temperature . the system is then described by the second - quantized hamiltonian @xmath4\hat{\psi } ( \mathbf{r})\nonumber\\ \hspace*{10mm}+\frac{1}{2 } \int d\mathbf{r}\int d\mathbf{r'}\ , \hat{\psi}^\dagger(\mathbf{r'})\hat{\psi}^\dagger(\mathbf{r } ) v_\mathrm{int}({\bf r}-{\bf r'})\hat{\psi } ( \mathbf{r})\hat{\psi } ( \mathbf{r'})\ , . \label{eq : hamiltonian } \end{aligned}\ ] ] since the pauli exclusion principle inhibits contact interaction , the long - range ddi between the polarized fermionic point dipoles is dominant . it is described by @xmath5 where @xmath6 denotes the relative position between the dipoles , @xmath7 stands for the angle between @xmath6 and the polarization axis of the dipoles , and @xmath8 represents the dipolar interaction strength , which depends on the nature of the dipoles . namely , for electric dipoles it is defined as @xmath9 , where @xmath10 is the vacuum permittivity , while for magnetic dipoles one has @xmath11 , where @xmath12 is the vacuum permeability . magnetic dipolar moments are usually measured in units of bohr magneton ( @xmath13 ) , and electric dipolar moments in units of debye ( @xmath14 ) . note that the ddi of polar molecules is about @xmath15 times stronger than that of dipolar atoms , as @xmath16 , with @xmath17 being the sommerfeld fine - structure constant . er @xcite . axes @xmath18 , @xmath19 , @xmath20 correspond to trap axes , while the orientation of the magnetic field @xmath21 and the atomic dipoles is determined by spherical coordinates @xmath22 and @xmath23 . earth s gravitational field is parallel to @xmath20 axis , while imaging axis is denoted by @xmath24 , lies in the @xmath25 plane , and forms an angle @xmath26 with @xmath19 axis.,width=253 ] due to the anisotropy in the dipolar interaction potential ( [ ddi ] ) , dipolar fermi gases tend to be stretched along the polarization direction , since this leads to a lower total energy . here we consider the parameters of a recent innsbruck experiment @xcite performed with the fermionic erbium atoms @xmath27er in the collisionless regime , which are confined into a three - dimensional optical dipole trap with frequencies @xmath28 . it contained @xmath29 atoms at a temperature @xmath30 of @xmath31 , with the fermi temperature being @xmath32 . the underlying geometry of the experiment is depicted in figure [ fig : fig1 ] . gravity is oriented along the @xmath20 direction . the atomic cloud is imaged along the @xmath24 axis , which forms an angle @xmath33 with respect to the @xmath19 axis . the magnetic field @xmath34 forms an angle @xmath22 with the @xmath20 axis and lies in the @xmath35 plane , which is rotated for an angle @xmath36 with respect to the @xmath37 plane . in the following we restrict ourselves to the general geometry of the anisotropic trap , where the dipoles are oriented in the direction of one of the trap axis , which reflects the experimental situation at the two limiting cases @xmath38 and @xmath39 . previous theoretical works have predicted that the degree of deformation of the fs depends on the fermi energy and the strength of the dipole moment @xcite , therefore we use a relative interaction strength of the ddi when comparing its effect on different species of ultracold fermi gases . the relative interaction strength is given by @xmath40 where @xmath41 denotes the geometric mean of the trap frequencies . .dipole moments ( @xmath42 for species with a magnetic dipole and @xmath43 for species with an electric dipole ) and relative interaction strengths of fermionic atoms and molecules currently used in ultracold experiments . [ cols="^,^,^,^,^,^,^",options="header " , ] the available dipolar fermi gases in current ultracold experiments are listed in table [ tab : tab1 ] , together with their corresponding dipole moments and relative interaction strengths considering the trap parameters and particle number given above . the first quantum degenerate fermi gas of the strongly magnetic atoms of dysprosium was produced in 2012 @xcite . later on , a degenerate fermi gas of erbium atoms @xcite and molecules @xcite was also realized . a dipolar fermi sea of degenerate @xmath44cr , together with a bec of @xmath45cr was produced in 2015 @xcite . few years ago a molecular dipolar gas of @xmath46k@xmath47rb was realized using a single step of stirap @xcite ( stimulated raman adiabatic passage ) with two - frequency laser irradiation @xcite , and with the same technique a new ultracold dipolar gas of fermionic molecules of @xmath48na@xmath46k was created in a recent experiment @xcite . the same technique can be also used for thermal @xcite and ultracold dipolar bose gases @xcite of heteronuclear molecues . we will consider the experimentally available range of relative strengths of the ddi and atom or molecule species given in table i in the following sections . therefore , the presented results are directly applicable to current and future experiments . a quantum many - body system can be described in terms of a wigner function @xmath49 , as it represents the wigner - weyl transform of the density matrix of the system and is equivalent to a quantum - mechanical wave function @xcite . the wigner function is a quasiprobability distribution function , and integrating it over the space or the momentum variables leads to the respective probability distribution functions . the quantum - mechanical expectation values of observables @xcite can be obtained as their phase - space averages , weighted by the wigner function . considering a trapped ultracold dipolar fermi gas , the equilibrium distribution function in the phase space will rapidly decrease to zero outside a certain closed surface , due to a combined effect of the pauli exclusion principle , which is responsible for a formation of the fs in the momentum space , and the trapping in real space . therefore , in order to model the global equilibrium distribution of the dipolar fermi gas we use an ansatz for the semiclassical wigner function , which resembles the form of the wigner - transformed fermi - dirac distribution of a noninteracting fermi gas . note that the temperature of the dipolar fermi gas in the experiment @xcite is that low that thermal fluctuations are expected to be of the order of @xmath50 due to the sommerfeld expansion . this justifies to use for the wigner - transformed fermi - dirac distribution of a noninteracting fermi gas the zero - temperature approximation : @xmath51 here @xmath52 is the heaviside step function . the variational parameters @xmath53 and @xmath54 represent the thomas - fermi ( tf ) radius and the fermi momentum in the @xmath55-th direction , respectively , and describe the extension of the equilibrium fermi surface in both coordinate and momentum space . with this ansatz , the normalization of the distribution @xmath56 to @xmath57 fermions leads to the condition @xmath58 where @xmath59 and @xmath60 denote the geometric means of the respective tf radii and momenta . the total energy of the system in the hartree - fock approximation for dipoles oriented along @xmath20 axis , i.e. , @xmath38 is given by @xmath61 where @xmath62 , while the function @xmath63 and its derivatives with respect to the first and second argument , @xmath64 and @xmath65 , respectively , are anisotropy functions defined in references @xcite . note that the corresponding expression for the hartree - fock energy of the second considered case of dipoles oriented along @xmath18 axis , i.e. , @xmath39 is obtained by a simple cyclic permutation of indices @xmath66 in equation ( [ eq : totalenergy ] ) . the same applies to all other equations throughout the paper . the tf radii and momenta @xmath53 and @xmath54 are determined by minimizing the energy ( [ eq : totalenergy ] ) with respect to them , which leads to the following set of algebraic equations : @xmath67=0\ , , \label{vstrx}\\ \hspace*{-15mm}\omega_y^2r_y^2-\frac{1}{3}\sum_j\frac{\hbar^2k_j^2}{m^2}-\frac{48nc_0}{m\bar{r}^3 } f\left ( \frac{k_z}{k_x},\frac{k_z}{k_y } \right ) \nonumber\\ \hspace*{-4.5mm}+\frac{48nc_0}{m\bar{r}^3}\left [ f\left ( \frac{r_x}{r_z},\frac{r_y}{r_z}\right)-\frac{r_y}{r_z } f_2\left(\frac{r_x}{r_z},\frac{r_y}{r_z}\right ) \right]=0\ , , \label{vstry}\\ \hspace*{-15mm}\omega_z^2r_z^2-\frac{1}{3}\sum_j\frac{\hbar^2k_j^2}{m^2 } -\frac{48nc_0}{m\bar{r}^3 } f\left ( \frac{k_z}{k_x},\frac{k_z}{k_y } \right ) \nonumber\\ \hspace*{-4.5mm}+\frac{48nc_0}{m\bar{r}^3}\left [ f\left ( \frac{r_x}{r_z},\frac{r_y}{r_z}\right)+\frac{r_x}{r_z } f_1\left(\frac{r_x}{r_z},\frac{r_y}{r_z}\right)+\frac{r_y}{r_z } f_2\left(\frac{r_x}{r_z},\frac{r_y}{r_z}\right ) \right]=0\ , . \label{vstrz } \end{aligned}\ ] ] note that equations ( [ 1])([3 ] ) are linearly dependent , and due to the symmetry of the anisotropy function @xmath68 they can be reduced to two independent equations @xcite , @xmath69,\label{k2 } \end{aligned}\ ] ] where @xmath70 denotes the diagonal part of the anisotropy function . this implies that the momentum distribution of a dipolar fermi gas in global equilibrium remains cylindrically symmetric despite a general triaxial harmonic confinement @xcite . due to the anisotropy of the dipolar interaction potential , dipolar quantum gases tend to be stretched along the polarization direction , i.e. , the direction of an external magnetic or electric field , since this leads to a lower total energy . we note that this is valid not only for fermions , but for bosons as well @xcite . equations ( [ vstrx])([k2 ] ) , together with equation ( [ eq : particlenumberconservation ] ) , represent a closed set of six algebraic equations , which fix all variational parameters @xmath53 and @xmath54 in global equilibrium . for fermi gases in global equilibrium for considered trap geometry with dipoles parallel to @xmath20 axis : ( a ) @xmath71 , ( b ) @xmath72 , and ( c ) @xmath73 . black up - pointing triangles represent aspect ratios for the limiting case of a noninteracting fermi gas : in real space @xmath74 and @xmath75 , while in momentum space @xmath76 ( fermi sphere ) . other symbols represent aspect ratios for dipolar atoms and molecules from table [ tab : tab1 ] : @xmath44cr ( gray down - pointing triangles ) , @xmath27er ( red circles ) , @xmath77dy ( blue diamonds ) , @xmath46k@xmath47rb ( purple squares ) , @xmath27er@xmath78er ( green stars ) . red vertical line corresponds to a critical value of the relative dipolar interaction strength @xmath79 for considered trap geometry ; for @xmath80 no stable stationary solution exists for a system of equations ( [ vstrx])([k2 ] ) , together with equation ( [ eq : particlenumberconservation]).,title="fig:",width=188 ] for fermi gases in global equilibrium for considered trap geometry with dipoles parallel to @xmath20 axis : ( a ) @xmath71 , ( b ) @xmath72 , and ( c ) @xmath73 . black up - pointing triangles represent aspect ratios for the limiting case of a noninteracting fermi gas : in real space @xmath74 and @xmath75 , while in momentum space @xmath76 ( fermi sphere ) . other symbols represent aspect ratios for dipolar atoms and molecules from table [ tab : tab1 ] : @xmath44cr ( gray down - pointing triangles ) , @xmath27er ( red circles ) , @xmath77dy ( blue diamonds ) , @xmath46k@xmath47rb ( purple squares ) , @xmath27er@xmath78er ( green stars ) . red vertical line corresponds to a critical value of the relative dipolar interaction strength @xmath79 for considered trap geometry ; for @xmath80 no stable stationary solution exists for a system of equations ( [ vstrx])([k2 ] ) , together with equation ( [ eq : particlenumberconservation]).,title="fig:",width=188 ] for fermi gases in global equilibrium for considered trap geometry with dipoles parallel to @xmath20 axis : ( a ) @xmath71 , ( b ) @xmath72 , and ( c ) @xmath73 . black up - pointing triangles represent aspect ratios for the limiting case of a noninteracting fermi gas : in real space @xmath74 and @xmath75 , while in momentum space @xmath76 ( fermi sphere ) . other symbols represent aspect ratios for dipolar atoms and molecules from table [ tab : tab1 ] : @xmath44cr ( gray down - pointing triangles ) , @xmath27er ( red circles ) , @xmath77dy ( blue diamonds ) , @xmath46k@xmath47rb ( purple squares ) , @xmath27er@xmath78er ( green stars ) . red vertical line corresponds to a critical value of the relative dipolar interaction strength @xmath79 for considered trap geometry ; for @xmath80 no stable stationary solution exists for a system of equations ( [ vstrx])([k2 ] ) , together with equation ( [ eq : particlenumberconservation]).,title="fig:",width=188 ] for weak enough interactions , a local minimum might exist to which the system would return after a small perturbation . the regions of system parameters satisfying this property are called stable and the mathematical criterion behind this classification is given by positive eigenvalues of the hessian matrix of the energy functional @xcite . figure [ fig : fig2 ] depicts aspect ratios of stable solutions , i.e. , the deformation of the fermi surface in real and momentum space in global equilibrium for the dipolar fermi gases given in table [ tab : tab1 ] . these results are obtained for the dipoles oriented in the direction of the @xmath20 axis , i.e. , for the angle @xmath38 ( see figure [ fig : fig1 ] ) . for the limiting case of a noninteracting fermi gas we know that the aspect ratios in real space are @xmath74 and @xmath75 , while in momentum space the fermi surface becomes the fermi sphere and therefore we have @xmath76 . in figure [ fig : fig2 ] corresponding aspect ratios for the noninteracting case are shown as black up - pointing triangles in comparison with aspect ratios for interacting fermi gases . for atomic gases of @xmath44cr , @xmath27er , and @xmath77dy the ddi is not that strong , and their aspect ratios in momentum space deviate less than @xmath81 from unity , see figure [ fig : fig2](c ) . actually , for @xmath44cr ( gray down - pointing triangles ) the aspect ratio in momentum space is just @xmath82 smaller than 1 , which is quite challenging to be observable in an experiment . nevertheless , for @xmath27er ( red circles ) the aspect ratio in momentum space turns out to be about @xmath83 less than 1 and has already been experimentally observed in reference @xcite , meaning that the @xmath81 deformation for @xmath77dy ( blue diamonds ) should also be observable . by using an external electric field and tuning the electric dipole moment of @xmath46k@xmath47rb ( purple squares ) to @xmath84 , one obtains the relative interaction strength @xmath85 , which leads to a large predicted fs deformation of about @xmath86 . furthermore , a molecule of @xmath78er@xmath27er ( green stars ) with @xmath87 would yield a fs deformation of nearly @xmath88 . red vertical lines in figure [ fig : fig2 ] represent a critical value of the relative interaction strength @xmath89 for the considered trap geometry . namely , for @xmath80 stable stationary solutions for equations ( [ vstrx])([k2 ] ) , together with equation ( [ eq : particlenumberconservation ] ) , do not exist @xcite for system parameters from the innsbruck experiment @xcite . note that the value of @xmath90 does not depend on the mass of the species and is universal for a given trap geometry , as can be shown by rewriting equations ( [ vstrx])([k2 ] ) in the dimensionless form . note that the molecule @xmath48na@xmath46k with the largest @xmath91 in table [ tab : tab1 ] is unstable for the used system parameters and therefore we will not consider it in the rest of this paper . the dynamics of a trapped ultracold dipolar degenerate fermi gas can be described in terms of the botzmann - vlasov ( bv ) equation , which was previously prominently used in the realm of nuclear @xcite and plasma @xcite physics . the bv equation determines the time evolution of the wigner function @xmath92 @xcite and for a dipolar fermi gas it reads : @xmath93({\bf{r , k}},t)\ , . \label{eq : bve}\end{aligned}\ ] ] here @xmath94 denotes the mean - field potential , which includes external trap potential , as well as the respective hartree and fock terms , where @xmath95 represents the ddi potential ( [ ddi ] ) and @xmath96 its fourier transform . note that this hartree - fock dynamic mean - field description is self - consistent and is of the first order in the interaction potential . on the right - hand side of equation ( [ eq : bve ] ) we have the collision integral @xmath97({\bf{r , k}},t)$ ] which is of second order of the interaction potential and describes collisions between two particles @xcite . instead of using a full expression for the collision integral , which would require a detailed modeling of scattering processes between atoms or molecules , we apply here the relaxation - time approximation @xcite in the form @xmath98=-\frac{\nu({\bf r , k } , t)-\nu^\mathrm{le}({\bf r , k})}{\tau}\ , . \label{eq : rta}\ ] ] here @xmath99 denotes the relaxation time , which is related to the average time between collisions , and @xmath100 stands for the distribution function corresponding to local equilibrium . the physical idea is that the particles interact via collisions and exchange energy and momentum , which eventually leads to a relaxation of the system into a local equilibrium state in which the collisions will no longer change the distribution function . in contrast to that , the local velocity field or the density can still be spatially dependent . the local thermodynamical equilibrium of a dipolar fermi gas is defined by @xmath101=0 $ ] . if the time - dependent distribution function @xmath102 is close to the global equilibrium @xmath56 , it can be approximately expressed by a suitable rescaling of the equilibrium distribution @xcite : @xmath103 with the rescaled variables defined by @xmath104 and @xmath105\ , , \label{eq : k}\ ] ] where @xmath106 and @xmath107 are time - dependent dimensionless scaling parameters . the normalization factor @xmath108 is given by @xcite @xmath109 the second term in the bracket of equation ( [ eq : k ] ) is proportional to the local velocity . namely , taking the derivative with respect to time in equation ( [ eq : r ] ) we get @xmath110 with @xmath111 @xcite . subtracting the drift velocity @xmath112 in the ansatz ( [ eq : k ] ) it is ensured that the momentum @xmath113 is not affected by the time dependence of the ansatz for @xmath114 . the time dependence of the distribution function is governed by the scaling parameters @xmath106 and @xmath107 , which denote the time - dependent deformations of the spatial and momentum variables , respectively . inserting the above ansatz into the boltzmann - vlasov equation ( [ eq : bve ] ) one obtains coupled ordinary differential equations of motion for the respective scaling parameters @xcite : @xmath115 \nonumber \\ \hspace*{-16mm}-\frac{48n c_0 } { mb_i r_i^2 \prod_j b_j r_j}\left [ f\left(\frac{\theta_z^{\frac{1}{2}}k_z}{\theta_x^{\frac{1}{2}}k_x } , \frac{\theta_z^{\frac{1}{2}}k_z}{\theta_y^{\frac{1}{2}}k_y}\right)+\theta_i^{\frac{1}{2}}k_i \frac{\partial}{\partial \theta_i^{\frac{1}{2}}k_i } f\left(\frac{\theta_z^{\frac{1}{2}}k_z}{\theta_x^{\frac{1}{2}}k_x } , \frac{\theta_z^{\frac{1}{2}}k_z}{\theta_y^{\frac{1}{2}}k_y}\right)\right]=0 \ , , \label{eq : bi}\\ \hspace*{-20 mm } \dot{\theta_i}+2\frac{\dot{b_i}}{b_i}\theta_i + \frac{1}{\tau}(\theta_i-\theta^\mathrm{le}_i)=0\ , \label{eq : theta_i}.\end{aligned}\ ] ] note that in the case of the global equilibrium the three equations ( [ eq : bi ] ) with the initial conditions @xmath116 and @xmath117 at @xmath118 reduce to equations ( [ vstrx])([vstrz ] ) , as expected . also , we remark that the initial conditions correspond to @xmath119 . the most ubiquitous method to study the physics of trapped ultracold gases is their absorption imaging after the release of the atomic or molecular cloud from the trap . turning off the trap potential allows the ultracold gas cloud to expand for tens of milliseconds and an absorption image is taken afterwards , when the cloud is large enough for the image to be recorded by a ccd camera . this technique , known as the tof imaging , is one of the most important probes of ultracold quantum systems and tof expansion experiments are a key diagnostic tool to study their properties . from the size of the expanded cloud and the known time of flight one can directly obtain , for instance , the fermi energy for a non - interacting degenerate fermi gas . in the case of free ballistic expansion , which is generically applied to theoretically model tof , the ellipsoidal fs deformation due to ddi is taken into account before tof , while all interactions between atoms during tof are neglected . in contrast to that , a nonballistic expansion model takes into account interactions for calculating both global equilibrium before tof and the subsequent expansion . in this section we show how quantitative information about the ellipsoidal fs deformation relevant for the current experiments can be determined from solving the bv equation for a tof expansion of the dipolar fermi gas . bearing in mind that the trap potential is turned off during tof , equations ( [ eq : bi ] ) and ( [ eq : theta_i ] ) can be used to describe the tof dynamics if we remove the terms @xmath120 which stem from the harmonic trap potential . within this formalism , the average sizes of the fermi gas cloud in real space are given by ( see appendix a for more details ) @xmath121 the deformation of the cloud shape is described in terms of the cloud aspect ratio @xmath122 , which is defined by the ratio of the root mean square of the transverse and longitudinal cloud radii , i.e. , the average sizes of the cloud in vertical @xmath123 and horizontal @xmath124 direction in the imaging plane . since the imaging axis in the innsbruck experiment @xcite has an angle of @xmath33 with respect to the @xmath19 axis , according to equation ( [ eq : ara ] ) from appendix a this leads to @xmath125 this aspect ratio in real space @xmath122 represents a directly measurable quantity in the tof dynamics experiments . in order to describe the influence of ddi on the fs we also use a corresponding aspect ratio in momentum space . in analogy to @xmath122 , the average sizes of the fermi gas cloud in momentum space read ( see appendix b for more details ) @xmath126 and the corresponding aspect ratio in momentum space is given by @xmath127 the relaxation time @xmath99 in equation ( [ eq : theta_i ] ) determines the regime of the dipolar fermi gas and , therefore , by solving the appropriate equations for varying values of @xmath99 , we are able to describe dynamic properties of the fermi gas all the way from the collisionless ( @xmath128 ) to the hydrodynamic ( @xmath129 ) regime . here , as before , @xmath130 represents the geometric mean of the trap frequencies . in section [ subsec : cl ] we will study the collisionless regime , in section [ subsec : hy ] the hydrodynamic regime , while in section [ subsec : cr ] we will investigate the system behavior in the intermediate , collisional regime . in section [ subsec : crsc ] we will improve the relaxation - time approximation in the collisional regime even further by determining the relaxation time in a self - consistent way . the value of the relaxation time @xmath99 determines the regime of the fermi gas during the expansion . in the low - density or collisionless regime , which is determined by the condition @xmath128 , the relaxation time @xmath99 can be taken to be infinite . in the limit @xmath131 the differential equations ( [ eq : theta_i ] ) for the scaling parameters @xmath132 decouple and the dynamic behavior in each direction is independent from the others . due to this , equations ( [ eq : theta_i ] ) can be solved analytically . with the respective initial conditions @xmath116 and @xmath117 we obtain @xmath133 . inserting this solution in equations ( [ eq : bi ] ) for the scaling parameters @xmath106 yields the equations of motion in the collisionless regime @xcite . we numerically solve them for a general system geometry , where the trap frequencies in the three directions are different and correspond to the values of the innsbruck experiment @xcite , and the magnetic field is oriented either in @xmath20 direction ( @xmath38 ) or in @xmath18 direction ( @xmath39 ) . although at @xmath39 the dipoles orientation forms an angle of @xmath36 ( see figure [ fig : fig1 ] ) with respect to the @xmath18 axis , we assume for simplicity in our calculations that the dipoles are parallel to the @xmath18 axis . graphs in the left - hand side column of figure [ fig : fig3 ] show aspect ratios @xmath134 , @xmath135 , as well the cloud aspect ratio @xmath136 in real space during tof in the collisionless regime . the black dotted line in the middle corresponds to the case of a noninteracting fermi gas , i.e. , @xmath137 , where the differential equations for the scaling parameters @xmath106 can be solved analytically , yielding @xmath138 with @xmath139 and @xmath140 denoting the global equilibrium radius and momentum in the @xmath55-th direction , respectively . these scaling parameters are solutions of equations ( [ vstrx])([k2 ] ) , together with equation ( [ eq : particlenumberconservation ] ) , for the case of a noninteracting fermi gas with @xmath141 where @xmath142 denotes the fermi energy . all aspect ratios for a noninteracting fermi gas in real space asymptotically approach one in the long tof limit . this shows that a cloud of noninteracting fermions becomes spherical after a long enough expansion , reflecting its isotropic momentum distribution even in the triaxial harmonic trap . as ddi is absent here , the orientation of the magnetic or the electric field , i.e. , of the dipole moments of atoms or molecules , has no influence on the fs deformation @xcite . graphs in the right - hand side column of figure [ fig : fig3 ] show the corresponding aspect ratios in momentum space . as expected , the black dotted line is constant and equal to one , as for the noninteracting fermions the fs is not deformed . furthermore , figure [ fig : fig3 ] also depicts the time dependence of the aspect ratios when the ddi is taken into account . dashed lines correspond to the ballistic expansion , when ddi is assumed to affect the initial ground state , but not later during the expansion . mathematically , this means that the ballistic expansion is also determined by equation ( [ eq : bip ] ) , but now with the parameters @xmath143 and @xmath144 instead of @xmath139 and @xmath140 , respectively : @xmath145 solid lines in figure [ fig : fig3 ] represent results for the nonballistic expansion , when we take ddi into account for calculating both the initial ground state and the subsequent expansion . to obtain these results , one has to solve numerically the coupled differential equations ( [ eq : bi ] ) together with @xmath133 . in figures [ fig : fig3](a ) to [ fig : fig3](d ) , top red solid and dashed lines correspond to the orientation of dipoles in the @xmath18 direction , and bottom blue solid and dashed lines correspond to dipoles orientation in the @xmath20 direction , while in figures [ fig : fig3](e ) and [ fig : fig3](f ) the position of lines turns out to be reversed : top blue lines give results for dipoles in the @xmath20 direction , and bottom red lines for dipoles in the @xmath18 direction . from the graphs in the right - hand column of figure [ fig : fig3 ] we read off that the aspect ratios in momentum space are constant if ballistic expansion approximation is used ( all dashed lines ) . this is not surprising , since here ddi is neglected during the expansion . this can also be shown mathematically if we insert the solution for @xmath146 from equation ( [ eq : bipbal ] ) into expression ( [ eq : k2 ] ) for @xmath147 , using @xmath133 , which is valid for the collisionless regime . with this we obtain @xmath148 , thus the momentum space aspect ratios @xmath149 for the ballistic expansion are clearly time - independent and are therefore determined by the initial ground state distribution . from figure [ fig : fig3 ] we see that the cloud aspect ratios in real space reach their corresponding plateaus after several tens of milliseconds . the asymptotic value of @xmath150 for @xmath38 for ballistic expansion is @xmath151 , whereas for nonballistic expansion it is @xmath152 , thus resulting in a @xmath82 difference due to ddi . for @xmath39 the asymptotic value of @xmath150 for ballistic expansion is @xmath153 , while for nonballistic expansion it is @xmath154 , representing again a @xmath82 difference . we also note that for @xmath38 the usual inversion of the cloud shape occurs , while for @xmath39 this is not the case . all these results are in excellent quantitative agreement with the experimental values reported in reference @xcite . aspect ratios in momentum space behave similarly , and again a difference of around @xmath155 between their asymptotic values in a ballistic and nonballistic expansion are observed . but one important difference is that here they are reached much faster , already after several milliseconds . a more detailed analysis reveals that the two terms in equation ( [ eq : k2 ] ) compete with each other during tof , but the second term becomes dominant quite fast . although the corresponding term in equation ( [ eq : r2mean ] ) has the same asymptotic behavior , the initial value of @xmath156 is much closer to its asymptotic value than in the case of @xmath150 and , as a consequence , all aspect ratios in momentum space converge faster . note that the aspect ratio in momentum space at the initial time @xmath118 coincides with the asymptotic aspect ratio in real space for ballistic expansion : @xmath157 the ballistic expansion aspect ratio in momentum space at @xmath118 can be calculated from equation ( [ eq : ak ] ) by using the initial conditions for the scaling parameters to yield @xmath158 on the other hand , the asymptotic value of the ballistic expansion aspect ratio in real space can be obtained if we insert the approximate expressions for the long - time behavior of the scaling parameters @xmath159 from equation ( [ eq : bipbal ] ) into equation ( [ eq : ar ] ) , which yields the same value as @xmath160 in equation ( [ eq : ak0bal ] ) . this fact was used in reference @xcite in order to observe the ellipsoidal deformation of the fs , as the real space aspect ratios can be readily measured during tof . however , this is only correct within the ballistic approximation , as for the truly nonballistic expansion such a relationship is no longer valid . er : ( a ) @xmath38 , ( b ) @xmath39 . solid ( dashed ) lines represent aspect ratios for nonballistic ( ballistic ) expansion of @xmath27er . the initially lower branch of blue lines corresponds to real space aspect ratios @xmath150 , while the initially upper branch of red lines corresponds to momentum space aspect ratios @xmath156.,title="fig:",width=264 ] er : ( a ) @xmath38 , ( b ) @xmath39 . solid ( dashed ) lines represent aspect ratios for nonballistic ( ballistic ) expansion of @xmath27er . the initially lower branch of blue lines corresponds to real space aspect ratios @xmath150 , while the initially upper branch of red lines corresponds to momentum space aspect ratios @xmath156.,title="fig:",width=264 ] but from figure [ fig : fig4 ] we read off that both for ballistic ( dashed lines ) and nonballistic ( solid lines ) expansion another relationship seems to hold . namely the aspect ratios in momentum space and the corresponding aspect ratios in real space turn out to have the same asymptotic values : @xmath161 the above is true for both considered orientations of dipoles , i.e. , @xmath38 and @xmath39 . a similar conclusion was reached in reference @xcite for a dipolar fermi gas that was initially in a cylindrically symmetric harmonic trap , but we see here that this is true even for a fully anisotropic harmonic trapping potential . note that this finding can not be directly used to determine the aspect ratio in momentum space at @xmath118 and the corresponding initial deformation of the fs , as for the ballistic expansion according to equation ( [ eq : ak0 ] ) . but this observation still allows to theoretically extract information on the momentum space distribution from experimental data . however , this requires that the corresponding equations are propagated backwards in time , so that the initial distribution in momentum space is calculated starting from the experimentally measured distribution in real space . here the numerical challenge is that this backward propagation has to be calculated for an infinitely long expansion time . in contrast to the previously considered collisionless regime , where collisions are completely neglected , we now turn to the hydrodynamic regime , where the system is supposed to have such a high density and , therefore , such a high collision rate , that it is always in local equilibrium . although realistic systems , even if initially in the hydrodynamic regime , eventually become collisionless during the expansion , we follow references @xcite and consider this theoretical limiting case for the sake of completeness . in the hydrodynamic regime , the scaling parameters @xmath162 always coincide with the local equilibrium values , i.e. , we have @xmath163 . however , since the limit @xmath164 holds , the last term in the left - hand side of equations ( [ eq : theta_i ] ) is undetermined . therefore , instead of equations ( [ eq : theta_i ] ) , the hydrodynamic regime is defined via the condition @xcite @xmath165 using this condition , minimizing the hartree - fock energy ( [ eq : totalenergy ] ) in the local equilibrium leads to the equations @xcite @xmath166\ , . \label{eq : hy2}\end{aligned}\ ] ] equations ( [ eq : bi ] ) , together with the identifications @xmath167 and @xmath168 , with equations ( [ eq : th_hy ] ) and ( [ eq : hy2 ] ) , and the normalization condition ( [ eq : hydef ] ) represent a closed set of six equations for the six respective scaling parameters in the hydrodynamic regime . these equations are solved numerically during the nonballistic tof expansion . for comparison , we have also solved the corresponding equations for the ballistic expansion , although the hydrodynamic regime implies that ddi can not be neglected at any point . figure [ fig : fig5 ] shows the corresponding aspect ratios in real and momentum space for @xmath27er . as expected , we see that there is a significant difference between the ballistic and the nonballistic expansion , in contrast to the collisionless regime in figure [ fig : fig3 ] . from the graphs in the left column of figure [ fig : fig5 ] we observe that the real space aspect ratios for @xmath38 behave generally similar to those in the collisionless regime , including the cloud shape inversion , although the asymptotic values differ more from their initial values for nonballistic expansion . on the other hand , for @xmath39 we see a qualitatively different behavior in figure [ fig : fig5](a ) , where the aspect ratio @xmath134 increases , while in figure [ fig : fig3](a ) it decreases . in figure [ fig : fig5](c ) for @xmath39 we read off that the aspect ratio @xmath150 even behaves nonmonotonously , with a local maximum at around 1 ms , while in the collisionless regime it only increases monotonously until it reaches its asymptotic value . however , again the inversion of the cloud shape is not present for @xmath39 , unlike for @xmath38 . we also note that the positions of ballistic expansion curves are reversed in all graphs compared to the collisionless regime , including those for momentum space aspect ratios . the behavior of momentum space aspect ratios in the right column of figure [ fig : fig5 ] is generally the same as in figure [ fig : fig3 ] for the collisionless regime , just with larger differences between initial and asymptotic values , for both cases @xmath38 and @xmath39 . the final cloud aspect ratio in real space for nonballistic expansion is twice as large as the corresponding collisionless value for @xmath38 , while for @xmath39 the asymptotic value of the aspect ratio is around 0.65 , which amounts to a decrease of around @xmath169 compared to the collisionless value . for the ballistic expansion , which we know to be unrealistic in the hydrodynamic regime , the corresponding increase and decrease amounts to both around @xmath170 . similar numbers are also obtained for the momentum space aspect ratio @xmath156 , as can be seen from the graphs on the right - hand side in figure [ fig : fig5 ] . since the corresponding values in the collisionless regime are all close to one , the above percentages also apply here , and represent the results for the ellipsoidal deformation of the fs in the hydrodynamic regime . the same conclusion can be also obtained from figure [ fig : fig6 ] , where we compare aspect ratios in real and momentum space . furthermore , with these graphs we confirm that the asymptotic values of the aspect ratios @xmath150 and @xmath156 also coincide in the hydrodynamic regime for both cases @xmath38 and @xmath39 , as stated by equation ( [ eq : akeqar ] ) for the collisionless regime . if we compare the convergence of aspect ratios to their asymptotic values in figures [ fig : fig4 ] and [ fig : fig6 ] , we see that in the hydrodynamic regime typical times to reach the plateau are similar in real and in momentum space , and have the value of several tens of milliseconds . this coincides with the corresponding convergence times for real space aspect ratios in the collisionless regime , where also a significant difference between the initial and the asymptotic value of aspect ratios occurred . only in the case of momentum space aspect ratios in the collisionless regime , where the deformation of the fs is small during the whole expansion , asymptotic values can be reached faster , namely in just few milliseconds . however , as already emphasized , even if initially in the hydrodynamic regime , the dipolar fermi gas becomes more and more dilute during the tof expansion , and the hydrodynamic regime continuously goes over into the collisional regime , and , finally , into the collisionless regime . therefore , we model the collisional regime in the remainder of this section , since it is relevant for experiments where the density of the fermi gas is high enough so that we can assume it is initially in the collisional or in the hydrodynamic regime . here we start considering the collisional regime and assume that the relaxation - time approximation ( [ eq : rta ] ) can be applied . furthermore , in this section we presume that the relaxation time @xmath99 remains constant during the tof . the latter assumption is only valid for short times of flight , before the density of the gas decreases significantly . we will improve upon this approximation in section [ subsec : crsc ] , where the relaxation time is determined self - consistently . however , provided that the relaxation time is constant , the tof dynamics can be obtained by directly solving equations ( [ eq : bi ] ) and ( [ eq : theta_i ] ) for a given value of @xmath99 . note that the values of the scaling parameters @xmath171 in local equilibrium are obtained according to section [ subsec : hy ] , i.e. , they represent the solutions of the equations for the hydrodynamic regime @xmath172 . the physical meaning of equations ( [ eq : theta_i ] ) is that dissipation occurs when the system is outside of a local equilibrium as long as there are collisions , i.e. , as long as the relaxation time @xmath99 remains finite . effects of collisions are therefore described through equations ( [ eq : theta_i ] ) , whereas equations ( [ eq : bi ] ) for the scaling parameters @xmath173 do not directly contain such terms . however , effects of collisions enter indirectly into equations ( [ eq : bi ] ) through the scaling parameters @xmath132 . here we numerically solve the coupled system of equations ( [ eq : bi ] ) and ( [ eq : theta_i ] ) during the nonballistic expansion for a fixed value of the relaxation time @xmath99 . varying the value of the relaxation time we are able to describe all regimes , from the collisionless , obtained in the limit @xmath174 , to the hydrodynamic , obtained in the limit @xmath175 . in particular , although the approximation of a fixed relaxation time is not realistic for longer expansion times , it allows us to understand and describe in more detail different collisional regimes , for finite values of @xmath99 , when the system undergoes a crossover from one limiting regime to the other . er : ( a ) @xmath38 , ( b ) @xmath39 . the pairs of curves in ( a ) from top to bottom and in ( b ) from bottom to top correspond to : hydrodynamic regime ( red ) , collisional regime ( black ) for fixed relaxation times @xmath176 and @xmath177 , and collisionless regime ( blue).,title="fig:",width=264 ] er : ( a ) @xmath38 , ( b ) @xmath39 . the pairs of curves in ( a ) from top to bottom and in ( b ) from bottom to top correspond to : hydrodynamic regime ( red ) , collisional regime ( black ) for fixed relaxation times @xmath176 and @xmath177 , and collisionless regime ( blue).,title="fig:",width=264 ] figure [ fig : fig7 ] shows the obtained aspect ratios for @xmath27er in real and momentum space for the two limiting cases considered previously , the collisionless and the hydrodynamic regime , as well as for the collisional regime with the fixed relaxation times @xmath176 and @xmath177 , where @xmath130 represents a geometric mean of the trap frequencies . depending on the respective geometry , asymptotic values of aspect ratios either decrease with increasing relaxation time , see figures [ fig : fig7](a ) for @xmath38 , or vice versa , see figure [ fig : fig7](b ) for @xmath39 . we also read off from these figures that the corresponding asymptotic values of aspect ratios in real and momentum space in the collisional regime are again equal according to relation ( [ eq : akeqar ] ) . motivated by the experiment reported in reference @xcite , in figure [ fig : fig8 ] we plot the aspect ratio in real space @xmath150 obtained after @xmath178 ms tof as a function of a fixed relaxation time @xmath99 for two different orientations of dipoles for the respective ultracold fermi gases given in table [ tab : tab1 ] . if the dipoles are oriented along @xmath20 axis , i.e. , figure [ fig : fig8](a ) , the corresponding aspect ratios for any fixed value of the relaxation time @xmath99 increase monotonously with the relative dipolar interaction strength @xmath179 , while for the dipoles along @xmath18 axis , i.e. , figure [ fig : fig8](b ) , the situation is opposite , as expected . note that the corresponding curves for the noninteracting case @xmath180 would be quite close to those for @xmath44cr , as can already be expected according to figure [ fig : fig2 ] . plots like those in figure [ fig : fig8 ] represent powerful diagnostic tools for estimating the relaxation time @xmath99 from experimentally measured values of aspect ratios @xmath150 for sufficiently short tof , when the fixed relaxation - time approximation is still applicable . furthermore , these graphs can be used for estimating the time scale @xmath181 to approach the asymptotic values of the aspect ratios from experimentally available tof expansion data . provided that it turns out for a tof @xmath181 that the corresponding relaxation time @xmath99 satisfies the condition @xmath182 , one has already reached the collisionless regime . this means that for longer times @xmath181 no further change of the aspect ratio is expected as one is already quite close to its asymptotic value . whereas in section [ subsec : cr ] we assumed that the relaxation time is constant , now we model the tof expansion of ultracold dipolar fermi gases more realistically and take into account that the relaxation time is also time - dependent . namely , during tof the gas rapidly expands , the distance between atoms grows , and as a consequence the relaxation time increases , thus eventually leading the system into the collisionless regime , even if initially it was in the hydrodynamic or in the collisional regime . in order to quantify this physical notion , we take into account that the elastic collisions of fermionic atoms or molecules arise purely from universal dipolar scattering and that the characteristic relaxation time is inversely proportional to the collision rate @xcite : @xmath183 here @xmath184 denotes a geometry - dependent , i.e. , @xmath22-dependent average number of collisions which is necessary to rethermalize ultracold dipolar fermions after a collision @xcite . furthermore , @xmath185 stands for a pauli suppression factor , which represents the reduction of the rethermalization rate in a degenerate fermi gas due to pauli blocking , and amounts to @xmath186 for non - degenerate gases . the pauli suppression factor depends on the degeneration level of fermions and is usually expressed as a function of the dimensionless temperature @xmath187 , where @xmath30 denotes the temperature and @xmath188 is the fermi temperature for the corresponding system @xcite . furthermore , @xmath189 in equation ( [ eq : tau ] ) represents the mean number density , where the volume @xmath190 of the fermi gas increases during the tof expansion according to @xmath191 the total elastic cross section @xmath192 is universally related to the dipole moment of fermions @xcite according to @xmath193 where @xmath194 represents a characteristic dipole length . the mean relative velocity @xmath195 is given by @xmath196 for the parameters of the experiment @xcite with atomic @xmath27er used throughout this paper , the universal dipolar scattering theory @xcite predicts the total elastic cross section value @xmath197 , which agrees with the value measured in reference @xcite . the temperature of the system was set to @xmath198 , with @xmath199 , which yields the pauli suppression factor @xmath200 @xcite , as well as the mean relative velocity @xmath195 according to equation ( [ eq : vmean ] ) . to completely fix all parameters which are necessary for a self - consistent determination of the relaxation time with equation ( [ eq : tau ] ) , we still need to take the appropriate value of the number of collisions @xmath184 for the given geometry , i.e. , for the given angle @xmath22 from reference @xcite . we remark that , although equation ( [ eq : tau ] ) represents a classical equation for the relaxation time , it is still applicable to quantum gases during the tof expansion . this is due to the fact that the expansion is so fast , that the volume of the gas increases several orders of magnitude in less than 1 ms , which can be checked using the obtained results for the tof dynamics as well as from the available experimental data . figure [ fig : fig9](a ) shows the corresponding aspect ratios in real and momentum space for @xmath27er during tof expansion for @xmath38 , for which the average number of collisions to rethermalize is @xmath201 @xcite . the red upper solid and dashed line in figure [ fig : fig9](a ) are obtained by numerically solving equations ( [ eq : bi ] ) and ( [ eq : theta_i ] ) , with the relaxation time determined self - consistently through equation ( [ eq : tau ] ) . in the same plot we also see for the sake of comparison the results for the collisionless regime in terms of the blue lower solid and dashed line . the difference between the corresponding lines is less than @xmath202 , which is certainly within the experimental error bars , and confirms that the system is indeed very close to the collisionless regime , as assumed in reference @xcite . furthermore , this justifies _ a posteriori _ the use of the classical equation ( [ eq : tau ] ) for the relaxation time . however , systems with a stronger ddi can easily reach the collisional regime , where a finite value for the relaxation time has to be taken into account . in order to demonstrate this , we analyze the tof expansion of a @xmath46k@xmath47rb dipolar fermi gas @xcite , whose relative dipolar interaction strength is @xmath85 , compared to @xmath203 for @xmath27er ( see table [ tab : tab1 ] ) . polar molecules have generically stronger electric dipole moments in comparison with the magnetic dipole moments of atoms , which is expected to yield a sensible difference in the respective aspect ratios . in figure [ fig : fig9](b ) we show the tof expansion dynamics for @xmath46k@xmath47rb for the same number of fermions and trap frequencies as in reference @xcite . the temperature of the system is assumed to be @xmath204 , as in reference @xcite , which yields the pauli suppression factor @xmath205 @xcite . the total elastic cross section in this case is @xmath206 @xcite , and the average number of collisions to rethermalize is again taken to be @xmath201 for @xmath38 @xcite . the difference between the aspect ratios calculated using the self - consistently determined relaxation time and those calculated assuming that the system is in the collisionless regime are here around @xmath207 , which could be clearly observed in future experiments . furthermore , for polar molecules with a stronger ddi the differences are expected to be even more pronounced . thus in experiments with such systems the relaxation time must be taken into account , for instance through the self - consistent approach presented here . we also note that the asymptotic values of aspect ratios in real and momentum space turn out to be again the same , as was already the case in the collisionless and in the hydrodynamic regime . ) as function of tof @xmath181 for ultracold fermi gas of @xmath27er ( red dashed line ) and @xmath46k@xmath47rb ( blue solid line ) for @xmath38 . the collisionless regime is achieved for @xmath182.,width=302 ] figure [ fig : fig10 ] shows the resulting time dependence of the self - consistently determined relaxation time during tof expansion for both analyzed species , i.e. , @xmath27er ( red dashed line ) and @xmath46k@xmath47rb ( blue solid line ) . as we can see , for an atomic gas of @xmath27er the relaxation time satisfies the condition @xmath182 right from the beginning , which further justifies the previous conclusion that the system is always in the collisionless regime @xcite . for a molecular gas of @xmath46k@xmath47rb , however , this condition is satisfied only after 1 - 2 ms , so initially the system is in the collisional regime . furthermore , we recognize that the relaxation time increases quite fast , namely faster than exponential , as we can see from the log - log plot of figure [ fig : fig10 ] . thus , the approximation of section [ subsec : cr ] with a fixed relaxation time would clearly not be suitable , and a self - consistent approach as presented here is indispensable . on the other hand , the obtained values of relaxation times are sufficiently large from the beginning to justify _ a posteriori _ the use of the classical formula ( [ eq : tau ] ) in both analyzed cases . in conclusion , we have explored the properties of trapped dipolar fermi gases at zero temperature in global equilibrium , as well as their dynamics during tof expansion by using the boltzmann - vlasov formalism in the relaxation - time approximation for the collision integral . we have studied the aspect ratios of the fermionic cloud in real and momentum space , including the deformation of the fermi sphere due to the presence of the dipole - dipole interaction . in particular , we have extended the existing theoretical models such that we could describe all experimentally relevant regimes : collisionless , collisional , and hydrodynamic . the obtained results for the global equilibrium and the tof expansion aspect ratios in the collisionless regime are in excellent agreement with experimental results of reference @xcite . in the collisional regime we have introduced an approach for self - consistently determining the relaxation time , which allows a detailed modeling of the global equilibrium and the tof expansion in cases when the collision integral can not be neglected . we have also shown that a strong dipole - dipole interaction , available for some experimentally accessible ultracold fermi species , could place the system into the collisional regime , which requires to use a self - consistent determination of the relaxation time presented here . furthermore , we find that in the collisional regime the tof dynamics can be accurately studied only if the nonballistic expansion is used , and the dipole - dipole interaction is properly taken into account not only to calculate the ground state , but also during the whole tof . therefore , the obtained theoretical results are relevant for designing future experiments with strongly dipolar fermi gases , for identifying results of the corresponding tof measurements , and for identifying effects of dipole - dipole interaction in general . furthermore , the approach developed here , based on the relaxation - time approximation for the boltzmann - vlasov equation , can also be applied to other fields of physics . the examples include nuclear physics , such as a study of viscosity of the quark - gluon plasma @xcite and ultra - relativistic heavy - ion collisions @xcite , as well as plasma physics @xcite , where , e.g. , transient regimes of degenerate electrons can be studied using the relaxation - time approximation @xcite . we acknowledge f. wchtler , a. r. p. lima , and i. vasi ' c , as well as inspiring discussions with f. ferlaino and l. chomaz . this work was supported in part by the ministry of education , science , and technological development of the republic of serbia under projects on171017 , oi1611005 , and ibec , by the german academic and exchange service ( daad ) under project ibec , by the german research foundation ( dfg ) via the collaborative research centers sfb / tr49 and sfb / tr185 , and by the european commission under h2020 project vi - seem , grant no . numerical simulations were run on the paradox supercomputing facility at the scientific computing laboratory of the institute of physics belgrade . to calculate aspect ratios in real space , we use the same geometry as in reference @xcite , see figure [ fig : fig1 ] . the imaging plane is @xmath208 , i.e. , the imaging is performed along the @xmath24 axis , which is rotated counterclockwise for an angle @xmath26 with respect to the @xmath19 axis . the tof absorption images correspond to density profiles of the system , so we first calculate the particle density @xmath209 from the wigner quasiprobability - distribution function , @xmath210 where expressions for @xmath211 and @xmath212 are given by equations ( [ eq : r ] ) and ( [ eq : k ] ) , respectively . changing the momentum variables @xmath213 to @xmath214 and switching to spherical coordinates yields @xmath215 the expectation value of a quantity @xmath216 in real space is given by @xmath217 so we immediately obtain that the expectation values of the coordinates vanish : @xmath218 . therefore , the size in the @xmath55-th direction of an atomic or molecular cloud in real space is described in terms of the root mean squares @xmath219 . using the expression ( [ eq : n ] ) for the particle density , the corresponding expectation values are found to be @xmath220 since the imaging is performed in the @xmath208 plane , the aspect ratio in real space is defined by @xmath221 so we also need to calculate the expectation value @xmath222 , where @xmath223 . after a straightforward but tedious calculation we get @xmath224\ , , \ ] ] and finally the aspect ratio ( [ eq : aar ] ) is given by @xmath225 note that in the innsbruck experiment @xcite the angle @xmath26 had the value @xmath33 . in order to describe effects of the ddi on the fermi surface , we use an aspect ratio in momentum space , which is defined similarly as the aspect ratio in real space . first , we calculate the particle density in momentum space @xmath226 from the wigner quasiprobability - distribution function , @xmath227 where again expressions for @xmath211 and @xmath212 are given by equations ( [ eq : r ] ) and ( [ eq : k ] ) , respectively . after a change of spatial variables @xmath228 according to @xmath229 with @xmath230 , we switch to spherical coordinates and obtain @xmath231 the expectation value of a variable @xmath232 in momentum space is given by @xmath233 so we get @xmath234 and the cloud sizes in momentum space are also defined by root mean squares @xmath235 . the corresponding expectation values can be explicitly calculated and yield @xmath236 where we have used the same variable change as above , as well as equation ( [ eq : particlenumberconservation ] ) . the aspect ratio in momentum space is defined as @xmath237 where @xmath238 . after a lengthy calculation we get @xmath239\ , , \ ] ] and finally the momentum space aspect ratio ( [ eq : bak ] ) reduces to @xmath240\cos^2\alpha+[\hbar^2k_y^2\theta_y(t)+ m^2r_y^2 \dot{b}^2_y(t)]\sin^2\alpha}}\ , .\ ] ]

an iterated refinement procedure for the guruswami sudan list decoding algorithm for generalised reed solomon codes based on alekhnovich s module minimisation is proposed . the method is parametrisable and allows variants of the usual list decoding approach . in particular , finding the list of _ closest _ codewords within an intermediate radius can be performed with improved average - case complexity while retaining the worst - case complexity . we provide a detailed description of the module minimisation , reanalysing the mulders storjohann algorithm and drawing new connections to both alekhnovich s algorithm and lee osullivan s . furthermore , we show how to incorporate the re - encoding technique of ktter and vardy into our iterative algorithm .

1404.3022

since the discovery of a polynomial - time hard - decision list decoder for generalised reed solomon ( grs ) codes by guruswami and sudan ( gs ) @xcite in the late 1990s , much work has been done to speed up the two main parts of the algorithm : interpolation and root - finding . notably , for interpolation beelen and brander @xcite mixed the module reduction approach by lee and osullivan @xcite with the parametrisation of zeh _ et al . _ @xcite , and employed the fast module reduction algorithm by alekhnovich @xcite . bernstein @xcite pointed out that an asymptotically faster variant can be achieved by using the reduction algorithm by giorgi _ et al . _ very recently chowdhury _ et al . _ @xcite used fast displacement - rank linear - algebraic solvers to achieve the fastest known approach . the gs approach was generalised by ktter and vardy to a soft - decision scenario @xcite , and the same authors also presented a complexity - reducing re - encoding transformation @xcite . cassuto _ et al . _ @xcite and tang _ et al . _ @xcite proposed modified interpolation - based list decoders with reduced average - complexity . for the root - finding step , one can employ the method of roth and ruckenstein @xcite in a divide - and - conquer fashion , as described by alekhnovich @xcite . this step then becomes an order of magnitude faster than interpolation , leaving the latter as the main target for further optimisations . for a given grs code , the gs algorithm has two parameters , both positive integers : the interpolation multiplicity @xmath0 and the list size @xmath1 . together with the code parameters they determine the decoding radius @xmath2 . to achieve a higher decoding radius for some given grs code , one needs higher @xmath0 and @xmath1 , and the value of these strongly influences the running time of the algorithm . in this work , we present a novel iterative method : we first solve the interpolation problem for @xmath3 and then iteratively refine this solution for increasing @xmath0 and @xmath1 . in each step of our algorithm , we obtain a valid solution to the interpolation problem for these intermediate parameters . the method builds upon that of beelen brander @xcite , but a new analysis of the computational engine alekhnovich s module minimisation algorithm reveals that each iteration runs faster than otherwise expected . # 1#1 the method therefore allows a fast _ multi - trial _ list decoder when our aim is just to find the list of codewords with minimal distance to the received word . at any time during the refinement process , we will have an interpolation polynomial for intermediate parameters @xmath4 , @xmath5 yielding an intermediate decoding radius @xmath6 , where @xmath7 is the minimum distance . if we perform the root - finding step of the gs algorithm on this , all codewords with distance at most @xmath8 from the received are returned ; if there are any such words , we break computation and return those ; otherwise we continue the refinement . we can choose any number of these trials , e.g. for each possible intermediate decoding radius between @xmath9 and the target @xmath2 . since the root - finding step of gs is less complex than the interpolation step , this multi - trial decoder will have the same asymptotic worst - case complexity as the usual gs using the beelen brander interpolation @xcite . however , its average - case complexity is better since due to the properties of the underlying channel it is more probable to have a small number of errors rather than a big one . this contribution is structured as follows . in the next section we give necessary preliminaries and state the gs interpolation problem for decoding grs codes . in section [ sec_modulemini ] we give a definition and properties of minimal matrices . we describe and reanalyse the conceptually simple mulders storjohann algorithm @xcite for bringing matrices to this form . we also give a fresh look at alekhnovich s algorithm @xcite simply as a divide-&-conquer variant of mulders storjohann , and our new analysis can carry over . our new iterative procedure is explained in detail in section [ sec_iterative ] and the incorporation of the re - encoding transformation @xcite is described in section [ sec_re - encoding ] . in section [ sec_simulation ] we present simulation results . parts of these results were presented at wcc 2013 @xcite ; compared to that article , this version contains the incorporation of the re - encoding scheme , a full example of the algorithm , simulation results , as well as a more detailed description of the module minimisation procedure . let @xmath10 be the finite field of order @xmath11 and let @xmath12}}}$ ] be the polynomial ring over @xmath10 with indeterminate @xmath13 . let @xmath14}}}$ ] denote the polynomial ring in the variables @xmath13 and @xmath15 and let @xmath16 be the @xmath17-weighted degree of @xmath18 . a vector of length @xmath19 is denoted by @xmath20 . if @xmath21 is a vector over @xmath12}}}$ ] , let @xmath22 . we introduce the leading position as @xmath23 and the leading term @xmath24 is the term at this position . an @xmath25 matrix is denoted by @xmath26 . the rows of such a matrix are denoted by bold lower - case letters , e.g. @xmath27 . furthermore , let the degree of such a polynomial matrix be @xmath28 . modules are denoted by capital letters such as @xmath29 . let @xmath30 be @xmath19 nonzero distinct elements of @xmath10 with @xmath31 and let @xmath32 be @xmath19 ( not necessarily distinct ) nonzero elements of @xmath10 . a grs code @xmath33 of length @xmath19 and dimension @xmath34 over @xmath10 is given by @xmath35}}},\ , \deg f(x ) < k \big\}.\end{aligned}\ ] ] grs codes are maximum distance separable ( mds ) codes , i.e. , their minimum hamming distance is @xmath36 . we shortly explain the interpolation problem of gs @xcite for list decoding grs codes up to the johnson radius @xcite in the following . [ thm_gsproblem ] let @xmath37 and @xmath38 be the corresponding information polynomial as defined in . let @xmath39 be a received word where @xmath40 . let @xmath41 denote @xmath42 . let @xmath43}}}$ ] be a nonzero polynomial that passes through the @xmath19 points @xmath44 @xmath45 with multiplicity @xmath46 , has @xmath15-degree at most @xmath1 , and @xmath47 . then @xmath48 . one can easily show that a polynomial @xmath49 that fulfills the above conditions can be constructed whenever @xmath50 , where @xmath51 is the difference between the maximal number of coefficients of @xmath49 , and the number of homogeneous linear equations on @xmath49 specified by the interpolation constraint . this determines the maximal number of correctable errors , and one can show that satisfactory @xmath0 and @xmath1 can always be chosen whenever @xmath52 . [ def_permiss ] an integer triple @xmath53 is _ permissible _ if @xmath50 . we define also the decoding radius - function @xmath54 as the greatest integer such that @xmath55 is permissible . it is well - known that @xmath50 for @xmath56 implies @xmath57 , which is half the minimum distance . therefore , it never makes sense to consider @xmath56 , and in the remainder we will always assume @xmath58 . furthermore , we will also assume @xmath59 since this e.g. holds for any @xmath2 for the closed - form expressions in @xcite . let us illustrate the above . the following will be a running example throughout the article . [ ex_rs164param ] a @xmath60 code over @xmath61 can uniquely correct @xmath62 errors ; unique decoding corresponds to @xmath63 and one can confirm that @xmath64 . to attain a decoding radius @xmath65 , one can choose @xmath66 and @xmath67 in order to obtain a permissible triple . also @xmath68 is permissible , though less interesting since it does not give improved decoding radius . however , one finds @xmath69 and @xmath70 are permissible . since @xmath71 , there are no permissible triples for greater decoding radii . let @xmath72}}}$ ] denote the space of all bivariate polynomials passing through the points @xmath73 with multiplicity @xmath0 and @xmath15-degree at most @xmath1 . we are searching for an element of @xmath74 with low @xmath75-weighted degree . following the ideas of lee and osullivan @xcite , we can first remark that @xmath74 is an @xmath12}}}$]-module . second , we can give an explicit basis for @xmath74 . define first two polynomials @xmath76 as well as @xmath77 in @xmath12}}}$ ] as the unique lagrange interpolation polynomial going through the points @xmath78 for @xmath79 . denote by @xmath80}}(x)$ ] the @xmath81-coefficient of @xmath49 when @xmath82 is regarded over @xmath12}}}[y]$ ] . [ lem_qdivg ] let @xmath83}}(x)y^i \in { \ifthenelse{\equal{{{s,\ell}}}{\empty } } { \ensuremath{m } } { \ensuremath{m_{{{s,\ell } } } } } } $ ] with @xmath84 . then @xmath85}}(x)$ ] . @xmath49 interpolates the @xmath19 points @xmath86 with multiplicity @xmath0 , so for any @xmath87 , @xmath88}}(x+\alpha_j)(y+r_j^{\prime})^j$ ] has no monomials of total degree less than @xmath0 . multiplying out the @xmath89-terms , @xmath80}}(x+\alpha_j)y^t$ ] is the only term with @xmath15-degree @xmath90 . therefore @xmath80 } } ( x+\alpha_j)$ ] can have no monomials of degree less than @xmath91 , which implies @xmath92}}(x)$ ] . as this holds for any @xmath87 , we proved the lemma . [ thm_mbasis ] the module @xmath74 is generated as an @xmath12}}}$]-module by the @xmath93 polynomials @xmath94}}}$ ] given by @xmath95 it is easy to see that each @xmath96 since both @xmath97 and @xmath98 go through the @xmath19 points @xmath78 with multiplicity one , and that @xmath97 and @xmath98 divide @xmath99 with total power @xmath0 for each @xmath90 . to see that any element of @xmath74 can be written as an @xmath12}}}$]-combination of the @xmath100 , let @xmath49 be some element of @xmath74 . then the polynomial @xmath101 } } ( x ) p{^{(\ell)}}(x , y ) $ ] has @xmath15-degree at most @xmath102 . since both @xmath49 and @xmath103 are in @xmath74 , so must @xmath104 be in @xmath74 . since @xmath100 has @xmath15-degree @xmath90 and @xmath105}}(x ) = 1 $ ] for @xmath106 , we can continue reducing this way until we reach a @xmath107 with @xmath15-degree at most @xmath108 . from then on , we have @xmath105}}(x ) = g(x)^{s - t}$ ] , but by lemma [ lem_qdivg ] , we must also have @xmath109}}(x)$ ] . therefore , we can reduce by @xmath110 . this can be continued with the remaining @xmath100 , eventually reducing the remainder to 0 . we can represent the basis of @xmath74 by the @xmath111 matrix @xmath112}(x)\|_{i=0,j=0}^{\ell , \ell}$ ] over @xmath12}}}$ ] ; more explicitly we have : @xmath113 any @xmath12}}}$]-linear combination of rows of @xmath114 thus corresponds to an element in @xmath74 by its @xmath90th position being the @xmath12}}}$]-coefficient to @xmath81 . all other bases of @xmath74 can similarly be represented by matrices , and these will be unimodular equivalent to @xmath114 , i.e. , they can be obtained by multiplying @xmath114 on the left with an invertible matrix over @xmath12}}}$ ] . extending the work of lee and osullivan @xcite , beelen and brander @xcite gave a fast algorithm for computing a satisfactory @xmath49 : start with @xmath114 as a basis of @xmath74 and compute a different , `` minimal '' basis of @xmath74 where an element of minimal @xmath115-weighted degree appears directly . in the following section , we give further details on how to compute such a basis , but our ultimate aim in section [ sec_iterative ] is different : we will use a minimal basis of @xmath74 to efficiently compute one for @xmath116 for @xmath117 and @xmath118 . this will allow an iterative refinement for increasing @xmath0 and @xmath1 , where after each step we have such a minimal basis for @xmath74 . we then exploit this added flexibility in our multi - trial algorithm . given a basis of @xmath74 , e.g. @xmath114 , the module minimisation here refers to the process of obtaining a new basis , which is the smallest among all bases of @xmath74 in a precise sense . we will define this and connect various known properties of such matrices . we will then show how to perform this minimisation using the mulders storjohann algorithm @xcite , reanalyse its performance and connect it to alekhnovich s algorithm @xcite . [ def_weakpopov ] a matrix @xmath119 over @xmath12}}}$ ] is in _ weak popov form _ if the leading position of each row is different . we are essentially interested in short vectors in a module , and the following lemma shows that the simple concept of weak popov form will provide this . it is a paraphrasing of ( * ? ? ? * proposition 2.3 ) and we omit the proof . [ lem_minrow ] if a square matrix @xmath119 over @xmath12}}}$ ] is in weak popov form , then one of its rows has minimal degree of all vectors in the row space of @xmath119 . denote now by @xmath120 the diagonal @xmath111 matrix over @xmath12}}}$ ] : @xmath121 since we seek a polynomial of minimal @xmath75-weighted degree , we also need the following corollary . [ cor_weightsol ] let @xmath122}}}^{(\ell+1 ) \times ( \ell+1)}$ ] be the matrix representation of a basis of @xmath123 . if @xmath124 is in weak popov form , then one of the rows of @xmath125 corresponds to a polynomial in @xmath123 with minimal @xmath75-weighted degree . let @xmath126 . now , @xmath127 will correspond to the basis of an @xmath12}}}$]-module @xmath128 isomorphic to @xmath123 , where an element @xmath129 is mapped to @xmath130 . by lemma [ lem_minrow ] , the row of minimal degree in @xmath127 corresponds to an element of @xmath128 with minimal @xmath13-degree . therefore , the same row of @xmath125 corresponds to an element of @xmath123 with minimal @xmath75-weighted degree . if for some matrix @xmath122}}}^{(\ell+1 ) \times ( \ell+1)}$ ] , @xmath124 is in weak popov form , we say that @xmath125 is in _ weighted weak popov form_. we introduce what will turn out to be a measure of how far a matrix is from being in weak popov form . [ def_orthogonality ] the _ orthogonality defect _ of a square matrix @xmath119 over @xmath12}}}$ ] is defined as @xmath131 [ lem_popovreduces ] if a square matrix @xmath119 over @xmath12}}}$ ] is in weak popov form , then @xmath132 . let @xmath133 be the rows of @xmath134}}}^{m \times m}$ ] and @xmath135 . in the alternating sum - expression for @xmath136 , the term @xmath137 will occur since the leading positions of @xmath138 are all different . thus @xmath139 unless leading term cancellation occurs in the determinant expression . however , no other term in the determinant has this degree : regard some ( unsigned ) term in @xmath136 , say @xmath140 for some permutation @xmath141 . if not @xmath142 for all @xmath87 ( as defined in ) , then there must be an @xmath87 such that @xmath143 since @xmath144 is the same for all @xmath141 . thus , @xmath145 . as none of the other terms in @xmath90 can have greater degree than their corresponding row s leading term , we get @xmath146 . thus , @xmath132 . the weak popov form is highly related to minimal grbner bases of the row space module , using a term order where vectors in @xmath12}}}^m$ ] are ordered according to their degree ; indeed the rows of a matrix in weak popov form _ is _ such a grbner basis ( though the opposite is not always true ) . similarly , the weighted weak popov form has a corresponding weighted term order . in this light , lemma [ lem_minrow ] is simply the familiar assertion that a grbner basis over such a term order must contain a `` minimal '' element of the module . see e.g. ( * ? ? ? * chapter 2 ) for more details on this correspondence . the language of grbner bases was employed in the related works of @xcite . [ def_rowred ] applying a _ row reduction _ on a matrix over @xmath12}}}$ ] means to find two different rows @xmath147 , @xmath148 and such that @xmath149 , and then replacing @xmath150 with @xmath151 where @xmath152 and @xmath153 are chosen such that the leading term of the polynomial @xmath154 is cancelled . algorithm [ alg_mulders ] is due to mulders and storjohann @xcite . our proof of it is similar , though we have related the termination condition to the orthogonality defect , restricting it to only square matrices . apply row reductions as in definition [ def_rowred ] on the rows of @xmath119 until no longer possible this matrix . introduce for the proof a _ value function _ @xmath155}}}^m \rightarrow { \mathbb n}_0 $ ] as @xmath156 . first let us consider the following lemma . [ lem_rowreddec ] if we replace @xmath157 with @xmath158 in a row reduction , then @xmath159 . we can not have @xmath160 since all terms of both @xmath157 and @xmath161 have degree at most @xmath162 . if @xmath163 we are done since @xmath164 , so assume @xmath165 . let @xmath166 . by the definition of the leading position , all terms in both @xmath157 and @xmath161 to the right of @xmath167 must have degree less than @xmath162 , and so also all terms in @xmath158 to the right of @xmath167 satisfies this . the row reduction ensures that @xmath168 , so it must then be the case that @xmath169 . [ lem_mulders ] algorithm [ alg_mulders ] is correct . for a matrix @xmath134}}}^{m \times m}$ ] , it performs fewer than @xmath170 row reductions and has asymptotic complexity @xmath171 where @xmath172 is the maximal degree of any term in @xmath119 . if algorithm [ alg_mulders ] terminates , the output matrix must be unimodular equivalent to the input since it is reached by a finite number of row - operations . since we can apply row reductions on a matrix if and only if it is not in weak popov form , algorithm [ alg_mulders ] must bring @xmath119 to this form . termination follows directly from lemma [ lem_rowreddec ] since the value of a row decreases each time a row reduction is performed . to be more precise , we furthermore see that the maximal number of row reductions performed on @xmath119 before reaching a matrix @xmath173 in weak popov form is at most @xmath174 . expanding this , we get @xmath175 where we use @xmath176 and that the @xmath177 are all different . for the asymptotic complexity , note that during the algorithm , no polynomial in the matrix will have larger degree than @xmath172 . the estimate is reached simply by remarking that one row reduction consists of @xmath178 times scaling and adding two such polynomials . let us consider an example to illustrate all the above . [ ex_weakpopov ] let us consider the following matrices @xmath179}}^{3 \times 3}$ ] . from matrix @xmath180 to @xmath181 we performed one row - operation : [ cols= " < , < , < , < , < , < , < " , ] where the indexes @xmath182 on the arrow indicated the concerned rows . the orthogonality defect is decreasing ; @xmath183 , and @xmath184 is in weak popov form . in @xcite , alekhnovich gave a divide-&-conquer variant of the mulders storjohann - algorithm : the same row reductions are performed but structured in a binary computation tree , where work is done on matrices of progressively smaller degree towards the bottom , ending with essentially @xmath10-matrices at the leaves . alekhnovich does not seem to have been aware of the work of mulders and storjohann , and basically reinvented their algorithm before giving his divide-&-conquer variant . for square matrices , we can improve upon the complexity analysis that alekhnovich gave by using the concept of orthogonality defect ; this will be crucial for our aims . [ lem_alekcompl ] alekhnovich s algorithm inputs a matrix @xmath134}}}^{m\times m}$ ] and outputs a unimodular equivalent matrix which is in weak popov form . let @xmath172 be the greatest degree of a term in @xmath119 . if @xmath185 then the algorithm has asymptotic complexity : @xmath186 the description of the algorithm as well as the proof of its correctness can be found in @xcite . we only prove the claim on the complexity . the method @xmath187 of @xcite computes a unimodular matrix @xmath173 such that @xmath188 or @xmath189 is in weak popov form . according to ( * ? ? ? * lemma 2.10 ) , the asymptotic complexity of this computation is in @xmath190 . due to lemma [ lem_popovreduces ] , we can set @xmath191 to be sure that @xmath189 is in weak popov form . what remains is just to compute the product @xmath189 . due to * lemma 2.8 ) , each entry in @xmath173 can be represented as @xmath192 for some @xmath193 and @xmath194}}}$ ] of degree at most @xmath195 . if therefore @xmath185 , the complexity of performing the matrix multiplication using the naive algorithm is @xmath196 . the beelen brander interpolation algorithm @xcite works simply by computing @xmath114 and then applying alekhnovich s algorithm on @xmath197 ; a minimal @xmath75-weighted polynomial in @xmath74 can then be directly retrieved as a row in the reduced matrix , after removing @xmath198 . the algorithm s complexity therefore follows from the above theorem once we have computed the orthogonality defect of @xmath197 : [ lem_odofa ] @xmath199 . we will compute first @xmath200 and then @xmath201 . for the former , we have @xmath202 , where the @xmath203 are as in theorem [ thm_mbasis ] . note that whenever @xmath204 is not a codeword then @xmath205 . therefore , @xmath206 and so @xmath207 this gives @xmath208 since @xmath114 is lower triangular , the determinant is : @xmath209 the orthogonality defect can then be simplified to @xmath210 in the earlier work of lee and osullivan @xcite , they construct basically the same matrix as @xmath114 , but apply their own grbner basis algorithm on this . they also does not seem to have been aware of the work of mulders and storjohann @xcite , but their algorithm is basically a variant of algorithm [ alg_mulders ] which keeps the rows in a specific order . using the results of the preceding section , we show in section [ ssec_steptypeone ] that , given a basis of @xmath74 as a matrix @xmath211 in weighted weak popov form , then we can write down a matrix @xmath212 which is a basis of @xmath213 and where @xmath214 is much lower than @xmath215 . this means that reducing @xmath212 to weighted weak popov form using alekhnovich s algorithm @xcite is faster than reducing @xmath216 . we call this kind of refinement a `` micro - step of type i '' . in section [ ssec_steptypetwo ] , we similarly give a way to refine a basis of @xmath74 to one of @xmath217 , and we call this a `` micro - step of type ii '' . if we first compute a basis in weighted weak popov form of @xmath218 using @xmath219 , we can perform a sequence of micro - steps of type i and ii to compute a basis in weighted weak popov form of @xmath123 for any @xmath220 with @xmath221 . after any step , having some intermediate @xmath222 , @xmath223 , we will thus have a basis of @xmath224 in weighted weak popov form . by corollary [ cor_weightsol ] , we could extract from @xmath225 a @xmath226 with minimal @xmath75-weighted degree . since it must satisfy the interpolation conditions of theorem [ thm_gsproblem ] , and since the weighted degree is minimal among such polynomials , it must also satisfy the degree constraints for @xmath227 . by that theorem any codeword with distance at most @xmath228 from @xmath204 would then be represented by a root of @xmath229 . algorithm [ alg_multitrial ] is a generalisation and formalisation of this method . for a given @xmath230 code , one chooses ultimate parameters @xmath231 being a permissible triple with @xmath58 . one also chooses a list of micro - steps and chooses after which micro - steps to attempt decoding ; these choices are represented by a list @xmath232 consisting of @xmath233 , @xmath234 and @xmath235 elements . this list must contain exactly @xmath236 @xmath233-elements and @xmath108 @xmath234-elements , as it begins by computing a basis for @xmath218 and will end with a basis for @xmath74 . whenever there is a @xmath235 element in the list , the algorithm performs root - finding and finds all codewords with distance at most @xmath227 from @xmath204 ; if this list is non - empty , the computation breaks and the list is returned . the algorithm calls sub - functions which we explain informally : @xmath237 and @xmath238 will take @xmath239 and a basis in weighted weak popov form for @xmath224 and return a basis in weighted weak popov form for @xmath240 respectively @xmath241 ; more detailed descriptions for these are given in subsections [ ssec_steptypeone ] and [ ssec_steptypetwo ] . @xmath242 finds a polynomial of minimal @xmath75-weighted degree in @xmath224 given a basis in weighted weak popov form ( corollary [ cor_weightsol ] ) . finally , @xmath243 returns all @xmath15-roots of @xmath49 of degree less than @xmath34 and whose corresponding codeword has distance at most @xmath2 from the received word @xmath204 . the correctness of algorithm [ alg_multitrial ] for any possible choice of @xmath244 and @xmath245 follows from our discussion as well as sections [ ssec_steptypeone ] and [ ssec_steptypetwo ] . before going to these two technical sections , we will discuss what possibilities the micro - steps of type i and ii offer , and in particular , do not , with regards to decoding radii . in the following two subsections we explain the details of the micro - steps . in section [ ssec_complanalysis ] , we discuss the complexity of the method and how the choice of @xmath245 influences this . the choice of @xmath245 provides much flexibility to the algorithm . the two extreme cases are perhaps the most generally interesting : the one without any @xmath235 elements except at the end , i.e. , usual list - decoding ; and the one with a @xmath235 element each time the intermediate decoding radius @xmath246 has increased , i.e. , a variant of maximum - likelihood decoding up to a certain radius . in section [ ssec_complanalysis ] , we discuss complexity concerns with regards to the chosen path ; it turns out that the price of either type of micro - step is very comparable , and the worst - case complexity is completely unchanged by the choice of @xmath245 . however , in the case where we have multiple @xmath235 elements we want to minimise the _ average _ computation cost : considering that few errors occur much more frequently than many , we should therefore seek to reach each intermediate decoding radius after as few micro - steps as possible . since we do not have a refinement which increases only @xmath0 , we are inherently limited in the possible paths we can choose , so the question arises if this limitation conflicts with our interest as given above . first and most important for any given final decoding radius @xmath2 , we mentioned in section [ ssec_ipdecoding ] that the corresponding parameters satisfy @xmath247 , and so we can reach these values using only micro - steps of type i and ii . for the intermediate steps , the strongest condition we would like to have satisfied is the following : let @xmath248 be the series of intermediate decoding radii where we would like to attempt decoding . let @xmath249 be chosen such that @xmath250 is permissible and either @xmath251 or @xmath252 is minimal possible for the given @xmath253 . can then the sequence of parameters @xmath249 be reached by a series of micro - steps of type i and ii ? unfortunately , we do not have a formal proof of this statement . however , we have verified for a large number of parameters that it is true . the function is based on the following lemma : [ lem_microstepone ] if @xmath255}}}$ ] is a basis of @xmath74 , then the following is a basis of @xmath256 : @xmath257 in the basis of @xmath256 given in theorem [ thm_mbasis ] , the first @xmath93 generators are the generators of @xmath74 . thus , all of these can be described by any basis of @xmath256 . the last remaining generator is exactly @xmath258 . in particular , the above lemma holds for a basis of @xmath256 in weighted weak popov form , represented by a matrix @xmath211 . the following matrix thus represents a basis of @xmath256 : @xmath259 \makebox[4cm][c]{$ { \ifthenelse{\equal{{{s,\ell}}}{\empty } } { \ensuremath{\mathcal{b } } } { \ensuremath{\mathcal{b}_{{{s,\ell } } } } } } $ } & \makebox[1.5em][r]{$\vec 0^t$ } \\[-0.2 cm ] \ \end{array } \\ \hline \\[-.3 cm ] \begin{matrix } 0 & \ldots & 0 & ( -r)^s & \binom s 1 ( -r)^{s-1 } & \ldots & 1 \end{matrix } \end{array}\right).\ ] ] [ lem_orthoi ] @xmath260 . we calculate the two quantities @xmath261 and @xmath262 . it is easy to see that @xmath263 for the row - degree , this is clearly @xmath264 plus the row - degree of the last row . if and only if the received word is not a codeword then @xmath205 , so the leading term of the last row must be @xmath265 thus , we get @xmath266 where the last step follows from lemma [ lem_popovreduces ] as @xmath267 is in weak popov form . [ cor_complmsi ] the complexity of @xmath268 is @xmath269 . follows by lemma [ lem_alekcompl ] . since @xmath270 we can leave out the @xmath0 in @xmath271-terms . the function is based on the following lemma : [ lem_microsteptwo ] if @xmath255}}}$ ] is a basis of @xmath74 , then the following is a basis of @xmath273 : @xmath274 denote by @xmath275 the basis of @xmath74 as given in theorem [ thm_mbasis ] , and by @xmath276 the basis of @xmath273 . then observe that for @xmath277 , we have @xmath278 . since the @xmath279 form a basis of @xmath74 , each @xmath280 is expressible as an @xmath12}}}$]-combination of these , and thus for @xmath281 , @xmath282 is expressible as an @xmath12}}}$]-combination of the @xmath283 . remaining is then only @xmath284 . as before , we can use the above with the basis @xmath211 of @xmath74 in weighted weak popov form , found in the previous iteration of our algorithm . recall that multiplying by @xmath15 translates in the matrix representation to shifting one column to the right , so the following matrix represents a basis of @xmath273 : @xmath285 \vec 0^t & \makebox[1cm][c]{$\vec 0 $ } \\[0.1 cm ] \end{array}\right ) + \left(\begin{array}{@{}c|c@ { } } 0 & \vec 0 \\\hline\\[-.2 cm ] \vec 0^t & \makebox[1cm][c]{$ { \ifthenelse{\equal{{{s,\ell}}}{\empty } } { \ensuremath{\mathcal{b } } } { \ensuremath{\mathcal{b}_{{{s,\ell } } } } } } $ } \\[0.1 cm ] \end{array}\right ) - r\cdot \left(\begin{array}{@{}c|c@ { } } \vec 0 & 0 \\\hline\\[-.2 cm ] \makebox[1cm][c]{$ { \ifthenelse{\equal{{{s,\ell}}}{\empty } } { \ensuremath{\mathcal{b } } } { \ensuremath{\mathcal{b}_{{{s,\ell } } } } } } $ } & \vec 0^t \\[0.1 cm ] \end{array}\right).\ ] ] [ lem : cortho ] @xmath286 . we compute @xmath287 and @xmath288 . for the former , obviously the first row has degree @xmath289 . let @xmath290 denote the @xmath87th row of @xmath211 and @xmath291 denote the @xmath87th row of @xmath267 . the @xmath292th row of @xmath293 has the form @xmath294 { \ifthenelse{\equal{\ell+1}{\empty } } { \ensuremath{\mathcal{w } } } { \ensuremath{\mathcal{w}_{\ell+1 } } } } = ( 0 \mid \vec{b}'_i)x^{k-1 } - r(\vec{b}'_i \mid 0).\ ] ] if and only if the received word is not a codeword , then @xmath205 . in this case , the leading term of @xmath295 must have greater degree than any term in @xmath296 . thus the degree of the above row is @xmath297 . summing up we get @xmath298 for the determinant , observe that @xmath299 where @xmath300 and @xmath301 is all but the zeroth column of @xmath211 . this means @xmath302 can be obtained by starting from @xmath211 and iteratively adding the @xmath303th column of @xmath211 scaled by @xmath77 to the @xmath304th column , with @xmath304 starting from @xmath305 up to @xmath102 . since each of these will add a scaled version of an existing column in the matrix , this does not change the determinant . thus , @xmath306 . but then @xmath307 and so @xmath308 by lemma [ lem_popovreduces ] since @xmath267 is in weak popov form . thus we get @xmath309 the lemma follows from the difference of the two calculated quantities . [ cor_complmsii ] the complexity of @xmath310 is @xmath311 . [ ex_algo164 ] we consider again the @xmath312 code over @xmath313 of example [ ex_rs164param ] , and specify now that @xmath314 and @xmath315 for @xmath316 . the aimed decoding radius is @xmath317 and therefore the permissible triple @xmath318 should be reached iteratively by algorithm [ alg_multitrial ] . to maximise the decoding radius during the procedure , we could choose the following sequence of intermediate parameters @xmath319 : @xmath320 we perform root - finding only if the decoding radius is increased . therefore , the list @xmath232 of operations becomes : @xmath321 with the information polynomial @xmath322 , we obtain with the following codeword : @xmath323 consider that @xmath324 was received , i.e. , that the error @xmath325 of weight 8 occurred . we will depict the degrees of the polynomials in the matrices in the iterative decoding process . these are for this particular received word , but for a generic received word , the degrees are the same . for some @xmath194}}}$ ] , we will write @xmath326 for @xmath327 if @xmath328 , and @xmath329 if @xmath330 , and we extend @xmath331 element - wise to matrices . to begin with , we have : @xmath332 according to and . we then apply alekhnovich s algorithm on @xmath333 to obtain @xmath334 which is in weak popov form . from this we can easily scale down the columns again to obtain @xmath335 . it took 11 row reductions , while @xmath336 was the upper bound , according to lemma [ lem_mulders ] . we obtain @xmath337 the first element in @xmath245 is @xmath235 , so we pick the second row of @xmath335 , since it has weighted degree less than @xmath338 , and we interpret it as a polynomial : @xmath339 root - finding of @xmath340 yields no results . the next element in @xmath245 is @xmath233 , so we move to the next intermediate parameters @xmath341 . from , we get @xmath342[c]{$ { \ifthenelse{\equal{1,1}{\empty } } { \ensuremath{\mathcal{b } } } { \ensuremath{\mathcal{b}_{1,1 } } } } $ } & \begin{matrix } 0 \\ 0 \end{matrix } \end{array } \\ \hline \\[-.4 cm ] \begin{matrix } 0 & -r & 1 \end{matrix } \end{array}\right ) \qquad \textnormal{\it and so } \qquad { \ifthenelse{\equal{1,2}{\empty } } { \ensuremath{\mathcal{c } } } { \ensuremath{\mathcal{c}_{1,2 } } } } { ^\mathrm i}\preceq \left(\begin{matrix } 10 & 6 & \bot \\ 9 & 6 & \bot \\ \bot & 15 & 0 \end{matrix}\right ) \quad \quad { \ifthenelse{\equal{1,2}{\empty } } { \ensuremath{\mathcal{c } } } { \ensuremath{\mathcal{c}_{1,2 } } } } { ^\mathrm i } { \ifthenelse{\equal{2}{\empty } } { \ensuremath{\mathcal{w } } } { \ensuremath{\mathcal{w}_{2 } } } } \preceq \left(\begin{matrix } 10 & 9 & \bot \\ 9 & 9 & \bot \\ \bot & 18 & 6 \end{matrix}\right).\end{aligned}\ ] ] running alekhnovich s algorithm on @xmath343 , we obtain : @xmath344 since @xmath345 , lemma [ lem_mulders ] gives @xmath346 as the upper bound on the number of row reductions , but it was done with only 24 . in the next iteration we again meet a @xmath235 . for our polynomial we can pick either the second or third row of @xmath347 since both have weighted degree @xmath348 ; we choose the second and obtain : @xmath349 again root - finding yields no results . the next element in @xmath245 is @xmath234 and we get intermediate parameters @xmath350 . we construct @xmath351 according to which gives : @xmath352 we needed 90 row reductions to reduce @xmath353 to weak popov form , while the upper bound is 160 , since we calculated @xmath354 . after row - reduction , we obtain @xmath355 : @xmath356 the next element in @xmath245 is @xmath234 , so we construct @xmath357 according to and get : @xmath358 we needed 86 row reductions for the module minimisation , while the upper bound was 145 since we calculated @xmath359 . the last iteration is again a @xmath235 , and we can use either of the two last rows of @xmath357 since they have weighted degree @xmath360 . using the last , the obtained polynomial is : @xmath361 indeed , @xmath362 and root - finding retrieves @xmath38 for us . as the example shows , performing module minimisation on a matrix can be informally seen to `` balance '' the row - degrees such that they all become roughly the same size . the complexity of this reduction depends on the number of row reductions , which in turn depends on the `` unbalancedness '' of the initial matrix . the matrices @xmath363 and @xmath357 are more balanced in row - degrees than using @xmath364 and @xmath365 directly . using the estimates of the two preceding subsections , we can make a rather precise worst - case asymptotic complexity analysis of algorithm [ alg_multitrial ] . the average running time will depend on the exact choice of @xmath245 but we will see that the worst - case complexity will not . first , it is necessary to know the complexity of performing a root - finding attempt . [ lem_rootcompl ] given a polynomial @xmath366}}}[y]$ ] of @xmath15-degree at most @xmath1 and @xmath13-degree at most @xmath172 , there exists an algorithm to find all @xmath12}}}$]-roots of complexity @xmath367 , assuming @xmath368 . we employ the roth ruckenstein @xcite root - finding algorithm together with the divide - and - conquer speed - up by alekhnovich @xcite . the complexity analysis in @xcite needs to be slightly improved to yield the above , but see @xcite for easy amendments . [ thm_complalg ] for a given @xmath230 code , as well as a given list of steps @xmath245 for algorithm [ alg_multitrial ] with ultimate parameters @xmath231 , the algorithm has worst - case complexity @xmath369 , assuming @xmath370 . the worst - case complexity corresponds to the case that we do not break early but run through the entire list @xmath245 . precomputing @xmath219 using lagrangian interpolation can be performed in @xmath371 , see e.g. @xcite , and reducing to @xmath335 is in the same complexity by lemma [ lem_alekcompl ] . now , @xmath245 must contain exactly @xmath236 @xmath233-elements and @xmath108 @xmath234-elements . the complexities given in corollaries [ cor_complmsi ] and [ cor_complmsii ] for some intermediate @xmath239 can be relaxed to @xmath0 and @xmath1 . performing @xmath372 micro - steps of type i and @xmath373 of type ii is therefore in @xmath374 . it only remains to count the root - finding steps . obviously , it never makes sense to have two @xmath235 after each other in @xmath245 , so after removing such possible duplicates , there can be at most @xmath1 elements @xmath235 . when we perform root - finding for intermediate @xmath239 , we do so on a polynomial in @xmath224 of minimal weighted degree , and by the definition of @xmath224 as well as theorem [ thm_gsproblem ] , this weighted degree will be less than @xmath375 . thus we can apply lemma [ lem_rootcompl ] with @xmath376 . the worst - case complexity of our algorithm is equal to the average - case ( and worst - case ) complexity of the beelen brander @xcite list decoder . however , theorem [ thm_complalg ] shows that we can choose as many intermediate decoding attempts as we would like without changing the worst - case complexity . one could therefore choose to perform a decoding attempt just after computing @xmath335 as well as every time the decoding radius has increased . the result would be a decoding algorithm finding all _ closest _ codewords within some ultimate radius @xmath2 . if one is working in a decoding model where such a list suffices , our algorithm will thus have much better average - case complexity since fewer errors occur more frequently than many . we now discuss how to adjust algorithm [ alg_multitrial ] to incorporate the re - encoding transformation proposed in @xcite . the basic observation is that we can correct @xmath204 if we can correct @xmath377 for any @xmath378 . if we chose @xmath379 such that @xmath377 for some reason is easier to handle in our decoder , we can save computational work . as in the original articles , we will choose @xmath379 such that it coincides with @xmath204 in the first @xmath34 positions ; this can be done since it is just finding a lagrange polynomial of degree @xmath380 that goes through these points . the re - encoded received word will therefore have 0 on the first @xmath34 positions . for ease of notation , assume that @xmath204 is this re - encoded received word with first @xmath34 positions zero , and we can reuse all the objects introduced in the preceding sections . define @xmath381 obviously @xmath382 so introduce @xmath383 . however , since @xmath384 for @xmath385 then also @xmath386 ; this will be the observation which will save us computations . introduce therefore @xmath387 . regard now @xmath114 of ; it is clear that @xmath388 divides every entry in the @xmath90th column for @xmath389 . this implies that the image of the following bijective map is indeed @xmath14}}}$ ] : @xmath390}}}\nonumber \\ q(x , y ) \quad & \mapsto \quad l(x)^{-s } q(x , l(x)y).\end{aligned}\ ] ] extend @xmath391 element - wise to sets of @xmath74 elements , and note that @xmath391 is therefore an isomorphism between @xmath74 and @xmath392 . the idea is now that the elements in @xmath392 have lower @xmath13-degree than those in @xmath74 , and we can therefore expect that working with bases of @xmath392 is computationally cheaper than with bases of @xmath74 . since we are searching a minimal @xmath75-weighted polynomial in @xmath74 , we need to be sure that this property corresponds to something sensible in @xmath392 . the following lemma and its corollary provides this : [ lem_wdeg_reenc ] for any @xmath49 in @xmath74 @xmath393 we have @xmath394 } } ( x ) + i(k-1 ) \big\}$ ] so we obtain : @xmath395 } } ( x ) - s \deg l(x ) + i \deg l(x ) - i \big\ } \\ & = \max_i \big\ { \deg { q_{[i ] } } ( x ) + i(k-1 ) \big\}- sk.\end{aligned}\ ] ] [ cor_wdeg_reenc ] @xmath49 has minimal @xmath75-weighted degree in @xmath74 if and only if @xmath396 has minimal @xmath397-weighted degree in @xmath392 . let us now describe the basis of @xmath392 corresponding to the one in : [ thm_mbasis_reenc ] the module @xmath392 is generated as an @xmath12}}}$]-module by the @xmath93 polynomials @xmath398}}}$ ] given by @xmath399 follows directly from theorem [ thm_mbasis ] and the mapping as defined . we can represent the basis of @xmath392 by the @xmath111 matrix over @xmath12}}}$ ] ( compare to ): @xmath400 we need an analogue of corollary [ cor_weightsol ] for the @xmath397-weighted degree , i.e. , we should find a diagonal matrix to multiply on @xmath401 such that when module minimising the result , we will have a row corresponding to a polynomial in @xmath392 with minimal @xmath397-weighted degree . we can not use @xmath402 , since multiplying with negative powers of @xmath13 might cause us to leave the polynomial ring ; however , we can to this add the same power to all the diagonal elements such that they become non - negative : @xmath403 therefore , for a vector @xmath404 in the row - space of @xmath405 corresponds a polynomial @xmath406 , and we will have the identity @xmath407 . obviously then , a minimal degree vector in @xmath405 is a minimal @xmath397-weighted polynomial in @xmath392 . finally , we need adjusted variants of the micro - steps i and ii . the necessary adaptions of algorithm [ alg_multitrial ] are summarised in the following lemma : [ lem_microstepsreenc ] let @xmath408}}}^{(\ell+1 ) \times ( \ell+1)}$ ] be the matrix representation of a basis of @xmath409 . if @xmath410 is in weak popov form , then one of the rows of @xmath411 corresponds to a polynomial in @xmath409 with minimal @xmath397-weighted degree . modified micro - steps of type i and ii can be obtained from the following . let @xmath412}}}$ ] be a basis of @xmath74 . then the following is a basis of @xmath256 : @xmath413 similarly , the following is a basis of @xmath273 : @xmath414 the first part follows from the previous discussion and analogously to corollary [ cor_weightsol ] . the recursive bases of and follow completely analogous to lemmas [ lem_microstepone ] and [ lem_microsteptwo ] , given theorem [ thm_mbasis_reenc ] . in the case of the @xmath312 code over @xmath313 with final decoding radius @xmath415 as shown in example [ ex_algo164 ] , then @xmath416 and the initial matrix satisfies : @xmath417 brander briefly described in his thesis @xcite how to incorporate re - encoding into the beelen brander interpolation algorithm by dividing out common powers of @xmath418 in the first @xmath0 columns of @xmath114 . here we construct instead @xmath419 where powers of @xmath418 are also multiplied on the latter @xmath236 columns , since we need the simple recursions of bases of @xmath392 which enables the micro - steps . however , before applying module minimisation , we could divide away this common factor from those columns and just adjust the weights accordingly ( i.e. , multiplying @xmath420 on the @xmath90th element of @xmath421 ) ; this will further reduce the complexity of the minimisation step . the micro - steps would then need to be modified ; the simplest way to repair this is to multiply back the powers of @xmath418 before applying a micro - step , and then remove them again afterwards . with a bit more care one can easily do this cheaper , though the details become technical . in asymptotic terms , the computational complexity of the iterative interpolation method stays exactly the same with re - encoding as without it , since @xmath422 under the usual assumption of @xmath423 being constant . the same is true for the original re - encoding scheme of ktter vardy @xcite . however , most of the polynomials that are handled in the matrix minimisation will be of much lower degree than without re - encoding ; for relatively high - rate codes this will definitely be noticeable in real computation time . in ( * ? ? ? * thm . 10 ) , it was shown that the root - finding procedure of roth ruckenstein @xcite , or its divide-&-conquer variant by alekhnovich @xcite can be directly applied to an interpolation polynomial in @xmath409 , so we can avoid to construct and work on the larger polynomial in @xmath123 . instead of finding @xmath38 , one will find the power series expansion of @xmath424 . the fraction in reduced form can be retrieved from the power series expansion using pad approximation : e.g. by the berlekamp massey algorithm or by module minimising a certain @xmath425 matrix . see e.g. ( * ? ? ? * section 2.5 ) for a general description of the latter . from the reduced fraction , @xmath38 can be obtained by re - extending the fraction . interestingly , one can easily calculate that the orthogonality defects stays the same , i.e. , @xmath426 , where @xmath427 is the matrix corresponding to a micro - step of type i in the re - encoded version . the analogue equality holds for type ii . this means that , roughly , the number of row operations carried out by the module minimisation algorithm is unchanged . the proposed algorithm has been implemented in sage , version 5.13 @xcite , using the mulders storjohann algorithm for module minimisation , and the roth ruckenstein root - finding procedure @xcite . for comparison , we also implemented construction of @xmath114 , leading immediately to the lee osullivan algorithm @xcite . figure [ fig_sim164 ] shows the total number of finite field multiplications performed for complete runs of the decoding algorithms , using the @xmath312 code over @xmath313 as considered in example [ ex_rs164param ] . for each algorithm , and for each number of errors @xmath428 , 1000 random codewords were generated and subjected to a random error pattern of weight precisely @xmath429 . the solid line gives the number of operations of the proposed multi - trial algorithm for any number of errors . for the lee osullivan decoder , one chooses the maximal decoding radius initially , and the figure depicts choosing both @xmath430 as dashed lines . the minimum - distance choice of @xmath431 coincides completely with the multi - trial algorithm since @xmath114 for @xmath432 and so is not shown . the figure demonstrate that the multi - trial algorithm provides a huge gain whenever there are fewer errors than lee osullivan s target , while not having a disadvantage in the matching case . the right - hand graph shows the complexity when using the re - encoding transformation , and we can observe a speedup of between 30% and 50% . the number of operations spent in constructing the matrices and for the root - finding step are relatively small compared to the number of operations needed for the row - reduction ( for the @xmath312 code , less than 5% ) . unfortunately , the performance of our operation - counting implementation for the decoding algorithm does not allow to run simulations with much larger codes . without counting , however , we can use optimised data structures in sage @xcite which run much faster . we have observed for the short @xmath312 code that the system - clock time spent with these data structures correspond very well to the operations counted . extrapolating , we can , albeit with larger uncertainty , make comparisons on larger codes using system - clock measurements . doing so , we have observed behaviour resembling that of the @xmath312 code . for example , decoding a @xmath433 code up to 23 errors ( requiring @xmath434 ) showed the multi - trial being slightly faster than lee osullivan in the worst case , while of course still giving a large improvement for fewer errors . when decoding a @xmath435 code up to 74 errors ( requiring @xmath436 ) , lee osullivan @xcite was slightly faster by about 15% in the worst case . it should also be noted that root - finding took up significantly more time for these larger codes , around 15% of the total time for lee osullivan and 20% for the multi - trial . as with any simulation , there are caveats to these results . in truth , only the wall - clock time spent by highly optimised implementations of various approaches can be fairly compared . also , we did not test the asymptotically fast alekhnovich s method for module minimisation . investigations performed by brander in magma indicated that the gain in using this algorithm in place of mulders storjohann might only be present once the code length exceeds about 4000 @xcite . the implementation of the algorithm , including the simulation setup , is freely available via http://jsrn.dk/code-for-articles . an iterative interpolation procedure for list decoding grs codes based on alekhnovich s module minimisation was proposed and shown to have the same worst - case complexity as beelen and brander s @xcite . we showed how the target module used in beelen brander can be minimised in a progressive manner , starting with a small module and systematically enlarging it , performing module minimisation in each step . the procedure takes advantage of a new , slightly more fine - grained complexity analysis of alekhnovich s algorithm , which implies that each of the module refinement steps runs fast . the main advantage of the algorithm is its granularity which makes it possible to perform fast multi - trial decoding : we attempt decoding for progressively larger decoding radii , and therefore find the list of codewords closest to the received . this is done without a penalty in the worst case but with an obvious benefit in the average case . the beelen brander approach for interpolation is not the asymptotically fastest : using the module minimisation algorithm by giorgi _ et al . _ @xcite , one gains a factor @xmath1 . by a completely different approach , chowdhury et al . @xcite further beat this by a factor @xmath437 , achieving @xmath438 . it is unclear for which sizes of the parameters these asymptotic improvements have concrete benefits , and whether a multi - trial approach can be developed for them . chowdhury , m.f.i . , jeannerod , c.p . , neiger , v. , schost , e. , villard , g. : faster algorithms for multivariate interpolation with multiplicities and simultaneous polynomial approximations .

absolute ages of young stars are important for many issues in pre - main sequence stellar and circumstellar evolution but are long recognized as difficult to derive and calibrate . in this paper , we use literature spectral types and photometry to construct empirical isochrones in hr diagrams for low - mass stars and brown dwarfs in the cha , cha , and tw hya associations and the pic and tuc - hor moving groups . a successful theory of pre - main sequence evolution should match the shapes of the stellar loci for these groups of young stars . however , when comparing the combined empirical isochrones to isochrones predicted from evolutionary models , discrepancies lead to a spectral type ( mass ) dependence in stellar age estimates . improved prescriptions for convection and boundary conditions in the latest models of pre - main sequence models lead to a significantly improved correspondence between empirical and model isochrones , with small offsets at low temperatures that may be explained by observational uncertainties or by model limitations . independent of model predictions , linear fits to combined stellar loci of these regions provide a simple empirical method to order clusters by luminosity with a reduced dependence on spectral type . age estimates calculated from various sets of modern models that reproduce li depletion boundary ages of the pic moving group also imply a myr age for the low mass members of the upper sco ob association , which is younger than the 11 myr age that has been recently estimated for intermediate mass members .

1505.06518

age estimates for pre - main sequence stars are critical in assessing the timescales for the formation of stellar systems and the evolution of molecular clouds . the fraction of embedded sources determines how long the dominant phase of stellar growth lasts and how quickly the envelope disperses . similarly , when combined with disk fractions , age measurements place strict constraints on the time available for giant planets to form . ages are also important in quantifying how the stellar contraction leads to rotational spin - up and changes in interior structure for solar and intermediate mass stars , factors which combine to determine the initial conditions for the evolution of magnetic activity for the duration of the stellar lifetime . members of young stellar associations are usually dated by comparing their temperatures and luminosities to pre - main sequence tracks . however , errors and biases in placing stars on hertzsprung - russell ( hr ) diagrams lead to uncertainty in ages of different clusters , while large luminosity spreads among members of individual clusters enhance uncertainty regarding age dating of any individual star ( see review by * ? ? ? further uncertainties in pre - main sequence age estimates from hr diagrams are traced directly to the varied stellar loci predicted from different stellar evolution models and to the lack of consistency between predicted and observed stellar loci . these differences have been noted between low and intermediate mass stars ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and among low mass stars of different mass ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? . components of young binary systems frequently have different estimates for age despite an expectation that they are co - eval @xcite . as a direct consequence , age estimates depend on the considered range in spectral type or stellar mass . @xcite found that in many regions g stars appears @xmath4 times older than k stars , depending on which pre - main sequence tracks were used . @xcite quantified these differences by measuring an age for the upper sco ob association of @xmath5 myr , larger than the 5 myr age measured previously from pre - main sequence low mass stars @xcite . similarly , in both younger and older clusters , analyses of main sequence turnoff and intermediate mass stars in color - magnitude diagrams yield ages that are twice as large as those obtained from low mass pre - main sequence stars @xcite . nearby stellar associations with ages of 540 myr provide us with a laboratory to establish empirical isochrones , which can be used as a model - independent clock and as a benchmark for testing pre - main sequence models . within a single association older than 10 myr , the age spread expected from star formation processes is small compared to the absolute age . member stars have traveled far from any remnant molecular material associated with their birth cloud and are unaffected by interstellar extinction . most disks have already dispersed by these ages , so observational uncertainties related to accretion and circumstellar extinction are negligible for the vast majority of stars in this age range . in recent years , the growing number of proper motions and parallax distances have led to a substantial improvement in the accounting of membership within young associations ( e.g. * ? ? ? * ; * ? ? ? the hr diagrams of these nearby moving groups should consequently have only a small age spread . at younger ages , placing stars on hr diagrams is more problematic because accretion and extinction affects temperature and luminosity measurements , and because the affect of any genuine age spread on luminosities decreases ( in dex ) with age . nevertheless , wide spreads in luminosities of members of young clusters have been interpreted as true age spreads . for example , in the onc a 1.5 dex luminosity spread ( da rio et al . 2010 ) led reggiani et al . ( 2011 ) to claim an age spread of 1.53.5 myr , although some of this spread may alternately be interpreted as a spread in radius from different accretion histories @xcite . improved spectral types and extinction measurements of two outliers in the onc sample indicate that the luminosity spread may be overestimated @xcite . if we consider @xmath6 myr as an approximate age spread , then that would lead to 0.16 dex and 0.08 dex spreads in luminosities of 5 myr old and 20 myr old clusters of single stars ( as assessed for 3500 k stars from the baraffe tracks , see 2.5 ) . these luminosity spreads are consistent with the 0.05 to 0.15 dex errors in placing young stars in the hr diagram ( see table 1 ; * ? ? ? * ; * ? ? ? * ) . in this paper , we use literature spectral types and photometry to construct empirical isochrones of luminosity versus temperature of diskless low mass stars in nearby associations and compare them to pre - main sequence isochrones . this approach is similar to the empirical isochrones derived by @xcite and by @xcite , but is focused exclusively on low mass stars in nearby young associations and moving groups and is motivated by a desire for deriving simple formulations that approximate empirical isochrones in hr diagrams . in 2 , we describe the literature data that is used to place stars on hr diagrams and the pre - main sequence evolutionary tracks used for age estimates . in 3 we develop empirical isochrones for a set of benchmark young associations . in 4 , the empirical isochrones are compared to isochrones predicted from models of pre - main sequence evolution to test evolutionary tracks and to obtain age estimates for the young clusters for each model . in 5 , we discuss how the ages from different tracks compare with independent age measurements . the @xmath2 pic and tuc - hor moving groups and the @xmath1 cha , @xmath0 cha , and tw hya associations are used in this paper as benchmark clusters to measure empirical isochrones of young regions . the results obtained from these benchmark clusters are then applied to the upper sco ob association . searches for membership in existing all sky surveys , along with analyses of proper motions and radial velocities , have substantially improved the confidence in membership for these associations . meanwhile , the proliferation of accurate parallax measurements for the nearest groups has significantly reduced the uncertainty in luminosities arising from errors in distance measurements . when combined , these advances allow for a robust analysis of the stellar loci of these clusters . in some analysis below , the two older associations ( the @xmath2 pic and tuc - hor moving groups ) and the three younger associations ( tw hya , @xmath0 cha , and @xmath1 cha ) are grouped together . gravity differences between these clusters are described in 2.2 - 2.3 , and the similarity in ages amongst the older and younger groups will be established in 3 . the binary census is discussed in this section , while the implementation of binaries is described in 2.4 . _ beta pic moving group : _ membership for the @xmath2 pic moving group is obtained from the analysis of @xcite , based on a @xmath7% probability calculated from proper motion , radial velocity , photometry , and parallaxes when available . spectral types are obtained from their compilation of literature spectral types . members with distances obtained from kinematic modelling are not used in this paper because many targets are close enough that the percentage uncertainties are too large@xmath8 . binarity was accounted for based on the compilation of @xcite , which relied heavily on @xcite for m - dwarfs . a few objects and binary information is added from @xcite . _ tuc - hor : _ the membership and spectral types for the tuc - hor association is obtained from @xcite . the distances used here are from their kinematic analysis@xmath9 . the spectral types used here are obtained from their spectral analysis rather than the seds . binary information is not included in our analysis of tuc - hor . _ @xmath1 cha association : _ membership , spectral types , and kinematic distances of @xmath1 cha members were obtained from @xcite . this membership list is mostly consistent with but more complete than that of @xcite . the targets listed as members in @xcite but excluded from @xcite are vw cha and cm cha , both of which have disks and would therefore not be included in our analysis , and 2mass j12074597 - 7816064 , which was identified by @xcite as a proper motion outlier . the kinematic distances from @xcite are adopted here and have a median of 111 pc . the known binaries are accounted for following the description in @xcite , though the binary census is poor . _ @xmath0 cha cluster : _ membership in the @xmath0 cha cluster is based primarily on the initial identification by @xcite and on a follow - up census by @xcite . the spectral types for those initial 18 members are obtained from @xcite . five dispersed very low mass stars were found and characterized by @xcite and are also included here . the additional members found by @xcite based on proper motions and the @xcite x - ray survey are excluded here because they are far from the cluster core and may therefore be located at a different distance . the distance to all members is assumed to be 94 pc , the average hipparcos parallax distance of @xmath0 cha and rs cha ( van leeuwen 2007 ; see discussion in murphy et al . an accounting of spatially resolved binaries was obtained by @xcite and is applied here . _ tw hya association : _ membership in the twa is based on kinematic analyses @xcite . spectral types of most members and binarity are obtained from the compilation in @xcite . the new equal luminosity binary twa member scr 1012 - 3124ab is also included here @xcite . although parallax distances are preferred , twa stars with kinematic distance measurements are included in our analyses . for twa 8a and 8b , a parallax distance distance of 44 pc is adopted here as the average of parallax distances measured by @xcite and @xcite . three members of the hd 98800 quadruple system are included here with data from @xcite , despite the presence of weak ir emission from hd 98800 n. _ upper sco ob association : _ the stellar census and spectral types for the upper sco ob association are obtained from the compilation of members by @xcite . multiplicity is corrected based on several incomplete companion searches @xcite . the luminosities estimated for the upper sco ob association members are corrected for extinction ( see 2.4 ) . the average extinction@xmath10 of @xmath11 magfor low mass stars and @xmath12 mag for brown dwarfs is intermediate between the average extinction of @xmath13 mag for a stars , @xmath14 mag for g stars , and @xmath15 mag for f stars @xcite . stars with @xmath16 mag are atypical for upper sco and are not included in the hr diagram analysis . all stars in upper sco are assumed to be located at a distance of 145 pc @xcite . the cluster is 30 pc wide , and a depth of similar length likely contributes to the large luminosity spread that has been measured for upper sco @xcite . stars without any wise w2 or w3 photometry ( 2 of 214 stars total between 32005000 k , 7 of 415 total brown dwarfs between 28003200 k ) are included in the upper sco sample because their exclusion could bias any results at low masses against faint objects . lcl parameter & @xmath17 range & uncertainty + temperature & @xmath18 k & 200 k or 2 spt subclasses + temperature & 37504100 k & 125 k or 1 spt subclass + temperature & 28003750 k & 75 k or 0.5 spt subclasses + luminosity & all & @xmath19 parallax distance + luminosity & all & 10 - 20% kinematic distance + luminosity & all & 2 - 5% bolometric correction@xmath20 + + photometry for members of the five benchmark clusters are obtained from the 2mass jhks survey @xcite , wise w1 , w2 , w3 , w4 survey @xcite , the ucac4 apass all - sky bvgri survey @xcite , and the galex nuv / fuv survey @xcite . this collection of photometry is used in the following subsections to identify and exclude disks , to ensure that extinction is negligible ( or correct luminosities for extinction , in the case of upper sco members ) , and to calculate bolometric corrections ( see also appendix a ) . all objects are detected in jhk . some objects lack photometry at other bands . broadband colors of objects in the younger tw hya , @xmath0 cha , and @xmath1 cha associations are mostly consistent with those from the older @xmath2 pic and tuc - hor moving groups , which all form a tight stellar locus in infrared and optical colors ( figure [ fig : jh ] and appendix ) . the contraction to the main sequence corresponds with a gradual increase in gravity . just as different gravities between main sequence and young stars yields color differences ( e.g. * ? ? ? * ; * ? ? ? * ) , corresponding differences should be present between 5 and 30 myr old pre - main sequence stars . h color is redder for the younger associations relative to the older moving groups ( table [ tab : bcor.tab ] and right panel of figure [ fig : jh ] ) , which is the result of gravity differences and possible mismatches between spt and temperature between the two samples . k color is also @xmath21 mag redder for early - to - mid m stars in younger associations . most empirical colors for the sample stars are also consistent with the intrinsic colors for the 530 myr stars as assessed by @xcite . the sample is restricted to diskless stars . disks are identified by at least one of a k - w3 excess or k - w2 of 0.5 mag above the photospheric level , as determined from the obseved stellar locus and from photospheric models ( figure [ fig : diskcolor ] ; see also , e.g. , luhman & mamajek 2012 ) . a few stars may retain transition or debris disks that do not appear as a k - w2 or k - w3 excess , but these disks are unlikely to affect the j - band brightness of the star . the sources 2mass j08014860 - 8058052 , 2mass j11550485 - 7919108 , and 2mass j02153328 - 5627175 fall below our disk criteria but are excluded in our analysis because of excess emission in the wise w4 band or the detection of accretion in optical spectra @xcite . c|cc|cc|ccc & & & + spt & hh14@xmath22 & pm13@xmath23 & here & pm13@xmath23 & old@xmath24 & young@xmath24 & pm13@xmath23 + f5&6600&6420&0.79&0.85&0.17&0.17&0.19 + f8&6130&6100&0.95&0.96&0.22&0.22&0.23 + g0&5930&6050&1.02&0.98&0.26&0.26&0.24 + g2&5690&5870&1.10&1.03&0.30&0.30&0.27 + g5&5430&5500&1.18&1.16&0.34&0.34&0.33 + g8&5180&5210&1.26&1.25&0.40&0.40&0.39 + k0&4870&5030&1.36&1.30&0.47&0.47&0.43 + k2&4710&4760&1.41&1.40&0.50&0.50&0.49 + k5&4210&4140&1.56&1.58&0.63&0.63&0.64 + k7&4020&3970&1.62&1.63&0.66&0.66&0.66 + m0&3900&3770&1.66&1.69&0.67&0.68&0.68 + m1&3720&3630&1.73&1.74&0.67&0.70&0.68 + m2&3560&3490&1.78&1.80&0.65&0.70&0.67 + m3&3410&3360&1.84&1.84&0.63&0.70&0.66 + m4&3190&3160&1.93&1.91&0.58&0.65&0.62 + m5&2980&2880&1.99&2.01&0.53&0.59&0.55 + m6&2860&&2.03&&0.50&0.55 & + m7&2770&&2.06&&0.48&0.53 & + + + + + + + + + the luminosities of stars in the five young , benchmark associations are calculated by applying bolometric corrections to 2mass j - band magnitudes @xcite . any bulk offset between the actual and implemented bolometric correction would lead to an offset in luminosity and therefore age , while any temperature dependence in bolometric correction offset would change the relative luminosities / ages of stars in the same association but with different temperatures . these systematic uncertainties introduce problems when comparing different methods and when comparing observations to models . in this subsection , we calculate bolometric corrections and subsequently assess temperature dependence in the uncertainties by comparing results from temperature and bolometric corrections derived here to those of @xcite . the appendix describes the color corrections and constants that are used here to calculate fluxes and stellar luminosities . our bolometric corrections are calculated from the bt - settl model atmospheres ( version cifits2011bc , which use abundances from caffau et al . 2011 ) with @xmath25 @xcite . minor offsets relative to the j - band flux are applied to several spectral bands to account for differences between synthetic and observed photometry ( see figure [ fig : colors ] ) . spectral types are converted to temperature from the conversion scale of @xcite , which were calculated by comparing low resolution optical blue - red spectra to 2011 bt - settl models . the model spectra and observed photometry are compared in magnitude space . the synthetic photometry is calculated by convolving the bt - settl model spectra in photon units with filter transmission curves ( see appendix a ) . the empirical colors are obtained from the photometry described in 2.2 . the primary uncertainties in these empirical corrections are introduced by the lack of photometric coverage in the z and y bands and in near - ir regions with poor telluric transmission , which are located between the zyjhk filters . table [ tab : bcor.tab ] compares our spectral type - temperature conversion and bolometric corrections to the 530 myr stars of @xcite . in general , temperatures in @xcite are 100200 k cooler than those presented here . these errors are roughly assessed from the uncertainty in assigning spectral types from low resolution optical spectra . @xcite also include k8 and k9 spectral types . their spt - temperature conversion extend only to m5 , so results for brown dwarfs ( 3.2 ) are limited to the conversions of @xcite . when evaluated at the same temperature , the bolometric corrections measured directly from the bt - settl models with the @xcite abundances reproduce the @xcite bolometric corrections to within 1% at all spectral types , even though @xcite used models with abundances from @xcite . differences between our bolometric corrections and those of pecaut et al . are introduced by differences in the spt - temperature conversion and by our tweaking of the bt - settl models to reproduce the measured colors , relative to j. the bolometric correction versus spectral type is also different because of the different spectral type - temperature conversion scale . as a consequence , when compared by temperature , bolometric corrections are similar for m - dwarfs and @xmath6% smaller here for k - dwarfs . the practical result of these differences is that temperatures are @xmath26 k higher and luminosities are @xmath27 brighter relative to those calculated from @xcite bolometric correction and temperature scales . the analysis of stellar temperatures and luminosities for 32005000 k stars ( 3.1 ) is repeated for the @xcite spt - temperature conversions and bolometric corrections . both @xcite and @xcite base their spt - temperature conversions on the bt - settl models , although @xcite find discrepancies between these models and near - ir colors at both early and late - m spectral types . while the spt - temperature conversion is still a significant source of uncertainty , the main results of this paper are robust to systematic uncertainties in these scales . stellar luminosities are calculated from 2mass j - band brightness , literature spectral types , and stellar distances calculated from parallax measurements or from association kinematics . the j - band is used for luminosity calculations because of the uniform reliability of the 2mass survey , and because the j - band is near the peak of the spectral energy distribution for late type stars , which minimizes the uncertainty in bolometric correction caused by spectral type errors . the stars are diskless , so accretion and disk extinction do not affect the observed j - band photometry . stars in each association are assumed to be co - eval . table [ tab : uncertainty ] lists the uncertainties applied here for luminosity and temperature estimates . luminosity uncertainties are calculated from an assumed 5% uncertainty from a combination of photometric and bolometric correction errors and from distance uncertainties , typically @xmath28% for parallax distances ( 5% distance uncertainty is assessed for all @xmath0 cha association members ) , 5% for kinematic distances for tuc - hor members ( see footnote 3 ) , and 10% for kinematic distances of other regions . bolometric corrections for the j - band are described in detail in the next section and are more uncertain for cooler stars with large opacities in molecules . the depth of upper sco adds uncertainty in the distance and therefore luminosity for individual members , which contributes to its luminosity spread . temperature uncertainties are based on estimated errors and methodological differences in spectral typing ( see discussion in herczeg & hillenbrand 2014 ) . systematic differences in spectral typing between different studies will also affect the hr diagrams constructed in this paper . in multiple systems , all members with resolved spectral types are included in this analysis . unresolved spectral types are included if the stars are spatially resolved in j , h , or k - bands . in these cases , the primary is assumed to have the system spectral type . the j - band magnitude of the primary is calculated by assuming that the brightness difference is constant in the near - ir ( the j h and j k colors roughly constant for m - dwarfs ) . optically resolved or double - lined spectroscopic binaries are included if both stars are similar brightness . single - lined binaries are excluded . extinction is negligible and assumed to be negligible for all diskless stars in the five benchmark associations ( see colors in 2.2 and the appendix ) . stars in the younger associations have a slightly bluer g - j color , which is the opposite of expectations if extinction were important and supports the assumption that extinction is negligible . the four stars in @xmath1 cha with @xmath29 mag , as calculated by @xcite , all have disks , which may either produce extinction or may alter the near - ir colors sufficiently to have led to extinction in their calculations . however , extinction affects luminosities of stars in the upper sco ob association . a color excess e(j - k ) is calculated for each star here from by measuring the difference between the observed j k color the j k color adopted for the young associations , as shown as the dashed purple line in figure [ fig : jh ] . the extinction @xmath30 is then calculated and corrected for by assuming a total - to - selective extinction of @xmath31 , and @xmath32 from @xcite . if the total - to - selective extinction @xmath33 instead of 3.1 , the @xmath30 values would increase by 0.07% , as calculated from the extinction curves of @xcite . young k and m stars without disks have extinctions accurate to @xmath34 mag , although the precise error in extinction depends on the error in spectral type . stars with extinctions @xmath16 mag are excluded from our analysis . negative values for extinction are retained and applied for statistical purposes . the empirical isochrones derived in 3 are compared to pre - main sequence isochrones in 4 . @xcite summarizes currently available pre - main sequence evolutionary tracks . the evolutionary models used here , dantona & mazzitelli ( 1997 , hereafter dm97 ) , siess et al . ( 2000 , hereafter siess ) , baraffe et al . ( 1998 , 2003 , hereafter bcah98 ) , dotter et al . ( 2008 , hereafter dartmouth ) , tognelli et al . ( 2011 , hereafter pisa ) , and parsecv1.2s ( bressan et al . 2012 , chen et al . 2014 ; herafter parsec , also frequently referred to as padova ) , are selected based on their availability and prevalence in the pre - main sequence literature . the bcah98 models have mixing lengths @xmath35 ( bcah98 ) or mixing length @xmath36 ( hereafter bcah98 - 1.9 ) . the parsec models are selected for their artificial tweaks to the @xmath37 relationship that reproduce the color - magnitude diagrams of very low mass stars in praesepe and m67 . for stars below 4000 k , stars of the same age and mass are much cooler and are only slightly less luminous ( because of larger radii ) in the parsecv1.2s models relative to the parsecv1.1 models @xcite . the parsecv1.1 models are used only in table [ tab : slopes.tab ] and are not discussed further . in addition to these sets of models , we include analysis of newly available tracks calculated by baraffe et al . ( 2015 , hereafter bhac-2015 ) using the bt - settl atmospheric models and a prescription of convection based on 2d/3d models , which is an improvement from the mixing length prescription for convection . we also include new dartmouth tracks , which improve the thermal structure in the atmosphere by calculating fitting to large optical depths ( feiden et al . in prep . , private communication , hereafter feiden ) the feiden models use the @xcite abundances with phoenix ames - cond atmospheres @xcite . @xcite include preliminary descriptions of the feiden tracks , while @xcite use a magnetic version of the feiden tracks . the bhac-2015 and feiden models predict very similar temperature / luminosity evolution for low mass stars and brown dwarfs , which then yields very similar isochrones ( figure [ fig : baraffefeiden ] ) . all model tracks use solar abundances@xmath38 , which is consistent with abundance measurements of young stars in the solar neighborhood ( e.g. * ? ? ? only the bcah98 and the dm97 pre - main sequence tracks are available for masses @xmath39 @xmath40 , while the pisa models are only available for masses @xmath41 m@xmath42 . the bcah98 models with @xmath35 are available for stars less massive than @xmath43 @xmath40 while only models between @xmath44 @xmath40 have been calculated with @xmath36 . the parsec models are only calculated to @xmath45 m@xmath42 , but extend to low temperatures ( 2800 k at 1 myr , 2500 k at 10 myr ) because the tweaks to the @xmath37 relationship reduce the effective temperature of very low mass stars . figure [ fig : age3900k ] shows the conversion to age at fixed luminosity and at fixed temperature for the different pre - main sequence evolutionary models . the dm97 model tracks are significant outliers ( see also , e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? their photospheric temperatures are hotter for stars of the same mass , so the corresponding mass of k and m stars is much lower than obtained from other evolutionary tracks . the measured luminosity for a star of that mass is therefore high , leading to a young age . at ages @xmath46 myr , the dartmouth tracks also yield ages that are younger than the other tracks . in contrast , the bcah98 tracks assessed for 4200 k stars are outliers in yielding older ages . all models predict similar luminosities for stars of the same mass for ages older than 2 myr , despite different temperatures ( fig . [ fig : massconv ] ) . with accurate age dating of young clusters , masses may be inferred from the luminosity alone and with consistent results between different model tracks . with an accurate theory of pre - main sequence evolution , the estimated luminosity of a star at a given estimated temperature could be directly converted to an age . however , scatter in @xmath47 and @xmath17 among young cluster members provokes concerns regarding the reliability of age estimates for single stars . moreover , differences in the slope between observed and model isochrones lead to ages that depend on spectral type . our approach here focuses on comparing the empirical stellar locus of different associations and interpreting the offsets as relative age differences . using the stellar loci constructed for young associations , we first focus on empirical trends in @xmath48 versus @xmath17 ( this section ) before conducting detailed comparisons to models ( 4 ) . figure [ fig : hrall ] shows the combined hr diagrams for the five benchmark young associations . the stellar locus between 35005000 k has a similar shape for all clusters , with offsets in luminosity and temperature . stars in the older associations have systematically fainter luminosities and warmer temperatures relative to young associations . the slope of the locus is steeper for very low mass stars and brown dwarfs@xmath49 ( @xmath50 k for the young clusters and @xmath51 k for older clusters ) relative to the locus for solar mass stars . the location where the break in slope of the temperature - luminosity isochrone occurs is referred to here as the inflection point in the isochrone . the observed hr diagrams are consistent with the basic shape predicted from isochrones obtained from pre - main sequence evolutionary models ( see comparisons in 4 and fig . [ fig : hrt2 ] ) . the gradual decrease in luminosity from 5000 to @xmath52 k is roughly consistent with all pre - main sequence evolutionary models . at temperatures cooler than @xmath52 k , the steep dropoff in luminosity is also roughly consistent with predictions from tracks that calculate pre - main sequence evolution for stars @xmath53 @xmath40 . based on the hr diagrams , the inflection point where the slope steepens occurs at @xmath54 k for the older two associations and @xmath55 k for younger associations . the temperature intervals are therefore different in fits to stellar and brown dwarf loci of the older and younger associations . the similarity in stellar loci suggests that the members may be analyzed together after accounting for age - dependent differences between the clusters . in the following subsections , we combine the loci from different associations into a single stellar locus to calculate an empirical isochrone , which can then be generically applied to any young cluster . combining data from different clusters yields results that are more robust than those calculated for individual , sparsely populated associations . when comparing cluster ages , scatter in data points and different spectral type distributions of members can introduce systematic errors in relative ages once a generalized fit to temperature and luminosity is obtained , scaling the slopes to each cluster in the luminosity direction then provide self - consistent comparisons of relative ages that furthermore are robust to certain types of scatter in the temperature - luminosity data points . for example , linear fits to the twa and tuc - hor populations yield median luminosities of @xmath56 and @xmath57 , respectively , for 4200 k stars , or @xmath58 and @xmath59 for 3400 k stars@xmath60 . these luminosities can be compared among clusters to infer relative ages in @xmath61 as a fuction of @xmath62 . the loci of stars above and below the inflection point ( 3200 k for young regions , 3400 k for the older regions ) are discussed separately in 3.1 and 3.2 . an underlying assumption to this approach is that the slopes of stellar loci are constant longward and shortward of the inflection point and that the slopes do not change with cluster age . these sections include a discussion of whether this simplifying assumption is correct . for brown dwarfs in particular , the @xmath63 slopes are different for well - populated loci of upper sco and tuc - hor . the location of the inflection point ranges from 30003500 k and is discussed in 3.3 . lccccc & & & + & slope & std & slope & std + & @xmath64 & 0.15 & @xmath65 & 0.18 + & @xmath66 & 0.16 & @xmath67 & 0.20 + & @xmath68 & 0.14 & @xmath69 & 0.18 + & @xmath70 & 0.14 & @xmath71 & 0.18 + & @xmath72 & 0.14 & @xmath73 & 0.16 + & @xmath74 & 0.17 & @xmath24 & + & @xmath75 & 0.17 & @xmath24 & + & @xmath76 & 0.06 & @xmath24 & + & @xmath77 & 0.18 & @xmath78 & 0.27 + model & age@xmath23 & 34005000 k & & 28003200 k & + bcah98 & 3 & 1.04 & & 1.97 & + bcah98 & 20 & 0.89 & & 1.62 & + bcah98 - 1.9 & 3 & 0.69 & & & + bcah98 - 1.9 & 20 & 0.76 & & & + bhac-2015 & 3 & 0.68 & & 1.49 + bhac-2015 & 20 & 0.72 & & 1.84 + pisa & 3 & 0.71 & & & + pisa & 20 & 0.76 & & & + dartmouth & 3 & 0.73 & & & + dartmouth & 20 & 0.74 & & & + feiden & 3 & 0.75 & & 1.57 & + feiden & 20 & 0.84 & & 2.02 & + siess & 3 & 0.87 & & & + siess & 20 & 1.00 & & & + dm97 & 3 & 0.70 & & 2.46 & + dm97 & 20 & 0.85 & & 2.94 & + parsec & 3 & 0.62 & & 1.15 & + parsec & 20 & 0.51 & & 1.32 & + parsecv1.1 & 3 & 0.83 & & & + parsecv1.1 & 20 & 1.09 & & & + + + + + + + in this subsection , we describe global fits to the 32005000 k region ( 34005000 k for the @xmath2 pic and tuc - hor moving groups ) , which corresponds to m3k0 spectral types . since the empirical loci in this temperature range are similar , a global linear fit to the entire dataset provides a consistent assessment of the relative luminosity evolution of a given cluster . in fits to the combined hr diagram of the five benchmark associations , the free parameters are the slope of the line and a different y - intercept for each cluster . for illustrative purposes , the different y - intercepts are described here as luminosity offsets . the lower left panel of figure [ fig : hrall ] shows the hr diagram , with each association shifted by the respective difference in y - intercept to match the location of the twa . the empirical isochrone for the combined hr diagram of nearby associations has a best fit slope of @xmath79 based on linear regressions that includes errors in @xmath80 and @xmath47 ( as described in 2.3 ) and calculated using mpfitexy.pro in idl @xcite . the inflection point @xmath81 used for the fits here is 3200 k for the younger groups and 3400 k for the older groups . the 0.07 dex uncertainty includes a 0.05 dex statistical uncertainty and a @xmath82 dex uncertainty based on different linear fitting techniques , how the data points are weighted , and the inclusion or exclusion of deviant data points . this uncertainty does not include possible errors in the spectral type - temperature conversion or in bolometric corrections , which may depend on spectral type . the standard deviation in luminosity between the observations and best linear fits is @xmath83 dex . the same analysis calculated from the @xcite spectral type - temperature conversion and bolometric corrections produces a slope of @xmath84 @xmath61 per 10@xmath85 k , which is formally consistent with the results when using the @xcite conversions . the different slopes would yield a change in @xmath61 of @xmath86 dex between the hottest and coolest stars in the applicable temperature range , which is smaller than other systematic errors . table [ tab : slopes.tab ] lists the best fit slopes for linear fits to the the combined data set , to individual clusters , and to the clusters combined into the young and old groups . when shifted based on the best fit offsets , the three youngest benchmark associations have steeper slopes than the older clusters , which suggests some evolution with age . the difference in slopes between the young and old clusters , 0.76 and 0.85 @xmath61 per 10@xmath85 k , may lead to systematic differences as large as 0.14 dex for stars at 3400 k compared with 5000 k. in this subsection , we describe global fits to objects cooler than the inflection point in the isochrones ( 28003400 k for the tuc - hor and @xmath2 pic moving groups , 28003200 k for the tw hya , @xmath0 cha , and @xmath1 cha associations ) . for illustrative purposes , the different y - intercepts are visualized as temperature offsets so that the loci overlap ( lower right panel of fig . [ fig : hrall ] ) . this visualization is practical because the inflection point occurs at different temperatures but approximately the same luminosity ( to within a few tenths of dex ; see also similar behavior in the model isochrones in figs . [ fig : baraffefeiden ] and [ fig : hrt2 ] ) . the total combined fit to our benchmark clusters yields a slope of @xmath87 where the temperature range goes from 28003200 k for the younger clusters and 28003400 k for the older clusters ( as also described above with @xmath81 ) . the error in the slope is calculated as in 3.1 . the fits are dominated by members of the tuc hor and @xmath2 pic moving groups because those regions have more identified brown dwarf members . fits are calculated for the combined locus of young associations but are not calculated for each individually because each association has only a few data points . table [ tab : slopes.tab ] lists slopes for the individual regions and the combined hr diagram . the coolest known objects in these regions ( @xmath88 k ) are fainter than expected from these linear fits . they are excluded in this analysis , but their inclusion would significantly steepen the slopes of this lines . this discrepancy suggests that these fits may not be robust to the selection of the low temperature cutoff for this fit . the measured uncertainty in the linear fit of 0.15 @xmath89 per @xmath90 k leads to a 0.06 dex uncertainty in relative luminosity between a 3200 k and 2800 k brown dwarf . the difference in slopes between the older and younger associations likely leads to 0.09 dex uncertainties over the same regions . this difference is likely real evolution in the brown dwarf loci and not related to uncertainty in the fits . in 3.4 , comparisons to the upper sco ob association demonstrate that the slope becomes shallower with age . the results from linear fits in producing a luminosity for a 3000 k brown dwarf may still be robust as long as the known membership is well distributed between 28003200 k. however , this relationship should only be used with caution . the fit to the brown dwarf locus is not recalculated for @xcite parameters because their bolometric corrections are not calculated for spectral types later than m5 . 3.1 and 3.2 describe empirical fits to hr diagrams , as illustrated in figure [ fig : hrall ] . the regions warmer and cooler than the inflection point in the temperature - luminosity locus were fit separately , with one slope but different y - intercepts for each association . the shifts obtained from the y - intercepts are visualized as luminosity shifts , while the fits to the cooler stars are visualized as temperature shifts . the location of the inflection point , which motivates the separate fits to the two temperature ranges , depends on the cluster age . in this subsection , we compare the independent results from the fits to low mass star and brown dwarf isochrones to describe how the empirical lines combine to describe the low mass locus of young clusters . for illustrative purposes only , the combined isochrones are plotted against the young and old associations in figure [ fig : hrt ] . figure [ fig : shift ] shows the luminosity of 3000 k brown dwarfs versus 4200 k stars from our independent fits to the young clusters with a best fit line of @xmath91 _ combining the line of equation 3 with the two slopes reported in equations 1 and 2 fully describes the evolution of the empirical locus of single low mass stars and brown dwarfs in clusters . _ the isochrone for low mass stars falls in luminosity more quickly than that for brown dwarfs . the inflection point where the temperature - luminosity slope changes is determined by the intersection of the stellar and brown dwarf locus described by this line . as a further test of our method , figure [ fig : shift ] also shows preliminary results of its application to younger clusters with proportionally higher luminosities . the data are obtained for ic 348 , chameleon i , lupus i , taurus , @xmath92 ori , @xmath93 ori , and the onc @xcite . although the errors in placement of individual stars on the hr diagram are larger for younger stars with more influence from disks and heavier extinction , their mean luminosities are encouragement that our method could be applied to younger regions in order to constrain the luminosity evolution for comparison to theoretical models . figure [ fig : usco ] shows how the fits derived above are applied to the hr diagram of the upper sco ob association . lines with slopes that best fit the loci of low mass stars ( 32005000 k ) and brown dwarf / very low mass stars ( 28003200 k ) are scaled to the measured isochrone . the initial sample includes stars with k - w3 color consistent with diskless stars and also stars without wise detections . an initial scaling to the 32005000 k region ( 186 stars ) has @xmath94 with a standard deviation of 0.26 dex , much larger than that obtained for the other clusters . several stars are severe outliers , defined here as 0.5 dex from the @xmath61 calculated in the initial fit . our adopted fit of 176 stars excludes these severe outliers . we calculate @xmath95 with a luminosity scatter of 0.18 dex . fitting stars from 34005000 k yields the same @xmath96 . this luminosity spread of 0.18 dex is slightly larger than that calculated for the other associations , with extra scatter that may be caused by a real age spread in the cluster or , as suggested by @xcite , a @xmath97 pc depth of the region . the brown dwarf locus of upper sco at 28003200 k has a much steeper slope than the linear fit to the 540 myr old sample , which points to problems in our simplifying assumption that the slope of the 28003200 k object locus does not evolve on these timescales . the fit to the slope of combined loci of the five benchmark clusters is dominated by tuc - hor members . however , the slopes from the younger associations are steeper than for the older associations . these differences suggest that the empirical brown dwarf locus is better described by a swinging gate rather than a single slope . this same swinging gate behavior is also suggested by the theoretical isochrones covering this low temperature regime ( figure 9 ) , where the predicted slopes appear to flatten with increasing age . in particular , the brown dwarf loci is remarkably well described by the dm97 isochrones and is also consistent with the bhac-15 isochrones . while a more in - depth analysis is beyond the scope of the simple - minded spirit of this paper , these problems with the brown dwarf isochrones lead us to focus subsequent analysis of the correspondence between stellar mean stellar luminosities and stellar ages on the 32005000 k stellar locus . in this section , we apply the analytical approximations of isochrones from 3.13.2 to the set of young clusters to order them in luminosity , as a proxy for age . the fits to luminosities for 32005000 k stars ( 3.1 ) are likely more accurate than for the 28003200 k brown dwarfs ( 3.2 ) because an analysis of the brown dwarf locus relies heavily on the two older moving groups and may not accurately describe the brown dwarf locus of younger clusters . lc|ccc|ccc stellar & & & + assn & d@xmath23 & n & @xmath96 & std & n & @xmath98 & std + upper sco & 145 & 176 & -0.25 & 0.18 & 286 & -1.51 & 0.27 + @xmath1 cha & 111 & 14 & -0.28 & 0.17 & 4 & -1.42 & 0.23 + @xmath0 cha & 94 & 7 & -0.32 & 0.06 & 5 & -1.61 & 0.16 + tw hya & ( 57 ) & 22 & -0.40 & 0.17 & 3 & -1.51 & 0.22 + @xmath2 pic & ( 30 ) & 16 & -0.65 & 0.14 & 11 & -1.91 & 0.15 + tuc hor & ( 51 ) & 48 & -0.72 & 0.14 & 89 & -2.20 & 0.18 + + + + table [ tab : lumorder.tab ] shows the ordering of clusters by luminosity for the 32005000 k fits . the five benchmark associations are ordered in luminosity as expected from previous studies ( e.g. , see discussions in mamajek 2008 ; hillenbrand 2009 ; kraus et al . 2011 and fang et al . the upper sco , @xmath1 cha and @xmath0 cha associations are the youngest , followed by the tw hya association , the @xmath2 pic moving group , and the tuc - hor moving group . use of @xcite temperatures and bolometric corrections would increase association @xmath96 by 0.02 - 0.03 dex . the formal uncertainties in the luminosity ( age ) ordering are calculated from a combination of ( 1 ) the standard deviation divided by the square root of the number of stars in the sample , and ( 2 ) systematic uncertainty in distances . methodological differences in luminosity calculations and in the census of multiplicity also can affect the luminosity ordering of clusters . for the specific benchmark clusters analyzed here , the age ordering is secure , with the possible exception of the @xmath0 cha association . to swap the twa and @xmath0 cha associations would require a systematic change in @xmath99 by 0.08 dex ( 20% ) . for young clusters , the primary culprits to explain such a luminosity change of 0.1 dex would be systematic errors in the distance ( 12% change ) , extinction ( change of @xmath100 mag ) , and spectral type ( change of 140 k ) . @xcite argue that twa 9a and 9b should be excluded from the twa census because they are much fainter than expected , which would increase the cluster luminosity by 0.04 dex . these comparisons are also affected by differences in binary accounting and by accretion , if samples include disk sources . the relationship between stellar luminosity and effective temperature has been measured for six young associations . because these stars are contracting pre - main sequence stars , the radius and therefore luminosity decreases with time . the luminosity of stars at a fixed temperature is therefore used in 3.5 as a proxy for age . the conversion from luminosity to age depends on the adopted choice among the many evolutionary models , as shown in fig . 3 , and on the stellar temperature . at a fixed temperature , the age estimates differ between the many evolutionary models . mismatches between empirical and model isochronal slopes lead to different age estimates along an empirical isochrone when a given model family is applied to stars with different temperatures . figures [ fig : lumslope]-[fig : hrt2 ] and table [ tab : slopes.tab ] compare the empirical isochrones to the pre - main sequence models discussed in this paper . the model isochrones are curved but can be well approximated with a two temperature fit within the 28005000 k region . the slopes presented in table [ tab : slopes.tab ] are intended as a guide , with a measured slope that is sensitive to the temperature region defined in the fit . for low mass stars ( 32005000 k ) , the luminosity slopes in the dartmouth , pisa , bcah98 - 1.9 , bhac-2015 , and feiden pre - main sequence tracks are consistent with the observations . these models produce similar ages for stars at 3800 and 4200 k. on the other hand , the isochronal slopes predicted from the dm97 , siess , bcah98 , and parsec models differ from empirical slopes . this discrepancy leads to differences in expected luminosities versus spectral type . the dm97 , siess , and bcah98 models all have a steeper slope than is observed , which leads to a temperature ( mass ) dependence in age estimates . as an example , if we normalize our empirical isochrone and the bcah98 isochrone to a young 4200 k star , a co - eval star of 3400 k would appear @xmath101 dex more luminous ( @xmath102 dex younger in age ) than expected from the bcah98 isochrone . the parsec isochrones conversely have a shallower slope than is observed , which leads to 3400 k stars appearing older than 4200 k stars . the new bhac-2015 and feiden models have isochrones that are similar to the empirical isochrones . the inflection point in both bhac-2015 and feiden occurs at temperatures @xmath26 k hotter than in the empirical isochrones , which leads to mismatches between empirical and model isochrones for very low mass stars and brown dwarfs . these two models predict isochrones that are close enough to empirical isochrones that the differences could be explained if effective temperatures below 3500 k are systematically underestimated by 50 - 100 k. lcc|ccc|cc|c stellar & & & dm97 & pisa & bcah98 & feiden & bhac-2015 & li depl.@xmath22 + association & d ( pc ) & @xmath103 & 3400 k & 4200 k & 4200 k & 3800 k & 3800 k & age + upper sco & @xmath104 & @xmath105 & @xmath106 & @xmath107 & @xmath108 & @xmath109 & @xmath110 & + @xmath1 cha & @xmath111 & @xmath112 & @xmath113 & @xmath114 & @xmath115 & @xmath116 & @xmath116 & + @xmath0 cha & @xmath117 & @xmath118 & @xmath119 & @xmath120 & @xmath121 & @xmath122 & @xmath122 & + twa & ( 57 ) & @xmath123 & @xmath124 & @xmath125 & @xmath126 & @xmath127 & @xmath128 & + bpmg & ( 30 ) & @xmath129 & @xmath130 & @xmath131 & @xmath132 & @xmath133 & @xmath134 & @xmath131 + tuc hor & ( 51 ) & @xmath135 & @xmath136 & @xmath137 & @xmath138 & @xmath139 & @xmath140 & @xmath141 + + + + given these mismatches between model and observed isochrones , even for the latest models for cool temperatures , comparisons between clusters should be made at a constant temperature . age estimates for each model are evaluated at 3400 , 3800 , and 4200 k and are plotted in fig . [ fig : ages_hist ] for three of the benchmark associations and for the upper sco ob association . because of the discrepancies in slope described above , for most models the 3400 k stars are 0.20.5 dex younger than 3800 k and 4200 k stars . the parsec models have the reverse dependence , with the 3400 k stars older than the 4200 k stars . the brown dwarf ages are not plotted . for the now - outdated bcah98 models , ages of 3000 k stars are another 0.3 dex younger than the 3400 k stars . on the other hand , for parsec models ages of 3000 k stars are typically 0.1 - 0.3 dex older than 3400 k stars . however , the differences between the brown dwarf loci of upper sco ob association and the tuc hor moving group demonstrates that the shape and slope of the brown dwarf isochrone evolves with time . fitting a single slope to the brown dwarf isochrones without respect to age is an overly simplistic approach to characterizing the location of brown dwarfs in the hr diagram . the dm97 models reproduce the evolution in the brown dwarf isochones especially well , however the absolute ages are anomalously young . converting the luminosity proxy into a consistent and accurate tool for age measurements requires solving two problems : ( 1 ) establishing whether the ages inferred from luminosities are consistent across a temperature range , and ( 2 ) comparing the model - dependent age estimates with independent age estimates to establish an absolute and correct age scale . the empirical isochrone fits presented here lead to a self - consistent luminosity ordering for nearby young associations . absolute ages are then estimated from comparisons to isochrones for model tracks , but these comparisons lead to a wide range of age estimates for different model tracks and within the same models for stars of different temperatures . if uncorrected , these discrepancies lead to significant spectral type ( mass ) dependences in ages of young stars . the temperature dependency of the inferred ages arises from the mismatch between empirical and theoretical slopes in @xmath61 vs temperature . the source for this mismatch may be the evolutionary models , although spectral type / temperature mismatches between different methodologies and uncertainties in the spectral type / temperature conversion also likely contribute to these discrepancies . the bhac-2015 models improved the prescription for convection , while the feiden models improve upon the thermal structure in the atmosphere . both the bhac-2015 and feiden models significantly improve the match between predicted and observed stellar loci . alternately , the parsec models apply an _ ad hoc _ change to the t-@xmath142 relationship to fit the color constraints of old clusters , and similar tweaks may be able to reproduce loci of low mass young stars . ultimately , despite recent advances , some model improvements are still necessary so that the evolution of radius and temperature better reflects the observational constraints of young stellar associations . missing physics in the evolutionary models , including magnetic fields and spots , may lead to mass dependent discrepancies or absolute errors in age dating young clusters . age comparisons between clusters are self - consistent only when compared over specific temperature ranges where the model and empirical isochrones match . the bhac-2015 and feiden models currently offer the best set of pre - main sequence tracks for relative age estimates , although mismatches at @xmath143 k still lead to temperature - dependent ages . the empirical isochrones derived in this paper offer a method to place stars of different temperatures on the same age scale , albeit with some uncertainty introduced by the possible temporal evolution of the isochronal slopes . table [ tab : ages.tab ] lists sets of ages for each cluster based on a consistent application of a single model family at a single stellar temperature . the ages are selected from models that produce young , intermediate , and old ages relative to the average model . uncertainties in these ages are calculated from the statistical uncertainties in the fits , as listed in table [ tab : lumorder.tab]@xmath144 , and an estimate for the systematic uncertainty in the median distance of the @xmath0 cha , @xmath1 cha , and upper sco ob associations . the reasonableness of these isochronal ages are testable by comparison to li depletion boundary ages@xmath145 . the tuc - hor moving group has a li depletion boundary age of @xmath146 myr from the bcah98 evolutionary models and @xmath147 myr from the dm97 models @xcite , so we adopt @xmath141 myr . the @xmath2 pic moving group has li depletion boundary ages of @xmath148 myr from @xcite and @xmath149 myr from the feiden models @xcite . we adopt an li depletion boundary age of @xmath131 myr for the @xmath2 pic moving group . figure [ fig : ages_hist ] and table [ tab : ages.tab ] compares the li depletion boundary ages to the isochronal ages estimated here from the different model tracks . no single model of low mass stars reproduces the li depletion boundary age for both the tuc - hor and @xmath2 pic moving groups . most models underpredict the age of the tuc - hor moving group by @xmath150 myr ( or 0.1 dex ) . on the other hand , most models reproduce the age of the @xmath2 pic moving group . the dm97 models yield especially young ages , likely because the isochrones of low - mass stars have temperatures that are initially too high ( see description in 2.5 ) . both the feiden and bhac-2015 models have similar isochrones that yield remarkably similar ages . when applied consistently , the li depletion boundary ages for the @xmath2 pic moving group implies a @xmath151 myr old age for low mass stars in the upper sco ob association , younger than the 11 myr age measured from intermediate and high mass stars by @xcite . the 11 myr age of upper sco would require ages of @xmath152 and 55 myr for the @xmath2 pic and tuc - hor moving groups . either the intermediate mass or low mass age ( based on li depletion boundary ages of older associations ) of the upper sco ob association is incorrect , or the identification and evaluation of the low mass and/or higher mass stars is biased in some way . errors in ages may be caused by either errors in the birthline or in the contraction rates . a bulk subtraction of 5 myr to intermediate mass isochrones would shift the intermediate mass isochrones sufficiently for upper sco and would not cause any significant discrepancy with the li depletion age of the @xmath2 pic moving group . the age estimate for the @xmath2 pic moving group from late f and g stars @xcite would also not change significantly because these stars are preferentially lower mass than the intermediate / high mass stars that led to the old age estimate for the upper sco ob association . an addition of 5 myr to low - mass isochrones would also lead to consistent ages , but would encounter severe problems by forcing the youngest known clusters to have ages of 5 myr . alternately , the contraction rates for either low - mass or intermediate mass stars may be in error . for low mass stars , the inclusion of magnetic fields and spots in evolutionary models will affect contraction rates . for intermediate and high mass stars , the inclusion of rotation leads to older ages for intermediate and high - mass stars . observational errors relative to the benchmark associations may also lead to an underestimate of the age . a bulk increase in temperatures by 250 k would lead to an age of 10 myr , but such a large systematic difference in spectral typing is unlikely . alternately , the luminosities could be overestimated by 0.2 dex if the distances are biased and overestimated by 25% . assuming @xmath153 mag , which disagrees with the colors of upper sco stars , would lead to only a 0.07 dex increase in the luminosities . in this paper , we calculate empirical isochrones in hr diagrams for low mass stars and brown dwarfs in young ( @xmath154 myr ) nearby associations . the sample consists of stars that have accurate distances , low extinction , and are not affected by the presence of disks or accretion . the simple formulation of the empirical isochrone assumes that two slopes , one applied to stars with temperatures 3200/34005000 k and one for stars between 28003200/3400 k , applies to the stellar locus of all young associations . this choice leads to reasonable fits to the five young benchmark clusters and to the stellar and brown dwarf locus of the upper sco ob association . a successful theory of pre - main sequence evolution should match these empirical fits . the dartmouth , pisa , and bcah98 - 1.9 pre - main sequence isochrones between 32005500 are consistent with the measured slopes . the slopes from dm97 , bcah98 , and siess isochrones are steeper than observed , so that very low mass stars will be assessed as younger than solar mass stars . the parsec isochrones have the opposite behavior , so that solar mass stars are assessed as younger than very low - mass stars . the parsec and bcah98 models typically lead to the oldest ages , while the ages obtained from the dm97 models are anomalously young . advances in prescriptions for convection and the thermal structure of the atmosphere in the bhac-2015 and feiden models , respectively , have improved the match to empirical stellar loci at @xmath155 k and match the decrease in luminosity at @xmath143 k , though with a temperature offset relative to established spectral type - temperature conversions . applying the empirical isochrone to associations provides a statistically robust age ordering , which is independent of spectral type . our age ordering is based on the 32005000 k region . the assumption that the 28003200 k isochrones have the same slope for both the young and old associations is likely overly simplistic . steeper slopes at younger ages , especially for the upper sco ob association , instead indicates some evolution in the brown dwarf locus . because the relationship between luminosity and temperature for brown dwarfs is steep , any use of brown dwarfs in age estimates requires accurate temperatures and should be used with caution . fitting these isochrones to the six associations discussed in this paper leads to a luminosity / age ordering mostly as expected from previous results , with the upper sco ( @xmath156 , @xmath0 cha ( @xmath157 ) , and @xmath1 cha ( @xmath158 ) , associations roughly co - eval , followed by the tw hya association ( @xmath159 ) and the @xmath2 pic ( @xmath160 and tuc - hor ( @xmath161 ) moving groups . for any given association , age estimates of different models evaluated at different temperatures span 0.8 dex . as a consequence , comparisons of ages between different associations should consistently apply a single luminosity - age scaling . the bhac-2015 and feiden pre - main sequence models yield very similar ages for low - mass stars . ages can be adopted to be consistent with the li depletion boundary age of the @xmath2 pic moving group , but these models imply a @xmath151 myr age for the upper sco ob assocation for low mass stars , younger than the 11 myr age obtained from intermediate mass and high mass stars stars . we thank the referee , cameron bell , for a careful read of the manuscript , which led to substantial improvements in the robustness and clarity of results and in the self - consistency within the paper . gjh thanks adam kraus for discussions on preparing the paper and jessy jose for valuable comments on the paper . gjh is also grateful for the role of the 2014 oort workshop on episodic accretion , jointly organized by ewine van dishoeck and neal evans , in motivating this analysis , and discussions with lee hartmann and isabelle baraffe at the workshop on early ideas for this paper . we appreciate gregory feiden sharing the new dartmouth tracks prior to their publication and for discussion of those tracks and magnetic fields . gjh is supported by a youth qianren grant and general grant # 11473005 awarded by the national science foundation of china . this research has made use of the vizier catalogue access tool , cds , strasbourg , france . in 2.22.3 , we describe the collection and use of photometry to identify disk sources , to tweak bolometric corrections calculated from models , and to measure extinctions for members of the upper sco ob association . this appendix describes how the photometry is implemented in this paper . filter curves and zero points are obtained from the virtual observatory filter profile service@xmath162 . the bt - settl models were convolved with the filter transmission profiles in photon space , which may lead to minor offsets relative to integrations in energy space @xcite . the number of counts is then converted into a magnitude from the filter zero point function . synthetic photometry is then calculated from the zero - point flux for the filter and the central wavelength of the filter . the listed central wavelengths are the average wavelength of the filter , calculated by integrating the transmission over the filter bandpass in wavelength space . figure [ fig : colors ] shows temperature - color plots that compare the observed photometry to synthetic photometry from the bt - settl models and , when possible , to photometry of @xcite . in cases where the observed and synthetic colors ( always relative to j ) are mismatched , the model fluxes at the corresponding wavelength are corrected to reproduce the observed colors . the j - band flux is always kept constant , so all corrections are relative to the j - band . the bolometric corrections are then calculated from the synthetic j - band photometry and the integrated emission in the full model spectrum . finally , the absolute bolometric magnitude @xmath163 is calculated from the distance - corrected j - band magnitude , @xmath164 , and the bolometric correction , bc , @xmath165 . the zero point for @xmath163 ( @xmath166 erg @xmath167 s@xmath168 ) and the assumed @xmath169 erg @xmath167 s@xmath168 are obtained from the 1999 iau zero point scale , as discussed at the webpage of eric mamajek listing basic astronomical data for the sun@xmath170 .

we have accumulated multiwavelength ( x - ray , optical , radio ) lightcurves for the eight black hole x - ray binaries which have been observed to enter a supposed ` soft x - ray transient ' outburst , but remained in the low / hard state throughout the outburst . comparison of the lightcurve morphologies , spectral behaviour , properties of the quasi - periodic oscillations and the radio jet provides the first study of such objects as a sub - class of x - ray transients . however rather than assuming that these hard state x - ray transients are different from the ` canonical ' soft x - ray transient , we prefer to consider the possibility that new analysis of both soft and hard state x - ray transients in a spectral context will provide a model capable of explaining the outburst mechanisms of ( almost ) all black hole x - ray binaries .

astro-ph0311152

with the list of known soft x - ray transients ( sxt ) growing longer every year and now totalling @xmath0 it is inevitable that we should find a few sources which do not follow the classical behaviour that we expect . in particular , there are a number of supposed black hole sxts which do not show a soft component in their x - ray spectra . instead they display power - law emission across the x - ray energy spectrum , despite some sources reaching high luminosities . models for sxt outbursts predict the presence of a thin accretion disc to explain the soft black - body spectrum emitted by the lower energy x - rays ( e.g. esin et al . 1997 ) . the power - law emission at higher energies is thought to be produced via reprocessing in some form of hot plasma ( e.g. corona or advection dominated accretion flow ) , which surrounds the compact object and upscatters disc and/or stellar photons to higher energies via the inverse compton emission mechanism . this soft spectrum has indeed been observed in the majority of soft x - ray transient outbursts . some sources , however , show power - law emission down to low x - ray energies and this has been the case for the sources considered in this paper ; despite entering outburst there has been no evidence for a soft blackbody component at x - ray wavelengths . in these objects it is assumed that all x - ray emission is produced via reprocessing in the ` corona ' and not in the accretion disc . the power - law x - ray energy spectrum is typical of black hole candidates in the low / hard state , most notably cyg x-1 which spends most of the time in this state ( brocksopp et al . corresponding x - ray power spectra for these sources show a high level ( relative to the other spectral states ) of low - frequency noise , a broken power - law and at least one quasi - periodic oscillation . ( see e.g. van der klis ( 1995 ) for further definition of spectral states . ) the state is also characterised by the presence of a low - intensity , but extremely powerful , radio jet which emits synchrotron radiation at radio and often higher frequencies ( fender 2001 ; corbel & fender 2002 ) . during the high / soft state we see the presence of the soft black - body component in the x - ray spectrum and the quenching of the radio jets , although rapid radio - emitting ejections appear to accompany changes in x - ray spectral hardness ( e.g. brocksopp et al . 2002 ) it is this behaviour that we expect in the classic black hole sxt events . in this paper we take a closer look at eight black hole ` sxts ' which , on the basis of their observed x - ray and radio properties , have been reported to have spent the whole of their outburst in the low / hard state . although observing coverage was incomplete for the less recent outbursts , each source ( other than 4u 1543@xmath1475 ) was observed at low x - ray energies for the whole outburst or at least a month after the initial rise , by which time the decay had started and no soft disc component had been discovered ; while it is common for sxts to spend some time in the low / hard state prior to softening , this period is generally only a few days ( the longest initial low / hard state period observed to date is that of xte j1650@xmath1500 which remained hard for @xmath2 month before softening but also continued to rise in flux ; wijnands priv . comm . ) . thus it is unlikely that any of the sources made an unobserved transition to a softer state . the sources are listed in table [ sources ] along with the date of the low / hard state outburst , the observations which took place and the sources from which we obtained data - points ; the observing coverage at the various different frequencies can be seen in the lightcurves plotted in fig . [ lightcurve1 ] . a brief summary of the properties of each source , in chronological order of low / hard state outburst , is given below . v404 cyg is probably one of the best - known x - ray transient sources on account of its high luminosity and levels of variability ( at various wavelengths ) during both outburst and quiescence ( e.g. tanaka & lewin 1995 , hynes et al . 2002 ) and many observations exist ( e.g. han & hjellming 1992 , shahbaz et al . 1994 ) . despite the high luminosity of the outburst , the lack of soft component present in the x - ray spectrum suggests that it was indeed a low / hard state transient ( e.g. oosterbroek et al . 1997 ) we note that this has been disputed and that it may have spent a very short period in the soft state ( ycki et al . 1999 ) but include it here for completeness . a1524@xmath162 has been observed in outburst twice@xmath3 , the first of which was a ` normal ' sxt event ( kaluzienski et al . the second outburst was observed in the hard x - ray regime by sigma ( barret et al . 1992 ) and by chance was also observed at soft x - rays during the _ rosat _ all sky survey ten days before the sigma observation . the soft x - ray spectrum could be equally well fit by either a cool black body or a power - law ; the presence of an ultra - soft component could not be confirmed or ruled out ( barret et al . we note , however , that the outburst was insufficient to trigger the soft x - ray all sky monitors on - board _ watch _ and _ ginga _ , the latter providing an upper limit of 50 mcrab which was consistent with the rosat detection ( barret et al . thus we include this source as a possible hard state candidate . 4u 1543@xmath1475 is another recurrent transient which has been observed in outburst four times and was a classical sxt during the first two of these , as well as the fourth and current outburst ( matilsky et al . 1972 , kitamoto et al . 1984 , miller & remillard 2002 ) . during the third event hard x - ray observations revealed a power - law spectrum ( harmon et al . 1992 ) but since there were no soft x - ray observations we can not confirm the presence or otherwise of a soft component at lower frequencies . we include this source as a possible hard state candidate . gro j0422 + 32 was first observed in 1992 when the x - ray source entered a ` canonical ' fast rise , exponential decay type outburst ( paciesas et al . the energy spectrum was a hard power - law ( sunyaev et al . 1994 ) and timing analysis revealed properties consistent with the low / hard state , although the expected anti - correlation between break frequency and power at the break was not present ( van der hooft et al . the radio counterpart and the evolution of its spectrum throughout the event are discussed in shrader et al . the optical counterpart was discovered by castro - tirado et al . ( 1993 ) and the intense optical monitoring which followed resulted in the detection of two additional and unusual ` mini - outbursts ' ( e.g. callanan et al . 1995 ) . gro j1719@xmath124 was detected simultaneously by batse and sigma and was notable for the stability of the emission at hard x - ray wavelengths ( harmon et al . 1993 , ballet et al . the low / hard state nature was confirmed through study of the x - ray energy spectra and found to be comparable with that of cyg x-1 in the low / hard state ( revnivtsev et al . 1998 ) ; the power spectra showed the presence of a qpo which varied in frequency ( van der hooft et al . the radio and optical counterparts were discovered by della valle et al . ( 1994 ) . grs 1737@xmath131 was discovered by sigma ( trudolyubov et al . 1999 ) and was also observed by @xmath4/pca , which confirmed the presence of the hard power - law spectrum at lower energies ( cui et al . further x - ray observations were made by _ bepposax _ and _ asca _ ( heise 1997 , ueda et al . 1997 ) which improved the positional accuracy and confirmed that the hard spectrum was still present ( at @xmath5 220 kev ) @xmath5 2 weeks after the initial detection ; there were no observations at optical or radio wavelengths . gs 1354@xmath164 is another recurrent transient and it has been observed during four different outbursts@xmath3 ( brocksopp et al . assuming that all events were indeed from the same source , this was first ever sxt outburst observed ( then named cen x-2 ) and was also one of brightest ( e.g. tanaka & lewin 1995 , and references therein ) . the first and third events were characterised by very soft x - ray spectra ; the second and fourth were hard events ( brocksopp et al . 2001 and references therein ) on account of the observing coverage we consider just the most recent event in this paper , during which a hard power - law energy spectrum was observed throughout by _ rxte _ ( revnivtsev et al . 2000 ) . xte j1118 + 480 was discovered in 2000 during an outburst ( remillard et al . 2000 ) ; the presence of the low / hard state was determined from study of the hard power - law x - ray energy spectrum and the characteristic power - law , high rms noise and qpo in the x - ray power spectrum ( revnivtsev et al . 2000a ) . the optical and radio counterparts were also discovered during the outburst ( uemura et al . 2000 , pooley 2001 , fender et al . lightcurves have been constructed for each of the eight sources that has exhibited a low / hard state outburst . the data points have been obtained from the literature and the _ rxte_/asm and _ cgro_/batse web - sites ( see table [ sources ] ) . we note that in a number of cases , datasets have been measured using the ads software dexter ; since there is a certain amount of error involved in making these measurements we recommend that the points shown here are used as a guide and the original papers consulted . for the same reason ( and also due to lack of availability ) a number of our plots do not show error bars again these can be seen in some of the original papers . although all data displayed here have been published previously , it has been uncommon for all wavebands to be analysed in conjunction with each other ( for combined x - ray / optical / radio analyses see e.g. brocksopp et al . 2001 , han & hjellming 1992 , mcclintock et al . 2001 ) . therefore for each source we have attempted to find data at x - ray ( in more than one energy band ) , optical and radio wavelengths , although this was not always possible . fig . [ lightcurve1 ] shows x - ray lightcurves for a1524@xmath162 , 4u 1524@xmath1475 and grs 1737@xmath131 and x - ray , optical and radio lightcurves for v404 cyg , gro j1719@xmath124 and gro j0422 + 32 . [ lightcurve2 ] ( contd . ) shows x - ray , optical and radio lightcurves for gs 1354@xmath164 and xte j1118 + 480 . the x - ray intensity values have all been converted into crab units according to the approximate conversions presented in table [ conversions ] . optical and radio points are shown in milli - janskys . it is immediately apparent that each source shows very different types of behaviour . the fast - rise exponential - decay ( fred ) lightcurve that is considered the ` classic ' signature of a soft x - ray transient ( see e.g. chen et al . 1997 ) is not predominant in a number of these lightcurves ; only in v404 cyg is the lightcurve morphology of both x - ray and optical emission ` canonical ' , although this being true also for a1524@xmath162 , 4u 1543@xmath1475 and grs 1737@xmath131 can not be ruled out due to lack of data . the x - ray , but not the optical , lightcurve of gro j0422 + 32 , shows a fred , as does the ` precursor ' x - ray lightcurve of xte j1118 + 480 ; the second and more significant outburst of xte j1118 + 480 ( x - ray and optical ) shows a ` plateau ' lightcurve morphology , as did the initial maximum of gro j1719@xmath124 . continuing to use the nomenclature of chen et al . ( 1997 ) , the x - ray lightcurves of gs 1354@xmath164 , the subsequent maxima of gro j1719@xmath124 and possibly a1524@xmath162 , 4u 1543@xmath1475 and grs 1737@xmath131 show ` triangular ' morphologies . the optical lightcurves of gro j0422 + 32 and gs 1354@xmath164 are indeterminate although possibly some variation on a fred shape . the radio lightcurves , although sometimes lacking in observing coverage , appear to display typical jet ejection lightcurves as discussed in fender ( 2001 ) and the references listed in table [ sources ] . each x - ray outburst , except gs 1354@xmath164 ( and a1524@xmath162 ? ) , is associated with at least one additional secondary outburst either during the decay ( i.e. v404 cyg , gro j0422 + 32 ) or following return to quiescence by a few tens to a few hundreds of days ( gro j1719@xmath124 , xte j1118 + 480 , grs 1737@xmath131 ) some of these secondary maxima have been classified as a ` glitch ' , ` bump ' or ` mini - outburst ' by chen et al . the main outburst of gro j1719@xmath124 was followed by 45 smaller peaks , reminiscent of the 1994 outburst of gro j1655@xmath140 ( e.g. harmon et al . 1995 ) , although the latter was much brighter and showed more dramatic variability . the optical lightcurves are not as well - correlated with the x - rays as we might expect considering that the optical light is thought to be due to the reprocessing of x - rays , although the general trends are often similar . it is not clear whether or not the secondary maximum seen in the v404 cyg optical lightcurve occurs also in the x - ray and the few points available suggest that the optical maximum of gro j1719@xmath124 was short - lived , unlike the x - ray although they appeared to rise ( quasi- ) simultaneously . the optical peak of gro j0422 + 32 may well have taken place prior to the x - ray peak and then decayed much more slowly than the x - rays ; following the x - ray return to quiescence , the optical flux density dropped more rapidly before increasing @xmath6 days later in the first of two additional peaks , unaccompanied by _ any _ x - ray behaviour . the partial optical lightcurve of gs 1354@xmath164 showed a decay while the x - rays were still rising , a peak plus its decay and then a further increase ( this is discussed in detail in brocksopp et al . conversely , the optical lightcurve of xte j1118 + 480 does seem to be reasonably well - correlated with the x - rays . finally , where coverage is available , the main outburst and the subsequent secondary maxima of the x - ray lightcurves tend to be associated with radio ejections , although this is _ not _ to say that the lightcurves correlate , even when taking time - lags into account ; they do not correlate and this would not necessarily be expected ( kuulkers et al . 1999 ) . in the case of xte j1118 + 480 the maximum radio intensity was maintained during the extended x - ray peak , reminiscent of the low / hard state jet of cyg x-1 and gx 339@xmath14 ( e.g. fender 2001 ) . the most notable feature of these lightcurves is their inconsistency particularly considering the similarities of their x - ray ( and broad - band , see section 5 ) spectral behaviour . if we had considered just the multiwavelength lightcurves then it is unlikely that these sources would have been labelled a ` class of object ' distinct from the genuinely soft x - ray transients . these plots reveal a number of questions yet to be answered by current models . why are some of the outbursts fred - type at one wavelength but not at others ? why are such different lightcurves produced in such ( x - ray ) spectrally similar sources ? why do the secondary outbursts take place on such variable timescales ? why are some of the outbursts ( un)correlated with other wavebands ? why are the durations of the peaks so variable ? most of these questions remain unanswered because , in applying the disc instability model ( or other models ) to x - ray transients , we have failed to consider the full multiwavelength aspect . the dynamical effect of jets ( such as removal of power and angular momentum the jets can potentially account for up to 90% of the accretion power ; fender 2001 ) on the models has often not been accounted for ; this is particularly important since the jet is thought to emit in the infrared and perhaps optical ( and x - ray ? ) regimes as well as the radio ( see section 5 ) . it has further been assumed that ` canonical ' x - ray transients follow the fred morphology whereas this is frequently not the case , particularly in the x - ray lightcurves ; even in the genuinely soft sources a precursor low / hard state phase is seen prior to the fast rise in all cases where data is available ( brocksopp et al . we investigate alternatives to these previous asssumptions in section 6 . it might be expected that x - ray transients would soften as they increase their flux during an outburst and then harden again during the decay similar to the spectrum of cyg x-1 during its well - known state - change of 1996 ( e.g. zhang et al . 1997 ) . therefore it was somewhat surprising to see that the sxt xte j1859 + 226 was very hard at the beginning of the outburst ( peaking first in the hard x - rays ) and gradually softened throughout the decay ( brocksopp et al . 2002 ) . superimposed upon this trend were small - scale temporary hardenings which coincided with radio jet ejections . likewise xte j1550@xmath1564 also showed a general softening as the luminosity decayed following the 1998 outburst ( wu et al . 2002 ) , although interestingly the 2000 outburst hardened during decay ( soria et al . thus a correlation between luminosity and spectral softness is not necessarily observed in all cases . while ideally we would have studied the variability of the xte / asm hardness ratio data for the low / hard state sources , this has proved unreliable due to the poor signal - to - noise ratio ; xte j1118 + 480 showed a possible softening at the onset of the outburst but this was non - conclusive however gro j1719@xmath124 did indeed soften as the flux rose above quiescence ( revnivtsev et al . revnivtsev et al . ( 2000b ) found a general softening in the spectra of gs 1354@xmath164 over the course of the outburst , with a sudden hardening in the final observation ( during the decay but prior to quiescence ) which coincided with a possible increase in flux . the ttm hardness ratio of gro j0422 + 32 remained approximately constant ( sunyaev et al . 1994 ) but we note that none of these observations took place during the rise to outburst ( which is where any significant spectral changes would be expected to take place ) . the spectral behaviour of v404 cyg is more complicated than the other sources and further discussion is beyond the scope of this paper instead we refer the reader to oosterbroek et al . ( 1997 ) and ycki et al . the presence of a quasi - periodic oscillation ( qpo ) in the x - ray power spectrum of an x - ray binary is a common feature of the low / hard state , as well as the intermediate and very high states ( e.g. wijnands & van der klis 1999 and references therein ) however , the frequencies at which they occur seem to be noticeably lower in the low / hard state . this has been the case in v404 cyg , gro j1719@xmath124 , gro j0422 + 32 , gs 1354@xmath164 and xte j1118 + 480 , all of which show a qpo at mhz frequencies ( see oosterbroek et al . 1997 , van der hooft 1996 , 1999 , brocksopp et al . 2001 and wood et al . 2000 respectively ) furthermore in each case other than v404 cyg ( for which a qpo at @xmath7 hz was detected during the brightest observation only ) the frequency of the qpo does not remain constant . instead it increases during the outburst , showing little or no apparent correlation with the x - ray intensity ; the presumed decrease in frequency following the return to quiescence has not been observed to date for any of the low / hard state sources . plots showing the frequency variability of the qpo in each source are shown in fig . 2 . we note that in the case of gro j0422 + 32 there were two different qpos detected , at approximately 40 and 200 mhz ; it is the latter which shows the variability . it is also the only source for which there is even the hint of a correlation between the qpo frequency and the x - ray flux however , cross - correlating the qpo data with the x - ray lightcurve does not confirm the presence of any correlation in _ any _ of the sources . furthermore we find that the duration of the qpo frequency increase is a factor of at least 1.52 times the duration of the x - ray flux rise ( with the possible exception of xte j1118 + 480 for which the rise times may be comparable , although there is insufficient data to be certain ) . the qpo frequencies of gs 1354@xmath164 are notably lower ( by a factor of @xmath8 ) than those of the other sources ; we note that the outburst of gs 1354@xmath164 was noticeably less dramatic that the others , although it is not at all clear whether this is related to the lower frequency of the qpo . it is also possible that , as in the case of gro j0422 + 32 , a second qpo was present ( but undetected ) in gro j1719@xmath124 , gs 1354@xmath164 and xte j1118 + 480 which would explain the difference between the frequencies . the remaining three sources have not been reported to display qpos . in the case of a1524@xmath162 and 4u 1543@xmath1475 this is because there is no published power spectrum . in the case of grs 1737@xmath131 a power spectrum has been computed but the data quality insufficient to determine whether or not a qpo was present ( cui et al . 1997 , cui 2001 priv . comm . ) . it has been suggested that the mhz qpo of x - ray transients may relate to the inner edge of the accretion disc ( e.g. revnivtsev et al . 2000b ) thus if the softening during the outburst is due to the inner edge of the disc moving inwards we would expect an increase of the qpo frequency , as indeed we see ( revnivtsev et al . this scenario seems feasible , especially since during genuinely soft outbursts the sources tend to display higher frequency ( hz - khz ) qpos ( e.g. cui et al . 2000 ) . however , we would also expect the inner edge of the disc to move outwards again at the return to quiescence , thus decreasing the qpo frequency back to its `` quiescent value '' ; while this behaviour has not been observed in any of these sources ( although hinted at in gro j0422 + 32 ) it has been observed during the low / hard state phases of the 2000 outburst of xte j1550@xmath1564 ( reilly et al . 2001 ) . an alternative origin for this qpo is presented in wood et al . they suggest that it is produced at a large radius within the disc and that its frequency is inversely proportional to the disc mass . while a decrease in qpo frequency after the outburst is still expected , their model predicts this to take place _ after _ the return to quiescence ; measurements of the qpo frequency have not yet taken place for long enough after the peak ( for any source ) for this to have been observed . however , if this model is correct then we might also expect a similar relationship between the disc mass and qpo frequency in canonical soft x - ray transients . in the case of these objects the qpo frequency is higher and so an even greater reduction in disc mass during the outburst would be required ; this would depend on the relative amounts of mass transfer and mass accretion taking place as the inner edge of the accretion disc moves inwards to the last stable orbit ( as is predicted to take place during the transition to softer states e.g. esin et al . evidence for the ubiquity of radio jets in most classes of x - ray binaries is mounting constantly . as well as probably all black hole candidates in the low / hard state and during outburst , the z - sources ( whilst on the horizontal branch ) and the transient neutron star systems ( whilst in outburst ) also show evidence for some form of jet behaviour ( e.g. fender 2001 , fender & hendry 2000 ) . however , it is also apparent that the jet is not purely a ` radio jet ' the synchrotron spectrum that is thought to be the jet signature is frequently seen up to higher frequencies in the infrared and possibly the optical . in the case of xte j1118 + 480 markoff et al . ( 2001 ) were able to fit a synchrotron spectrum to the full range of frequencies up to the hard x - rays ( with an additional black - body disc component in the optical ) . more recently a similar model has been successfully applied to thirteen broad - band spectra of differing luminosities observed from gx 339@xmath14 during the low / hard state . the jet model also analytically predicts the slope of the radio / x - ray correlations seen in this source ( markoff et al . 2003 , corbel et al . if the power - law x - ray emission is indeed emitted by the jet then these jets are considerably more powerful and use a much greater fraction of the accretion energy than most current models allow . a flat or inverted synchrotron spectrum has already been well - established at radio frequencies in each of the low / hard state sources studied here ( where radio observations were available , fender 2001 ) . we have therefore compiled the full spectrum from radio to x - rays in order to determine whether the synchrotron spectrum of xte j1118 + 480 is an exception or the rule . the resulting broad - band spectra for v404 cyg , gro j1719@xmath124 , gro j0422 + 32 and gs 1354@xmath164 are shown in fig . [ spectrum ] for three different epochs per source . the epochs were chosen so as to make optimum use of simultaneous observations and also to sample a range of different types of activity for example , the peak of the initial outburst , a secondary outburst and a period of low luminosity ( this was not possible for gs 1354@xmath164 for which the three epochs were the rise , peak and decay ) . all optical points have been de - reddened according to the extinction values quoted in liu et al . ( 2001 ) and the interstellar extinction curve of whittet ( 1992 ) . initial inspection of the resultant broad - band spectra suggests that despite the very different lightcurve properties of each outburst , their spectral properties were similar . it is also notable that the spectra at different epochs do not vary considerably . the flat / inverted radio behaviour is clearly seen for each source , although some epochs are contaminated by optically thin ejections ( e.g. gro j1719@xmath124 : jd 2449777 ) , but as is so often the case ( fender 2001 ) it is not possible to determine the high energy cut - off to the synchrotron spectrum . while the frequency coverage is not good enough to confirm that the synchrotron spectrum extends to the x - ray regions , neither can we state categorically that it is impossible ; the spectra of these four sources here resemble that of xte j1118 + 480 and formal fitting of these broad - band spectra with a synchrotron spectrum is in progress . it is important to note that while much of this work is speculative on account of insufficient observations in the millimetre regions of the spectrum , it is not always possible to fit a disc spectrum to the optical emission . the most notable example is xte j1118 + 480 which requires an additional flat - spectrum component , possibly synchrotron ( hynes et al . 2000 ) . likewise gro j0422 + 32 was poorly fit by a disc spectrum and was _ better _ fit by a self - absorbed synchrotron spectrum ( hynes & haswell 1999 ) . although in this latter case the fit was determined when the source was close to quiescence , as we noted above this need not have much effect since the spectrum was approximately constant throughout the outburst period ( fig . [ spectrum ] ) . we acknowledge that an adaf is also expected to emit optical and uv synchrotron emission ( e.g. esin et al . 1997 ) but it is remarkable that the flux density of both jet and adaf should be apparently so similar . there must be a very close relationship between the two components , if indeed they co - exist ( e.g. meier 2001 ) . in support of this we note that in the case of gs 1354@xmath164 , xte j1118 + 480 and v404 cyg the optical : soft x - ray luminosity ratio was surprisingly low ( @xmath9 compared with an average of @xmath10 ; van paradijs & mcclintock 1995 ) relative to that of soft x - ray transients , thus suggesting that x - ray processing was not the only ( or even the dominant ) source of optical emission , particularly as some sources peak in the optical _ before _ the x - ray ( brocksopp et al . 1999 , chen et al . inspection of the broad - band spectra would suggest that the optical : soft x - ray luminosity ratio was probably also low in gro j1719@xmath124 and gro j0422 + 32 . while it is possible that the excess optical emission could be due to the companion star and/or accretion stream impact point , again it is a remarkable coincidence that the radio and optical flux densities of each source are so comparable ( fender 2001 ) , particularly in those sources where there is also some degree of correlation between the optical and radio lightcurves . however , there are further features which may indicate that the broad - band spectrum is not always dominated by the jet , particularly as the optical and radio lightcurves are not necessarily well - correlated . brocksopp et al . ( 2001 ) show that in the case of gs 1354@xmath164 the various peaks of the optical lightcurve implied that the increased mass flow ( and/or instability ) which caused the outburst was observed directly as it crossed the optical - emitting regions of the disc , suggesting that the optical luminosity excess ( relative to the x - ray ) is not necessarily synchrotron emission . furthermore the qpo discovered in xte j1118 + 480 at optical frequencies was found to be correlated with the x - ray data ( kanbach et al . 2001 ) but with an optical dip preceding the x - ray peak that is not well - explained by reprocessing of x - rays ( spruit & kanbach 2002 ) ; instead it was suggested that a relatively slow , dense outflow from the inner regions of the accretion disc was present , producing optical cyclosynchrotron emission . if we are to explain the broad - band spectra in terms of a dominating synchrotron jet then further investigation of the nature of the qpo would be necessary , particularly at optical wavelengths . the aim of this paper is not necessarily to ` drive a wedge ' between the ` canonical ' black hole sxts and this sub - class of low / hard state x - ray transients ( lhxt ; we note that the term ` hard x - ray transient ' has been applied previously to the be + neutron star binary systems ) . while we can see that the sources studied here are not classical fred sources , neither are a large number of sxts , on account of both lightcurve morphology ( chen et al . 1997 ) and x - ray spectral state ( e.g. brocksopp et al . 2002 ) ; thus it is becoming timely for a re - definition of the ` canonical ' x - ray transient which incorporates the now numerous radio jet and spectral ( x - ray and broad - band ) observations . models for sxt outbursts have been largely based on the disc instability model ( dim ) , initially developed for the outbursts of dwarf novae but later adapted for sxts ( e.g. lasota 2001 and references therein ) . the model s success has been due to its ability to produce fred - shaped lightcurves , thus having the potential to describe observations of x - ray transient outbursts ( e.g. king & ritter 1998 ; dubus et al . the original model has been adapted to include additional physical mechanisms of relevance in x - ray binaries ; by invoking disc irradiation and truncation of the inner disc , the basic rise , decay and recurrence times can be reproduced ( lasota 2001 ) . there are a number of problems with the dim , particularly as a large number ( @xmath11 ; chen et al . 1997 ) of sxt outbursts do not display ` canonical ' fred behaviour . these problems are well - known ( e.g. chen et al . 1997 ) and it is thought that they may be rectified with the inclusion of accretion disc coronae and/or warps ( lasota 2001 ) . additionally there are difficulties in describing the quiescent and low / hard states without the inclusion of an adaf or coronal outflow ( e.g. esin et al . 1997 , merloni & fabian 2002 ) . however it is also of vital importance that the broad - band , and particularly the x - ray spectra are taken into account , especially if a truncated inner disc is required in order to model the lightcurves . in particular the dim assumes that the x - ray outburst takes place in the disc and that the optical emission is produced via reprocessing of these soft x - ray photons ( e.g. lasota 2001 ) but in the case of the outbursts which remained in the low / hard state all x - ray photons were produced in the corona ; the dim does not consider the production of this power - law x - ray emission . more recently the outburst of xte j1118 + 108 was modelled in terms of a diffusion model instead of the dim ( wood et al . 2001 ) . rather than assuming that an outburst is caused by a build - up of mass in the disc during quiescence which then becomes unstable ( as in the dim ) , this diffusion model is based on the assumption that sporadic / variable mass transfer from the companion takes place . while a fred morphology for the first of the two x - ray peaks of xte j1118 + 108 was generated by this model , as well as the non - fred second peak , it should also be noted that this model does not yet include a trigger and as such is currently incomplete . the most desirable model ( or perhaps combination of models ) would not only explain all sxt and lhxt outbursts , but would also be consistent with models for the state changes of persistent x - ray binaries such as cyg x-1 and lmc x-3 ( including the ` failed state changes ' of cyg x-1 during which the source remains in the low / hard state brocksopp et al . 1999 ) . these are all supposedly the result of variable mass transfer and/or accretion ; could they all be produced by the same mechanism as x - ray transient outbursts ? in particular , wilms et al . ( 2001 ) showed that the long modulation of lmc x-3 was produced , not by the assumed precession of the accretion disc , but by quasi - periodic changes in the mass transfer rate . likewise fender et al . ( 1999 ) suggest variable mass transfer for outbursts in cyg x-3 . we note also that , in the be + neutron star binary systems , outbursts occur at periastron of an eccentric orbit ( e.g. van paradijs & mcclintock 1995 and references therein ) again suggestive of some outbursts being the result of variable mass transfer rather than instabilities within the disc . it has also been suggested that the ` superoutbursts ' seen in the su uma class of cataclysmic variable stars are the result of a ` normal ' disc instability outburst triggering additional mass transfer from the companion , perhaps due to increased irradiation ; the dim alone can not provide sufficient power for an outburst of such magnitude and duration ( frank , king , raine 1992 ) . a similar mechanism has been suggested for xte j1118 + 480 ( kuulkers 2001 ) . however this does not explain the outburst of gro j1719@xmath124 in this case the ` superoutburst ' _ preceded _ the ` normal ' outburst . if irradiation is not the sole cause of the increased mass transfer then an alternative mechanism may be necessary ; models of an accretion disc wind instability limit cycle or a variable roche lobe filling factor have been suggested for lmc x-3 ( wilms et al . 2001 , wu et al . 2001 ) and should be considered as potential mechanisms in other sources . the suggestion that variable mass transfer from the companion might be important for x - ray transient outbursts is not new ; e.g. hameury et al . ( 1986 ) considered the possibility of unstable mass transfer caused by illumination of the stellar companion . this was later discounted in favour of the dim ( e.g. mineshige & wheeler 1989 , gontikakis & hameury 1993 ) , the main problem being that the mass transfer instability is insufficient to produce the high levels of variability observed . this problem may be avoided if the broad - band spectrum is considered . if , as appears to be the case in the low / hard state of ( at least ) xte j1118 + 480 and gx 339@xmath14 ( markoff et al . 2001 , 2003 ; corbel & fender 2002 ) , a significant proportion of the x - ray luminosity is produced via the synchrotron mechanism then acceleration within the jet is likely to be its source , thus reducing the magnitude of the required mass transfer variability . while this is particularly important in the case of the low / hard state sources , the power in the jet ejections also has significant implications on determination of the mass accretion rate in soft events . furthermore sxt outbursts of xte j1859 + 226 , xte j1550@xmath1564 and many others have been observed to pass through the low / hard state prior to softening ; they have _ not _ made a direct transition from quiescence to the high / soft ( or very high ) state . all recent black hole sxt outbursts have shown this behaviour and those less recent sources which have not appeared to exhibit this behaviour have not actually been observed in sufficient detail to be certain either way ( brocksopp et al . 2002 ) . there has been no attempt to fit the dim to any source with the inclusion of this initial hard state and it is important that the power requirements of the initial hard state ( and its associated jet ) are incorporated into models . with the low / hard state shown to be such a ubiquitous behaviour of black hole x - ray binaries it is extremely important that these lhxt ` mini - outbursts ' and the initial hard state prior to a black hole sxt outburst are not forgotten about . does the same mechanism produce both the off lhs and lhs vhs transitions ( i.e. lhs = low / hard state and vhs = very high state ) ? can we assume that the off lhs transition is produced by the same mechanism regardless of whether the outburst is a black hole sxt or lhxt ? ( we note that it is not yet certain whether or not the ` off ' and ` low / hard ' states are actually two distinct spectral states . ) what conditions are present to cause a source to then make the lhs vhs transition as well ? these questions are particularly interesting for a source such as gs 1354@xmath164 which has displayed both types of outburst . in summary , while we do not intend this paper to necessarily favour one model over others , the low / hard state outbursts studied in this paper do not appear to be explained in terms of the dim as it stands currently ; either the dim needs significant modification to include the properties of the jet and broad - band spectrum , or some new model is required , in which case a sporadic mass transfer model ( such as the diffusion model of wood et al . 2001 ) may be a viable alternative , subject to further testing and inclusion of the jet and outburst trigger . furthermore , given the ubiquity of low / hard state behaviour in black hole x - ray binaries , it is possible that some form of variable mass transfer model could apply to black hole sxt outbusts as well , at least in providing the off lhs transition at the onset of the outburst . if this is the case then we need to determine the conditions and mechanism ( disc instability ? ) by which a black hole sxt then proceeds to the very high state . the points raised in the previous sections can be well - illustrated by using xte j1550@xmath1564 as a case - study . it is particularly useful since its observing coverage has been excellent at all frequencies and all behaviours studied in this paper have been observed in this single source . xte j1550@xmath1564 was discovered when it entered a sxt outburst in 1998 , passing through all of the known x - ray spectral states over the course of this outburst , including an initial low / hard state phase ( e.g. homan et al . 2000 ) . the outburst has been studied at x - ray radio wavelengths and a mass transfer instability invoked as the trigger of the extremely hard `` spike '' , which took place at the onset of the outburst ( wu et al . 2002 ) . radio observations revealed a jet with apparent superluminal motion , as well as the ubiquitous synchrotron spectrum ( hannikainen et al . 2001 ) . more recent observations have shown that more than two years after the event the ejecta were still resolvable at radio _ and at x - ray _ wavelengths and with a broad - band synchrotron spectrum ( corbel et al . 2002 ) . furthermore there were additional outbursts in 2000 , 2001 and 2002 ; an initial low / hard state was observed in 2000 before the source softened and the two more recent outbursts remained in the low / hard state throughout ( belloni et al . radio observations in 2002 confirmed the presence of a typical low / hard state jet spectrum ( corbel et al . 2002 ) . the qpo behaviour during the 2000 outburst was particularly well - observed ; both rise and fall of qpo frequency were seen during the lhs phases of the outburst . the frequency reached a plateau briefly during the lhs vhs transition before disappearing for most of the vhs and reappearing again during the transition back to the lhs ( reilly et al 2001 ) . thus study of this one source suggests ( confirms ? ) that ( i ) there are problems in explaining the power - law x - ray emission with conventional outburst models , ( ii ) the proposed broad - band jet spectrum can be imaged and ( iii ) a single source can enter outbursts of different spectral properties . we have accumulated multiwavelength datasets for the eight black hole x - ray binaries which have been observed to enter an outburst but remained in the low / hard state throughout . we show that , despite very similar x - ray spectral properties , the lightcurve morphologies of these sources are very different , as are the relationships between the x - ray , optical and radio lightcurves . however , the qpo and jet properties are comparable in each system . this comparison of the low / hard state transients suggests that some mechanism to explain the outbursts in the context of their spectral properties is necessary ; the disc instability model currently lacks the flexibility to do this and needs modification in order to allow for the the observed jet . we also suggest that in light of the more recent observations of both low / hard state and soft x - ray transients ( e.g. xte j1550@xmath1564 ) , the various models of sporadic mass transfer should not be ruled out , at least during the low / hard state phase of the outburst testing of these suggestions is work in progress . we are extremely grateful to the many people who have contributed the various flux conversions and data for this paper particularly guy pooley , erik kuulkers , mike mccollough and jean int zand . rxte_/asm data used in this paper was obtained from the public asm database ; the _ cgro_/batse data from the web pages of the compton science support center . the green bank interferometer was operated by the national radio astronomy observatory for the u.s . naval observatory and the naval research laboratory during the time - period of these observations . finally , this research has made much use of nasa s astrophysics data system abstract service and its dexter software . cb acknowledges a pparc grant and starlink facilities at jmu . a portion of this work was performed while rmb was supported by a national research council research associateship at the naval research laboratory . ballet j. , denis m. , gilfanov m. , sunyaev r. , 1993 , iauc 5784 barret d. et al . , 1992 , apj , 392 , l19 barret d. , motch c. , pietsch w. , voges w. , 1995 , a&a , 296 , 459 belloni t. , columbo a.p . , homan j. , campana s. , van der klis m. , 2002 , a&a , 390 , 199 brocksopp c. , fender r.p . , lariononv v. , lyuty v.m . , tarasov a.e . , pooley g.g . , paciesas w.s . , roche p. , 1999 , mnras , 309 , 1063 brocksopp c. , jonker p.j . , fender r.p . , groot p.g . , van der klis m. , tingay s.j . , 2001 , mnras , 323 , 517 brocksopp c. et al . , 2002 , mnras , 331 , 765 callanan p.j . et al . , 1995 , apj , 441 , 786 casares j. , charles p.a . , jones d.h.p . , rutten r.g.m , callanan p.j . , 1991 , mnras 250 , 712 castro - tirado a.j . , pavlenko e.p . , shlyapnikov a.a . , brandt s. , lund n. , 1993 , a&a , 276 , l37 ortiz , j. l. chen w. , shrader c.r . , livio m. , 1997 , apj , 491 , 312 corbel s. et al . , 2001 , apj , 554 , 43 corbel s. , fender r.p . , 2002 , apj , 573 , 35 corbel s. , fender r.p . , tzioumis a. , tomsick j.a . , orosz j.a . , miller j.m . , wijnands r. , kaaret p. , 2002 , science , 298 , 196 corbel s. , nowak m.a . , fender r.p . , tzioumis a. , markoff s. , 2003 , a&a , 400 , 1007 cui w. , heindl w.a . , swank j.h . , smith d.m . , morgan e.h . , remillard r. , marshall f.e . , 1997 , apj , 487 , l73 cui w. , shrader c.r . , haswell c.a . , hynes r.i . , 2000 , apj , 538 , 307 della valle m. , mirabel i.f . , rodrguez l.f . , 1994 , 290 , 803 dubus g. , hameury j - m . , lasota j - p , 2001 , a&a , 373 , 251 esin a.a . , mcclintock j.e . , narayan r. , 1997 , apj , 489 , 865 fender r.p . , 2001 , mnras 322 , 31 fender r.p . , hanson m.m . , pooley g.g . , 1999 , mnras , 308 , 473 fender r.p . , hendry m.a . , 2000 , mnras , 317 , 1 fender r.p . , hjellming r.m . , tilanus r.p.j . , pooley g.g . , deane j.r . , ogley r.n . , spencer r.e . , 2001 , mnras , 322 , l23 frank j. , king a. , raine d. , 1992 , ` accretion power in astrophysics ' , cup garcia m.r . , callanan p.j . , mcclintock j.e . , zhao p. , 1996 , apj , 460 , 932 hameury j.m . , king a.r . , lasota j.p . , 1986 , 162 , 71 han x. , hjellming r.m . , 1992 , apj , 400 , 304 hannikainen d. et al . , 2001 , esa sp-459 , p291 , proceedings of the fourth integral workshop , 4 - 8 september 2000 , alicante , spain . a. gimenez , v. reglero , c. winkler . harmon b.a . , paciesas w.s . , fishman g.j . , 1993 , iauc 5874 harmon b.a . , wilson r.b . , finger m.h . , paciesas w.s . , rubin b.c . , fishman g.j . , 1992 , iauc 5504 harmon b.a . et al . , 1995 , nature , 374 , 703 heise j. , 1997 , iauc 6606 hjellming r.m . , rupen m.p . , shrader c.r . , campbell - wilson d. , hunstead r.w . , mckay d.j . , 1996 , apj , 470 , l105 homan j. , wijnands r. , van der klis m. , belloni t. , van paradijs j. , klein - woldt m. , fender r. , mendez m. , 2000 , apjs , 132 , 377 hynes r.i . , haswell c.a . , 1999 , mnras , 303 , 101 hynes r.i . , mauche c.w . , haswell c.a . , shrader c.r . , cui w. , chaty s. , 2000 , apj , 539 , l37 hynes r.i . , zurita c. , haswell c.a . , casares j. charles p.a . , pavlenko e.p . , shugarov s.y . , lott d.a . , 2002 , mnras , 330 , 1009 int zand j.j.m . , pan h.c . , bleeker j.a.m . , skinner g.k . , gilfanov m.r . , sunyaev r.a . , 1992 , a&a , 266 , 283 kaluzienski l.j . , holt s.s . , boldt e.a . , serlemitsos , eadie g. , pounds k.a . , ricketts m.j . , watson m. , 1975 , apj , 201 , l121 kanbach g. , straubmeier c. , spruit h.c . , belloni t. , 2001 , nature , 414 , 180 king a.r . , ritter h. , 1998 , mnras , 293 , l42 kitamoto s. , miyamoto s. , tsunemi h. , makishima m. , 1984 , pasj , 36 , 799 kuulkers e. , 2001 , an , 322 , 9 lasota j - p . , 2001 , new astron . , 45 , 449 levine et al . , 1996 , apj , 469 , l33 liu q.z . , van paradijs j. , van den heuvel e.p.j . , 2001 , a&a , 368 , 1021 markoff s. , falcke h. , fender r.p . , 2001 , a&a , 372 , l25 markoff s. , nowak m.a . , corbel s. , fender r.p . , falcke h. , 2003 , a&a , 397 , 645 masetti n. , bianchini a. , bonibaker j. , della valle m. , vio r. , 1996 , a&a , 314 , 123 matilsky t.a , giacconi r. , gursky h. , kellogg e.m . , tanabaum h.d . , 1972 , apj , 174 , l53 mcclintock j.e . et al . , 2001 , apj , 555 , 477 miller j.m . , remillard r.a . , 2002 , iauc 7920 mineshige s. , wheeler j.c . , 1989 , apj , 343 , 241 oosterbroek t. et al . , 1997 , a&a , 321 , 776 paciesas w.s . , briggs m.s . , harmon b.a . , wilson r b. , finger m.h . , 1992 , iauc 5580 pietsch w. , haberl f. , gehrels n. , petre r. , 1993 , a&a , 273 , l11 pooley g.g . , 2001 , mnras , 324 , l23 reilly k.t . et al . , 2001 , apj , 561 , l183 remillard r. , morgan e. , smith d. , smith e. , 2000 , iauc 7389 revnivtsev m.g . , borozdin k.n . , priedhorsky w.c . , vikhlinin a. , 2000b , apj , 530 , 955 revnivtsev m.g . , sunyaev r. , borozdin k.n . , 2000a , a&a , 361 , l37 revnivtsev m.g . , 1998 , a&a , 331 , 557 shabaz t. , ringwald f.a . , bunn j.c . , naylor t. , charles p.a . , casares j. , 1994 , mnras , 271 , l10 shrader c.r . , wagner r.m . , hjellming r.m . , han x.h . , starrfield s.g . , 1994 , 434 , 698 soria r. , wu k. , hannikainen d. , mccollough m. , hunstead r. , 2001 , proceedings of the jhu / lhea workshop : `` x - ray emission from accretion onto black holes '' ( baltimore , june 2023 , 2001 ) , ( astro - ph/0108084 ) spruit h.c . , kanbach g. , 2002 , a&a , 391 , 225 sunyaev r.a . et al . , 1994 , astl , 20 , 777 tanaka y , lewin w.h.g . , 1995 , ` x - ray binaries ' chap 3 , cup , eds . lewin w.h.g . , van paradijs j. , van den heuvel e. terada k. , miyamoto s. , kitamoto s. , egoshi w. , pasj , 46 , 677 trudolyubov s. et al . , 1999 , a&a , 342 , 496 tsunemi h. , kitamoto s. , manabe m. , miyamoto s. , yamashita k. , nakagawa m. , 1989 , pasj , 41 , 391 ueda y. et al . , 1997 , iauc 6627 uemura m. et al . , 2000 , pasj , 52 , l15 van der hooft f. , et al . 1996 , apj , 458 , l75 van der hooft f. , et al . 1999 , apj , 513 , 477 van der klis m. , 1995 , ` x - ray binaries ' , chapter 6 , eds . lewin w.h.g . , van paradijs j. , van den heuvel e. , cup van paradijs j. , mcclintock j.e . , 1995 , ` x - ray binaries ' chap 1 , cup , eds . lewin w.h.g . , van paradijs j. , van den heuvel e. van paradijs j. , telesco c.m . , kouveliotou c. , fishman g.j . , 1994 , 429 , l19 wagner r.m . , starrfield s.g . , howell s.b , kreidl t.j . , , cassatella a. , bertram r. , fried r. , 1991 , apj , 378 , 293 whittet d.c.b . , 1992 , ` dust in the galactic environment ' , iop : new york wijnands r. , van der klis m. , 1999 , 514 , 939 wilms j. , nowak m.a . , pottschmidt k. , heindl w.a . , dove j.b . , begelman m.c . , 2001 , mnras , 320 , 327 wood k.s . et al . , 2000 , apj , 544 , l45 wood k.s . , titarchuk l. , ray p.s . , wolff m.t . , lovellette m.n . , bandyopadhyay r. , 2001 , apj , 563 , 246 wu k. , soria r. , page m.j . , sakelliou i. , kahn s.m . , de vries c.p . , 2001 , a&a , 2001 , 365 , l267 wu k. , soria r. , campbell - wilson d. , hannikainen d. , harmon b.a . , hunstead r. , johnston h. , mccollough m. , mcintyre v. , 2002 , apj , 565 , 1161 zhang s.n . , cui w. , harmon b.a . , paciesas w.s . , remillard r.e . , van paradijs j. , 1997 , apj , 477 , l95 ycki p.t . , done c. , smith d.a . , 1999 , mnras , 309 , 561 table 1 : list of sources which entered an outburst but remained in the low / hard state throughout . bullet points indicate the frequencies at which observations were attempted during the outburst . + table 2 : conversion table for instrumental count rates to crab units . + fig . 1 : lightcurves for six of the low / hard state transient sources . the top left plot shows x - ray lightcurves for a1524@xmath162 , 4u 1543@xmath1475 and grs 1737@xmath131 . the other three plots show x - ray , optical and radio lightcurves for v404 cyg ( top right ) , gro j0422 + 32 ( bottom left ) and gro j1719@xmath124 ( bottom right ) . data have been obtained from the sources listed in table [ sources ] . arrows represent upper limits . thick vertical lines on the x - ray panels indicate the epochs plotted in fig . [ spectrum ] . + fig . 1 : ( contd . ) x - ray , optical and radio lightcurves for gs 1354@xmath164 ( left ) and xte j1118 + 480 ( right ) . data have been obtained from the sources listed in table [ sources ] . arrows represent upper limits ; note that the optical error - bars are smaller than the symbols in the case of gs 1354@xmath164 . thick vertical lines on the x - ray panels indicate the epochs plotted in fig . [ spectrum ] . 2 : the variability of the low frequency qpo in gro j1719@xmath124 , gro j0422 + 32 , gs 1354@xmath164 and xte j1118 + 480 . the increase in frequency does not correlate well with the x - ray lightcurve . figures have been adapted from van der hooft ( 1996 , 1999 ) , brocksopp et al . ( 2001 ) and wood et al . + fig . 3 : broad - band spectra for v404 cyg , gro j1719@xmath124 , gro j0422 + 32 and gs 1354@xmath164 , which qualitatively resemble fig . 1 of markoff et al . three epochs are plotted corresponding , where possible ( i.e. _ not _ in the case of gs 1354@xmath164 ) to outburst peak ( cross ) , secondary peak ( star ) and low luminosity ( square ) but the level of variability appears to have little effect on the shape of the spectrum . optical data have been de - reddened as described in the text . an additional point has been included in the gro j0422 + 32 plot , which is the 10@xmath12 m detection by van paradijs et al . errors on the points are smaller than the size of the symbols ( for those points for which error information is available ) and downwards arrows indicate upper limits . + llccccl v404 cyg&gs 2023 + 338&1989&@xmath13&@xmath13&@xmath13&terada et al . 1994 , int zand et al . 1992 , wagner et + & & & & & & al . 1991 , casares et al . 1991 , han & hjellming 1992 + a1524@xmath162&tra x-1 , v*ky tra&1990&@xmath13&&&barret et al . 1992 , 1995 + 4u 1543@xmath1475&v*il lup & 1992&@xmath13&&&batse archive + gro j0422 + 32&nova per , v518 per&1992&@xmath13&@xmath13&@xmath13&batse archive , sunyaev et al . 1994 , + & & & & & & garcia et al . 1996 , shrader et al . 1994 + gro j1719@xmath124&grs 1716@xmath1249&1993&@xmath13&@xmath13&@xmath13&batse archive , sunyaev et al . 1994 , masetti et al . + & nova oph , v2293 oph&&&&&1996 , hjellming et al . 1996 , della valle et al . 1994 + grs 1737@xmath131&&1997&@xmath13&&&_rxte_/asm archive , trudolyubov et al . 1999 + gs 1354@xmath164&cen x-2 , bw cir&1997&@xmath13&@xmath13&@xmath13&_rxte_/asm & batse archives , brocksopp et al . 2001 + xte j1118 + 480&&2000&@xmath13&@xmath13&@xmath13&_rxte_/asm archive , kuulkers 2001 , pooley 2001 , + & & & & & & wood et al . 2000 + [ sources ] llll _ rosat_/pspc&0.12.4&303 cts / s&pietsch et al . 1993 + _ granat_/sigma&3575&0.108 ph / s/@xmath14&trudolyubov et al . granat_/sigma&75150&0.044 ph / s/@xmath14&trudolyubov et al . 1999 + _ ginga_/asm&120&2.7 cts / s/@xmath14&tsunemi et al . 1989 + _ kvant_/ttm&210 & 1.44 cts / s/@xmath14&j . in t zand , kvant_/ttm&10 - 27 & 0.32 cts / s/@xmath14&j . in t zand , cgro_/batse&20100&0.3 ph / s/@xmath14&m.mccollough , priv . comm . + _ rxte_/asm&210&75cts / s&levine et al . 1996 + _ argos_/usa&2 - 12&3300 cts / s&p . ray , priv . + [ conversions ]

we report on the mixed - orbital ground state compounds analogous to the mixed - valence ones . orbital doublet and singlet ground states of the fe ion displayed in mssbauer spectra of the prussian blue analogues afe[co(cn)ho ( ) are interconverting to each other as temperature changes for a = k , na . relative weight of orbital singlet ground states increases with lowering temperature . in the alcali - free cobalticyanides , the mssbauer spectra are dominated by quadrupole - splt doublets of the same origin with a large splitting , characteristic of the fe species coordinated by two ( at least ) oxygen atoms of water molecules in _ cis_-configuration . single - type octahedral fe coordination is isolated in a - filled cobalticyanides for a = rb , cs to be characterized by the narrower quadrupole spectrum associated with the orbital doublet ground states . ionic exchange of k for cs in kfe[co(cn) ho results in predominance of the orbital singlet , i.e. , in enhanced anisotropy of the fe valence electrons . we found hence that the fecharge distribution can be modified together with crystal size at the step of synthesis owing to distortion isomerism . specialpagefalseevenheadoddhead evenfootoddfoot[sec : intro ]

1302.2192

materials enclosing the transition metal elements are characterized by a wide diversity of ground states whether this be the quantum states of collective electron systems or local states . in latter case , the transitions between disparate ground states of 3d ions can be driven by changes in temperature , chemical doping , or by external stimuli . phase transitions are thus induced that involve the spin , charge , lattice and orbital degrees of freedom . charge transfer , spin crossover and linkage isomerism are the three types of transitions requiring a complex elucidation of the roles of corresponding degrees of freedom to be distinguished from each other@xcite . the long - range structural aspects display their role via succession of transition steps as they evidence the interconversions between various local states mediated by elastic interactions@xcite . charge transfer from one site to another could be a prerequisite of the concomitant spin - state transition@xcite . orbital degree of freedom is responsible for a long - range - distortion symmetry lowering , provided that orbital ordering is coupled with the mixed - valence charge ordering transition@xcite . orbital ground state reversal @xcite is an example of yet another type of transitions strongly coupled to the lattice and , possibly , to some other degrees of freedom . simultaneous presence of the orbital singlet and orbital doublet ground states , familiar in case of fe@xmath0 ion@xcite , could be an indication of proximity of such a transition . prussian blue ( pb ) analogues constitute a family of simple cubic compounds which excited a great interest recently owing to the family members exemplifying the spin crossover , charge transfer and linkage isomerism @xcite . here , in the same family , a@xmath1fe@xmath2[co(cn)@xmath3@xmath4 @xmath7h@xmath4o ( @xmath6 , a = k , na , rb , cs ) , we report on mixing and intercoverting the orbital ground states of fe@xmath0 . metal - organic framework of the pb analogues is formed by the 3d metal cations and the cyanometallate complex anions ( of the [ co(cn)@xmath3@xmath8-type ) arranged into a nacl - like 3d checkerboard . in major group of these compounds , the valence and spin differ between the states of metal ions occupying the cationic node and the anion core . the framework complex anions are typically having higher valences than the cations , in this case , the framework is bearing a net negative charge , which is balanced either by insertion of alcali ( a ) cations into the framework interstitials or by spontaneous generation of the large - size vacancies on the anionic nodes . crystallization and zeolitic water molecules ( c - water and z - water ) fill the cavities at the vacant sites of complex anions and a - cations , respectively . the local electronic configurations are probed by the quadrupole interactions of a @xmath9fe mssbauer nuclei sensing the electric field gradient from valent electrons and lattice charges surrounding the fe@xmath0 ion . mssbauer spectra of ferrous hexacyanocobaltates were reported previously only for the alcali - free phases ( @xmath10)@xcite,@xcite . the spectra can be fitted by two or more quadrupole doublets implying the existence of a number of different states of the high - spin fe@xmath0 ions . paired doublets , both originating from the cationic sites outside the low - spin complex [ m(cn)@xmath3 were also observed in mssbauer spectra of some other ferrous hexacyanometallates , e.g. , for m = cr ( @xmath10)@xcite . the early mssbauer study@xcite has proposed that the water molecules in _ cis- _ and _ trans- _ configurations could contribute to the appearance of multiple sites for fe@xmath11 in case of m = cr , the spin - state transition of fe@xmath0was observed only for @xmath12 ( a = cs ) @xcite . it was conjectured@xcite that filling the structure with a ions induces a stronger ligand field on the fe@xmath0 ion that causes the spin transition . in this study of the local electronic configurations in afe[co(cn)@xmath3@xmath13h@xmath4o ( a = k , na , rb , cs ) and related alcali - free form we show that the ligand field stabilizes the singular ground state ( orbital doublet ) only for a = cs and rb . in the cases of a = na , k , and @xmath10 we report on the coexisting orbital singlet and orbital doublet ground states for fe@xmath0 and on the interconversions between them induced by changes of temperature . rbfe[co(cn)@xmath3@xmath14h@xmath4o and csfe[co(cn)@xmath3@xmath14h@xmath4o were prepared mixing the 0.025 m aqueous solution of k@xmath15co(cn)@xmath16with 0.025 m solutions of fecl@xmath174h@xmath4o in excess of rbcl or cscl . similarly synthesized samples in excess of ki and nacl gave the compounds k@xmath18fe@xmath19[co(cn)@xmath3@xmath20h@xmath4o and na@xmath21fe@xmath22[co(cn)@xmath3@xmath23h@xmath4o , respectively . the a - free compounds were synthesized from solutions of similar molarity ( sample 8 in table i ) and from 10-fold diluted ( sample 6 ) . analytic methods of inductively - coupled plasma and x - ray - fluorescent analysis gave the results consistent with the site occupancies obtained from rietveld analysis . the same compounds with different crystal size were synthesized for a = rb and cs via the ionic exchange using kfe[co(cn)@xmath3@xmath20h@xmath4o as a sorbent . transformation of this sorbent into csfe[co(cn)@xmath3@xmath14h@xmath4o was detected by rietveld analysis and confirmed by analytic techniques . * table i. * lattice parameters in the precipitated and ion - exchanged hexacyanocobaltates at room temperature . [ c]l + + + + + + + + + [ c]|l|l|compound & parameter * , nm + na@xmath21fe@xmath22[co(cn)@xmath3@xmath23h@xmath4o & + k@xmath18fe@xmath19[co(cn)@xmath3@xmath20h@xmath4o & + kfe[co(cn)@xmath3@xmath14h@xmath4o & + rbfe[co(cn)@xmath3@xmath14h@xmath4o & + csfe[co(cn)@xmath3@xmath14h@xmath4o & + fe[co(cn)@xmath3@xmath24h@xmath4o & + cs - exchanged ( b ) at 20@xmath25c & + @xmath9fe[co(cn)@xmath3@xmath24h@xmath4o & + * refined according to fig.1 ( c ) and ( d ) structure models using the fm@xmath26 m space group . x - ray diffraction ( xrd ) patterns were measured on a pw3040/60 xpert pro ( panalytical ) diffractometer using cu ka source . the atomic ratio fe / co and k / co were determined by inductively coupled plasma atomic emission spectrometer ( icp - aes ) on an iris interpid ii xsp instrument ( thermo electron corporation ) . the x - ray fluorescent analysis was carried out using the xrf axios x ( panalytical ) spectrometer . the water content in the samples was estimated by thermogravimetric analysis carried out using a setaram setsys 16/18 analyser measuring the weight loss in an air flow of 25 ml / min with a heating rate of 10 k / min . magnetization of the paramagnetic compounds was measured with a squid magnetometer in the range between 2 k to 350k . mssbauer spectra were measured with a topologic 500 a spectrometer at the temperatures between 78 k and room temperature . the measurements at heating the sample above room temperature were performed in a wissel gmbh furnace . isomer shifts ( is ) are reported relative to @xmath27-fe at room temperature . [ ptb ] fff1.eps the a - filled and a - free prussian blues are two types of the cubic frameworks , described by structure models of keggin and miles@xcite and ludi@xcite , respectively . both of them exhibit the variable degrees of randomness in the distribution of the a ions over the framework interstitials and in the distribution of the water - filled anionic vacancies . in both cases , the _ random _ structures are described by the symmetry group fm@xmath26 m . when the a - cations and z - waters occupy the interstitials in an _ ordered _ fashion the symmetry turns into f@xmath283 m . the f@xmath283 m symmetry appears , for example , in csfe[cr(cn)6]@xmath14h@xmath4o @xcite . it means in the ideal case that the system of framework interstitials splits into two sublattices ( 1:1 ) , as shown in fig 1(a ) . in reality , each of these ( 1:1 ) sublattices is not represented by solely a - ions and z - waters , but occupied just unequally with these two species . in the a - free structure , another sublattice may split into two subsystems that is the anionic sublattice . when one proceeds from the disordered to ordered placement of anionic vacancies the symmetry turns from fm@xmath26 m to pm@xmath26 m giving birth to two dissimilar sublattices@xcite . in the pm@xmath26 m structure , two born sites are abundant with the ratio of 3:1 . four relevant lattice cells are shown in fig.1 for the pairs of a - free and a - filled structures . rietveld refinements were carried out using the above mentioned four structural models and three symmetry groups known from the literature @xcite . these treatments have shown that the most symmetric fm@xmath26 m model is sufficient to fit all the observed reflections . when fitted with lower symmetries the xrd least - square analysis has resulted in stagnant goodness of fit , unimproved , despite larger number of fitting parameters . the intensity ratio @xmath29 changes dramatically when k , rb or cs ions enter into the structure , as reported previously@xcite . the change becomes very large as the number of electrons in an alcali ion increases , allowing for cs a rough estimation of its content even prior to rietveld analysis . the refined lattice parameters at ambient temperature are listed in table i. in all compounds , the lattice parameter is close to that of the low - spin phase of csfe[cr(cn)@xmath3@xmath14h@xmath4o@xcite , however , it is found from magnetic measurements that the addressed in table i compounds enclose the fe@xmath0 ions in the high - spin state that is unchanged down to 2 k. in the fully a - filled pb analogues , the fe@xmath0 ions are coordinated with six nitrogens , however , at partial filling at least one c - water molecule enters the first coordination sphere of fe@xmath0 . from the combined scheme of compensating the charge misbalance between fe@xmath0cations and co(cn)@xmath3@xmath8 anions this is evidently the case of our samples ( table i ) . electronic configuration of a high - spin d@xmath30 ion implies only one electron extraneous with respect to isotropic half - closed d - shell composed of five spins - up electrons . in an octahedral environment , the group of three t@xmath31 orbitals is available for this sixth electron . we show in fig . 2 the orientations of these orbitals relative to the water molecule . in case of octahedral high - spin fe@xmath0 , the magnetic susceptibility of an ion is contributed by the orbital motion of the t@xmath31 valence electron that is added to the spin s=2 throgh the spin - orbit coupling . the valence contribution of the electric field gradient ( efg ) is predominant and exceeds the lattice efg term by an order of magnitude@xcite . the t@xmath31 group is the source of valence term of efg that is the main contribution of the observed quadrupole splitting . [ ptb ] fff2.eps one of the orbitals of the t@xmath31 group in fig . 2 ( a , b , c ) is perpendicular to the direction pointing towards the water molecule . corresponding energy level would not be affected by the ligand replacement . the lobes of two other orbitals ( d@xmath32 and d@xmath33 in fig.2 ) are at 45@xmath34 with respect to oxygen direction . owing to reduced repulsion between oxygen and d@xmath32 or d@xmath33 electrons corresponding levels would have lower energy . in the ground state of fe@xmath0 ion , these d@xmath32 and d@xmath33 orbitals share the sixth d - electron ( spin - down electron ) , while five spin - up d - electrons span isotropically . the third ( d@xmath35 ) orbital would split off its energy level upwards from the ground state , formed by two degenerate d@xmath32 and d@xmath33 orbitals as shown in fig . 2 ( d ) is relevant to zero spin - orbit coupling . in case of combined effect of an axial field and of spin - orbit interaction nine energy levels are produced , namely three singlets and six doublets@xcite . in the hexacyanocobaltates , where the high - spin fe@xmath0 species are surrounded by a low - spin complex anions [ co(iii)(cn)@xmath3@xmath8 , the fe@xmath0 site is a single source of paramagnetism . orbital motion contributing to the magnetic moment of fe@xmath0 is unfrozen in the extent determined by the degree of bonding covalence . ligand replacement affects the degree of valent electron delocalization . variation of magnetic moment of fe@xmath0 ions over the sample is expected caused by the effect of a number of c - waters entering the first coordination sphere of fe@xmath0 . there appears a distribution of the fe@xmath0 ions over the number of the c - water ligands . for the magnetic measurements the samples 4 - 6 of the table i were selected . the compounds rbfe[co(cn)@xmath3@xmath14h@xmath4o and csfe[co(cn)@xmath3@xmath13h@xmath4o were selected because they are fully a - filled , enclosing only the z - water , and showing the single - site mssbauer spectra . as a counterpart from the opposite family the compound fe[co(cn)@xmath3@xmath24h@xmath4o was selected ( the sample no . 6 ) . a convenient method to display the results of magnetic measurements is to plot the experimental curves @xmath36 fitted with theoretical ones @xcite . accordingly , in fig . 3 , the experimental data ( 1 ) and ( 2 ) for @xmath36 are matched to a single set of fitting parameters ( @xmath37 ) . the data ( 3 ) are approximated by a mean squared @xmath38 with two magnetic moments @xmath39 @xmath40 and abundances @xmath41 , @xmath42 adjusted to the parameters of subspectra observed in corresponding mssbauer spectra . [ ptb ] fff3.eps from the theory of magnetic properties of transition metal ions in low - symmetry ligand field@xcite the @xmath43 is predicted to culminate around 100 k exceeding the spin - only value of 4.9@xmath44 and depending on the crystal - field splitting energy , the energy of spin - orbit coupling @xmath45 , and the orbital reduction factor @xmath46 . the results tabulated in the reports of two groups@xcite coincide@xcite . the curves @xmath36 can be described in terms of low - temperature dip ( below maximum ) , the negative slope ( above maximum ) , and the overall level . the width of the low-@xmath47 dip in the @xmath36 curves scales with @xmath45 , therefore , the value of @xmath45 can be most accurately fitted , using the accurate claim that we can exclude mixing antiferromagnetism into such a low-@xmath47 dip . on the other side , the negative slope of the @xmath48 vs. @xmath47 curve is steeper when the orbital doublet is lowest ( @xmath49 ) . in this case ( rb and cs - samples ) , large slope absolute value evidences the orbital doublet ground state . the slope is almost unvaried with @xmath50 , while the level of @xmath48 depends strongly on @xmath50 . however , the level of @xmath48 depends also strongly on @xmath46 . the correlation between these two dependencies reduces much the accuracy of determination of @xmath51 nevertheless , the choice between the cases of @xmath49 and @xmath52 is unambiguous . with increasing the degree of t@xmath31 electron delocalization both values of @xmath45 and @xmath46 are reduced from their ideal free - ion values of @xmath53 @xmath54 and @xmath55 , respectively@xcite . a concommitent reduction of @xmath56 and @xmath57 in a series of coordination compounds was evidenced by figgis et al@xcite . our results in fig . 3 confirm such a correlated reduction , except in rb - based compound . however , if one assume that the molar weight of our sample is slightly reduced by rb - deficiency , the value of @xmath46 would be reduced as well . the deficiency of 0.06 ( three times smaller than in our k@xmath18fe@xmath19[co(cn)@xmath3@xmath14h@xmath4o sample ) would correspond to the reduction of @xmath46 from 1 to 0.8 . in the samples 5 and 6 , the observed @xmath36 decreases with increasing temperature even more steeply than does the maximally steep theoretical curve ( fig . this result can be explained@xcite by the change of the covalence degree with temperature as the overlap between the iron(ii ) 3d- and ligand charge clouds changes owing to thermal expansion@xcite . we observe two doublets , both of them having the temperature - dependent quadrupole splitting . in addition , the ratio of doublet areas is also strongly temperature - dependent . the values of chemical shifts for both doublets in fig . 4 are in the range 1.1 mm / s to 1.15 mm / s allowing us to identify both doublets with the high - spin fe@xmath0 . a tiny spectral asymmetry indicates a trace of oxidized ( fe@xmath58 ) component . the external doublet is dominating at low temperatures . as the temperature is raised an inverted doublet area ratio is observed . mssbauer spectra suggest that the triple degeneracy of t@xmath31 is lifted over either in favor of the orbital singlet or in favor of orbital doublet ground states . the charge distribution is less anisotropic when the spin - down electron is shared between two orbitals as shown in fig . we ascribe such a charge state to the internal quadrupole doublet of fig.4 . larger charge anisotropy originates from localizing the sixth electron of fe@xmath0 on a single @xmath59-orbital ( external qs doublet ) . from the qs ratio the valence efg term is expected to be twice larger for the orbital singlet than for the orbital doublet . the literature overview confirms indeed that in fe@xmath0 the orbital singlet produces twice larger quadrupole splitting than the orbital doublet ground state@xcite . [ ptb ] fff4.eps in the bottom of fig . 4 , we fitted the curves @xmath60 using the approach developed by ingalls@xcite . the crystal - field splitting , the spin - orbital interactions , the lattice term and covalency effects are taken into account in a step - by - step procedure@xcite allowing to estimate the triplet t@xmath31 splitting energies @xmath50 , @xmath61 , the spin - orbit coupling constant @xmath45 scaled - down by covalency effects from the free - ion value of 100 @xmath54 , and the lattice term @xmath62e@xmath63 . in the first step , the so - called preliminary results@xcite were derived from fitting the curves @xmath64 with the theoretical expression@xcite : @xmath65 & \delta e_{\text{q}}^{0}\frac{(1+e^{\frac{-2\delta_{1}}{k_{\text{b}}t}}+e^{\frac{-2\delta_{2}}{k_{\text{b}}t}}-e^{\frac{-\delta_{1}}{k_{\text{b}}t}}-e^{\frac{-\delta_{2}}{k_{\text{b}}t}}-e^{\frac{-(\delta_{1}+\delta_{2})}{k_{\text{b}}t}})^{\frac12 } } { 1+e^{\frac{-\delta_{1}}{k_{\text{b}}t}}+e^{\frac{-\delta_{2}}{k_{\text{b}}t}}}\ ] ] @xmath66 here @xmath67 is the quadrupole splitting at zero temperature , which is for fe@xmath0 as high as 4 mm / s @xcite or even 4.5 mm / s@xcite . the eq.(1 ) shows that the value of @xmath68 is reached at @xmath69 if @xmath50=@xmath61 ( orbital singlet ) but only the value of @xmath70 is reached at @xmath69 if @xmath49 ( orbital doublet ) . the preliminary fitting results imply the orbital doublet state for the inner mssbauer doublet and the orbital singlet state for the outer mssbauer doublet . next step of the fitting procedure consist of taking into account the spin - orbit coupling @xmath45 . covalence effects reduce the value of @xmath45 compared to the free - ion value of @xmath71 -100 @xmath54 . in this approximation , the boltzmann distribution of the valence electron [ eq.(1 ) ] over three levels must be replaced with that over nine levels@xcite . we used in fig.4 the theoretical results reported in figs . 3 and 5 of the ingall s work @xcite , assuming an axially symmetric local distortions of the fen@xmath16 octahedra within the entirely cubic structure . finally , it is tempting to take into account the temperature - independent lattice term . most frequently the lattice term is opposite in sign to the valence term and smaller by an order of magnitude@xcite . if the lattice term is subtracted from valence term , the overal @xmath60 slope would decrease , thus producing the underestimate of the crystal - field splitting . the contribution @xmath72 of -10% of @xmath68 leads to the underestimate of @xmath50 of the order of @xmath45 . lacking the calculation of lattice sums we assumed @xmath73 in fig.4 , however , if the ratio @xmath74 is indeed @xmath75 , then the expermental @xmath60 curves may be steeper than the theoretical ones . the divergence of this type is observed in all @xmath64 curves of fig . 4 . at approaching 300 k the experimental @xmath60 curves become very steep . there could exist , however , other source of such behavior . yet another possibility is the steep change @xmath60 in a proximity of a phase transition@xcite , to be sought in these compounds presumably above 300 k. also , one may attribute the narrow doublet to a large fraction of the supercooled high - temperature phase supposing the high- and low - temperature phases isostructural . indeed , from xrd no evidence of biphasic character of samples were seen . however , if the transition is glasslike in its nature , a single phase with disordered cubic structure may be hosting the residual spatial inhomogeneity that implies the mixed orbital ground states coexisting in a way by which different local structures coexist in a glass . thermally - induced change of areas of two mssbauer doublets shows that the population of orbital doublet states grows with increasing temperature . it appears that different local distortions substitute for one another with preserving the global cubic symmetry . this gives us a reason to believe that the coexisting orbital states originate from the _ local _ distortions of the fen@xmath16 octahedra that may have either static or dynamic character . in the low - temperature region , the orbital singlet is lowest because the specific distortion of the fen@xmath16 octahedra might be dynamically more stable . with increasing temperature these polyhedra repose more time in a state of converse distortion . the latter hosts the more isotropic distribution of the fe@xmath0 valence electrons . dynamic equilibrium shifts towards the reduced electronic anisotropy with increasing temperature . therefore , the orbital doublet states appear to be partly thermally - induced . parameters of mssbauer spectra ( fig . 4 ) in na@xmath21fe@xmath22@xmath23h@xmath4o : isomer shift ( is ) @xmath76 , quadrupole splitting ( qs ) @xmath77 , linewidths @xmath78(mm / s ) , and spectral abundance ( % ) . [ c]|c|c|c|c|c| & @xmath76(mm / s ) & @xmath79 & @xmath78(mm / s ) & % + & & ( mm / s ) & & + & 1.142(2 ) & 1.56(1 ) & 0.48(1 ) & 44 + @xmath80k & 1.110(2 ) & 0.82(1 ) & 0.46(1 ) & 51 + & 0.46(1 ) & 0.3(2 ) & 0.27(2 ) & 5 + & & & & + & 1.161(8 ) & 2.06(2 ) & 0.50(2 ) & 40 + 240 k & 1.168(2 ) & 0.99(3 ) & 0.51(2 ) & 54 + & 0.35(3 ) & 0.5(2 ) & 0.4(1 ) & 6 + & & & & + & 1.209(3 ) & 2.25(1 ) & 0.54(2 ) & 50 + 200 k & 1.169(4 ) & 1.19(1 ) & 0.56(2 ) & 46 + & 0.4(1 ) & 0.5(2 ) & 0.4(4 ) & 3 + & & & & + & 1.218(2 ) & 2.467(9 ) & 0.67(1 ) & 65 + 140 k & 1.240(3 ) & 1.287(1 ) & 0.51(1 ) & 31 + & 0.5(1 ) & 0.5(2 ) & 0.5(3 ) & 4 + & & & & + & 1.236(6 ) & 2.63(3 ) & 0.69(2 ) & 62 + 80 k & 1.283(8 ) & 1.47(3 ) & 0.58(3 ) & 35 + & 0.5(1 ) & 0.5(2 ) & 0.4(2 ) & 3 + orbital - ground - state reversal in high - spin fe@xmath0 compounds has been a well - known phenomenon since the pioneering work of dzsi and keszthelyi@xcite , in which a pair of doublets was observed in fe(clo@xmath81)@xmath176h@xmath4o within a rather narrow temperature interval . it was argued in this case , that two different orbital ground states coexists within the transition range owing to the first - order phase transition@xcite . in this system , the transition was observed by calorimetry , neutron scattering and magnetic susceptibility measurements@xcite . order - disorder transformation in the remote perchlorate ion bonding system built up of hydrogen bonds was speculated to underlie the crystallographically undetected second - order transition switching the feo@xmath16 octahedra from compression to elongation along the trigonal axis@xcite . in the isostructural fluoroborate hydrate , brunot@xcite observed not so precipitous change in hydrogen bond dynamics and deduced an inference of a different transition mechanism . the temperature- and pressure - driven orbital ground state reversal phenomena were proven to occur at the first - order phase transitions in iron(ii ) fluorosilicate hydrate@xcite . in thiourea high - spin fe(ii ) compound , no direct evidence of phase transition was found , while singlet and doublet orbital ground states were found to coexist in a wider temperature range @xcite . the authors ruled out the interpretation based on classical phase transition after they have conducted the kinetic study@xcite that allowed them to specify the transforming system as an intermediate between fully homogeneous and biphasic heterogeneous . table iii . parameters of mssbauer spectra ( fig . 4 ) in k@xmath18fe@xmath19@xmath20h@xmath4o : isomer shift ( is ) @xmath76 , quadrupole splitting ( qs ) @xmath77 , linewidths @xmath78(mm / s ) , and spectral abundance ( % ) . [ c]|c|c|c|c|c| & @xmath76(mm / s ) & @xmath79 & @xmath78(mm / s ) & % + & & ( mm / s ) & & + & 1.135(2 ) & 1.76(1 ) & 0.44(1 ) & 44 + 280k & 1.098(2 ) & 1.007(7 ) & 0.468(8 ) & 54 + & 0.4(1 ) & 0.3(2 ) & 0.4(2 ) & 2 + & & & & + & 1.173(2 ) & 2.207(8 ) & 0.38(2 ) & 36 + 240 k & 1.152(2 ) & 1.185(9 ) & 0.59(1 ) & 62 + & 0.4(1 ) & 0.5(2 ) & 0.4(1 ) & 2 + & & & & + & 1.204(3 ) & 2.45(1 ) & 0.47(2 ) & 44 + 200 k & 1.182(5 ) & 1.19(1 ) & 0.62(2 ) & 54 + & 0.4(2 ) & 0.5(2 ) & 0.6(4 ) & 2 + & & & & + & 1.234(3 ) & 2.68(2 ) & 0.63(2 ) & 62 + 140 k & 1.229(7 ) & 1.54(3 ) & 0.60(3 ) & 37 + & 0.5(2 ) & 0.5(2 ) & 0.7(3 ) & 1 + & & & & + & 1.260(6 ) & 2.86(2 ) & 0.65(2 ) & 60 + 80 k & 1.253(1 ) & 1.72(2 ) & 0.66(3 ) & 39 + & 0.5(2 ) & 0.5(2 ) & 0.7(3 ) & 1 + in the same way , we avoid interpreting the data on fig . 4 as a classical thermodynamic first - order or second - order phase transition such as phonon - mode condensation . in the compounds of the pb family , the ubiquitous cubic porous framework allows for variable site and cell symmetries to be accommodated ( cf . structures in fig.1 ) . as the steric hindrances released in low - density frameworks the coherence in atomic displacements ceases to exist . additional factor preventing the cooperativity of the local distortions is the disorder related to incomplete a - filling . as temperature decreases we observe in tables ii and iii that the linewidths of both major doublets increases . this is because each of the mssbauer doublets is contributed by several fe@xmath0 sites with different composition of the 1@xmath82 coordination sphere and with difference of positioning relative to each other of the n and h@xmath4o ligands . depending on the number and isomeric configuration ( trans , cis ) of the h@xmath4o ligands around fe@xmath0 site the site contributes either to orbital doublet or to orbitals singlet ground states . two mssbauer doublets correspond to two types of the ground state . therefore , the values of parameters displayed on fig . 4 ( @xmath83 ) have the meaning of averages over the orbital singlet and orbital doublet ground states . subtle differences between the crystal - field and spin - orbit parameters of the particlar environments give rise to the low - temperature linewidth broadening of both doublets . next , we classify all the possible ligand configurations by the manner in which the d - orbitals are filled depending on the symmetry of environment . our rietveld refinements in both a - filled and a - free series of cobalticyanides have revealed that their structures are best described by the symmetry fm@xmath26 m , allowing for the fe@xmath0 ion to have a number of different coordination environments . the fe@xmath0 sites can be classified by the number and configuration of n - ligands replaced with h@xmath4o . in these fe@xmath0 sites , the lowest in energy state depends on the h@xmath4o inclusion in first coordination sphere , however , when h@xmath4o is not included , the ground state is determined by the local distortion . owing to so - called distortion isomerism@xcite the ground state becomes not uniquely defined . shown in fig.5 are the cases of distortion isomerism ( a ) , orbital singlet ( b ) , and orbital doublet ( c ) . [ ptb ] fff5.eps the distortion isomerism is the phenomenon closely related to the inability for the sixth ( spin - down ) electron to be distributed isotropically over the triply degenerate t@xmath31 orbital . such an overmuch triple degeneracy is universally lifted over by a local distortion of one or another type . in a fen@xmath16 octahedron , the compression along tetragonal axis would stabilize the orbital singlet lowest ( a-1 ) , while elongation along tetragonal axis stabilizes the orbital doublet as a ground state ( a-2 ) . opposite effects are expected for the compression and elongation along the trigonal axis of the octahedron@xcite . self - coupling between electron localization and local distortion eliminates the case of zero efg valence term that is never observed . this physical picture is a key to understanding the temperature variation of the abundance of two mssbauer doublets in fig . 4 . assuming that the 6n - coordinated fe@xmath0 sites are contributing to both doublets , we observe that the relative abundance of the orbital doublet and orbital singlet states is varying with temperature owing to varied lattice distortion . matsuda et al@xcite have reported that the pb lattice possess an intrinsic instability with respect to rotations of the rigid units of the [ co(cn)@xmath3@xmath8-type . the rotational displacements generate spontaneously to stabilize the excited charge - transferred states , which arises owing to strong lattice - electron coupling . the same rotational displacements cause also the negative thermal expansion@xcite . recently , the thermal expansion in fe@xmath15[co(cn)@xmath3@xmath84h@xmath4o was reported to be indeed negative , at least above 200 k@xcite . we observe thus that the orbital ground state reversal accompanies the distortion isomerism . one or another of coexisting distortion isomers prevails in a particular temperature range . largest charge anisotropy occurs at lower temperatures owing to localizing the sixth electron of fe@xmath0 on a single @xmath59-orbital . the charge distribution is less anisotropic at higher temperatures when the spin - down electron is shared between two orbitals . the temperature variation of the abundance of two mssbauer doublets is very smooth as the distortions are local and broadly distributed . with decreasing the content of a - ions the anionic vacancies ( water - filled pores ) are generated for charge compensation . when the pore concentration is small only populated are the fe@xmath0 sites which contain one h@xmath4o ligand as shown in figs . 2 and 5(c ) . two pores in a vicinity of a fe@xmath0 species occur either in _ trans- _ or in _ cis- _ configurations . consider in fig . 5(c ) , for example , the ground state of the _ trans- _ fe@xmath0 . if we denote by @xmath5 the direction from the central fe@xmath0 ion towards two ligands then the lobes of both @xmath85 and @xmath86 orbitals would be pointing to directions having an angle of 45@xmath25 with the @xmath5 axis . the lobes of the third t@xmath31 orbital ( @xmath87 ) would be pointing to the direction orthogonal to the @xmath5-axis . therefore the ground state of the fe@xmath0 ion in such fen@xmath81(h@xmath4o)@xmath4 octahedron would be the orbital doublet . this is the same ground state as shown in fig . 2 for the fe@xmath0 ion in the fen@xmath88h@xmath4o octahedron with single water in first coordination sphere . orbital singlet in fig . 5(b ) is the ground state corresponding to the _ cis_-configuration of fen@xmath81(h@xmath4o)@xmath4 and to the octahedron with three in - plane waters . the h@xmath4o placement into the first coordination sphere of fe@xmath0 suppresses the distortion isomerism and stabilizes a singular ground state of fe@xmath0 . in our samples na@xmath21fe@xmath22[co(cn)@xmath3@xmath23h@xmath4o and k@xmath18fe@xmath19[co(cn)@xmath3@xmath20h@xmath4o the density of anionic vacancies is rather small and the population of the fe@xmath0 ions surrounded by two vacancies is also small . in these samples the inner doublet comes from the fe@xmath0 orbital doublet ground state associated with the fen@xmath88h@xmath4o octahedron and with one of distortion isomers of fen@xmath16 . the outer doublet comes mainly from another distortion isomer of fen@xmath16 having the orbital singlet ground state . another manifestation of distortion isomerism is the difference between the mssbauer spectra of the a - filled cobalticyanides obtained for rb and cs by two different procedures . using the standard technique , the a - filled cobalticyanides were obtained by precipitation method . these compounds have demonstrated the single - component mssbauer spectra ( a ) and ( b ) in fig.6 , identified with orbital doublet ground state . on the other hand , when the method of ionic exchange was applied to replace k by cs in kfe[co(cn)@xmath3@xmath14h@xmath4o , the second qs subspectrum , identified with the orbital singlet , did not disappear . on the contrary , we observed the strong enhancement of the outer doublet with @xmath77=1.8 mm / s , from 32% to 64% . in addition , there appears a component ( 13% ) with extremely large @xmath77 of 3 mm / s . it is assigned to another orbital singlet state with larger energy level splitting parameters @xmath50 , @xmath61 . the difference between two external doublets of the sample ( d ) is a manifestation another type of distortion isomerism unrelated to switching between different orbital ground states@xcite . [ ptb ] fff6.eps rietveld refinements resulted in similar compositions for two different syntheses . the intensity of x - ray diffraction has very similar patterns , except linewidth difference . larger - size crystallites of csfe[co(cn)@xmath3@xmath14h@xmath4o are obtained via ionic exchange . in this method , the crystal size of csfe[co(cn)@xmath3@xmath14h@xmath4o is determined by the size of the kfe[co(cn)@xmath3@xmath14h@xmath4o precursor crystals . the target compound csfe[co(cn)@xmath3@xmath14h@xmath4o with such large crystals could not be obtained by the standard method of precipitation from a solute . despite of similarity of the x - ray patterns , the qs of the ion - exchanged and precipitated samples of csfe[co(cn)@xmath3@xmath14h@xmath4o indicates that the valent spin - down electron has a very different distributions over the orbitals of t@xmath31 group ( table iv ) . * table iv . * parameters of mssbauer spectra ( fig . 6 ) in as - synthesized ( a - c ) and ( e ) and cs - sorbed(d ) hexacyanocobaltates : isomer shift ( is ) @xmath76 , quadrupole splitting ( qs ) @xmath77 , linewidths @xmath78(mm / s ) , and spectral abundance ( % ) . [ c]|c|c|c|c|c|compound & @xmath76(mm / s ) & @xmath77 & @xmath78(mm / s ) & % + & & ( mm / s ) & & + ( a):rbfe[co(cn)@xmath3 & 1.091(5 ) & 0.992(8 ) & 0.60(1 ) & 100 + @xmath14h@xmath4o & & & & + ( b):csfe[co(cn)@xmath3 & 1.116(3 ) & 1.077(4 ) & 0.71(1 ) & 100 + @xmath14h@xmath4o & & & & + ( c):kfe[co(cn)@xmath3 & 1.144(3 ) & 1.77(1 ) & 0.44(2 ) & 32 + @xmath14h@xmath4o & 1.104(2 ) & 1.03(1 ) & 0.52(2 ) & 64 + & 0.39(1 ) & 0.66(2 ) & 0.23(3 ) & 5 + & & & & + ( d ) : ion - exchanged & 1.111(5 ) & 1.80(3 ) & 0.69(4 ) & 64 + k for cs in ( c ) & 1.084(9 ) & 0.94(3 ) & 0.45(4 ) & 23 + & 1.123(8 ) & 2.98(2 ) & 0.33(4 ) & 13 + & & & & + ( e):fe[co(cn)@xmath3@xmath89 & 1.114(2 ) & 1.65(1 ) & 0.45(1 ) & 64 + @xmath90h@xmath4o & 1.104(2 ) & 0.86(1 ) & 0.38(1 ) & 31 + & 0.33(1 ) & 0.23(2 ) & 0.26(3 ) & 5 + we observe in the standard ( not ion - exchanged ) compounds that only one type of the octahedron distortion is preferred when the large - size a - ions occupy the interstitials , and this type of distortion stabilizes the orbital doublet lowest . on the other hand , in the case of a = k or a = na the appearance of the outer doublet comes into play . in this series of pb analogues , several related aspects of structure and properties turn out to be a - size - dependent . whereas in the case cs(rb)-containing pb analogue compounds the cs(rb ) ions are situated in the centers of the interstitial cubes , in the case of some k(na)-containing compounds the k(na ) ions are reported to be eccentrically located @xcite . the mobility of alkali metal ions in the channels of the hexacyanometalate structure decrease dramatically with increasing the alcali ionic radius@xcite . the irreversible fixation of the less mobile ions ensures the cs - selectivity for the cobalticyanide sorbents , allowing the ionic exchange to be used in practice for @xmath91cs removal from radioactive wastes@xcite . although rather large linewidths were observed in the y=1 compounds ( cf . in table iv @xmath92 for a = rb and @xmath93 for a = cs ) , there is no doubts in attribution of the whole mssbauer spectra to the orbital doublet ground states . in these cases , the xrd linewidths associated with the smallness of crystal size@xcite and the mssbauer linewidths correlate well to each other . the distribution of efg tensor component values that causes the line broadening arises from the small size of crystals . these compounds contains only the z - water molecules , therefore , they are closer to anhydrous salts rather than to crystal hydrates wherein the crystallization water helps the precipitation of larger crystals . similarly large mssbauer linewidths were observed previously by kosaka et al@xcite in the high - spin phase of the analogous chromate compound csfe[cr(cn)@xmath3@xmath14h@xmath4o devoid of crystallization water as well . the isomer shifts in the mssbauer spectrum of the a - free fe[co(cn)@xmath3@xmath89 ( fig.6e ) indicates predominantly ( 95% in best sample ) the high - spin fe@xmath0 . the ground states of the major part of the fe@xmath0 species ( 64% ) are the orbital singlet states . second place in population is taken by the orbital doublet states . third minor doublet is assigned to fe@xmath58 according to its isomer shift and magnetic measurements . refinements of the structure of fe@xmath94[co(cn)@xmath3@xmath14h@xmath4o were useful to confirm the assignments of three fe@xmath0 mssbauer subspectra to the defined structural sites . [ ptb ] fff7.eps let us consider first the structure models of archetype prussian blue investigated previously@xcite . the concentration of pores in the a - free hexacyanocobaltate is very large , therefore , the pores can not be considered as isolated ones . all kinds of associations between the water - filled pores underlie the resulting abundance of the fe@xmath0 species in various coordinations . interactions between the pores at the stage of synthesis may result in various distributions of the pores . for the prussian blue fe@xmath81[fe(cn)@xmath3@xmath95h@xmath4o , depending on the growth conditions , buser et al@xcite could synthesize the single crystals with varied degree of disorder . structures of several crystals were refined using the fm@xmath26 m model , others were best refined using the pm@xmath96 m model . _ a priori _ , fully randomized distribution of the pores is consistent with the fm@xmath26 m model ( fig . low symmetry ( pm@xmath26 m ) becomes applicable when one finds unequal occupancies for the cell body center and edge centers ( fig . these sites are abundant with the ratio of 1:3 . if @xmath97 is the occupancy of the body center , and @xmath98 is the occupancy of the edge centers , the conditions @xmath99 and @xmath100 correspond to fully ordered and random pb structures , respectively@xcite . in ferrous hexacyanocobaltate , owing to different valence ratio the same conditions correspond to @xmath99 and @xmath101 . intermediate partially - ordered structures can be characterized by the order parameter @xmath102 . as @xmath103 decreases , the integral value of @xmath104 shows in fig . 7 the percentages of 63% , 52% , 44% , 39% and 38% for @xmath105 , respectively . main trend of reduction of @xmath104 is determined by the reduction of the _ trans_-subpopulation of fen@xmath81(h@xmath4o)@xmath4 . the orbital singlet ground state prevails in the disordered structure described by the fm@xmath26 m model . in this case ( @xmath106 ) , the expected area of the external doublet @xmath107 reaches the value of 62% similar to experimental result of 64% in table iv . in fig . 7 , we have calculated the abundance of various coordinations of fe@xmath0 for several values of @xmath103 between 0 and 1 . the total abundance of the fe@xmath0species in the orbital doublet ground state @xmath104 is made up mainly of the abundances of fen@xmath88h@xmath4o and _ trans_-fen@xmath81(h@xmath4o)@xmath4 . relatively small number of water - free octahedra fen@xmath16 is assumed to be halved between two distortion isomers . [ ptb ] fff8.eps abundance of the orbital singlet @xmath107 is composed _ cis_-fen@xmath81(h@xmath4o)@xmath4 , fen@xmath15(h@xmath4o)@xmath15 and remaining fen@xmath16(distortion isomer ) . both in - plane and 3d ligand configurations in the 3:3 octahedron fen@xmath15(h@xmath4o)@xmath15 contribute to @xmath107 . first , when 3 oxygens coordinating to iron are in one plane the @xmath5 , @xmath108 and @xmath109 axes become all nonequivalent . in this case , the spin - down electron cloud loses the axial symmetry , and the t@xmath31 group splits into three singlets . second , when three oxygens are in 3d configurations , all three axes @xmath5 , @xmath108 and @xmath109 are equivalent ; nevertheless , the cubic symmetry is lost . since h@xmath4o is a weaker ligand than n this geometry can be viewed as an octahedron with three ligands expelled to longer distance than other three . equivalent spatial distortion is the elongation along the trigonal axis . such a distortion was shown to stabilize the orbital singlet lowest@xcite . while there are just two sorts of the ground states , the inspection of fig.7 for p=2/3 shows that there are eight sites of fe@xmath0 having different ligand configurations with the abundance of 5% or more . among the octahedra with n=0,1,2,3 and 4 those with n=0 , 2 and 3 are found in two isomeric configurations : compression - elongation ( n=0 ) , cis - trans ( n=2 ) and 2d-3d ( n=3 ) . these eight sites are to be completed with ninth site if the oxidized component ( fe@xmath58 ) is present in the sample . it appears that the most sophisticated way of spectra evaluation consist of partitioning the spectral area into nine doublets . even if the method would reproducibly fit the multiple spectra , it would be uneasy task to assign the resulting subspectra to a particular isomer . highly valuable remains the models in which less than nine subspectra are assumed . in the a - free cobalticyanide ( fig.8 ) , a plausible model of 4 subspectra is constructed through breaking up the external doublet with the excessive linewidth into two doublets . resulting intermediate quadrupole doublet is assigned to the 3-dimensional 3:3 configuration of the ligands ( 3n+3h@xmath4o ) . corresponding crystal fields have a trigonal symmetry axis . we estimate this symmetry to reduce slightly the ensuing quadrupole splitting . thus , in this model , the orbital singlet states stabilized in the crystal fields of trigonal symmetry are distinguished from the orbital singlet settled in the 4:2 ( cis ) and 3:3 ( 2d ) environments . in the experiment with enriched sample of @xmath9fe[co(cn)@xmath3@xmath24h@xmath4o the spectra were collected with the step @xmath110 k ( fig.8 ) the evaluation of these spectra with five , and six doublets were also attempted . these preliminary analyses showed that the model of four doublets , while providing already strong improvement of @xmath111 , ensures also the most reproducible and stable solution . the model is changing its parameters smoothly as temperature changes . orbital ground state reversal observed above in the a - filled compounds ( fig.5 ) is not seen in the a - free ones ( fig.8 ) . this is naturally explained by the small contribution of the n=0 sites susceptible to the distortion isomerism . it is shown in fig.7 that the number of such sites does not exceed 8% . other sites strongly perturbed by the cn ligand replacement exhibits persistent ground states independent of temperature . parameters of crystal - field splitting and spin - orbit coupling derived from fitting the temperature dependencies of @xmath77 are in agreement with the averaged curve of @xmath36 on fig . the splitting of 340 @xmath54 estimated previously in fe@xmath15[co(cn)@xmath3@xmath112h@xmath4o by rasmussen and meyers@xcite corresponds closely to one of our results , namely , to the crystal fields in the fe@xmath0 sites contributing to the external doublet . the concentration of pores in archetypal prussian blue , ferric ferrocyanide fe[fe(cn)@xmath3@xmath1133.5h@xmath4o , is @xmath114 , so that the ratio @xmath115 is 3 : 1 . the ordered pm@xmath26 m structure proposed for prussian blue in ref . @xcite contains the high - spin cations only in coordinations fen@xmath16 ( 25% ) and trans - fen@xmath81(h@xmath4o)@xmath4 ( 75% ) . in ferrous hexacyanocobaltate , the ratio @xmath115 is 2 : 1 . therefore , for a vacancy - ordered phase of such a stoichiometry one can not expect the same set of coordinations as for the pm@xmath26 m structure of the ref . @xcite . in order to construct an ordered phase with the ratio @xmath115 of 2 : 1 we try to fill 3 sites with fe , 2 sites with co , 1 site with co vacancies , and 0 sites with fe vacancies . we find that among cubic structures of nacl - type there is no superstructure with exact site ratio 3:2:1:0 , however , a unique superstructure with the ratio of 30:20:12:2 exists . using the body - centered im@xmath26 m model we can develop the proper arrangement of the vacancies with a variable order parameter . table v shows the details of the fully ordered ( @xmath116 ) im@xmath26 m structure . the order parameter @xmath117 is defined via the occupancy @xmath97 of the ( 12e ) position that is the core of complex anion [ co(cn)@xmath3@xmath8 . * table v. * wyckoff positions , atom type , relative coordinates and occupancies of the na and cl nodes in the superstructure im@xmath26 m model on the basis nacl - type framework . [ c]|llllll|wyckoff & atom & x & y & z & occ . + 2a & fe & 0 & 0 & 0 & 0 + 6b & fe _ trans _ & 0 & 1/2 & 1/2 & 1 + 8c & co & 1/4 & 1/4 & 1/4 & 1 + 12d & co & 1/4 & 0 & 1/2 & 1 + 12e & co & 1/4 & 0 & 0 & 0 + 24h & fe _ cis _ & 0 & 1/4 & 1/4 & 1 + among total 14 vacancies per cubic cell of the im@xmath26 m structure ( table v ) we place 12 vacancies of the complex anion co(cn)@xmath16 and 2 cationic ( fe@xmath0 ) vacancies . we obtained the ordered structure of fe[co(cn)@xmath3@xmath118h@xmath4o which contains 2 very large cross - shaped pores ( 6@xmath1191-pores , or @xmath119-pores ) . each cationic vacancy is surrounded by 6 anionic vacancies . the hydration degree @xmath120 is calculated assuming 14 water molecules per standard pore and an additional water molecule per fe@xmath0 vacancy . thus , the 6@xmath1191-pore is composed of 6 standard@xmath121pores aggregated into 3d crosslike body . the edge of the im@xmath26 m cell is twice longer than the edge of the pm@xmath26 m cell and the cell volume is 8-fold . the body - centered im@xmath26 m model exhibits only two kinds of fe@xmath0 sites with the abundance ratio of 3:1 , similar to that of the pm@xmath26 m model . in both models , the major site ( 75% ) is coordinated with two waters . however , in the pm@xmath26 m model the major site is coordinated by waters in _ trans_-conformation , whereas in the im@xmath26 m model the major site is coordinated by waters in _ cis_-conformation . the minor site in pm@xmath26 m model is coordinated by six nitrogens , and in im@xmath26 m model by four nitrogens and two waters in _ trans_-conformation . it is important that we were able to construct the superstructure model , which strongly favors the _ cis_-conformation in agreement with prevailing external doublet in mssbauer spectra . although the random fm@xmath96 m model agrees as well with the experiment , the lack of any short - range order between pores appears to be a severe limitation of this model because the vacancies bearing a charge interact with each other repulsively . we must note also that the luck of any superstructure reflections in diffraction patters may not be an evidence of a missing short - range order , but only the evidence of lacking _ long - range _ order . the suggested im@xmath26 m structure is a plausible model of the short - range order in the mixed - valence 2:3 pb analogues . although the superstructure reflections , predicted by either pm@xmath26 m or im@xmath26 m models , were not observed by x - ray diffraction , using the probabilities of local structure patterns observed by such a local technique as mssbauer spectroscopy one can identify the proper model of short - range order . the im@xmath26 m model predicts the strong predominance ( 3:1 ) of _ cis_- over _ trans_- population among the fe@xmath0 environments . the model describes the situation when concentration of elementary vacancies is so high that they must be in small clusters . in contrast , the pm@xmath26 m model describes the limit when all vacancies are still isolated , however , no more vacancies can be added without clustering . if they are actually added to proceed from @xmath114 to @xmath122 ( as in the bottom panel of fig.7 ) , there will result the extra - holey pm@xmath26 m model that fails to describe the mssbauer spectral data . in our experiments , we observe the prevailing abundance of the _ _ cis-__conformations related to the extended crosslike pores . the aggregation of small pores into extended pores underlies the large ( up to 750 m@xmath123/g ) specific surface and unique gas sorption properties of these materials@xcite . the charge - transferred states which were discussed above to induce the distortion isomerism are seen in the third small doublet with is of fe@xmath58 in figs . 6 ( c , e ) and 8 . a slight asymmetry appears also in spectra of fig.4 , although these spectra can be just a little less well fitted without introducing explicitly the fe@xmath58 component . the pairs of fe@xmath0/co@xmath58 and fe@xmath58/co@xmath0 can be considered as valence tautomers , and we explain the asymmetry in these spectra by electron transfer from fe@xmath0 to co@xmath58 . the formation and rupture of the chemical bonds implies no migration of atoms but only migration of electrons . the cobalt ions remain in low - spin state as the sixth electron of fe@xmath0 transfers from t@xmath31 orbital of fe@xmath0 to e@xmath124 orbital of co@xmath0 . other compound susceptible to valence tautomerism were known previously@xcite . for example , the prussian blue analogue rbmn[fe(cn)@xmath3@xmath14h@xmath4o , when synthesized in rb excess that fills enough framework interstitials with rb , crystallizes in the structure shown in fig . ( 1a ) . in this system , the tautomers fe@xmath0/mn@xmath58and fe@xmath58/mn@xmath0 are able to form different crystal phases which give way to each other as temperature changes . in the ferrous cobalticyanides , no abrupt phase change was found to be caused by the charge - transferred states . on the other hand , we observed that the content of fe@xmath58 can be changed depending on the excess of fecl@xmath4 and dilution degree in the course of precipitation of fe[co(cn)@xmath3@xmath125h@xmath4o . rapidly synthesized isotope - enriched compound contained up to 15 - 20% of fe@xmath58 , however , the content of fe@xmath58 was reduced down to just 5% through reducing the reaction rate in dilute solutions . most probably , the lack or misplacement of water molecules is needed to stabilize locally the charge - transferred states and make them observable in mssbauer spectra . it appears in this case that the rapidly prepared samples are not saturated in c - water . the sites of fe@xmath0 nearest to the nodes lacking the coordination water are subject of local tautomer transformation . in a slowly synthesized fe@xmath94[co(cn)@xmath3@xmath90h@xmath4o such crystallization defects are reduced . measuring the mssbauer spectra above room temperature supports this concept . with raising temperature the area of the fe@xmath58 subspectrum starts to increase irreversibly . thermogravimetric analysis showed that the loss of sample weight occurs at heating practically starting from room temperature . the mssbauer spectrum measured _ in situ _ in a furnace at 318 k shows the area of the fe@xmath58 doublet increased up to 37% . the same sample measured _ ex - situ _ after prolonged ( 1 week ) heat treatment at 60@xmath25c showed the area of the fe@xmath58 increased up to 57% . the equilibrium hydration degree at 60@xmath25c is related with nearly half iron species transformed to fe@xmath58 . longer exposure in air at at 60@xmath25c induces no further increase of the spectral asymmetry and the area of spectral component with is of 0.39 mm / s remains at the level of 60% . the isomer shift of 0.39 mm / s is attributable , in principle , either to high - spin fe@xmath58 or to low spin fe@xmath0 . the latter was observed below the spin - state transition in csfe[cr(cn)@xmath3 @xcite . theoretical approaches to lattice - dynamic calculations@xcite,@xcite show that the change in the electronic state is closely related to the structure expansion as the lattice parameter grows between the low - spin and high - spin forms from 1.03 nm to 1.07 nm . in fe@xmath15[co(cn)@xmath3@xmath84h@xmath4o , the thermal lattice expansion is negative@xcite . to verify that we do not deal with a _ reverse _ spin transition @xcite , we have compared the magnetic susceptibilities of the samples having 5% and 57 % of the subspectrum with is of 0.39 mm / s . in the heat - treated sample , the susceptibility increased too little compared to what is expected for the increase of @xmath48 upto 5.9 @xmath44 . with the weight loss taken into account this result would rather indicate the formation of an intermediate - spin state of fe@xmath58 which is known for strongly distorted octahedral coordinations with two alternate ligands in _ cis_-configuration@xcite . mssbauer spectroscopy was applied to study the mixed - valence ( fe@xmath0-co@xmath58 ) pb analogues which displayed the mixed - orbital ground states of fe@xmath0 ions . the ground state mixing originates from both the distortion isomerism and the varied composition of fe@xmath0 nearest - neighbor environment . in the a - free pb analogues , water molecules enter the first coordination sphere of the fe@xmath0 ions . there appears two kinds of anisotropic ligand configurations which are the intrinsic generant of the singlet and doublet orbital ground states . the fe@xmath0 sites are thus classified into two types . distribution of the anionic vacancies ( water - filled pores ) over the framework is characterized by the order parameter for two long - range order models pm@xmath26 m or im@xmath26 m . these models are relevant to the pb analogues of the 4:3 and 3:2 stoichiometries , respectively . two water molecules within the first coordination sphere of fe@xmath0 is the most populated configuration of the random model fm@xmath26 m and ordered models pm@xmath26 m or im@xmath26 m . the short - range order event occurrence in pb analogues of the 3:2 and 4:3 stoichiometries favors the cis- and trans- conformations , respectively . orbital singlet is the lowest in energy state for the cis - conformation . accordingly , in the a - free ferrous cobalticyanide pb analogue , the mssbauer spectra of fe@xmath0 are dominated by the larger quadrupole - splitting component . in the a - filled pb analogues , where the the water ligands are withdrawn from the fe@xmath0 coordination sphere , the least anisotropic charge distribution attainable for the cases of a = rb and cs is the orbital doublet ground state . for a = na and k , the thermally - induced interconversions between the doublet and singlet ground states were observed attributed to distortion isomerism . charge - transfered states intrinsic in the pb analogues with negative thermal expansion@xcite are suggested to be the source of the lattice distortions associated with the distortion isomerism . asymmetry of mssbauer spectra is attributed to the admixture of a small content ( at the level of @xmath1265% ) of the fe@xmath58 species associated with the charge - transferred states . since the octahedral sites fen@xmath16 possess a higher symmetry than the sites fen@xmath127(h@xmath4o)@xmath128 we observe upon the insertion the a - ions into a free compounds that the value of mssbauer qs drops in ferrous cobalticyanides similarly to the archetypal prussian blue fe@xmath81[fe(cn)@xmath3@xmath95h@xmath4o@xcite . however , when the large - size a - ions ( cs , rb ) replace the small - size ones ( na , k ) through ionic exchange we observe the larger qs associated with lower symmetry local distortion isomer ; the sixth d - electron of the fe@xmath0 cation is self - localized on a single sublevel ( orbital singlet ) of the t@xmath31 group owing to strong lattice distortions around the implanting sites of the low - mobility ions . this work was partly supported by the chinese academy of sciences visiting professorships for senior international scientists . 2011t1g15 . financial support obtained from the chinese academy of sciences for 100 talents project , the national natural science foundation of china ( no . 11079036 ) and the natural science foundation of liaoning province ( no . 20092173 ) is also greatly acknowledged . e. coronado , m.c . gimnez - lpez , t. korzeniak , g. levchenko , f.m . romero , a. segura , v. garca - baonza , j.c . cezar , f.m.f . de groot , a. milner . and m. paz - pasternak , j. am . chem . soc . * 130 , * 15519 - 15532 * * ( 2008 ) . the results tabulated in the above works@xcite are reported with different conventions for the sign of crystal field splitting @xmath76 . we avoid using the negative numbers for the crystal - field splitting energy . instead of @xmath129 , and sign of @xmath76 we use explicitly the notations of @xmath50 and @xmath61 , such that @xmath130 @xmath61 when orbital singlet is lowest , and @xmath49 , when the orbital singlet is lowest .

the plastic flow of a polycrystal is analyzed assuming grains as fine that the rate limiting process is grain boundary sliding , and grains readily accommodate their shapes by slip to preserve spatial continuity . it is shown that thinking of a polycrystal with randomly oriented grains as an homogeneous and isotropic continuum when dealing with it as a dynamical medium , even in a scale much larger than the grain size , leads to gross errors . the polyhedral nature of grains influences the plastic flow in a radical manner , as the relative velocity of adjacent grains is constrained to the common boundary plane , and only the in plane shear stress contributes to their relative motion . this constriction determines that , where is the velocity field of the material medium , and plastic deformation necessarily involves grain volume variations , which can only be elastic . as this has a limit , fracture follows as a necessary step of plastic flow . a theoretical approach to plastic flow is developed , and emphasis is done in superplastic deformation , from zero to fracture strain . the theory allows to quantitatively explain the observed features of superplastic materials and responds to many open questions in the field .

0911.3649

superplasticity is an anomalously high tensile ductility occurring in a growing class of fine grained polycrystalline solids . at rather precise conditions of composition , grain refinement , temperature and strain rate , they can undergo uniform neckless plastic elongations that may reach thousands of percent prior to fracture . these large strains occur with no significant change in the grain shapes or sizes , nor in any other fine attribute . as well , different bodies in the superplastic regime of deformation become prone to diffuse one into the other when their surfaces make contact ( diffusion bonding ) . among the many singular features of the phenomenon is the fairly narrow range of temperatures over which it happens . for instance , tensile tests for the superplastic titanium alloy ti6al4v give maximum ductility at temperatures close to 1153 k , and elongation to fracture rapidly decays to half the optimal figure within a neighborhood not wider than 200 k around @xcite . these temperatures are well below the melting point ( 1930 k ) , so we are talking about a genuine solid state effect . beside the scientific problem it poses , superplasticity has acquired considerable technical importance . after the pioneer work of backofen on the superplastic forming of zn - al sheets by a simple air pressure operation @xcite , as in the fabrication of plastic bottles , and the discovery of a number of high strength superplastic alloys , the subject moved into the scope of engineering materials for advanced applications . since then , the list of high strength superplastic alloys , intermetallic compounds and ceramics has increased notably @xcite , together with the technical knowledge for producing superplastic structures . a number of present day aluminium superplastic alloys , like al-7475 , al-8090 and supral 220 , or titanium ti-6al-4v and sp700 , have outstanding importance as aeronautic materials @xcite . superplastic forming combined with diffusion bonding is now a standard technique for producing both engine and airplane structural parts . the development of light ductile materials of high strength at elevated temperatures has a principal role in the race for more efficient aeroengines . superplastic titanium and nickel aluminides are drawing great attention in this respect @xcite . having half the density of the current materials , tial - based alloys are promising candidates to replace the nickel - based superalloys in combustion chambers and turbine blades . after a large plastic deformation with no significant microstructural changes , and resembling a steady flow , superplastic solids collapse at a rather well - defined strain , which depends on temperature and strain rate . the regular behaviour of superplastic elongation to fracture suggests that the phenomenon is governed by a specific variable , not identified so far , which keeps account for the total strain and produces failure when reaching a precise value . certainly , the development of cavities may contribute to precipitate collapse @xcite but seems not to be the main cause . maximum elongation to fracture strongly depends on the strain rate , while the evolution of cavities , when occurring , has been observed to be almost independent of that variable @xcite . moreover , titanium superplastic alloys do not cavitate significantly @xcite . for decades , people have been interpreting their observations on the basis of constitutive equations of the form @xmath2 , relating the uniaxial applied stress @xmath3 with the superplastic strain rate @xmath4 . the coefficient @xmath5 depends on the microstructure and follows an arrhenius dependence with temperature . a variety of theories give this kind of equation with constant @xmath6 , @xcite . nevertheless , experiments show that @xmath7 and @xmath8 are never linearly related , as implied by the power law , but the curves display a sigmoidal shape , which is a characteristic feature of the superplastic deformation , and the value of the strain rate sensitivity @xmath6 that fits the data near the inflexion point is usually a non - integral number which vary strongly with strain rate and strain . in many superplastic materials the large ductility is observed even for @xmath8 far from the inflexion point , where the slope @xmath6 has changed , and the strain rate at the inflexion point shifts considerably with strain @xcite . despite the success of some previous work in accurately predicting the temperature dependent stress _ vs. _ strain rate curves at zero or small strain for a number of superplastic solids @xcite , it rests to explain why are they superplastic . knowing the mechanisms that make the grains to slide past each other and the successful prediction of plastic properties at small strains are not enough to answer this question , and it is necessary to investigate how grains flow over long paths when the solid undergoes a finite deformation . the literature shows several different theoretical attempts to explain superplastic phenomena , but no one addresses the point of the modifications occurring inside the material when strained up to fracture @xcite , and the dependence of the stress _ vs. _ strain rate curves with strain . we put forward here a general theoretical approach for the plastic deformation of fine grained solids up to fracture strain . we assume the grain boundary sliding mechanisms of refs . and , but refine the study of the grain collective motions in the spirit of ref . . the latter is also modified in order to interpret better the physics of the sliding grains . as a general conclusion , our results prove that the deformation and fracture properties of superplastic solids are governed by general laws amenable of analytical mathematical description , simple and accurate enough to constitute a practical tool for interpreting material tests . though the assorted conglomerate of possible structural defects may affect in specific situations , the main mechanical behaviour of solids is determined by a few parameters associated to the most basic structural properties . at a scale much larger than the grain size , polycrystalline matter lacks symmetry constrictions and periodicity , and displays same average packing and properties in all directions , and over its whole extention . however , despite this , assimilating a polycrystal to an homogeneous and isotropic continuum may lead to gross errors , no matter the scale , when dealing with it as a dynamical medium . it has been proven before @xcite , and is revised in this section , that the faceted nature of the structural constituents of a polycrystal , the grains , influences the macroscopic plastic flow in a radical manner . to explain briefly how this does happen , assume that plastic flow takes place by just grain sliding , and that grains readily accomodate their shapes by slip , so the rate limiting process is the former . sliding of two adjacent grains occurs when the shear stress in the plane of the common boundary exceeds a critical value @xmath9 . however , no matter how big the externally applied forces may be , shear stresses vanish in planes whose normals are in the principal directions of the stress tensor . by continuity , shear stresses in planes whose normals are inside well defined finite solid angles around the principal directions are smaller than @xmath9 . hence , many grains are impeded to slide simply because of the orientation of their surfaces @xcite . by this effect , the material medium ceases to be dynamically isotropic , and cones are created by the very external forces within which no displacement is allowed . these forbidden cones destroy the kinematic balance and @xmath10 , where @xmath1 represents the velocity field of the flowing medium . hence , mass density is not locally conserved in the overall plastic flow . the consequent internal pressure build up helps deformation and , at a critical strain , the material becomes mechanically unstable , undergoing brittle or ductile fracture , depending on the precise physical circumstances . the internal pressure @xmath11 , which operates in the local microscopic scale , keeps account of cumulated strain and triggers fracture at a precise value of it . the proper configuration variables in the macroscopic scale are the components @xmath12 ( @xmath13 ) of the strain tensor , which , together with their time derivatives @xmath14 , give a complete picture of the dynamical state of the system . however , the evolution of @xmath12 in time is governed by forces excerted between each pair of adjacent grains , and hence the dynamical analysis demands to go first to the grain level . as the objective is to return to the larger scale , the first task is to establish the connection between the two descriptions . given the statistics of spatial orientations of the grain boundaries , which may be assumed isotropic from the beginning , one can derive the relation between @xmath14 and the set of the relative velocities @xmath15 between pairs of adjacent grains . to show this , temporarily place the origin o of the coordinate system at a point of a grain surface . then call @xmath1 the velocity of a point p in the positive @xmath16-coordinate axis , which is in another grain boundary distant @xmath17 from o , where @xmath18 is a large natural number and @xmath19 the mean grain size . ( in what follows it will be denoted either @xmath13 or @xmath20 , with @xmath21 ) . as there are @xmath18 contiguous grains between points o and p , @xmath1 is the sum @xmath22 where @xmath11 correlatively numbers the adjacent grains along the @xmath16-axis and @xmath23 is the velocity of grain @xmath11 relative to grain @xmath24 at the interface between them . the second term in the right hand side of eq . ( [ ec1 ] ) accounts for the assumed deformability of the grains , @xmath25 is the unit vector along the @xmath16axis and @xmath26 stands for the expansion rate of grain @xmath11 along the @xmath16direction . however , we are considering plastic deformation processes in which grains remain equiaxed in the mean and do not vary significantly their average size . therefore , on this basis , as @xmath18 is large we can set @xmath27 which means that , statistically speaking , the grains neither stretch nor shrink . hence @xmath28 if the solid is being deformed , the components of the displacement of p in a time @xmath29 are @xmath30 substituting @xmath17 and eq . ( [ ec2 ] ) in eqs . ( [ ec3 ] ) , it is obtained that @xmath31 where the @xmath16 dependence in the right hand side is implicit in the direction along which the relative velocities are sampled . ( notice that the present arguments leading to eq . ( [ ec4 ] ) do not assume that the grains are rigid , as seems implicit in the demonstration given in ref . , but that they conserve mean size and shape in the plastic flow ) . if the material is homogeneous and isotropic in the macroscopic scale , the @xmath18 plane grain interfaces intersecting op are oriented in any direction with the same probability . therefore , one may interpret the sum in eq . ( [ ec4 ] ) as an average which , as @xmath18 is large , runs over all possible orientations of the grain boundaries intersecting op . however , vectors @xmath15 represent relative velocities between consecutive grains ordered along the positive @xmath16-coordinate axis . thus , to take this into account , the average must be performed over all orientations in the hemispace @xmath32 . equivalently , denoting @xmath33 the unit vector normal to the interface between two grains , the sum in eq . ( [ ec4 ] ) can be replaced by an integral over all directions @xmath33 satisfying the condition @xmath34 . therefore , we can write @xmath35 where symbol @xmath36 means the average over all orientations in the semispace @xmath32 . if the material is not isotropic in the larger scale , for instance because the grains are not equiaxed , the average ( [ ec5 ] ) should incorporate the proper weigth factors . the stress tensor @xmath37 is diagonal in the main frame of reference @xmath38 , with @xmath39 ( @xmath40 ) being the principal stresses . to properly express the physics at the grain scale , define a local frame of reference @xmath41 whose @xmath42 plane is in the common grain boundary plane of two adjacent grains ( fig . [ fig1 ] ) . in the local reference system the stress tensor reads @xmath43 where @xmath44 is the rotation matrix that makes the transformation from the @xmath38 to the @xmath45 frames , and @xmath46 is the transposed of @xmath44 . local reference system @xmath45 , with the @xmath47 axis normal to the plane of the common boundary of two adjacent grains . the axes of the @xmath38 frame of reference are in the principal directions of the stress tensor.,width=207 ] as pointed out in ref . , superplastic grain boundary sliding seems closely related with elastic instabilities occurring at interfaces in solids when subjected to overcritical stresses @xcite . monte carlo simulation and analytical treatment show that , in general , planar structures inside the bulk of crystalline matter become rough when subjected to high enough in plane shear forces @xcite . after the occurrence of the elastic instability the locally induced normal stress field drives an active atomic transport between the interface and the two adjacent crystals . this bulk effect is closely related with the asaro tiller grinfeld instability , _ i. e. _ stress induced roughening of solid surfaces @xcite , and with the spontaneous roughening of thin films produced by strong in plane stresses arising from lattice mismatch with the substrate @xcite . it has been shown that the active atomic transport between the buckled grain boundary and the two adjacent grains follows closed paths , which makes the two grain surfaces to slide @xcite . the relative velocity between the two sliding grains turns out to be proportional to the in plane components , @xmath48 and @xmath49 , of the shear stress operating in the shared grain boundary plane , provided that the total in plane shear stress @xmath50 be higher than the threshold stress @xmath9 for the elastic instability of the grain boundary @xcite . hence @xmath51 where @xmath52 is a coefficient depending on the normal stresses @xmath53 which will be discussed in detail in the next section . the factor @xmath54 ensures that @xmath55 for @xmath56 . the solid line represents the linear relation assumed here between @xmath57 and the total in plane shear stress . the dotted line represents the model of ref . @xcite.,width=226 ] the coefficient @xmath52 must not depend on the orientation of the local frame of reference @xmath45 , therefore its dependence on the normal stresses is only via the invariant @xmath58 this is also a necessary and sufficient condition for considering @xmath59 as independent of the shear stresses for any orientation of the grain boundary . but for the critical condition incorporated in eqs . ( [ ec9 ] ) , @xmath52 may be seen as an inverse grain boundary viscosity coefficient @xcite . notice that @xmath60 in eqs . ( [ ec9 ] ) is related with an irreversible shift between the two grains , and hence can not describe the variation of an elastic degree of freedom . in what follows , elastic and plastic strains will be independent mechanical variables ( which nevertheless do interact ) . grain elastic distortions are always implicit in the subsequent developments , but plastic strains are the dynamical variables because grains are the flowing entities . elastic strains of the grains will remain implicit in the stresses , which are linearly related with elastic strains through the equations of the theory of elasticity . to put eqs . ( [ ec9 ] ) in a more practical way , we must perform the matrix transformation ( [ ec7 ] ) , choose the proper components of @xmath61 to insert them in eqs . ( [ ec9 ] ) and write @xmath15 in the local @xmath41 reference system , and then make the inverse vector transformation to express @xmath15 in the main @xmath38 coordinate system . hence , for @xmath62 the components @xmath63 are @xmath64_{i^\prime z^\prime } \left ( 1-\displaystyle\frac{\tau_c}{\tau_{(x'y')}}\right ) , \ ; i^\prime = x^\prime , y^\prime , \\ & \delta v_{z^\prime } = 0 , \label{ec10 } \end{aligned}\ ] ] where @xmath65_{i^\prime z^\prime}$ ] selects the @xmath66 component of the tensor . when expressed in the @xmath38 frame , the relative velocities ( [ ec10 ] ) are @xmath67 or , more explicitly , @xmath68 the local @xmath45 frame can be arbitrarily rotated around the @xmath69 axis without breaking the condition that @xmath69 be normal to the grain boundary plane . then , the @xmath70 axis can always be chosen on the @xmath71 plane , as in fig . [ fig3 ] . calling @xmath72 the euler angle between the @xmath70 and @xmath73 axes , and @xmath74 the other euler angle , going from the @xmath75 to the @xmath47 axis , the rotation matrix reads @xmath76 angles between the axes of the local reference system @xmath45 and the principal directions . the @xmath70 axis is in the @xmath71 plane.,width=188 ] transforming @xmath77 with the matrix ( [ ec12 ] ) one obtains , in particular , that @xmath78 and , making the inverse vector transformation ( [ ec11 ] ) , the components of the relative velocity between grains takes the explicit form @xmath79 \left ( 1-\frac{\tau_c}{\tau ( \theta , \phi)}\right)\\ \delta v_y & = \mathcal{q } \sin\theta \big [ ( \sigma_x -\sigma_y)\sin^2 \phi \cos\phi -(\sigma_x \sin^2 \phi\\ & + \sigma_y \cos^2 \phi -\sigma_z)\cos\phi\cos^2\theta\big ] \left ( 1-\frac{\tau_c}{\tau ( \theta , \phi)}\right)\\ \delta v_z & = -\mathcal{q } ( \sigma_x \sin^2 \phi\\ & + \sigma_y \cos^2 \phi -\sigma_z)\sin^2 \theta \cos\theta \left ( 1-\frac{\tau_c}{\tau ( \theta , \phi)}\right ) . \end{aligned } \label{ec14}\ ] ] eqs . ( [ ec14 ] ) hold when @xmath80 with @xmath81^{1/2 } , \end{aligned } \label{ec16}\ ] ] otherwise @xmath55 . to simplify the equations , let us particularize to the important case of uniaxial external stress @xmath82 . far from the sample surfaces , the material has cylindrical symmetry and @xmath83 , which will be assumed not to vanish for the sake of generality . ( [ ec14 ] ) and ( [ ec16 ] ) greatly reduce and give @xmath84 where the unit vector @xmath85 in the @xmath86 system is @xmath87 the choice of the @xmath70 axis in the @xmath71 plane proves to be particularly convenient because , by the cylindrical symmetry , any direction in the @xmath71 plane is a principal direction , and then @xmath88 for any @xmath74 and @xmath72 . the condition ( [ ec15 ] ) on the minimal shear stress @xmath9 becomes @xmath89 which determines a critical angle @xmath90 , given by @xmath91 such that @xmath92 for @xmath93 . therefore , under the restriction that @xmath94 , plastic flow is not allowed for @xmath95 , @xmath96 and @xmath97 , as shown in fig . [ fig4](a ) . ( a ) grain boundaries whose normals are in the spherical shells are the only ones allowed to slide , if the stress field has cylindrical symmetry with respect to the @xmath75 axis . grain boundaries whose normals make angles @xmath95 , @xmath96 and @xmath98 with the @xmath75 axis do not contribute to plastic flow . ( b ) , ( c ) and ( d ) integration domains for integral ( [ ec22]).,width=340 ] to calculate the averages @xmath99 , appearing in eq . ( [ ec5 ] ) , we must integrate over all directions @xmath100 such that the unit vector @xmath101 , normal to the grain boundary plane , has positive projection on the @xmath102 axis . vector @xmath33 is the unit vector @xmath103 along the @xmath69 axis of the local coordinate system , that is , @xmath104 in general , @xmath105 and the integration domains @xmath106 can be determined by examining the geometry of the @xmath107 transformation . they are ( fig . [ fig4](b),(c ) and ( d ) ) @xmath108\cup[\pi/2+\theta_c,\pi-\theta_c],\\ & \phi\in [ 0,\pi],\\ d_y : \ ; & \theta\in [ \theta_c,\pi/2-\theta_c]\cup[\pi/2+\theta_c,\pi-\theta_c],\\ & \phi\in [ \pi/2,3\pi/2],\\ d_z : \ ; & \theta\in [ \theta_c,\pi/2-\theta_c],\\ & \phi\in [ 0,2\pi ] . \label{ec23 } \end{aligned}\ ] ] integrating and combining with eq . ( [ ec5 ] ) one obtains @xmath109 , \label{ec24}\ ] ] and @xmath110 \\ & + \frac{\mathcal{q}\ , \tau_c}{\pi d}\cos ( 2\theta_c ) . \label{ec25 } \end{aligned}\ ] ] eqs . ( [ ec24 ] ) and ( [ ec25 ] ) hold for @xmath111 . when @xmath112 they must be modified reversing the sign of the terms multiplied by @xmath9 . by eq . ( [ ec20 ] ) , @xmath113 can be written in terms of just @xmath114 and the constant @xmath9 . replacing in eqs . ( [ ec24 ] ) and ( [ ec25 ] ) , the strain rates turn out to be entirely determined by @xmath114 , which becomes a natural and convenient auxiliary variable which varies in the range @xmath115 for finite @xmath9 , @xmath116 corresponds to the asymptotic situation in which the external forces applied on the sample are very strong . for @xmath117 , the forbidden directions extend over the whole sphere and no grain is allowed to slide . as the relative velocities , the strains @xmath12 are purely plastic , but elastic strains are not disregarded and remain implicit in the stresses , which are linearly related with elastic strains through the equations of the theory of elasticity . from eqs . ( [ ec24 ] ) and ( [ ec25 ] ) , the dilation rate @xmath118 turns out to be @xmath119\ , , \label{ec27 } \end{aligned}\ ] ] where @xmath120 stands for the volume of the sample . ( [ ec27 ] ) discloses the amazing feature that the dilation rate does not vanish in general if the threshold stress for grain boundary sliding @xmath121 . if @xmath122 eqs . ( [ ec24 ] ) and ( [ ec25 ] ) reduce to @xmath123 @xmath124 and the material flows as a viscous liquid . the dilation rate ( [ ec27 ] ) vanishes only in the extreme cases @xmath116 and @xmath125 , and for @xmath126 is negative for any other value in between . this means that plastic stretching involves a spontaneous compression of the grains , which can only be elastic , and demands the ability of the material to support some elastic volume reduction . as this ability is essentially limited , plastic strain is limited as well , and a finite strain to failure is implicit in eq . ( [ ec27 ] ) . the single - crystal elastic compressibility will determine how much the polycrystal could be plastically elongated . another general consequence of eq . ( [ ec27 ] ) is that , in general , the transversal stresses @xmath127 do not vanish in the plastic deformation , even when the applied external force is strictly uniaxial , and the material is isotropic in a scale much larger than the grain size . this may seem rather odd if one has in mind a static , or quasistatic , picture of the deformation process . such a picture is , however , misleading . plastic flow is an essentially dynamic process . it must be noted that eq . ( [ ec27 ] ) deals with volume changes , which are necessarily elastic , but has nothing to do with grain shape variations . our primary hypothesis is that grains are plastic , and they readily accommodate their shapes to preserve matter continuity , but grain boundary sliding is the rate limiting process . this is consistent with accomodation by slip , since crystal deformation by slip does not demand much additional stress once the critical resolved shear stress ( crss ) has been exceeded . then , if @xmath9 is greater than the crss of the material then our hypothesis about grain plasticity should hold . we are not very specific about the accommodation mechanisms because grain plastic deformation is not the rate limiting process , diffusional creep or dislocation creep may contribute , but we assume that the main accomodation mechanism is by slip processes . settling @xmath128 for @xmath129 is a natural initial condition . however , it will be shown that , as the plastic stretching along the @xmath75 axis increases with time , the magnitude of the transversal stresses @xmath127 monotonically increases as well , with negative sign . hence , the plastic elongation of the sample produces a spontaneous compression in the transversal plane . neck formation prior to ductile fracture of polycrystalline bars can be attributed to this circumstance . the forbidden directions for @xmath15 . vectors @xmath15 in the solid angles a and c correspond to grain boundaries with @xmath33 in the solid angle b , and vice versa . then , there are more vectors supressed in a and c than in b.,width=226 ] the dilation rate ( [ ec27 ] ) can be understood in a qualitative manner by a simple geometric argument . the relative speed @xmath130 between adjacent grains depends only on @xmath74 and is symmetric with respect to @xmath131 in the top hemisphere , and with respect to @xmath132 in the bottom one ( eq . ( [ ec17 ] ) ) . [ fig5 ] depicts the solid angles a , b and c in which @xmath133 . grain boundaries whose normals @xmath33 are in b correspond to banned relative velocities @xmath134 in a or c , and vice versa . as the solid angle b is larger than the sum of a and c , there are more supressed vectors @xmath15 in a and c than in b. therefore , grain displacements in the @xmath75 direction , or close to it , are less favoured than those occuring in directions near the @xmath71 plane . if no dilation takes place when @xmath135 ( i. e. @xmath136 ) , some volume change should occur for @xmath137 . the same kind of argument serves to show that no dilation does occur in two dimensions , and the area is conserved in the plastic distortion of a strictly two dimensional layer . we will examine the case of the uniaxial plastic strain of a bulk sample along the @xmath75 axis at a strain rate @xmath138 . the only externally applied stress is @xmath139 . we will denote also @xmath140 for the pressure . the transversal stresses @xmath141 are assumed finite to give a chance for internal stresses to develop . of course , cylindrical symmetry demands that the point into consideration be far enough from the surfaces . with this notation @xmath142 recalling hooke s law @xmath143 where @xmath144 is the bulk elastic modulus , the dilation rate reads @xmath145 replacing the identity @xmath146 and eq . ( [ ec27 ] ) in eq . ( [ ec30 ] ) , @xmath147 , \label{ec31 } \end{aligned}\ ] ] where use was made of eq . ( [ ec20 ] ) to eliminate @xmath148 and put the right hand side in terms of just @xmath114 . the factor @xmath149 is @xmath150 for traction ( @xmath151 ) , the right hand side of eq . ( [ ec31 ] ) is positive for @xmath152 , and hence the pressure @xmath11 increases monotonically with strain . elastic degrees of freedom , implicit in @xmath153 , are brought to the fore once more by eq . ( [ ec27 ] ) . eq . ( [ ec30 ] ) expresses the volume variation ( the trace of the elastic strain rate tensor ) in terms of the pressure @xmath11 ( essentially the trace of the stress tensor ) . hence it allows to return to the scheme in which the elastic distortions are kept implicit in the stresses . the variation of @xmath11 in the plastic deformation indicates that elastic and plastic degrees of freedom are coupled . written in terms of just the critical angle @xmath114 , eq . ( [ ec24 ] ) for the strain rate becomes @xmath154 , \label{ec33}\ ] ] with @xmath155 in what follows we will consider just traction , and @xmath149 will be omitted . at the start of the deformation process @xmath156 and @xmath157 . then the initial condition written in the new notation is @xmath158 the coefficient @xmath52 , governing grain boundary sliding , was studied in detail in refs . @xcite and @xcite . it was shown there that the stress dependence of @xmath52 is only through the pressure @xmath11 , and varies with temperature and grain size . it obeys the rather simple law @xmath159 where @xmath160 is the boltzmann constant , @xmath161 the temperature , the coefficient @xmath162 depends only on the grain size @xmath19 , the constant @xmath163 is the energy necessary for evaporating a crystal vacancy from the grain boundary , and @xmath164 measures the sensitivity of this energy to stress . the derivation of expression ( [ ec35 ] ) has been already published @xcite , but a brief explanation of the physics involved is in order here . the mechanical analysis of a stressed polycrystaline material must distinguish intergranular and crystalline matter because they have different mechanical properties @xcite . a shear stress greater than a critical value applied in the plane of a grain boundary should buckle it , producing either a boundary corrugation or the formation of a periodic series of trenches . this localized deformation induces a periodic normal stress field that alternates compression and traction on the adjacent crystal surfaces , as shown schematically in fig . [ fig6](a ) . grain boundaries are efficient sinks and sources for vacancies and the periodic stress field induced by the buckled boundary yields a periodic variation of the equilibrium value for the concentration of crystal vacancies . the grain boundary then evaporates and condenses the point defects in alternate sectors , producing streams of the defects in closed loops that cross the boundary and involve the two adjacent crystals ( fig . [ fig6](b ) ) . these closed loops perform as the driving pulleys of a conveyor belt and the phenomenon provides either an accommodation mechanism and the driving force for crystal sliding @xcite . ( a ) schematic representation of a buckled grain boundary , and the induced periodic stress field alternating compression and traction regions in the adjacent crystals . ( b ) the grain boundary releases and captures crystal vacancies in the tractioned and compressed regions , respectively , producing this way closed paths of the point defects . the diffusion of vacancies , equivalent to an atomic counterflow , have opposite senses in the two grains.,width=283 ] as explained , the grain boundary elastic instability causes a strongly fluctuating stress field , which in the macroscopic scale averages to zero . be @xmath165 the energy necessary for the grain boundary to release a vacancy into one of the two crystals adjacent to it , when the system is unstrained . when the grains are strongly stressed , this energy is expected to change linearly with stress , as long as @xmath165 is much larger than its expected variations . denoting @xmath166 the characteristic amplitude of the induced stress field spatial fluctuations , the relative concentration of defects in two neighbouring regions of maximal and minimal stress is proportional to @xmath167 -\exp \left [ -\frac{\epsilon_b+\omega^*(p - p^1)}{k_bt}\right ] , \label{ec36}\ ] ] where @xmath168 is the contribution from the induced stress field to the pressure . expression ( [ ec36 ] ) depends only on the normal stresses @xmath169 , @xmath170 and @xmath171 because it is expected that the energy of a vacancy will not be significantly affected by the shear stresses , which do not change the volume of the hole , but only its shape . the flow of defects between these two regions is proportional to the spatial concentration variation ( [ ec36 ] ) , which can be written as @xmath172 for @xmath173 . in eq . ( [ ec37 ] ) @xmath174 $ ] is the diffusion coefficient for crystal vacancies , with @xmath175 being the corresponding activation energy . no matter the details of the process , one can expect that the speed of the boundary sliding be proportional to the rate at which the assisting vacancies flow along the crystal boundaries , and expression ( [ ec37 ] ) leads to the main eq . ( [ ec35 ] ) with @xmath176 . by the time , direct experimental evidence was supplied by vetrano _ @xcite for the deformation induced supersaturation of vacancies in the vicinity of many grain boundaries of an al mg mn alloy during superplastic deformation . rapid quenching of the sample while being superplastically deformed revealed the formation of very unstable nano cavities in a great fraction of the grain boundaries , which rapidly coalesce and disappear on moderate heating . the authors show they start forming after cooling and subsequent mounting of the sample in the transmission electron microscope , and conclude that the cavities provides evidence of a supersaturation of crystal vacancies near the sliding grain boundaries . the defects condense into voids when sliding abruptly ceases , and the just released vacancies are not recaptured by the grain boundaries . hence grain boundary sliding is associated to the observed vacancy excess , which provides evidence of the capital role of these defects in the deformation process . some aspects of the mechanism for grain boundary sliding introduced in refs . and , and reviewed above in a more compact approach , resembles the one put forward earlier by raj and ashby @xcite . in both the shared grain boundary is taken as a material medium able to release and capture crystal vacancies , to or from the two grains , causing a relative motion of them . also , in both theoretical approaches the in plane shear force induces a strongly varying normal stress field localized in the boundary region and , as a consequence , the less ordered intergrain matter evaporates and condenses vacancies in a periodic series of sources and sinks . the main difference is in the physical nature of the sources and sinks for the point defects . in refs . and the sources and sinks are originated by the buckling and trenching of the initially plane grain boundary ( with eventual steps ) by effect of the overcritical shear stress applied to it . raj and ashby @xcite assume an uneven grain boundary subjected to a shear force . sources and sinks are caused by the variation of the normal stresses induced in the opposite sides of the permanent grain boundary irregularities . when the plastic distortion just starts , @xmath177 and holds the initial condition ( [ ec34 ] ) , @xmath178 . then @xmath179 and , replacing this and eq . ( [ ec35 ] ) in eq . ( [ ec33 ] ) , @xmath180\\ & \times\exp\left ( -\frac{\epsilon_0-\omega^*\sigma /3}{k_bt}\right ) \qquad ( \varepsilon\approx 0 ) . \label{ec39 } \end{aligned}\ ] ] eq . ( [ ec39 ] ) gives the plastic strain rate @xmath4 of an axially symmetric sample at temperature @xmath161 under the action of the uniaxial tensile stress @xmath3 . hence it is the constitutive equation of the material for the established conditions . the discussion on grain sliding of sections [ strainrates ] and [ dilationrate ] is valid for any polycrystalline solid . also , there is no reason to think that the grain boundary sliding mechanisms leading to eq . ( [ ec35 ] ) for the coefficient @xmath52 be exclusive to superplastic grain sliding . hence , in principle , eq . ( [ ec39 ] ) is expected to hold for the deformation of any fine grained polycrystalline solid , fine enough to ensure that grain boundary sliding be the rate limiting deformation process . however , the large amount of experimental data available in the literature on the superplastic materials makes them specially suited for testing eq . ( [ ec39 ] ) . the data for very small strain rates , in the range @xmath181 to @xmath182 , are specially valuable because the effect of the critical stress @xmath9 on the stress strain rate curves is particularly notorius in it . such small strain rates are not usual in tensile tests of normal , not superplastic , materials . to appreciate how @xmath9 affects the strain _ vs. _ strain rate curves , consider eq . ( [ ec39 ] ) in the asymptotic limit of very large tensile stress @xmath3 , such that @xmath183 . the equation reduces to @xmath184 which is the result given in ref . , in which the effect of the critical stress @xmath9 discussed in sections [ strainrates ] and [ dilationrate ] was passed over . this produces no significant consequence when comparing the theoretical results with data on alloys having small @xmath9 , as is the case of the aluminium al7475 samples whose mechanical testing was reported by hamilton @xcite . however , the data on the same alloy al7475 prepared by pilling and ridley @xcite , aluminium al8090 spf @xcite or titanium ti6al4v @xcite , do show some disagreement with eq . ( [ ec40 ] ) for strain rates below @xmath185 . as an illustrative example , fig . [ fig7 ] shows the excellent fit given by eq . ( [ ec39 ] ) to the data of cope _ et al . _ on ti6al4v @xcite together with the curve representing eq . ( [ ec40 ] ) with the same values for the parameters . the latter has negative concavity over the whole range of @xmath4 and does not exhibit the characteristic inflexion point , which is traditionally taken as a distinctive feature of superplasticity . however , both curves become practically the same for @xmath186 . curves given by eqs . ( [ ec39 ] ) and ( [ ec40 ] ) , showing the effect of the critical shear stress @xmath9 . values for the parameters are shown in the insets . circles represent data of cope _ et al . . , on the superplastic titanium alloy ti6al4v , strained at constant true strain rates and @xmath187.,width=302 ] figure [ fig7 ] also illustrates the usefulness of very low strain rate data . as will be seen in the next sections , the critical shear stress @xmath9 has important consequences in the plastic deformation at finite strains and high or low strain rates , as fracture , for example , but does not produce any easily recognizable feature in the test curves . on the contrary , at low enough @xmath4 , tensile stresses are close to @xmath188 , and the solid angles a , b and c in fig . [ fig5 ] have significant magnitudes , producing the inflexion point shown in fig . [ fig7 ] . as the deformation mechanisms considered here are in principle not exclusive to superplastic materials , we guess that the strain _ vs. _ strain rate curves for normal ductile materials should exhibit and inflexion point as well , at low enough strain rates . in a subsequent paper @xcite , eq . ( [ ec40 ] ) was modified in the way @xmath189 and an excellent agreement with the experimental results was achieved at low , intermediate and high @xmath4 . however , the introduction of the threshold tensile stress @xmath190 was based on just intuitive physical arguments , and justified only by the good results it produced . the pre exponential factor in eq . ( [ ec39 ] ) behaves like @xmath191 with @xmath192.,width=245 ] fig . [ fig8 ] explains the success of eq . ( [ ec41 ] ) . it shows that , despite its more involved appearance , the pre exponential factor in eq . ( [ ec39 ] ) is almost linear in @xmath3 , and vanishes for @xmath193 , which plays the role of @xmath190 in eq . ( [ ec41 ] ) . therefore , eq . ( [ ec41 ] ) constitutes a very good approximation of the exact eq . ( [ ec39 ] ) , and the same accurate fit of the experimental data on al7475 , al8090 spf @xcite , and ti6al4v @xcite attained with the former @xcite is to be expected for the latter , but now with no empirically justified modification . [ fig9 ] shows the data of cope _ et al . _ @xcite on the superplastic titanium alloy ti6al4v at six temperatures ( circles ) . the solid lines represent eq . ( [ ec39 ] ) . the constants @xmath164 , @xmath163 and @xmath162 have the same values for the six curves , and are shown in one of the insets . notice that the values for @xmath164 , @xmath163 and @xmath162 used here in relation to eq . ( [ ec39 ] ) are the same than those used in ref . to insert in eq . ( [ ec41 ] ) , with the same success . the critical shear stress @xmath9 evidences some dependence with temperature , as shown in the other inset . strain rate curves for ti6al4v at six temperatures . circles represent data of cope _ et al . _ , refs . , and solid lines depict eq . ( [ ec39 ] ) for the parameters shown in the inset . the parameter @xmath9 exhibits some temperature dependence.,width=302 ] the critical shear stress @xmath9 is expected to be temperature dependent . it may be equal to either the smaller of the two critical stresses associated to the grain boundary elastic instabilities studied in refs . and , or the critical resolved shear stress ( crss ) of the material when it is greater than the critical stresses for the elastic instabilities . the critical stresses for the elastic instabilities depend on the elastic modulus of the material , which varies with temperature , and the crss is temperature dependent as well when the temperature is high enough . for ti-6al-4v it seems that @xmath9 is the crss . deformation up to a finite strain @xmath194 can be accomplished in an indefinite number of inequivalent ways because of the other two variables involved in the process , @xmath3 and @xmath4 . establishing the time dependence of one of them , or its dependence on the intermediate values assumed by @xmath194 , would be enough to determine the process . the simplest conditions are to keep @xmath195 , as is usually done when testing superplastic materials , or set @xmath196 , which is the common practice for normal ductile materials . in this section we will assume the former . the adimensional coefficient appearing in eq . ( [ ec48 ] ) establishes the scale for @xmath194 and is a very small number . the bulk elastic modulus @xmath144 is close to @xmath213 for aluminium and @xmath214 for titanium at normal room temperature . from the comparison of our curves with experiments for @xmath177 , we know that @xmath164 is @xmath215 for al8090 and @xmath216 for ti6al4v at rather high temperatures , however , @xmath164 is not expected to vary with @xmath161 . hence , at @xmath217 , because of the small value of the coefficient ( [ ec51 ] ) , any significative strain @xmath194 demands that either @xmath204 or @xmath114 , or both , be in one of the two asymptotic regions @xmath219 or @xmath220 . the asymptotic region @xmath220 can be associated to creep , while @xmath221 is related to ductile and superplastic deformation . for example , to attain a strain @xmath222 it must be @xmath223 . observing fig . [ fig10 ] one readily realizes that such big differences can only exist in the asymptotic regions of @xmath206 . the procedure outlined in the preceding subsection for determining the exact dependence of the flow stress on strain , strain rate and temperature relies on solving a set of trascendental equations . however , a rather simple closed form constitutive equation @xmath231 can be written if one finds acceptable to restrict the scope to the strain rates for which eq . ( [ ec40 ] ) is valid . this means @xmath232 for ti6al4v , which is a typical figure of more or less general validity . the coefficient @xmath233 appearing in eq . ( [ ec48 ] ) is in general a very small adimensional quantity . as @xmath206 is an universal function exhibiting large variations only close to its sigularities at @xmath209 and @xmath210 , significant strains @xmath194 demand @xmath234 or @xmath235 . in the latter eventuality the stress is near the minimum necessary to produce plastic flow , which progresses at a very low rate and corresponds to _ creep_. the former case is the ductile deformation we are interested in . where @xmath238 is an integration constant determined by the condition ( [ ec34 ] ) . isolating @xmath114 from eq . ( [ ed2 ] ) and inserting in eqs . ( [ ec52 ] ) and ( [ ec53 ] ) , both rewritten for @xmath239 , it gives which is the general constitutive equation @xmath245 we were searching for . if one replaces in eq . ( [ ed6 ] ) @xmath246 and @xmath247 , the approximate constitutive equation ( [ ec40 ] ) for @xmath246 is recovered . the domain in which the approximation holds is apparent in fig . [ fig7 ] , where the curves given by eq . ( [ ec40 ] ) and the exact eq . ( [ ec39 ] ) are compared . derivating and combining properly the equations written in subsections [ strainfunction ] and [ straindependentstresses ] one can deduce a closed form equation for the slope @xmath249 at @xmath156 . in general , for any @xmath194 , which may be considered as an analytical condition for superplasticity . although this conclusion is mostly empirical , it is quite reasonable because , as a polycrystal can not flow steadily by the monotonic increase of the internal pressure @xmath11 , eq . ( [ er24 ] ) expresses the situation which best emulate a steady regime . the next example illustrates the point . the uniaxial flow stress @xmath3 for a modified al 5083 alloy deformed at @xmath253 and constant strain rate @xmath4 . circles represent the data of kaibyshev _ et . al . _ ( ref . ) and solid line depicts eq . ( [ ec39 ] ) with the material constants shown in the insets.,width=264 ] circles represent stress strain data at constant strain rates @xmath254 and @xmath255 for the same system of fig . [ fig11 ] . the straight lines are the tangents to the stress strain curves at @xmath156 and the same strain rates , calculated with eq . ( [ er23 ] ) . the effective bulk modulus is @xmath256.,width=264 ] fig . [ fig11 ] shows the variation of flow stress with strain rate for a modified 5083 aluminium alloy , deformed at @xmath257 and constant true strain rate . the experimental points are due to kaibyshev _ et . @xcite and the solid line depicts eq . ( [ ec39 ] ) with the constants shown in the insets . these authors observed a sharp maximum in elongation to failure of 1150% for @xmath257 and a strain rate @xmath258 . inserting in eq . ( [ er23 ] ) the constants obtained from the fit of the stress strain rate data , shown in fig . [ fig11 ] , the slope @xmath249 of the stress strain curves at @xmath156 can be calculated for any initial stress @xmath3 . the solid dashed straight lines drawn in fig . [ fig12 ] represent the calculated tangents to the stress strain curves at @xmath156 starting from @xmath259 , and @xmath260 . according to eq . ( [ ec39 ] ) these three stresses correspond to the strain rates @xmath261 and @xmath255 . the circles in fig . [ fig12 ] represent the experimental stress strain data taken by kaibyshev _ et . al . _ at constant strain rates @xmath262 and @xmath255 . the consistency between theory and experiment is apparent in fig . [ fig12 ] . the effective bulk modulus , which fixes the scale for @xmath194 , was taken as @xmath256 . best superplastic response of the material is attained at a strain rate @xmath263 , for which @xmath264 at @xmath156 . in a recently published experimental study of the plastic tensile deformation of ti6al4v , vanderhasten _ et . _ find out that positive concavity seems to be the fingerprint of grain boundary sliding @xcite . these authors conducted tests starting from room temperature to 1323 k , recording tensile properties , microstructural evolution and crystallographic texture changes at a constant true strain rate @xmath265 . the tests seem not intended to attain optimal superplastic performance , but clearly show a dramatic increase of the strain to fracture at @xmath266 ( @xmath267 , which reaches the optimum ( @xmath268 ) at @xmath269 ( @xmath270 ) . starting from @xmath271 , grain boundary sliding begins to be dominant and , coincidently , the corresponding stress strain curve exhibits positive concavity over the most of the strain domain . the lower temperature range is characterized by negative concavity and grain stretching . at temperatures between @xmath269 and @xmath272 the stress strain curves have positive concavity over the whole range , from @xmath156 to fracture at @xmath273 . for higher temperatures the strain to fracture decreases and the stress strain curves develop a final sector with small negative curvature before failure . circles represent data for ti6al4v of vanderhasten _ et . _ at the the temperatures for which grain sliding dominates over grain stretching . solid lines were obtained from eq . ( [ ed6 ] ) and the same set of parameters , which are shown in the insets . just @xmath162 shows some variation with @xmath161 , probably due to its grain size dependence . the effective bulk modulus is @xmath256.,width=302 ] the open circles in fig . [ fig13 ] represent experimental data taken by vanderhasten _ on the stress strain response of the titanium alloy ti6al4v at the three temperatures for which the deformation is observed to take place by grain sliding @xcite . the strain rate was kept at @xmath274 in all runs . the samples exhibit high ductility but not properly superplasticity . the solid lines represent the asymptotic expression ( [ ed6 ] ) for the three temperatures , with the parameters shown in the insets and @xmath256 . ( [ ed6 ] ) can describe with high precision the experimental situation , which can be shown by comparing the values for @xmath3 at @xmath156 given by eq . ( [ ed6 ] ) with the exact value @xmath190 , obtained from eq . ( [ ec39 ] ) to introduce it into the same eq . ( [ ed6 ] ) as a one of its constants . the agreement between theory and experiment is remarkable up to @xmath275 . [ fig14 ] shows the stress strain rate curves given by eq . ( [ ec39 ] ) and the same material constants used to fit the data of fig . [ fig13 ] . there is no experimental points to compare with since vanderhasten _ et . _ employed a unique strain rate . the curves exhibit the typical sigmoidal shape . same as fig . [ fig13 ] , but with @xmath162 replaced by @xmath276 , as given by eq . ( [ er25 ] ) with the values of @xmath277 shown in the inset . @xmath276 models the effect of grain size variations.,width=302 ] in general , @xmath162 depends on the grain size @xmath19 as @xmath278 and the rest of the parameters do not vary with @xmath19 @xcite . [ fig15 ] shows an attempt to explain the departure between theory and experiment , starting at @xmath275 , by the dynamic modification of the grain size . the curves represent eq . ( [ ed6 ] ) with @xmath162 substituted by where @xmath277 is a constant . the values of @xmath277 giving the best fit to experiment are shown in one of the insets of fig . [ fig15 ] . amazingly , @xmath277 turns out to be such that @xmath280 at the fracture strain . this means that failure occurs because the grains become unable to slide , which can hardly be explained by just their size variations . the explanation may be as follows : the mean grain sizes measured by vanderhasten _ al . _ at 1123 and 1173 k show a rather irregular reduction on strain . the consequent division of old grains involves the creation of new grain surfaces which do not have necessarily the composition and structure of the old ones . grain boundaries are efficient traps for impurities , and can condense atomic species having a low bulk concentration . however , the process is governed by bulk diffusion and is not rapid . therefore , concerning their composition and consequent sliding ability , fresh grains may be very different than well aged ones . from this viewpoint , grain size variations may simply reduce the number of grains able to slide . this explains also why the smooth stress strain curves do not reflect the irregularities observed in the grain size strain curves . the theoretical approach to finite strain superplasticity put forward here permits to explain quantitatively the main features observed in the superplastic deformation . however , the theory lacks an important element : the effect of heat generation at the sliding grain boundaries . in our scheme , the whole work made by the external forces is dissipated into heat at the sliding grain boundaries , which occupy a very small volume when compared with the grain bulk . hence one may expect that grain boundaries will be at a higher temperature than bulk matter , being the nominal temperature the bulk one . grain boundaries whose normals are within solid angles , determined by @xmath114 , around the principal stress directions do not slide because the in plane shear stress is smaller than @xmath9 , and then remain at the bulk temperature . bending downward of the stress strain curves at high strains and strain rates seems to be a grain boundary temperature effect . the scheme described thus far should hold for single phase alloys . materials combining two or more allotropic phases in thermodynamic equilibrium may display a more complex phenomenology . in particular , superplastic materials are generally two phase or pseudo single phase alloys . probably the most characteristic example of the former is ti6al4v , which combines an @xmath281 hcp solid phase with a @xmath282 phase with bcc structure , in comparable volume fractions up to the @xmath282 transus temperature . cope el al . @xcite observed best superplastic response of ti6al4v for a temperature @xmath283 , for which the @xmath282 phase proportion was 0.42% . in pseudo single phase materials the second phase occurs as small inclusions between the grains constituting the main one . the new feature introduced by the coexistence of two allotropic phases in thermodynamical equilibrium is that a volume reduction is not necessarily related with just the elastic properties of the material . if the two phases ( which we will call generically as @xmath281 and @xmath282 ) have different specific volumes , the total volume may change also by the transformation of part of one of the phases into the other one . in general , the atomic fraction of , for instance , the @xmath282 phase where @xmath285 is the number of atoms of the main constituent of the alloy in the @xmath282 phase and @xmath286 is the total , is a funtion of the temperature @xmath161 and pressure @xmath11 . if @xmath11 changes with time , the atomic fraction of the phases changes , as well as the total volume @xmath120 . hence , the dilation rate is no more given by eq . ( [ ec30 ] ) but by eq . ( [ ec56 ] ) assumes that the atomic fractions have their equilibrium values in any instant , and hence a quasi static process , otherwise the transformation kinetics would come into the fore . the total area of the interfaces between the phases is maximal when their volumes have the same value , and then comparable volume fractions contribute to increase the phase transformation rates , and the accuracy of eq . ( [ ec56 ] ) . also , the new isothermal coefficient appearing in eq . ( [ ec56 ] ) is in general not a constant , and it is expected to be much more sensitive to temperature than @xmath290 . with this in mind we define the elastic modulus @xmath144 takes in general large values ( typically @xmath295 for metals ) , which is not necessarily the case of @xmath293 . for two phases coexisting in thermodynamical equilibrium one may expect that the volume fraction of each one would vary significantly for pressure variations much smaller than the yield stress . then , values of less than 1 mpa for @xmath293 are quite reasonable . in such a situation @xmath296 and @xmath297 is several orders of magnitude larger than @xmath290 . by this reason solids exhibiting two allotropic phases in equilibrium may display a special plastic behaviour , of which superplasticity seems to be an example . the abnormally large strains to failure displayed by superplastic materials have been attributed to the better stability they have against necking in tensile tests . the plastic flow can be represented by the relation where @xmath6 is a function of @xmath4 and @xmath194 , generally known as the strain rate sensitivity , which is always smaller than unity . when @xmath299 any irregularity in the cross section of the sample is not accentuated by the deformation in a tensile test , necking has no effect and the material behaves as a newtonian viscous fluid . necking instability is high for @xmath6 below 0.5 and decreases markedly when @xmath300 . indeed , the condition of high values of @xmath6 is a necessary requirement , but not enough to reach elongations as large as those termed superplastic . our previous discussions show that plastic elongation has a limit because demands a volume reduction , and failure is not a matter of just a mechanical instability of geometric origin . therefore , in our view , superplasticity has its real origin in three concurrent features of the material and physical conditions of the deformation process : ( a ) a high strain sensitivity @xmath6 . ( b ) the ability of the material to reduce the internal pressure @xmath11 by undergoing a phase transformation to a phase of higher density . ( c ) the material must be in the verge of a ductile to brittle transition . this latter requirement is explained below , taking the superplastic behaviour of titanium ti6al4v to illustrate the argument . the general properties of the stress _ vs. _ strain curves can be visualized from just examining eq . ( [ ed6 ] ) for the strain dependent flow stress @xmath3 . it has a first term linear in @xmath194 , with slope proportional to @xmath301 , and a negative logarithmic term whose coefficient increses with @xmath161 . hence , at low temperatures @xmath3 is linear in @xmath194 and , as @xmath161 increases , the slope decreases and the function develops positive concavity . beyond a critical temperature for which the slope at the origin vanishes the curve starts at @xmath156 whith negative slope and , after surpassing a minimum , or second yield point , begins to increase monotonically . hence , eq . ( [ ed6 ] ) predicts strain hardening at any strain for temperatures below a critical temperature , which roughly coincides with the temperature which optimizes superplastic properties . at higher temperatures the material should weaken for strains between @xmath156 and the second yield point . this configures a mechanically unstable situation which can not occur in practice in a smooth and homogeneous way . the equations written so far describe the mechanical response at a point , which does not necessarily can be straightforwardly extended to the whole sample . homogeneity is a quite reasonable assumption to study averaged properties in the macroscopic scale , but may fail in critical situations . we can safely assume that @xmath164 and @xmath163 are uniform and constant throughout the material , but @xmath162 depends on the grain size as @xmath278 @xcite , and hence may fluctuate or have a multimodal distribution , following grain size . there is a class of mild steels @xcite and aluminium alloys @xcite exhibiting stress strain curves with two yield points . the plastic deformation of these materials is inhomogeneous and takes place by a series of succesive sudden band formations ( lders bands @xcite ) , transversal to the tensile axis and making a rather precise angle with it ( of about @xmath302 ) , which concentrate the strain . the flow stress keeps constant ( though displaying a serrated aspect ) until the bands , increasing in number and extension , fill the whole sample and the complete material becomes strained beyond the second yield point strain . just then the stress starts to increase . in the intermediate steps , samples have sections with no strain and sections strained beyond the second yield strain . the relevant question is then how a system with the stress strain curves given by eq . ( [ ed6 ] ) will evolve with strain , and what kind of effective stress strain curves could be expected from such evolution . if our material is not exactly homogeneous the critical temperature for the onset of the second yield point is not the same in its whole extension . hence , at temperatures close to these critical temperatures and small strain one may expect the solid will be partitioned in a network of sectors hardening on strain , and another one which weakens . as the stress is uniform , the weakening sectors will deform much faster , and will be responsible for the main part of the effective strain before reaching their second yield point . meanwhile the hardening sectors will furnish mechanical stability against sudden yield steps , like lders band generation . anyway , the condition for superplastic behaviour seems related with the onset of the second yield point , producing a transition to a potentially unstable system . such a transition begins at the temperature for which @xmath249 at @xmath156 vanishes and changes from positive to negative sign . on the other hand , generally speaking , the most basic condition for a steady flow , as superplastic deformation tends to be , is that the dynamical variables remain constant and independent of @xmath194 , which here represents time . as @xmath3 varies with @xmath194 , all what can be done is to emulate a steady flow by providing the conditions that keep @xmath248 small and steady in a range as large as possible , which incorporates @xmath156 . j. s. vetrano , c. h. henager and e. p. simonen , in _ superplasticity current status and future potential _ , edited by p. b. berbon , m. z. berbon , t. sakuma and t. langdon . ( materials research society , warrendale , pa , 2000 ) .

usual approach to the foundations of quantum statistical physics is based on conventional micro - canonical ensemble as a starting point for deriving boltzmann - gibbs ( bg ) equilibrium . it leaves , however , a number of conceptual and practical questions unanswered . here we discuss these questions , thereby motivating the study of a natural alternative known as quantum micro - canonical ( qmc ) ensemble . we present a detailed numerical study of the properties of the qmc ensemble for finite quantum systems revealing a good agreement with the existing analytical results for large quantum systems . we also propose the way to introduce analytical corrections accounting for finite - size effects . with the above corrections , the agreement between the analytical and the numerical results becomes very accurate . the qmc ensemble leads to an unconventional kind of equilibrium , which may be realizable after strong perturbations in small isolated quantum systems having large number of levels . we demonstrate that the variance of energy fluctuations can be used to discriminate the qmc equilibrium from the bg equilibrium . we further suggest that the reason , why bg equilibrium commonly occurs in nature rather than the qmc - type equilibrium , has something to do with the notion of quantum collapse .

1010.4673

this article is about the properties of a quantum statistical ensemble , which is , in a sense , opposite to the conventional micro - canonical ensemble . it is an extension of the previous work of one of us@xcite . the present work contains a discussion of underlying conceptual issues and detailed analytical and numerical investigations of finite quantum systems . these two aspects of the article are complementary but otherwise quite different in scope and style . sections [ narrow ] and [ prospects ] are dealing with conceptual issues . they motivate and outline a broader agenda . the remaining sections ( [ qmc]-[spins ] plus appendices ) are more specialized . we hope that the conceptual part would motivate the readers unfamiliar with the above agenda to look into the rest of the article . the standard quantum - mechanical derivation of boltzmann - gibbs ( bg ) equilibrium is based on the conventional micro - canonical postulate formulated as follows . one should consider a subsystem in a macroscopic environment , assume , that the two form an isolated macroscopic system , neglect the interaction between them , and then select the statistical ensemble amounting to an incoherent mixture of eigenstates of the total system in a narrow energy window around a given value of energy , @xmath0 . the width of this micro - canonical energy window @xmath1 should be large enough to include a very large number of quantum levels , but , at the same time , small enough to imply the negligible change of temperature across this energy window . this condition can be formally expressed as follows : @xmath2 and @xmath3 \over \partial e } \right\vert_{e = e_{\hbox{\scriptsize av } } } \gg - \delta_{mc } \left . { \partial^2 \hbox{log } [ \nu(e)/\nu_0 ] \over \partial e^2 } \right\vert_{e = e_{\hbox{\scriptsize av}}}. \label{t}\ ] ] here @xmath4 is the density of energy states of the isolated macroscopic system , which should be approximately equal to the density of states of the environment at the same energy , @xmath5 is unimportant normalization constant , and @xmath6 is temperature in energy units . given the exponentially large number of levels of a macroscopic system , the above two inequalities can be easily satisfied for @xmath1 ranging over many orders of magnitude . recently , further progress was made by showing that a single pure state of a macroscopic system formed by a coherent superposition of eigenstates randomly selected from the above energy window would also lead to a density matrix of the subsystem exhibiting the bg distribution@xcite . it has been further conjectured that , if the isolated macroscopic system is chaotic in the classical limit , then even a single eigenstate would lead for the bg distribution in the density matrix of the subsystem@xcite . yet the question remains : what is the justification of the narrow energy window in the microcanonical postulate ? the origin , the limits and the implications of this condition are not well understood at present . the microcanonical postulate as such remains a rather formal construction , because in reality we do not encounter macroscopic systems isolated from the rest of the world . even , if we were to find one , the non - relativistic quantum - mechanical description of such a system normally leads to unrealistically small spacings between adjacent eigenstates much smaller than the minimal natural width of an energy level estimated as @xmath7 , where @xmath8s is the big - bang age of the universe . the real width of the energy levels should , of course , be much larger due to relativistic effects . although the notion of an energy eigenstate of a macroscopic system is not well - defined , one can still hope that the statistical treatment based on this notion would produce an outcome , whose validity extends beyond the systems with well - defined energy eigenstates in the same way as bg statistics derivable for classical systems extends to quantum systems . considering systems with well - defined eigenstates , one can proceed by giving the conventional microcanonical ensemble with its narrow energy window the status of a fundamental postulate . however , such an approach to justifying bg equilibrium is not quite satisfactory . if instead of simply postulating the bg equilibrium for quantum systems , one were to claim its derivation from a more fundamental postulate , then the introduction of this extra postulate would be justified only if , in addition to the bg equilibrium , it would lead to an explanation of some other previously unexplained facts about nature and predict new ones . the micro - canonical postulate , however , does not lead to any other observable prediction besides bg equilibrium itself . it could have been that the narrow energy condition is only a simplification , while the bg distribution emerges for a broader class of ensembles . however , as discussed in refs.@xcite , the departure from the narrow energy condition easily spoils the bg statistics for a subsystem . it leads to a mixture of different thermal distributions , which does not produce a distribution characterized by a single temperature . yet one could have asked the question : if it is not a narrow energy window , then what is the most likely shape of the participation function for the energy eigenstates of the entire system ? can it be that , if the most likely shape is found , then the resulting statistics for the subsystems is of bg type ? one can , in particular , recall that the conventional canonical ensemble does not appear to have narrow energy window for participating eigenstates . to each eigenstate with energy @xmath9 , it assigns probability weight @xmath10 , which follows either from maximizing entropy @xmath11 subject to the total energy constraint , or from the practical consideration that , if one slowly isolates a large quantum system from its environment , which initially was in bg equilibrium with temperature @xmath12 , then the diagonal elements of the resulting density matrix will have the above form . here , it should be noted , however , that the quantity of practical interest in not the probability weight function @xmath13 as such but rather its product with the density of states @xmath14 . since @xmath13 decays exponentially with constant rate `` @xmath15 , while for a macroscopic system @xmath4 increases exponentially with decreasing ' ' rate " @xmath16 , where @xmath17 is defined by eq.([t ] ) , the product would have a narrow maximum at energy @xmath18 corresponding to @xmath19 . the width of that maximum @xmath20 is determined by @xmath21 \over \partial e^2 } \right\vert_{e = e_{\hbox{\scriptsize av } } } = \left . { \partial [ 1/t(e ) ] \over \partial e } \right\vert_{e = e_{\hbox{\scriptsize av } } } \sim - { 1 \over t_0 ^ 2 n_p } , \label{plog}\ ] ] where @xmath22 is the number of particles in the system . this gives @xmath23 . as a result , @xmath24 \over \partial e^2 } \right\vert_{e = e_{\hbox{\scriptsize av } } } \sim { 1 \over t_0 \sqrt{n_p } } \ll { 1 \over t_0 } , \label{dlog}\ ] ] i.e. condition ( [ t ] ) is satisfied . thus the canonical ensemble for an isolated macroscopic quantum system , in effect , imposes a narrow energy window condition . the idea of maximization entropy at a given energy is closely related to the notion of chaos and ergodicity in the phase space of classical systems . for quantum systems , the notion of phase space trajectories is not defined because of the heisenberg uncertainty principle . hence the maximization of conventional entropy can not be justified on the basis of the dynamics of isolated quantum systems . to address the issue of what may happen beyond the narrow energy window imposed by the conventional ensembles , an ensemble with unrestricted participation of eigenstates and without the requirement of the maximization of conventional entropy , referred to below as quantum micro - canonical ensemble@xcite , is a conceptually necessary limit to consider . ( similar quantum canonical ensembles have also been defined refs.@xcite . ) in addition to the above conceptual issues , the basic concern in the context of the foundations of quantum statistical physics remains about the actual behavior of an isolated quantum system perturbed from equilibrium and not interacting with a large thermal bath . it is overwhelming intuition of many researchers , including the present authors , that an isolated macroscopic classical system with generic non - linear interaction between particles would evolve dynamically to exhibit the bg statistics for small subsystems . the situation is , however , not so trivial for quantum systems , because they can be in a superposition of states having different total energies . if , after the perturbation , an isolated quantum system is in a broad superposition of eigenstates violating the narrow energy condition , then the occupations numbers of eigenstates will not change with time under the dynamics guided by linear schrdinger equation , and hence the narrow energy condition will never emerge . generic wave function @xmath25 of a quantum system is a superposition of eigenstates @xmath26 : @xmath27 where @xmath28 are complex amplitudes , and @xmath29 the total number of eigenstates . for an isolated quantum system , fixing energy @xmath0 associated with wave function @xmath25 implies @xmath30 where @xmath31 are the occupation numbers of quantum eigenstates , and @xmath32 are the corresponding energies . we shall refer to @xmath0 as `` average energy '' [ not to be confused with the uniform average of all @xmath32 ] . in this work , we focus on quantum micro - canonical ( qmc ) ensemble@xcite , where all quantum wave functions satisfying average energy constraint ( [ epsav ] ) are equally probable in the hilbert space of the system . the qmc ensemble amounts to a natural generalization of the equipartition postulate for energy shells in classical phase spaces . despite this analogy , the qmc ensemble is quite different from the conventional micro - canonical ensemble , because it does not limit the participating eigenstates to a narrow energy window around @xmath0 . it was shown in ref.@xcite that the qmc ensemble does not lead to the bg statistics , at least in the absence of interaction between the subsystem and the environment . this negative result is important , because it forces one to think seriously about justifying the narrow energy window in the conventional microcanonical ensemble . as discussed in refs.@xcite and in section [ prospects ] , quantum collapse may play here a fundamental role . further study of the qmc ensemble is also important , because this ensemble may be realizable in small isolated quantum systems with a large number of quantum levels . beyond the subject of foundations of quantum statistical physics , the properties of the qmc ensemble are also relevant to the studies of various properties of hilbert spaces , e.g. in the context of quantum computing or many - particle entanglement . it is finally plausible on general grounds that such a natural quantum ensemble would find many realizations not discovered so far . for example , it was demonstrated in ref.@xcite that the expansion of the basis states of sparce random matrices in terms of eigenstates exhibits the qmc type of statistics . the description of the qmc ensemble involves two steps : ( 1 ) characterization of the shape of the energy window function in this case , the participation function for the eigenstates of the entire isolated system ; and ( 2 ) obtaining the density matrix for the subsystem of interest . sections [ general]-[results ] below deal with step ( 1 ) , while section [ spins ] deals with step ( 2 ) . in section [ general ] , we summarize the analytical results obtained in ref.@xcite for the qmc ensemble in large-@xmath29 isolated quantum systems . in section [ ansatz ] , we propose a method of introducing finite-@xmath29 corrections to the large-@xmath29 approximation . in section [ results ] , we present the numerical tests of the resulting description for a 12-level isolated quantum system . in section [ spins ] , we examine numerically the density matrix for a subsystem of a 12-level system , and compare the qmc - based result with that of the conventional canonical ensemble . section [ prospects ] discusses open questions motivated by our results . [ algorithmsl ] describes a practical numerical algorithm for finding an important characteristic function appearing in the course of the analytical approximation procedure . [ algorithmmc ] gives the details of our monte - carlo sampling algorithm for the qmc ensemble . we consider quantum spectum @xmath34 consisting of @xmath29 energy levels . the values of energies are assumed to be ordered with @xmath35 and @xmath36 . the analytical approach and the analytical results of ref.@xcite can be described as follows . the qmc ensemble is defined by specifying uniform joint probability distribution for complex amplitudes @xmath37 in the hilbert space subject to average energy constraint ( [ epsav ] ) and to the normalization constraint @xmath38 the two constraints are of the second order in variables @xmath39 and do not depend on @xmath40 . to avoid dealing with curved manifolds , we now change from variables @xmath39 to variables @xmath31 . completely random choice of states in hilbert space implies that the probability measure is proportional to the volume element : @xmath41 since constraints ( [ epsav ] , [ norm ] ) do not depend on @xmath40 , the assignment of variables @xmath40 is totally random . the resulting uniform distribution of @xmath40 can be just factorized out and then integrated . what remains is the euclidean subspace of variables @xmath42 with uniform joint probability measure on the manifold constrained by conditions ( [ epsav ] , [ norm ] ) together with the positivity condition @xmath43 in the @xmath42-space , due to the linear nature of constraints ( [ epsav],[norm],[positiv ] ) , the boundaries of the resulting manifold are flat , i.e. the manifold itself is an @xmath44-dimensional polyhedron . we will refer to it as `` qmc polyhedron '' . our goal is to obtain marginal probability distribution @xmath45 for a given occupation number @xmath46 and the corresponding average value @xmath47 representing the participation of the @xmath48th eigenstate in the qmc ensemble . probability distribution @xmath45 is proportional to the @xmath49 dimensional volume of the intersection of the qmc polyhedron with hyperplane @xmath50 . ( here we use variable @xmath46 as a label of the coordinate axis , while in @xmath45 and everywhere below , @xmath46 refers to the value of @xmath51 . ) the intersection manifold is to be denoted as @xmath52 , and its volume as @xmath53 . this volume plays the role of unnormalized probability distribution such that @xmath54 and the average value of @xmath46 is @xmath55 in the space of all variables excluding @xmath46 to be denoted as @xmath56 , manifold @xmath52 is defined by conditions : @xmath57 and @xmath58 together with positivity conditions ( [ positiv ] ) . the most relevant limit to consider now is @xmath59 , because , for @xmath60 , it is improbable that a single occupation number @xmath46 in a randomly chosen superposition of all quantum states becomes comparable to 1 . we refer to this limit as `` small-@xmath46 approximation . '' it gives@xcite @xmath61}. \label{vkpk}\ ] ] where @xmath62 is the volume renormalization parameter associated with the shift of energy hyperplane ( [ epsavk0 ] ) . it was further shown in ref.@xcite that , in the limit @xmath60 all parameters @xmath62 are approximately equal to the same value to be denoted as @xmath63 , i.e. in this limit @xmath64}. \label{vkpk1}\ ] ] the corresponding average value of the occupation number @xmath46 is @xmath65}. \label{pav2}\ ] ] the value of parameter @xmath63 can now be found numerically by substituting eq.([pav2 ] ) into either averaged normalization condition @xmath66 or averaged energy condition @xmath67 the two values of @xmath63 are guaranteed to be the same for any @xmath29 , probably because formula ( [ pav2 ] ) is analytically accurate leading term in the large-@xmath29 expansion . [ for finite-@xmath29 systems , the actual behavior of @xmath53 can exhibit noticeable deviations from the exponential dependence of eq.([vkpk1 ] ) ] . behind the geometry - based discussion one should not overlook that eq.([vkpk1 ] ) represents grand - canonical gibbs ensemble defined in the hilbert space , or the space of occupation numbers @xmath46 . the problem is , in fact , equivalent to finding the probability distributions for a system of @xmath29 classical states characterized by one - particle energies @xmath68 , with @xmath46 being the number of classical particles occupying the @xmath48-th state . in such a problem both the total energy and the total number of classical particles is conserved , which leads to the grand canonical ensemble in the hilbert space . this analogy , however , does not imply conventional observable results , because it pertains to the probability distribution @xmath45 , while , in the original quantum problem , @xmath46 itself is the probability of observing the system in the @xmath48th state , i.e. the prediction of the observable outcome requires finding the average value @xmath47 . it is at this last step that the departure from the conventional statistics occurs . in a broader context , this kind of result associated with double - averaging is known as superstatistics@xcite . the grand - canonical analogy becomes evident , once formula ( [ vkpk1 ] ) is rewritten as @xmath69 where @xmath70 equation ( [ vkpk_grand ] ) implies that @xmath71 is the inverse temperature in the hilbert space , and @xmath72 is the chemical potential . the value of @xmath63 is zero , when @xmath0 is equal to @xmath73 , the uniform average of all energies in spectrum @xmath34@xcite : @xmath74 we also recall that @xmath75 for @xmath76 , and @xmath77 for @xmath78 . here we follow the convention of ref.@xcite , that , unless specified otherwise , the origin of the energy axis is set at @xmath73 , i.e. , @xmath79 . the chemical potential @xmath72 is simultaneously the pole of function @xmath80 given by eq.([pav2 ] ) , which implies that for the small-@xmath46 approximation to be valid for all levels , @xmath72 should be sufficiently below the lowest energy level for positive @xmath63 or above the highest energy level for negative @xmath63 . in the former case , `` sufficiently below '' means that@xcite @xmath81 it is shown in ref.@xcite that the above condition is violated for a typical macroscopic system at a typical value of @xmath0 , which leads to condensation into the lowest energy state , i.e. @xmath82 becomes comparable to one . condition ( [ elemin ] ) can also be easily violated for finite - n systems leading to the departure from the small-@xmath46 results ( [ vkpk1],[pav2 ] ) for several lowest levels . in the both cases , it is necessary to consider what happens beyond the small-@xmath46 approximation . when typical values of @xmath46 are not small , it is necessary to recall @xcite that parameters @xmath62 are defined as functions of average energy @xmath0 as follows : @xmath83 = { 1 \over v_k } \left . { \partial v_k \over \partial v } \right|_{v=0 } , \label{lk}\ ] ] where [ in this equation only ] @xmath84 is the volume of the manifold constrained in space @xmath85 by conditions @xmath86 @xmath87 together with positivity conditions ( [ positiv ] ) . here @xmath88 is the energy shift parameter . we follow the convention of ref.@xcite to use square brackets in functions @xmath89 $ ] and later in @xmath90 $ ] . given the above definition of @xmath89 $ ] , the volume of manifold @xmath52 defined by eqs.([positiv],[normk],[epsavk0 ] ) can be expressed as @xmath91 d e \right ] \right\}. \label{vkpk2a}\ ] ] [ this equation does not appear explicitly in ref.@xcite , but otherwise it is an obvious intermediate step in the calculation described there . ] in the limit @xmath60 , all @xmath92 $ ] are approximately equal to a single function @xmath93 $ ] defined in interval @xmath94 $ ] . therefore , eq.([vkpk2a ] ) becomes@xcite @xmath95 d e \right ] \right\}. \label{vkpk2}\ ] ] functions @xmath53 given by both eq.([vkpk2a ] ) and ( [ vkpk2 ] ) are not necessarily defined in the entire range @xmath96 , because , for too large @xmath46 , and , in the case of @xmath97 or @xmath98 , for too small @xmath46 , the intersection manifold defined by eqs.([positiv],[normk],[epsavk0 ] ) may simply not exist . the practical rule is that @xmath53 is defined , when @xmath92 $ ] or @xmath93 $ ] are defined at the upper integration limit @xmath99 . this limit should fall within interval @xmath94 $ ] for eq.([vkpk2 ] ) and within sometimes different boundaries ( given in section [ ansatz ] ) for eq.([vkpk2a ] ) . the upper cutoffs for @xmath46 are given in appendix d of ref.@xcite and the lower cutoffs in section [ ansatz ] . the mean-@xmath63 approximation represented by eq.([vkpk2 ] ) neglects the existence of lower cutoffs of @xmath46 for @xmath97 and @xmath98 . these cutoffs , however , can be incorporated , at the level of finite-@xmath29 corrections based on eq.([vkpk2a ] ) . if a lower cutoff is present , then @xmath100 is not defined , and the prefactor in eq.([vkpk2a ] ) should be viewed as simply a constant , which is then canceled by normalization in eq.([pkpk ] ) . also , when the lower cutoff is present , the lower integration limit `` @xmath0 '' in eq.([vkpk2a ] ) should be replaced with the minimum or the maximum value of energy , where @xmath92 $ ] is defined , for @xmath97 and @xmath98 , respectively . in general , function @xmath53 becomes exponentially small as it approaches either of its cutoffs . therefore , as @xmath29 increases , the practical significance of these cutoffs decreases . in eq.([vkpk2 ] ) , we took , at first sight inconsistently , the large-@xmath29 limit @xmath92\approx \lambda[e]$ ] but did not approximate @xmath101 . the reason is that we are aiming at making numerical tests for hilbert space with @xmath102 , where condition @xmath103 is not well fulfilled , and , since the term @xmath104 is an exact one@xcite , we thereby limit the finite-@xmath29 error only to the integral in eq.([vkpk2 ] ) and deal with this error later in section [ ansatz ] . it is also worth noting that the small-@xmath46 formulas ( [ vkpk ] ) and ( [ vkpk1 ] ) can be obtained by linearizing exponents in eqs.([vkpk2a ] ) and ( [ vkpk2 ] ) with respect to @xmath46 , and approximating @xmath101 . equation ( [ vkpk2 ] ) was the basis for proving the condensation of the qmc ensemble into the lowest energy state of macroscopic systems@xcite . it indicated that the character of function @xmath105 was not exponential but rather narrowly peaked around @xmath106 . this condensation is different from the bose - einstein condensation , in the sense that it occurs in the many - particle ground state independently of the type of the particles composing the system . while we can not propose a controllable procedure for finding corrections for @xmath89 $ ] with respect to @xmath90 $ ] , we introduce here an ansatz , which makes these corrections in a conceptually meaningful way and , at the same time , significantly improves the agreement with monte - carlo results . the ansatz is based on the following considerations . function @xmath90 $ ] is defined in interval @xmath94 $ ] . it has two universal properties in the large-@xmath29 limit@xcite : \(i ) @xmath107 = 0 $ ] ; \(ii ) for @xmath108 , @xmath90 $ ] quickly approaches the asymptotic form : @xmath109 \approx { 1 \over e_{\hbox{\scriptsize av } } - e_{\hbox{\scriptsize min } } } , \label{lambdaa}\ ] ] while for @xmath110 , the asymptotic form is @xmath109 \approx { 1 \over e_{\hbox{\scriptsize av } } - e_{\hbox{\scriptsize max } } } . \label{lambdab}\ ] ] in other words , the behavior of the entire function @xmath90 $ ] is almost entirely determined just by the values of three parameters @xmath73 , @xmath111 and @xmath112 . now , function @xmath90 $ ] characterizes the entire quantum spectrum @xmath113 , while functions @xmath89 $ ] characterize spectra @xmath114 obtained from @xmath113 by removing energy level @xmath68 . as a result , the uniform average of all energies for spectrum @xmath114 is shifted to @xmath115 while the end points of the spectrum remain unchanged , _ unless _ @xmath97 or @xmath98 , when , respectively , the minimum or the maximum energy levels are removed : @xmath116 @xmath117 apart from the above specific differences @xmath89 $ ] should be reminiscent of @xmath90 $ ] . therefore , in order to approximate @xmath89 $ ] from the knowledge of @xmath90 $ ] we adopt the following two - step procedure : \(i ) shift of @xmath90 $ ] to move @xmath73 to @xmath118 : @xmath119 = \lambda[e_{\hbox{\scriptsize av } } + e_{\hbox{\scriptsize av}0 } - e_{\hbox{\scriptsize av}0k } ] . \label{lshift}\ ] ] this also shifts @xmath111 and @xmath112 ( the end points of @xmath90 $ ] ) to @xmath120 and @xmath121 , respectively . \(ii ) uniform contraction or expansion of energy differences above and below @xmath118 to shift @xmath122 and @xmath123 to , respectively @xmath124 and @xmath125 given by eqs.([emink],[emaxk ] ) and simultaneously rescale @xmath126 itself to make sure that , if @xmath127 $ ] has asymptotic forms @xmath128 above @xmath122 and @xmath129 below @xmath123 , then the resulting function @xmath89 $ ] has asymptotic forms @xmath130 above @xmath124 and @xmath131 below @xmath125 . this transformation has form @xmath83 = \left\ { \begin{array}{l@{\quad:\quad}l } { 1 \over \alpha_- } \lambda_k^{\prime}[e_{\hbox{\scriptsize av}0k } + \alpha_- ( e_{\hbox{\scriptsize av } } - e_{\hbox{\scriptsize av}0k } ) ] & e_{\hbox{\scriptsize av } } < e_{\hbox{\scriptsize av}0k } , \\ & \\ { 1 \over \alpha_+ } \lambda_k^{\prime}[e_{\hbox{\scriptsize av}0k } + \alpha_+ ( e_{\hbox{\scriptsize av } } - e_{\hbox{\scriptsize av}0k } ) ] & e_{\hbox{\scriptsize av } } > e_{\hbox{\scriptsize av}0k } , \end{array } \right . \label{lambdak}\ ] ] where @xmath132 and @xmath133 the resulting function @xmath89 $ ] is then used in eq.([vkpk2a ] ) . the last step automatically corrects the drawback of the mean-@xmath63 formula ( [ vkpk2 ] ) mentioned in section [ general ] . namely , the mean-@xmath63 approximation does not have lower cutoffs for @xmath105 and @xmath134 , while the exact solution should have them . indeed , if @xmath0 falls between @xmath111 and @xmath135 , then @xmath105 has the lower cuttoff @xmath136 which corresponds to only two lowest levels @xmath111 and @xmath135 being occupied with the occupation numbers determined by the value of @xmath0 . once higher energy levels become occupied , the value of @xmath137 becomes greater than @xmath138 to fulfill average energy condition ( [ epsav ] ) . likewise , if @xmath0 falls between @xmath139 and @xmath112 , then lower cutoff for @xmath134 is @xmath140 in our approximation procedure , @xmath53 is defined only , when @xmath141 $ ] is defined at the upper integration limit in eq.([vkpk2a ] ) , i.e. when @xmath99 falls in the interval between @xmath124 and @xmath125 . the appearance of @xmath142 and @xmath143 in eqs.([alpha-],[alpha+ ] ) then implies the correct lower cuttoff values @xmath138 and @xmath144 , respectively . these lower cutoffs are unimportant in the large-@xmath29 limit , but become important for finite-@xmath29 systems , like the one considered numerically in this work . here we compare analytical and numerical results for a quantum spectrum that has @xmath145 energy levels . this is nearly the maximum number of quantum levels , whose statistics we can access in reasonable time with the monte - carlo algorithm described in [ algorithmmc ] . the values of the spectrum energies are \{@xmath146 , @xmath147 , @xmath148 , @xmath149 , @xmath150 , @xmath151 , @xmath152 , @xmath153 , @xmath154 , @xmath155 , @xmath156 , @xmath157 } ( also shown in fig . 1 ) . this spectrum is obtained as a strongly discretized version of the gaussian density of states with maximum at @xmath158 . ( nearly gaussian density of states is generically expected for macroscopic systems@xcite . ) ( 0,0)(157,80 ) ( 0,0 ) = 4.5 in ( 2.4,5 ) = 1.55 in figure 1 presents curves for @xmath90 $ ] obtained in various theoretical approximations : namely , small-@xmath46 approximation based on eqs.([vkpk1],[pav2 ] ) ; mean-@xmath63 approximation based on eq.([vkpk2 ] ) ; and the calculation with finite-@xmath29 corrections based on eq.([vkpk2a ] ) and the ansatz of section [ ansatz ] . the numerical algorithms for computing the latter two curves are described in [ algorithmsl ] . figure 1 also includes curves representing the large-@xmath29 asymptotic behavior ( [ lambdaa],[lambdab ] ) . the calculations of all three curves are based on satisfying energy constraint ( [ epsavav ] ) . the small-@xmath46 approximation , guarantees that the normalization constraint ( [ normav ] ) is satisfied at the same time . however , the other two approximations do not guarantee it _ a priori_. therefore , the check of how the normalization constraint is satisfied constitutes an indication of the accuracy of the approximations . this check is presented in the inset of fig . 1 as a function of @xmath0 . it indicates that the normalization for the mean-@xmath63 approximation falls short of 1 by about 10 per cent ( roughly @xmath159 ) . the normalization for the finite-@xmath29 corrected calculation deviates from 1 within @xmath160 per cent . ( 0,0)(157,190 ) ( 0.7,17.5 ) ( 5.9,17.5 ) ( 11.1,17.5 ) ( 0,13.5 ) = 1.9 in ( 5.2 , 13.5 ) = 1.94 in ( 10.4 , 13.5 ) = 1.94 in ( 0.7,13 ) ( 5.9,13 ) ( 11.4,13 ) ( 0,9 ) = 1.92 in ( 5.2 , 9 ) = 2 in ( 10.4 , 9 ) = 2 in ( 0.7,8.5 ) ( 5.9,8.5 ) ( 11.4,8.5 ) ( 0,4.5 ) = 2 in ( 5.2,4.5 ) = 2 in ( 10.4,4.5 ) = 2 in ( 0.7,4 ) ( 5.9,4 ) ( 11.4,4 ) ( 0,0 ) = 2 in ( 5.2,0 ) = 2 in ( 10.4,0 ) = 2 in figure 2 presents detailed comparison between monte - carlo results and the above analytical approximations for three different values of @xmath0 , @xmath161.the numbers of monte carlo points accepted by the algorithm from [ algorithmmc ] for the three values of @xmath0 were , respectively , 2000 , 5000 and 5000 . the first row of frames in fig . 2 represents the comparison for the values of the average occupation numbers @xmath47 corresponding to twelve energies @xmath68 in the spectrum . the remaining three rows represent the comparison for the three ( out of twelve ) probability distributions @xmath45 for each of the above three average energies . in the case of @xmath162 , the monte - carlo result qualitatively corresponds to the prediction of ref.@xcite that , for macroscopic systems ( meaning gaussian density of states in the large-@xmath29 limit ) at realistic positive hilbert space temperatures @xmath163 ( i.e. with @xmath108 ) , the qmc ensemble leads to condensation into the lowest energy state with @xmath164 having the character of @xmath165-function peaked at @xmath82 , while all other states have small average occupations @xmath47 and exhibit exponential shapes of @xmath45 . what one sees in figs . 2(a , b , c , d ) is a finite-@xmath29 realization of the above condensation . the monte - carlo results indicate that all @xmath45 except for @xmath164 corresponding to the lowest spectral level @xmath166 have nearly exponential shape well describable by all three of the approximations considered [ fig . 2(c , d ) ] . the differences , however , become pronounced once @xmath164 is considered . figure 2(b ) indicates , that the small-@xmath46 approximation fails completely to describe @xmath164 , the mean @xmath63 approximation reproduces the monte - carlo results qualitatively , but with large quantitative discrepancy , while the finite-@xmath29 corrected calculation is in a good quantitative agreement with the monte - carlo results . despite failing to predict @xmath164 , the small-@xmath46 approximation predicts accurately the value of @xmath82 in fig 2(a ) . as discussed in ref.@xcite , it is the consequence of the fact that this approximation accurately describes all @xmath45 and @xmath47 for @xmath167 , but then , since it automatically satisfies the normalization condition ( [ normav ] ) , it guarantees that the value of @xmath82 is also accurate . the example of @xmath168 corresponds to the negative hilbert space temperature @xmath163 . in this case , the qmc ensemble tends to condense to the highest energy state . the facts , that , in this case , the distribution of @xmath169 [ fig . 2(l ) ] is rather broad , and that the form of @xmath170 [ fig . 2(k ) ] for the second highest level also noticeably deviates from exponential , are the finite-@xmath29 effects . here again , the small-@xmath46 approximation fails to describe @xmath170 and @xmath169 , the mean-@xmath63 calculation gives correct qualitative trends with quantitative errors , and the finite-@xmath29-corrected calculation exhibits a good quantitative agreement with the monte - carlo results . for @xmath171 all three approximations give nearly exponential decays for @xmath45 , in good agreement with the monte - carlo results [ see , e.g. , fig . since the small-@xmath46 approximation fails for three highest levels , it is no longer guaranteed to produce the accurate values of @xmath172 , @xmath173 and @xmath174 and , in fact , the noticeable discrepancy can be seen in fig . the example of @xmath175 corresponds to the infinite hilbert space temperature . in this case , the small-@xmath46 approximation predicts that all @xmath45 have the same exponential shape , and that , independently of @xmath68 , @xmath176 . this would indeed be the case in the limit @xmath177 . however , here the symmetric position of @xmath175 with respect to the quantum spectrum leads to a more systematic character of finite-@xmath29 corrections , which , in turn , leads to a larger discrepancy with the monte - carlo results for all @xmath48 . the origin of this systematic discrepancy is in the corrections to the linearized version of integral in eq.([vkpk2 ] ) . the linearization of that integral with respect of @xmath46 leads to the small-@xmath46 approximation . the higher order corrections are , in general , partly canceled , because @xmath93 $ ] in the integral changes sign , when the integration range includes @xmath178 somewhere in the middle . however , when @xmath175 , this sign - change point coincides with the lower integration limit . as a result , @xmath93 $ ] does not change sign within the integration range for any @xmath48 , which leads to the above larger discrepancy . two other approximations capture the above systematic trend , but here again , the finite-@xmath29 corrected calculation exhibits a better quantitative agreement with the monte - carlo results . it should be emphasized that the finite-@xmath29 corrected calculation is based on an ansatz of section [ ansatz ] , which , although meaningful and apparently very effective , does not amount to an exact calculation . therefore , it is expected to show small deviations from the monte - carlo results , which can , indeed , be seen in figs.2(k , l ) . the statistics of eigenstate participation for the entire isolated system reported in this section is not a direct test of the presence or the absence of bg equilibrium for a small subsystem . however , in this respect , the results of significance are those shown in figs . 2(a , e , i ) . they represent , in a sense , the participation functions for conventional micro - canonical ensembles with different positions of micro - canonical energy windows . these participation functions have algebraic dependence on energy mostly captured by the small-@xmath46 result ( [ pav2 ] ) . this slow non - exponential dependence precludes one from treating the participation of eigenstates as a conventional canonical ensemble . as a result , the broad participation of eigenstates in the qmc ensemble implies a broad mixture of conventional thermal states , which , for a small subsystem , does not result in a single thermal bg distribution the conclusion confirmed by direct calculation in ref.@xcite . in the next section , we present numerical evidence corroborating this conclusion . reference @xcite derived analytical results for the density matrix of a small subsystem within a larger isolated quantum system described by the qmc ensemble . when the whole system is macroscopic ( i.e. the number of levels @xmath29 is exponentially large ) , the qmc - based density matrix of the small subsystem is a weighed sum of two terms : the zero - temperature density matrix and the infinite temperature density matrix . in the finite-@xmath29 case , one has to solve a system of equations obtained in @xcite not to be attempted in the present work . instead , in this section , we demonstrate numerically that the following qualitative property of the macroscopic limit also appears in the finite-@xmath29 case . namely , for a subsystem within an isolated quantum system , the lowest energy state and the states on the high - energy end of the spectrum have larger occupations when the isolated system is sampled according to the qmc ensemble than when it is sampled according to the conventional canonical ensemble . correspondingly , the intermediate energy states have smaller occupations for the qmc ensemble . below we consider a 12-level system consisting of two spins 1/2 and spin 1 , and examine numerically the implications of the qmc ensemble for the subsystem of two spins 1/2 . as discussed in the previous section , the 12-level limitation is due to the computational difficulty of sampling high - dimensional hilbert spaces . in principle , in order to exhibit deviations from the bg statistics , the subsystem should have at least three levels , and therefore one could think that it is more natural to consider the spin 1 as a subsystem . however , with only 3 levels in the subsystem the differences between the qmc - based results and the bg statistics would be less instructive . we assume the the environment represented by spin 1 does not interact with the subsystem of two coupled spins 1/2 . hence the hamiltonian of the entire system is : @xmath179 where @xmath180 and @xmath181 are , respectively , the hamiltonians for the two spins 1/2 and for spin 1 . here @xmath182 are the operators of for the @xmath183-components ( @xmath184 ) of the two spins 1/2 ( @xmath185 ) , @xmath186 is the operator of the @xmath187-component of spin 1 , @xmath188 is the external magnetic field , @xmath189 and @xmath190 are the gyromagnetic ratios for spins 1/2 and for spin 1 , respectively ; @xmath191 , @xmath192 , @xmath193 are anisotropic coupling constants . since spin 1 plays the role of the environment , the parameters of the hamiltonian were chosen such that the spread of the spin-1 levels is significantly broader than that of the two - spin-1/2 subsystem . the energy spectrum of the hamiltonian ( [ h ] ) with the above choice of parameters is shown in fig . [ fig - spectrum ] . it has three clearly identifiable 4-level groups . ( 0,0)(157,25 ) ( 3,0 ) = 4 in we compare the subsystem density matrices for the qmc ensemble and for the canonical ensemble at the same value of the average energy of the entire isolated system . in order to do this , we first fixed the temperature of the canonical ensemble t=1 , then calculated the corresponding average energy @xmath194 , and then generated 2000 sample wave functions from the qmc ensemble for the above value of @xmath195 . afterwards , we calculated the average subsystem density matrices corresponding to the both ensembles . the results for the diagonal elements of the density matrix @xmath196 in the eigenbasis of the hamiltonian @xmath197 are presented in fig . [ fig - rho ] . as expected from the large-@xmath29 limit@xcite , the values of @xmath196 for the qmc ensemble are larger than those for the canonical ensemble for the lowest and the highest subsystem energy levels , and correspondingly smaller for the middle two levels . ( 0,0)(157,80 ) ( 3,0 ) = 4.5 in ( 9.3,3.8 ) = 1.9 in if one can have only limited experimental access to the properties of the subsystem , the qmc ensemble can be identified by looking at the variance of energies for the subsystem states : @xmath198 where index @xmath183 labels subsystem s eigenstates , @xmath199 denotes the corresponding subsystem eigenenergies , and @xmath200 is the average energy of the subsystem . in fig . [ fig - de ] , we plot subsystem energy variances for the qmc and the canonical ensembles as a function of @xmath201 . for @xmath202 ( see fig . [ fig - spectrum ] ) and @xmath203 , the predictions of the two ensembles are identical , and hence the variances are the same . ( in the first case , only the ground state is occupied and @xmath204 $ ] , while in the second case all subsystem s states are equally occupied and @xmath205 . ) in - between the energy variance for the qmc ensemble is always larger . this is true not only for the subsystem , but also for the entire system . ( 0,0)(157,80 ) ( 3,0 ) = 4.5 in an interesting fact is that for the parameters considered , the subsystem average energies for the qmc and the canonical ensembles are different even though the energy of the entire system is the same for the both ensembles . the subsystem energies for the two ensembles are , respectively , @xmath206 and @xmath207 , i.e. , in the qmc case , the subsystem is in a sense colder " . we have investigated this issue further , and in fig . [ fig - ebar ] have plotted the result for @xmath201 for the both ensembles as a function of @xmath0 of the total system . our findings indicate that there exists a critical value @xmath208 such that @xmath209 for @xmath210 , and @xmath211 for @xmath212 . however for @xmath212 , the results for the two ensembles are very close ( the high - temperature limit ) and hence the difference between @xmath213 and @xmath214 is insignificant and difficult to resolve statistically . ( 0,0)(157,90 ) ( 3,0 ) = 4.5 in ( 8.8,1.8 ) = 1.9 in the difference between @xmath213 and @xmath214 may , perhaps , be used as another experimentally identifiable signature of the qmc ensemble . it should be noted , however , that the above difference can only appear when the subsystem particles are different from the environment particles . the subsystem energies for the two ensembles are guaranteed to be the same , when , for example , the whole system is describable by a translationally invariant hamiltonian of spins 1/2 , and the interaction between the subsystem of a few spins 1/2 and the environment is negligible . here we review relevant open questions . it appears now very likely that the micro - canonical narrow energy window condition is crucial for the justification of bg equilibrium in macroscopic systems . what , however , can justify this condition itself ? the only argument known to us is associated with quantum collapse also referred to as quantum measurement or quantum projection@xcite . eigenstates with significantly different energy values should have macroscopically distinguishable characteristics . hence the mixture of macroscopically distinguishable states , coherent or not , would not be tolerated in nature , and instead the mixture would collapse to one macroscopic state with usual probabilities predicted by quantum mechanics . the disturbing part of the above argument is that the quantum collapse is an essentially non - linear process , and hence can not appear in the quantum - mechanical theory in the course of linear evolution of a wave function or a density matrix . therefore , it implies that bg equilibrium can not be justified in the framework of linear quantum mechanics it requires a non - linear quantum collapse into the narrow micro - canonical energy window . one can still ask a different question : if the quantum collapse occurred only once in the evolution of a macroscopic system , is it realistic to expect that external perturbations on otherwise isolated macroscopic quantum system would force the spread of participating eigenstates over an energy window violating condition ( [ t ] ) ? is it further realistic to expect that the qmc ensemble can be realized in an isolated macroscopic quantum system that has once experienced quantum collapse ? for a truly macroscopic system consisting of @xmath215 weakly interacting parts , the answer to both questions appears to be `` extremely unlikely '' . the reasoning here is based on the central limit theorem : if such a system is subjected even to a drastic external perturbation , e.g. explosion , it is still likely that different parts of such a system , while being strongly perturbed , will continue to interact weakly , end hence the resulting total energy distribution will have gaussian form , which would be sufficiently narrow to satisfy condition ( [ t]).(see also refs.@xcite . ) yet a single perturbation makes the energy window of participating eigenstates somewhat broader . the next level of this argument would be to ask , how many sequential perturbations are necessary to force an isolated macroscopic system to violate condition ( [ t ] ) . this question requires further investigation . realistic macroscopic systems are continuously subjected to the fluctuations of external potentials of electromagnetic or gravitational origin . can these fluctuations drive an isolated system out of the micro - canonical energy window ? if they can , but we do not observe any manifestation of it , does it mean that quantum collapse to the narrow energy window occurs continuously ? if this happens , does it imply observable energy fluctuations not explainable by linear quantum mechanics ? the second line of investigation is related to dealing with quantum systems that do not yet have large number of degrees of freedom , but already have a very large number of quantum levels , e.g. @xmath216 clusters of interacting spins 1/2 or a few cold atoms in an optical trap . in this case , it appears realistic to isolate these systems against energy dissipation and , possibly , against decoherence . the central - limit - theorem - based considerations are not applicable to these systems . therefore , it is clear that such systems can be driven out of condition described by conventional micro - canonical or canonical ensembles . will the qmc ensemble generically emerge in this case , possibly , after several perturbations@xcite ? if such an ensemble emerges , what are the observable signatures of it , e.g. in terms of measurable single particle properties ? what would be the effects of decoherence and subsequent quantum collapse in this case ? it is , in particular , possible that even if the broad qmc energy window collapses towards the canonical or micro - canonical shape , then the qmc - based calculations would predict the fluctuations of the resulting temperature . these questions require numerical investigations beyond the scope of this work . it is , in fact , likely that non - micro - canonical ensembles , and possibly qmc , routinely occur in the numerical studies of finite quantum systems . in conclusion , we have demonstrated that the large-@xmath29 analytical description of the qmc ensemble developed in ref.@xcite allows one to make good predictions for the monte - carlo sampling of the qmc ensemble for finite-@xmath29 systems . the description of ref.@xcite is amenable to finite-@xmath29 corrections , and once these corrections are introduced , the theory produces accurate quantitative agreement with the monte - carlo results for systems with @xmath217 . we have also studied numerically the implications of the qmc ensemble for a subsystem of a small spin system . our results indicate that , already in this rather artificial case , the behavior of subsystem s density matrix qualitatively follows the analytical result for the large-@xmath29 case . we have further suggested that the subsystem energy variance , as well as the average value of subsystem s energy itself , can be used to discriminate experimentally between the qmc and the bg statistics . in the context of foundations of quantum statistical physics , the apparent impossibility of obtaining the boltzmann - gibbs equilibrium from the qmc ensemble indicates that one should be looking closer at the basis and the applicability limit of the narrow energy window condition of the conventional micro - canonical ensemble . in this respect , quantum collapse may play a fundamental role . _ note added : _ as we were finishing editing this manuscript for resubmission , we discovered a series of papers by fresch and moro@xcite very closely related to the scope of the present work and ref.@xcite . in particular , in ref.@xcite , the above authors have pursued similar monte - carlo sampling of the qmc ensemble in finite dimensional hilbert spaces . their results appear to be consistent with and , in some aspects , complementary to ours . in particular , they have used metropoilis - hastings algorithm , as opposed to the direct sampling routine of our paper , and thereby were able to explore significantly larger hilbert spaces . on the other hand , the present paper advances further the subject of finite-@xmath29 corrections to the large-@xmath29 qmc results . we have also learned from ref.@xcite that , already in 1990 , wootters have considered the qmc ensemble and arrived to what we call the small-@xmath46 approximation@xcite . here we describe the algorithm for finding @xmath90 $ ] from the mean-@xmath63 integral equation ( [ vkpk2 ] ) `` algorithm a '' , and the modification of algorithm a including finite-@xmath29 corrections based on eq.([vkpk2a ] ) incorporating ansatz from section [ ansatz ] `` algorithm b '' . function @xmath90 $ ] is defined in the interval @xmath218 $ ] . we use a non - uniform discretization of this interval with 228 grid points less dense in the middle and more dense near @xmath111 and @xmath112 , were @xmath90 $ ] tends to diverge . this near - divergence is , however , manageable , because , it is compensated by the overall exponential decrease of @xmath53 and affects only the range of @xmath46 , where @xmath53 is exponentially small . _ algorithm a : outline _ the algorithm first finds @xmath90 $ ] in the small-@xmath46 approximation on the basis of eqs.([pav2 ] ) and ( [ epsavav ] ) and then iteratively improves it by ( i ) calculating @xmath53 by substituting @xmath90 $ ] into eq.([vkpk2 ] ) , ( ii ) finding the corresponding @xmath47 , ( iii ) checking averaged energy condition ( [ epsavav ] ) ; ( iv ) improving the value @xmath90 $ ] by matching condition ( [ epsavav ] ) and then returning to step ( i ) . as explained below , step ( iv ) is done `` locally '' , i.e. , for every @xmath0 , on the grid it improves @xmath90 $ ] independently of the values of @xmath90 $ ] at other grid points , but then once all grid values of @xmath90 $ ] are improved , the algorithm returns to step ( i ) with the entire improved set of @xmath90 $ ] . _ algorithm a : details and explanations _ at the small-@xmath46 approximation step , we substitute @xmath47 from eq.([pav2 ] ) into eq.([epsavav ] ) , and then find numerically the value of @xmath63 such that @xmath219 or @xmath220 . at the first iteration step , we start by substituting @xmath90 $ ] obtained in the small-@xmath46 approximation into the integral formula ( [ vkpk2 ] ) , thereby obtaining the generally non - exponential function @xmath53 , then use eq.([pkav ] ) to obtain @xmath47 and then check condition ( [ epsavav ] ) , which shows deviation from zero . in response to this deviation , we modify the function @xmath90 $ ] . the value of @xmath90 $ ] at one grid point affects , according eq.([vkpk2 ] ) , the values of @xmath53 and hence condition ( [ epsavav ] ) at other grid points . therefore , the proper iteration requires finding the whole matrix of linear responses of function @xmath221 at each grid point to the change of @xmath63 at each other grid point . if the @xmath0-grid has @xmath222 points , then one has to solve a system of @xmath222 linear equations to find the modification @xmath223 $ ] for each grid point . from our experience such an approach often produces spurious solutions and , therefore , is difficult to manage . instead , we adopt a different approach based on our understanding , why the small-@xmath46 approximation works so well . in the small-@xmath46 limit , the algorithm for finding @xmath90 $ ] is `` local '' , i.e. it does not involve explicit influence of @xmath90 $ ] at a given grid point on @xmath224 at other grid points . in this case , the value of @xmath63 is tried , and if it does not give @xmath225 , then it is modified to @xmath226 , which , according to eq.([vkpk1 ] ) , implies that @xmath53 is also modified to become @xmath227}. \label{vkpkp}\ ] ] this equation can now be rewritten as @xmath228}. \label{vkpkp1}\ ] ] in the above form , the iteration step is generalizable to the case , when @xmath53 itself deviates from the exponential form . we find the value of @xmath229 such that @xmath230 calculated with @xmath47 corresponding to @xmath231 substituted into eq.([pkav ] ) equals zero . such an ansatz for @xmath231 is not fully consistent with eq.([vkpk2 ] ) , but it is a good approximation of it , and more importantly leads to a good improvement in the value of @xmath63 . the version of @xmath53 consistent with eq.([vkpk2 ] ) is then obtained at the beginning of the next iteration step by substituting the entire improved function @xmath90 $ ] into eq.([vkpk2 ] ) , which , in turn , leads to the deviation of @xmath230 from zero , which is again canceled at each @xmath0 by replacing @xmath53 by @xmath231 with the new value @xmath229 , etc . in our experience , the form of @xmath90 $ ] stops noticeably changing after three such iteration cycles . the calculated results presented in section [ results ] are based on six iterations . ultimately , it is the observed quick convergence that justifies the validity of the present algorithm . the rationale for updating the values of @xmath90 $ ] on the basis of eq.([vkpkp1 ] ) is that ( i ) change of @xmath63 at one value of @xmath0 affects mostly @xmath224 for @xmath232 reasonably close to @xmath0 ; and ( ii ) for @xmath232 close to @xmath0 , the value of @xmath233 $ ] should not be very different from @xmath223 $ ] hence @xmath223 $ ] well approximates @xmath233 $ ] once the corrections are substituted in eq.([vkpk2 ] ) . the correction @xmath223 $ ] obtained with the help of eq.([vkpkp1 ] ) accurately anticipates the linear term in the expansion of integral in eq.([vkpk2 ] ) in powers of @xmath46 and the resulting exponential change of @xmath53 for sufficiently small values of @xmath46 . for the larger values of @xmath46 , the exponential factor in eq.([vkpkp1 ] ) introduces a larger relative error , but , in almost all cases , this is a larger error to the small and hence less important tail of @xmath53 . the tail is then corrected at the beginning of the next iteration step once the updated function @xmath90 $ ] is substituted into eq.([vkpk2 ] ) . _ algorithm b _ algorithm a can now be modified to incorporate the finite-@xmath29 corrections associated with the deviations of individual functions @xmath89 $ ] from the `` mean '' function @xmath90 $ ] . algorithm b continues searching for mean function @xmath90 $ ] , but at each step of algorithm a that used eq.([vkpk2 ] ) , it uses eq.([vkpk2a ] ) with values of @xmath89 $ ] obtained from @xmath90 $ ] with the help of ansatz described in section [ ansatz ] . the algorithm continues using the `` local '' step involving eq.([vkpkp1 ] ) , and , here again , the fast convergence of the iteration procedure is the evidence of the adequacy of that step . without energy constraint ( [ epsav ] ) , there exist a straightforward and efficient algorithm for the sampling of hilbert space . namely , one can first choose all values of re@xmath234 and im@xmath234 according to the same gaussian probability distribution . this gives unnormalized wave function . after it is normalized , the resulting monte - carlo points have the required isotropic probability density on the normalization manifold ( provided one can control the effects of the cutoff on the tail of gaussian distribution necessarily appearing in the numerical implementation of the sampling ) . with energy constraint ( [ epsav ] ) , we are not aware of any algorithm , which would be as efficient as the above one and yet rigorously implement uniform sampling on the energy - normalization intersection manifold . we used the same algorithm as ref.@xcite . namely , we do the sampling in the @xmath235-euclidean space , where we define an orthogonal basis in the @xmath44 dimensional subspace formed by the intersection of energy and normalization hyperplanes ( [ epsav ] ) and ( [ norm ] ) . the orientation of these hyperplanes are determined by the unit vectors , which are orthogonal to them : @xmath236 and @xmath237 . the basis within the intersection manifold is then generated by gramm - schmidt orthogonalization procedure . namely , we make the first basis vector @xmath238 orthogonal to both @xmath239 and @xmath240 ; then @xmath241 orthogonal to @xmath239 , @xmath240 and @xmath238 ; etc . each of @xmath44 vectors @xmath242 then gives rise to a new coordinate axis , which we denote as @xmath243 . one has , of course , much freedom in choosing this basis . in the above @xmath44-dimensional subspace , the qmc polyhedron is further constrained by the positivity conditions @xmath244 . as a result , it has @xmath245 vertices@xcite , where @xmath246 , and @xmath247 are the numbers of energy levels in spectrum @xmath34 above and below @xmath0 , respectively . in the original basis of space @xmath235 , each vertex has only two non - zero coordinates @xmath248 corresponding to one possible pair of energy levels such that @xmath249 and @xmath250 . the values of these coordinates are @xmath251 and @xmath252 . we choose the origin of the @xmath243-coordinate system on one of the above vertices . the @xmath42 coordinates of each vertex are then transformed into @xmath243 coordinates . then , in the @xmath253-coordinate system , we choose an @xmath44-dimensional hyperrectangle ( orthotope ) with edges parallel to the @xmath243 axes and with the extension along each axis limited by @xmath254 and @xmath255 such that @xmath243-coordinate of each polyhedron vertex falls between @xmath254 and @xmath255 , i.e. the polyhedron of interest is completely inside the above hyperrectangle . the monte - carlo sampling is then straightforwardly implemented in the hyperrectangle by choosing each coordinate @xmath243 randomly in the interval @xmath256 $ ] . the monte - carlo point is accepted , when , after transformation to the original @xmath42-coordinate system , all of its coordinates satisfy the positivity condition @xmath244 . the acceptance rate of this algorithm decreases exponentially with increasing @xmath29 , because high - dimensional polyhedron typically occupies exponentially small part of the volume of a hyperrectangle that covers it . the above ratio of volumes has a further strong dependence on the relative orientation of the hyperrectangle and the qmc polyhedron . it was our guess confirmed empirically that this ratio is strongly increased , once one of the faces of the polyhedron is parallel to one of the faces of the hyperrectangle . the faces of the polyhedron are determined in the @xmath42-space by the intersection of three hyperplanes : @xmath257 , energy hyperplane ( [ epsav ] ) and normalization hyperplane ( [ norm ] ) . if one of vectors @xmath242 is orthogonal to one of the polyhedron faces , then that face will be parallel to the corresponding face of the hyperrectangle . for example , if we want @xmath238 to be orthogonal to the face determined by hyperplane @xmath258 , then we choose @xmath238 to be parallel to @xmath259 , where @xmath260 in the original @xmath42-space . for @xmath145 , the resulting increase in algorithm s acceptance rate was , approximately , by a factor of 50 . we have not yet systematically optimized the choice of the `` parallel '' polyhedron face as a function of the corresponding energy @xmath32 from the viewpoint of increasing the polyhedron - to - hyperrectangle volume ratio , but , so far , our experience indicates that , for @xmath108 , the face corresponding to the lowest energy , i.e. originating from hyperplane @xmath261 , is one of the best . the algorithm can , perhaps , be further improved by optimizing the subsequent choice of @xmath241 , @xmath262 , etc . from the viewpoint of increasing the above volume ratio . 10 g. jona - lasinio and c. presilla . on the statistics of quantum expectations for systems in thermal equilibrium . in a. bassi , d. durr , t. weber , and n. zanghi , editors , _ quantum mechanics : are there quantum jumps ? and on the present status of quantum mechanics _ , volume 844 of _ aip conference proceedings _ , pages 200205 , 2006 .

we present a comprehensive study of vicsek - style self - propelled particle models in two and three space dimensions . the onset of collective motion in such stochastic models with only local alignment interactions is studied in detail and shown to be discontinuous ( first - order like ) . the properties of the ordered , collectively moving phase are investigated . in a large domain of parameter space including the transition region , well - defined high - density and high - order propagating solitary structures are shown to dominate the dynamics . far enough from the transition region , on the other hand , these objects are not present . a statistically - homogeneous ordered phase is then observed , which is characterized by anomalously - strong density fluctuations , superdiffusion , and strong intermittency .

0712.2062

collective motion phenomena in nature have attracted the interest of scientists and other authors for quite a long time @xcite . the question of the advantage of living and moving in groups , for instance , is a favorite one among evolutionary biologists @xcite . in a different perspective , physicists are mostly concerned with the mechanisms at the origin of collective motion , especially when it manifests itself as a true , non - trivial , emerging phenomenon , i.e. in the absence of some obvious cause like the existence of a leader followed by the group , a strong geometrical constraint forcing the displacement , or some external field or gradient felt by the whole population . moreover , the ubiquity of the phenomenon at all scales , from intra - cellular molecular cooperative motion to the displacement in group of large animals , raises , for physicists at least , the question of the existence of some universal features possibly shared among many different situations . one way of approaching these problems is to construct and study minimal models of collective motion : if universal properties of collective motion do exist , then they should appear clearly within such models and thus could be efficiently determined there , before being tested for in more elaborate models and real - world experiments or observations . such is the underlying motivation of recent studies of collective motion by a string of physicists @xcite . among them , the group of tamas vicsek has put forward what is probably the simplest possible model exhibiting collective motion in a non - trivial manner . in the `` vicsek model '' @xcite , point particles move off - lattice at constant speed @xmath0 , adjusting their direction of motion to that of the average velocity of their neighbors , up to some noise term accounting for external or internal perturbations ( see below for a precise definition ) . for a finite density of particles in a finite box , perfect alignment is reached easily in the absence of noise : in this fluctuation - less collective motion , the macroscopic velocity equals the microscopic one . on the other hand , for strong noise particles are essentially non - interacting random walkers and their macroscopic velocity is zero , up to statistical fluctuations . vicsek _ et al . _ showed that the onset of collective motion occurs at a finite noise level . in other words , there exists , in the asymptotic limit , a fluctuating phase where the macroscopic velocity of the total population is , on average , finite . working mostly in two space dimensions , they concluded , on the basis of numerical @xcite simulations , that the onset of this ordered motion is well described as a novel non - equilibrium continuous phase transition leading to long range order , at odds with equilibrium where the continuous xy symmetry can not be spontaneously broken in two space dimensions and below @xcite . this brought support to the idea of universal properties of collective motion since the scaling exponents and functions associated to such phase transitions are expected to bear some degree of universality , even out of equilibrium . the above results caused a well - deserved stir and prompted a large number of studies at various levels @xcite . in particular , two of us showed that the onset of collective motion is in fact discontinuous @xcite , and that the original conclusion of vicsek _ et al . _ was based on numerical results obtained at too small sizes @xcite . more recently , the discontinuous character of the transition was challenged in two publications , one by the vicsek group @xcite and another by aldana _ et al . _ @xcite . here , after a definition of the models involved ( section [ models ] ) , we come back , in section [ transition ] , to this central issue and present a rather comprehensive study of the onset of collective motion in vicsek - style models . in section [ ordered ] , we describe the ordered , collective motion phase . section [ discussion ] is devoted to a general discussion of our results together with some perspectives . most of the numerical results shown were obtained in two space dimensions , but we also present three - dimensional results . wherever no explicit mention is made , the default space dimension is two . similarly , the default boundary conditions are periodic in a square or cubic domain . let us first recall the dynamical rule defining the vicsek model @xcite . point particles labeled by an integer index @xmath1 move off - lattice in a space of dimension @xmath2 with a velocity @xmath3 of fixed modulus @xmath4 . the _ direction _ of motion of particle @xmath1 depends on the average velocity of all particles ( including @xmath1 ) in the spherical neighbourhood @xmath5 of radius @xmath6 centered on @xmath1 . the discrete - time dynamics is synchronous : the direction of motion and the position of all particles are updated at each timestep @xmath7 , in a driven , overdamped manner : @xmath8 \label{evicsek}\ ] ] where @xmath9 is a normalization operator ( @xmath10 ) and @xmath11 performs a random rotation uniformely distributed around the argument vector : in @xmath12 , @xmath13 is uniformely distributed around @xmath14 inside an arc of amplitude @xmath15 ; in @xmath16 , it lies in the solid angle subtended by a spherical cap of amplitude @xmath17 and centered around @xmath14 . the particles positions @xmath18 are then simply updated by streaming along the chosen direction as in @xmath19 note that the original updating scheme proposed by vicsek _ _ in @xcite defined the speed as a backward difference , although we are using a forward difference . the simpler updating above , now adopted in most studies of vicsek - style models , is not expected to yield different results in the asymptotic limit of infinite size and time . the `` angular '' noise term in the model defined above can be thought of as arising from the errors committed when particles try to follow the locally - averaged direction of motion . one could argue on the other hand that most of the randomness stems from the evaluation of each interaction between particle @xmath1 and one of its neighbors , because , e.g. , of perception errors or turbulent fluctuations in the medium . this suggests to replace eq.([evicsek ] ) by : @xmath20 \label{en2}\ ] ] where @xmath21 is a random unit vector and @xmath22 is the number of particles in @xmath5 . it is easy to realize that this `` vectorial '' noise acts differently on the system . while the intensity of angular noise is independent from the degree of local alignment , the influence of the vectorial noise decreases with increasing local order . in the original formulation of the vicsek model as well as in the two variants defined above , the only interaction is _ alignment_. in a separate work @xcite , we introduced a two - body repulsion / attraction force , to account for the possibility of maintaining the cohesion of a flock in an infinite space ( something the vicsek model does not allow ) . here , we only study models without cohesion . nevertheless , we have considered , in the following , the case of a pairwise repulsion `` force '' , to estimate in particular the possible influence of the absence of volume exclusion effects in the basic model , which leaves the local density actually unbounded . we thus introduce the short ranged , purely repulsive interaction exerted by particle @xmath23 on particle @xmath1 : @xmath24^{-1 } \ ; , \label{eforce2}\ ] ] where @xmath25 is the unit vector pointing from particle @xmath1 to @xmath23 and @xmath26 is the typical repulsion range . equations ( [ evicsek ] ) and ( [ en2 ] ) are then respectively generalized to @xmath27 \label{evicsek2}\ ] ] and @xmath28\ ; , \label{en3}\ ] ] where @xmath29 measures the relative strength of repulsion with respect to alignment and noise strength . the natural order parameter for our polar particles is simply the macroscopic _ mean velocity _ , conveniently normalized by the microscopic velocity @xmath0 @xmath30 where @xmath31 stands for the average over the whole population . here , we mostly consider its modulus @xmath32 , the _ scalar order parameter_. in the following , we set , without loss of generality , @xmath33 and @xmath34 , and express all time- and length - scales in terms of these units . moreover repulsive force will be studied by fixing @xmath35 and @xmath36 . this leaves us with two main parameters for these models : the noise amplitude @xmath37 and the global density of particles @xmath38 . recently , the microscopic velocity @xmath0 has been argued to play a major role as well @xcite . all three parameters ( @xmath37 , @xmath38 , and @xmath0 ) are considered below . ( color online ) typical behavior across the onset of collective motion for moderate size models ( @xmath39 , @xmath40 , @xmath41 ) with angular noise ( black circles ) and vectorial noise ( red triangles ) . the reduced noise amplitude @xmath42 is shown in abscissas ( transition points estimated at @xmath43 vectorial noise and @xmath44 angular noise ) . ( a ) time - averaged order parameter @xmath45 . ( b ) binder cumulant g. ( c ) variance @xmath46 of @xmath47 ; ( d ) order parameter distribution function @xmath48 at the transition point . bimodal distribution for vectorial noise dynamics ( red dashed line ) , unimodal shape for angular noise ( black solid line ) . time averages have been computed over @xmath49 timesteps.,width=325 ] as mentioned above , the original vicsek model attracted a lot of attention mostly because of the conclusions drawn from the early numerical studies @xcite : the onset of collective motion was found to be a novel continuous phase transition spontaneously breaking rotational symmetry . however , it was later shown in @xcite that beyond the typical sizes considered originally the _ discontinuous _ nature of the transition emerges , irrespective of the form of the noise term . recently , the discontinuous character of the transition was argued to disappear in the limit of small @xmath0 @xcite . we now address the problem of the nature of the transition in full detail . ( color online ) fss analysis of angular noise dynamics ( @xmath39 , @xmath40 , time averages computed over @xmath50 timesteps ) . time - averaged order parameter ( a ) and binder cumulant ( b ) as a function of noise for various system sizes @xmath51 . ( c ) piece of an order parameter time series close to the transition point ( @xmath52 , @xmath53).,width=325 ] even though there is no rigorous theory for finite - size - scaling ( fss ) for out - of - equilibrium phase transitions , there exists now ample evidence that one can safely rely on the knowledge gained in equilibrium systems @xcite . the fss approach @xcite involves the numerical estimation of various moments of the order parameter distribution as the linear system size @xmath51 is systematically varied . of particular interest are the variance @xmath54 and the so - called binder cumulant @xmath55 where @xmath56 indicates time average . the binder cumulant is especially useful in the case of continuous phase transitions , because it is one of the simplest ratio of moments which takes a universal value at the critical point @xmath57 , where all the curves @xmath58 , obtained at different system sizes @xmath51 , cross each other . at a first - order transition point , on the other hand , the binder cumulant exhibits a sharp drop towards negative values @xcite . this minimum is due to the simultaneous contributions of the two phases coexisting at threshold . moreover , it is easy to compute that @xmath59 in the ordered phase , while for a disordered state with a continuous rotational simmetry one has @xmath60 in @xmath12 and @xmath61 in @xmath16 . ( color online ) transition to collective motion with short range repulsive interactions . left panels : angular noise . right panels : vectorial noise . ( a , b ) : order parameter vs noise amplitude at different system sizes . ( c , d ) : binder cumulant @xmath62 as a function of noise amplitude . ( @xmath39 , @xmath63 , time averages carried over @xmath64 timesteps).,width=325 ] as an overture , we analyze systems of moderate size in two dimensions ( @xmath65 particles ) at the density @xmath39 , typical of the initial studies by vicsek _ et al . _ , but with the slightly modified update rule ( [ evol ] ) and for both angular and vectorial noise . the microscopic velocity is set to @xmath40 . for angular noise , the transition looks indeed continuous , as found by vicsek _ et al._. on the other hand , the time averaged scalar order parameter @xmath66 displays a sharp drop for vectorial noise , and the binder cumulant exibits a minimum at the transition point , indicating a discontinuous phase transition ( fig . [ fcompare]a - b ) . simultaneously , the variance is almost delta - peaked . the difference betwen the two cases is also recorded in the probability distribution function ( pdf ) of @xmath47 which is bimodal ( phase coexistence ) in the vectorial noise case ( fig . [ fcompare]c - d ) . ( color online ) transition to collective motion in three spatial dimensions . left panels : angular noise . right panels : vectorial noise . ( a , b ) : time - averaged order parameter vs. noise amplitude at different system sizes . ( c , d ) : binder cumulant @xmath62 as a function of noise amplitude . ( @xmath67 , @xmath40 , time averages carried over @xmath68 timesteps).,width=325 ] the qualitative difference observed upon changing the way noise is implemented in the dynamics is , however , only a finite - size effect . as shown in @xcite , the transition in the angular noise case reveals its asymptotic discontinuous character provided large - enough system sizes @xmath51 are considered ( fig . [ fig2]a - b ) . remaining for now at a qualitative level , we show in fig . [ fig2]c a typical time - series of the order parameter for the angular noise case in a large system in the transition region . the sudden jumps from the disordered phase to the ordered one and vice - versa are evidence for metastability and phase coexistence . note that the system size beyond which the transition reveals its discontinuous character for the angular noise case at density @xmath39 and velocity @xmath40 the conditions of the original papers by vicsek _ et al._ is of the order of @xmath69 , the maximum size then considered . it is clear also from fig . [ fcompare ] that the discontinuous nature of the transition appears earlier , when increasing system size , for vectorial noise than for angular noise . thus , finite - size effects are stronger for angular noise . the same is true when one is in presence of the repulsive interactions ( fig.[ffssvm2 ] ) . finally , the same scenario holds in three space dimensions , with a _ discontinous _ phase transition separating the ordered from the disordered phases for both angular and vectorial noise ( fig . [ fig3d ] ) . before proceeding to a study of the complete phase diagram , we detail now how a comprehensive fss study can be performed on a particular case . correlation time @xmath70 of the order parameter near the transition point for vectorial noise dynamics with repulsion . system parameters are @xmath39 , @xmath40 and @xmath71 . ( a ) time correlation function @xmath72 at @xmath69 . the lin - log inset shows the exponential decay . ( b ) correlation time @xmath70 as a function of system size @xmath51 . the dashed line marks linear growth with @xmath51 . correlation functions were computed on samples of @xmath73 realizations for typically @xmath74 timesteps.,width=325 ] for historical reasons , the following study has been performed on the model with vectorial noise and repulsive force ( eq . [ en3 ] ) . it has not been repeated in the simpler case of the `` pure '' vicsek model because its already high numerical cost would have been prohibitive due to the strong finite size effects . as a first step , we estimated the correlation time @xmath75 , whose knowledge is needed to control the quality of time - averaging : the duration @xmath76 of numerical simulations has been taken much larger than @xmath75 ( @xmath77 in the largest systems , but typically @xmath78 for smaller sizes ) . moreover , @xmath70 is also useful to correctly estimate the statistical errors on the various moments ( as @xmath66 , @xmath46 , and @xmath62 ) of the pdf of the order parameter , for which we used the jackknife procedure @xcite . the correlation time was estimated near the transition ( where it is expected to be largest ) as function of system size @xmath51 measuring the exponential decay rate of the correlation function ( fig . [ ftau]a ) @xmath79 we found @xmath70 to vary roughly linearly with @xmath51 ( see fig . [ ftau]b ) . it is interesting to observe that at equilibrium , one would expect @xmath70 to scale as @xcite @xmath80 where @xmath81 is the surface tension of the metastable state . therefore , our result implies a very small or vanishing surface tension @xmath82 , a situation reminiscent of observations made in the cohesive case @xcite , where the surface tension of a cohesive droplet was found to vanish near the onset . + following borgs and kotecky @xcite , the asymptotic coexistence point @xmath57 ( i.e. the first order transition point ) can be determined from the asymptotic convergence of various moments of the order parameter pdf . first , the observed discontinuity in @xmath45 , located at @xmath83 , is expected to converge exponentially to @xmath57 with @xmath51 . second , the location of the susceptibility peak @xmath84 which is the same as the peak in @xmath46 provided some fluctuation - dissipation relation holds , see the appendix also converges to @xmath57 , albeit algebraically with an exponent @xmath85 . third , the location of the minimum of @xmath62 , @xmath86 , is also expected to converge algebraically to @xmath57 with with an exponent @xmath87 . interestingly , the value taken by these exponents actually depends on the number of phases and of the dimension @xmath2 of the system : for two - phase coexistence one has @xmath88 , while for more than two phases @xmath89 . in figure [ fdisc ] , we show that our data are in good agreement with all these predictions . the three estimates of @xmath57 are consistent with each other within numerical accuracy . moreover @xmath83 is found to converge exponentially to the transitional noise amplitude , while both @xmath84 and @xmath86 show algebraic convergence with an exponent close to 2 . this agrees with the fact that due to the continuous rotational symmetry , the ordered phase is degenerate and amounts to an infinite number of possible phases . , @xmath63 ) . convergence of the finite - size transition points measured from different moments of the order parameter fss to the asymptotic transition point @xmath57 ( see fig . [ ffssvm2 ] ) . upper panels : finite size transition points estimated from ( a ) time average , ( b ) variance and ( c ) binder cumulant . the horizontal dashed line marks the estimated asymptotic threshold @xmath90 . lower panels : scaling of the finite size reduced noise @xmath42 transition point . ( d ) exponential convergence for the jump location in the time - averaged order parameter . ( e ) power - law behavior of the variance peak position . ( f ) power - law behavior of the binder cumulant minimum . the dashed lines in ( e - f ) mark the estimated exponents @xmath91.,width=325 ] one of the classical hallmarks of discontinuous phase transitions is the presence , near the transition , of the hysteresis phenomenon : ramping the control parameter at a fixed ( slow ) rate up and down through the transition point , a hysteresis loop is formed , inside which phase coexistence is manifest ( see fig . [ hyst3d]a for the @xmath16 case with vectorial noise ) . the size of such hysteresis loops varies with the ramping rate . an intrinsic way of assessing phase coexistence and hysteresis is to study systematically the nucleation time @xmath92 needed to jump from the disordered phase to the ordered one , as well as @xmath93 the decay time after which the ordered phase falls into the disordered one . [ hyst3d]b shows , in three space dimensions , how these nucleation and decay vary with @xmath37 at two different sizes . a sharp divergence is observed , corresponding to the transition point . at a given time value @xmath70 , one can read , from the distance between the `` up '' and the `` down '' curve , the average size of hysteresis loops for ramping rates of the order of @xmath94 . per time step ( @xmath95 , @xmath40 , @xmath96 ) . empty circles mark the path along the adiabatic increase of noise amplitude , full triangles for adiabatic decrease . ( b ) nucleation times from the disordered phase to the ordered phase ( @xmath92 , left curves ) and vice - versa ( @xmath93 , right curves ) for two system sizes ( other parameters as in ( a ) ) . each point is averaged over @xmath97 realizations . , width=325 ] the above detailed fss study would be very tedious to realize when varying systematically the main parameters @xmath37 , @xmath38 , and @xmath98 , as well as the nature of the noise and the presence or not of repulsive interactions . from now on , to characterize the discontinuous nature of the transition , we rely mainly on the presence , at large - enough system sizes @xmath51 , of a minimum in the variation of the binder cumulant @xmath62 with @xmath37 ( all other parameters being fixed ) . we call @xmath99 the crossover size marking the emergence of a minimum of @xmath100 . we are now in the position to sketch the phase diagram in the ( @xmath37 , @xmath38 , @xmath0 ) parameter space . the numerical protocol used is , at given parameter values , to run a large - enough system so that the discontinuous character of the transition is seen ( i.e. @xmath101 ) . for larger sizes , the location of the transition point typically varies very little , so that for most practical purposes , locating the ( asymptotic ) transition point from systems sizes around @xmath99 is satisfactory . ( color online ) asymptotic phase diagrams for the transition to collective motion . ( a ) two space dimensions : threshold amplitude @xmath57 for angular noise as a function of density @xmath38 at @xmath40 . inset : log - log plot to compare the low density behavior with the mean field predicted behavior @xmath102 ( dashed red line ) . ( b ) as in panel ( a ) , buth with vectorial noise dynamics . ( c ) noise - density phase diagram in three dimensions for vectorial noise dynamics at fixed velocity @xmath40 . in the log - log inset the transition line can be compared with the predicted behavior @xmath103 ( dashed red line ) . ( d ) two space dimensions : threshold amplitude @xmath57 for angular noise as a function of particle velocity @xmath0 at fixed density @xmath95 ( black circles ) and @xmath104 ( red triangles ) . the horizontal dashed line marks the noise amplitude considered in ref . @xcite ( see section [ sumtransition]).,width=325 ] ( color online ) first - order transition for angular noise dynamics at high ( left panels ) and low velocity @xmath0 ( right panels ) . typical averaging time is @xmath73 timesteps . ( a ) time - averaged order parameter and ( c ) binder cumulant at large particle velocity for angular noise in two spatial dimensions at increasing velocities and @xmath105 ( @xmath39 ) . ( b ) time averaged order parameter and ( d ) binder cumulant for @xmath106 and two increasing system sizes ( @xmath95).,width=325 ] the results presented below are in agreement with simple mean - field - like arguments in the diluted limit : in the small-@xmath38 regime , one typically expects that the lower the density , the lower the transitional noise amplitude @xmath57 . indeed , for @xmath107 of the order of or not much smaller than the interaction range @xmath6 and in the low - density limit @xmath108 , the system can be seen as a dilute gas in which particles interact by short range ordering forces only . in this regime , the persistence length of an isolated particle ( i.e. the distance travelled before its velocity loses correlation with its initial direction of motion ) varies like @xmath109 . to allow for an ordered state , the noise amplitude should be small enough so that the persistence length remains larger than the average inter - particle distance , _ i.e. _ @xmath110 . thus the transition noise amplitude is expected to behave as @xmath111 in @xcite , it was indeed found that @xmath112 with @xmath113 in two dimensions . our own data ( fig . [ phase]a - c ) now confirm eq . ( [ meanfield ] ) for both the angular and vectorial noise in two and three spatial dimensions , down to very small @xmath38 values . the data deviate from the square - root behavior as the average inter - particle distance gets of the order of or smaller than the interaction range . finally , we also investigated the transition line when @xmath0 is varied ( fig . [ phase]d ) . for the vectorial noise case , at fixed density , the threshold noise value @xmath57 is almost constant ( data obtained at @xmath114 , not shown ) . for the angular noise , in the small @xmath0 limit where the above mean - field argument does not apply , we confirm the first - order character of the phase transition down to @xmath115 for both angular and vectorial noise ( fig . [ specialcases]b , d ) . for even smaller values of @xmath0 , the investigation becomes numerically too costly ( see section below ) . note that @xmath57 seems to be finite when @xmath116 , a limit corresponding to the xy model on a randomly connected graph . still for angular noise , the large velocity limit is also difficult to study numerically . again , we observe that the transition is discontinuous as far as we can probe it , i.e. @xmath117 ( fig . [ specialcases]a , b ) . we now discuss particular limits of the models above together with the relative importance of finite - size effects . recall that these are quantified by the estimated value of the crossover size @xmath99 beyond which the transition appears discontinuous . all the following results have been obtained for @xmath12 . partial results in three dimensions indicate that the same conclusions should hold there . keep in mind that in all cases reported , the transition is discontinuous . we are just interested here in how large a system one should use in order to reach the asymptotic regime . ( color online ) crossover system size @xmath99 above which the discontinuous character of the transition appears ( as testified by the existence of a minimum in the @xmath100 curve ) . black circles : angular noise . red triangles : vectorial noise . ( a ) @xmath99 vs @xmath38 for @xmath40 ; inset : the low density behavior in log - log scales ; the dashed line marks a power - law divergence proportional to @xmath118 . ( b ) @xmath99 vs @xmath0 at fixed density ( open symbols : @xmath119 ; filled symbols @xmath120).,width=325 ] fig . [ fs]a shows that finite - size effects are stronger for angular noise than for vectorial noise for all densities @xmath38 at which we are able to perform these measurements . note in particular that at @xmath39 , the density originally used by vicsek _ et al . _ , @xmath121 for angular noise , while it is very small for vectorial noise , confirming the observation made in section [ overture ] . in the small-@xmath38 limit , the discontinuous character of the transition appears later and later , with @xmath99 roughly diverging as @xmath118 ( inset of fig . note that this means that in the small-@xmath38 limit one needs approximately the same number of particles to start observing the discontinuity . the large-@xmath38 limit reveals a difference between angular and vectorial noise : while @xmath99 remains small for vectorial noise , it seems to diverge for angular noise ( fig . [ fs]a ) , making this case difficult to study numerically . we also explored the role the microscopic velocity @xmath0 in the strength of finite - size effects . qualitatively , the effects observed are similar to those just reported when the density is varied ( fig . [ fs]b ) . in the small @xmath0 limit , we record a strong increase of @xmath99 as @xmath122 for both types of noise . in the large velocity limit , @xmath99 decreases for vectorial noise , whereas it increases for angular noise . the summary of the above lengthy study of the order - disorder transition in vicsek - like models is simple : for any finite density @xmath38 , any finite velocity @xmath0 , and for both types of noise introduced , the transition is discontinuous . this was observed even in the numerically - difficult limits of large or small @xmath38 or @xmath0 . these results contradict recent claims made about the angular noise case ( original vicsek model ) . we now comment on these claims . vicsek and co - workers @xcite , showed that keeping the density and the noise intensity fixed , a qualitative change is observed when decreasing @xmath0 : for not too small @xmath0 values , in the ordered phase , particles diffuse anisotropically ( and the transition is discontinuous ) , while diffusion becomes isotropic at small @xmath0 , something interpreted as a sign of a continuous transition in this region . rather than the convoluted arguments presented there , what happens is in fact rather simple : decreasing @xmath0 at fixed @xmath38 and @xmath37 , one can in fact cross the transition line , passing from the ordered phase ( where particles obviously diffuse anisotropically due to the transverse superdiffusive effects discussed in section [ superdiff ] ) to the disordered phase . our figure [ phase]d , obtained in the same conditions as in @xcite ( apart from harmless change of time updating rule ) , shows that if one keeps @xmath123 ( as in @xcite ) , one crosses the transition line at about @xmath124 , the value invoked by vicsek and co - workers to mark a crossover from discontinuous to `` continuous '' transitions . in a recent letter @xcite , aldana _ et al . _ study order - disorder phase transitions in random network models and show that the nature of these transitions may change with the way noise is implemented in the dynamics ( they consider the angular and vectorial noises defined here ) . arguing that these networks are limiting cases of vicsek - like models , they claim that the conclusions reached for the networks carry over to the transition to collective motion of the vm - like systems . they conclude in particular that in the case of `` angular '' noise the transition to collective motion is continuous . we agree with the analysis of the network models , but the claim that they are relevant as limits of vicsek - like models is just wrong : the data presented there ( figure 1 of @xcite ) to substantiate this claim is contradicted by our figures [ specialcases]a , c ( see also @xcite ) obtained at larger system sizes . again , for large - enough system sizes , the transition is indeed discontinuous . thus , at best , the network models of aldana _ et al . _ constitute a singular @xmath125 limit of vicsek - like models . we now turn our attention to the ordered , symmetry - broken phase . in previous analytical studies , it has often been assumed that the density in the ordered phase is spatially homogeneous , albeit with possibly large fluctuations ( see , e.g. @xcite ) . this is indeed what has been reported in early numerical studies , in particular by vicsek _ et al . _ @xcite . in the following , we show that this is not true in large enough systems , where , for a wide range of noise amplitudes near the transition point , density fluctuations lead to the formation of localized , travelling , high - density and high - order structures . at low enough noise strength , though , a spatially - homogeneous ordered phase is found , albeit with unusually strong density fluctuations . , @xmath40 , @xmath126 and increasing system sizes - respectively @xmath41 , @xmath52 and @xmath127 . sharp bands can only be observed if @xmath51 is larger than the typical band width @xmath128 . ( d ) vectorial noise : @xmath95 , @xmath40 , @xmath129 and @xmath41 : bands appear at relatively small system sizes for this type of noise . for clarity reasons , only a representative sample of 10000 particles is shown in ( b ) and ( c ) . boundary conditions are periodic.,width=302 ] but in different geometries / boundary conditions or space dimensions . ( a)-(b ) vectorial noise ( @xmath130 , @xmath104 , and @xmath40 ) , boundary conditions are periodic along the @xmath131 ( vertical ) axis and reflecting in @xmath132 . ( a ) : a long single band travels along the periodic direction . ( b ) : the domain size along the periodic direction is too small to accomodate bands , and a single band bouncing back and forth along the non - periodic direction is observed . ( c ) angular noise , repulsive force , and periodic boundary conditions ( @xmath39 , @xmath133 and @xmath63 ) . ( d ) high - density sheet traveling in a three - dimensional box with periodic boundary conditions ( angular noise with amplitude @xmath134 , @xmath95 and @xmath40).,width=325 ] numerical simulations of the ordered phase dynamics ( @xmath135 ) , performed at large enough noise amplitudes , are characterized by the emergence of high - density moving bands ( @xmath12 ) or sheets ( @xmath16 ) . typical examples are given in figs . [ photo1][photo2 ] . these moving structures appear for large - enough systems after some transient . they extend transversally with respect to the mean direction of motion , and have a center of mass velocity close to @xmath0 . while particles inside bands are ordered and , in the asymptotic regime , move coherently with the global mean velocity , particles lying outside bands in low density regions are not ordered and perform random walks . as shown in fig . [ photo1]a - c for angular noise dynamics ( [ evicsek ] ) , there exists a typical system size @xmath136 , below which the bands or sheets can not be observed . numerical simulations indicate that @xmath136 depends only weakly on the noise amplitude and is of the same order of magnitude as the crossover size marking the appearance of the discontinuous character of the transition : @xmath137 . it is therefore numerically easier to observe bands in the ordered phase of vectorial noise dynamics ( [ en2 ] ) , as in fig [ photo1]d . bands may be observed asymptotically without and with a repulsive interaction ( fig [ photo2]c ) and for both kinds of noise . they appear for various choices of boundary conditions ( see for instance fig . [ photo2]a - b , where reflecting boundary conditions have been employed ) , which may play a role in determining the symmetry - broken mean direction of motion . for instance , bands travelling parallel to one of the axis are favoured when periodic boundary conditions are employed in a rectangular box ( they represent the simplest way in which an extended structure can wrap around a torus , and are thus reached more easily from disordered initial conditions ) , but bands travelling in other directions may also appear , albeit with a smaller probability . bands can be described quantitatively through local quantities , such as the local density @xmath138 , measured inside a domain @xmath139 centered around @xmath140 , and the local order parameter @xmath141 further averaging these local quantities perpendicularly to the mean velocity ( [ globalop ] ) one has the _ density profile _ @xmath142 and the _ order parameter profile _ @xmath143 , where @xmath144 indicates the longitudinal direction w.r.t . mean velocity . bands are characterized by a sharp kink in both the density and the order parameter profiles ( see fig . [ fband]a and [ fband]c - d ) . they are typically asymmetric , as it can be expected for moving structures , with a rather sharp front edge , a well defined mid - height width @xmath128 which typically is of the same order as @xmath145 and an exponentially decaying tail with a characteristic decay length of the order of the @xmath128 ( fig . [ fband]b ) . , @xmath146 and @xmath147 ) . ( b ) tail of the density profile shown in ( d ) ( black line ) and its fit ( blue dashed line ) by the formula : @xmath148 , with @xmath149 ( lin - log scales ) . ( c - d ) traveling sheet in three dimensions ( angular noise , @xmath119 , @xmath150 , and @xmath147 ) . ( c ) projection of particle positions on a plane containing the global direction of motion ( marked by red arrow ) . ( d ) density ( black line ) and order parameter ( dashed red line ) profiles along the direction of motion @xmath144.,width=302 ] large systems may accomodate several bands at the same time , typically all moving in the same direction ( see for instance fig . [ photo1]c and the density profile in fig . [ longband]e ) . however , they do not form well - defined wave trains , but rather a collection of solitary objects , as hinted by the following numerical experiments . ) from a spatially - homogeneous ( uniformely distributed random positions ) initial condition with all particle velocities oriented along the major axis of a @xmath151 domain with periodic boundary conditions . vectorial noise of amplitude @xmath146 , density @xmath39 , and @xmath40 . ( a ) : space - time plot of the density profile ( time is running from left to right from @xmath152 to @xmath153 , while the longitudinal direction is represented in ordinates . colour scale from blue ( low values ) to red ( high values ) . ) . ( b ) : same as ( a ) but at later times ( from @xmath154 to @xmath155 ) . ( c ) : spatial fourier power spectrum @xmath156 of an early density profile ( @xmath153 ) . ( d ) : early time evolution of selected fourier modes ( chosen between spectrum peaks of ( c ) ) , @xmath157 ( inset : average over 50 different runs ) . ( e ) : density profile at a late time ( @xmath155 , final configuration of ( b ) ) . ( f ) : same as ( c ) but for the late time density profile of ( e ) . the quality of this figure has been reduced to meet the arxiv size constraint . , title="fig:",width=325 ] ) from a spatially - homogeneous ( uniformely distributed random positions ) initial condition with all particle velocities oriented along the major axis of a @xmath151 domain with periodic boundary conditions . vectorial noise of amplitude @xmath146 , density @xmath39 , and @xmath40 . ( a ) : space - time plot of the density profile ( time is running from left to right from @xmath152 to @xmath153 , while the longitudinal direction is represented in ordinates . colour scale from blue ( low values ) to red ( high values ) . ) . ( b ) : same as ( a ) but at later times ( from @xmath154 to @xmath155 ) . ( c ) : spatial fourier power spectrum @xmath156 of an early density profile ( @xmath153 ) . ( d ) : early time evolution of selected fourier modes ( chosen between spectrum peaks of ( c ) ) , @xmath157 ( inset : average over 50 different runs ) . ( e ) : density profile at a late time ( @xmath155 , final configuration of ( b ) ) . ( f ) : same as ( c ) but for the late time density profile of ( e ) . the quality of this figure has been reduced to meet the arxiv size constraint . , title="fig:",width=325 ] we investigated the instability of the density - homogeneous , ordered state in a series of numerical simulations starting from particles uniformly distributed in space but strictly oriented along the major axis in a large rectangular domain . figure [ longband]a , b show space - time plots of the density profile : initially flat , it develops structures with no well - defined wavelength ( fig . [ longband]c ) . density fluctuations destroy the initially - ordered state in a rather unusual way : a dynamical fourier analysis of the density profile show a weakly - peaked , wide band of wavelengths growing _ subexponentially _ ( fig . [ longband]d ) . this is at odds with a finite - wavelength supercritical instability , which would lead to a wavetrain of traveling bands . furthermore , the asymptotic ( late time ) power spectra of the density profiles are not peaked around a single frequency either , but rather broadly distributed over a large range of wavenumbers ( fig . [ longband]f ) . in the asymptotic regime , bands are extremely long - lived metastable ( or possibly stable ) objects , which are never equally - spaced ( a typical late - time configuration is shown in fig . [ longband]e ) . to summarize , the emerging band or sheet structure in the asymptotic regime is not a regular wave train characterized by a single wavelength , but rather a collection of irregularly - spaced localized traveling objects , probably weakly interacting through their exponentially decaying tails . as the noise amplitude is decreased away from the transition point , bands are less sharp , and eventually disappear , giving way to an ordered state characterized by an homogeneous local order parameter and large fluctuations of the local density . a quantitative measure of the presence , in the ordered phase , of structures spanning the dimension transverse to the mean motion ( i.e. bands or sheets ) is provided by the variances of the density and order parameter profiles : @xmath158 where @xmath159 indicates the average of the profile in the longitudinal direction with respect to mean velocity . indeed , these profile widths vanish in the infinite - size limit except if band / sheet structures are present . domain , global motion along the major axis , @xmath39 , and @xmath40 ) . the dashed vertical blue line marks the order - disorder transition . ( c - d ) typical instantaneous profiles along the long dimension of the system described in ( a - b ) for intermediate noise value ( ( b ) @xmath160 ) and in the bandless regime ( ( c ) @xmath161).,width=325 ] in fig [ bandgnf]a , b , we plot these profile widths averaged over time as a function of noise amplitude . both quantities present a maximum close to the transition point in the ordered phase , and drop drastically as soon as the disordered phase is entered . lowering the noise away from the transition point , these profiles decrease steadily : bands / sheets stand less sharply out of the disordered background ( fig [ bandgnf]c ) . at some point ( @xmath162 for the parameters values considered in fig [ bandgnf]a , b ) , bands rather abruptly disappear and are no more well - defined transversal objects . it is difficult to define this point accurately , but it is clear that for lower noise intensities the local order parameter is strongly homogeneous in space . nevertheless , fluctuations in the density field are strong ( fig [ bandgnf]d ) , but can no more give rise to ( meta)stable long - lived transverse structures . density fluctuations in the bandless regime are in fact anomalously strong : measuring number fluctuations in sub - systems of linear size @xmath163 , we find that their root mean - square @xmath164 does _ not _ scale like the square root of @xmath165 , the mean number of particles they contain ; rather we find @xmath166 with @xmath167 both in @xmath12 and in @xmath16 ( fig . [ sdf - gnf]a ) . this is reminiscent of the recent discovery of `` giant density fluctuations '' in active nematics @xcite . however , the theoretical argument which initially predicted such fluctuations @xcite can not be invoked directly in the present case . ( indeed , the above value of @xmath168 , although needing to be refined , does not seem to be compatible with the prediction @xmath169 made in @xcite . ) more work is needed to fully understand under which circumstances the coupling between density and order in systems of `` active '' , self - propelled particles gives rise to such anomalous density fluctuations . scales approximately like @xmath170 ( the dashed line has slope 0.8 ) both in two dimensions ( black circles , @xmath52 , @xmath39 , @xmath40 , angular noise amplitude @xmath171 ) and in three dimensions ( red triangles , @xmath41 , @xmath95 @xmath40 , vectorial noise amplitude @xmath123 , values shifted for clarity ) . ( b ) average doubling time @xmath172 of the transverse ( w.r.t . mean velocity ) interparticle distance @xmath173 . black circles : ordered bandless regime ( @xmath174 , angular noise amplitude @xmath175 in a rectangular box of size @xmath176 ) . the black dashed line marks the expected growth @xmath177 . red squares : same but deep in the disordered phase ( @xmath174 , angular noise amplitude @xmath178 , @xmath179 ) . the dot dashed red line shows normal diffusive behavior : @xmath180.,width=325 ] according to toner and tu predictions @xcite , the dynamics of the symmetry - broken ordered phase of polar active particles should be characterized by a superdiffusive mean square displacement @xmath181 in the direction(s ) transversal to mean velocity . in particular , in @xmath12 one has @xcite @xmath182 with @xmath183 . while this analytical result has been succesfully tested by numerical simulations of models with cohesive interactions @xcite , numerical simulations in models without cohesion present substantial difficulties , mainly due to the presence of continuously merging and splitting sub - clusters of particles moving coherently ( as discussed in section [ cluster ] ) . as a consequence , an ensemble of test particles in a cohesionless model is exposed to different `` transport '' regimes ( w.r.t . center of mass motion ) which are not well separated in time . when the mean displacement is averaged at fixed time , this tends to mask the transverse superdiffusion . to overcome this problem , we chose to follow @xcite and to measure @xmath172 , the average time taken by two particles to double their transverse separation distance @xmath173 . from eq . ( [ superd ] ) one immediately has @xmath184 with @xmath185 in @xmath12 . in order to easily separate the transverse from the parallel component , we considered an ordered system in a large rectangular domain with periodic boundary conditions and the mean velocity initially oriented along the long side . the mean direction of motion then stays oriented along this major axis , so that we can identify the transverse direction with the minor axis . furthermore , a high density and a small ( angular ) noise amplitude ( corresponding to the bandless regime ) have been chosen to avoid the appearance of large , locally disordered patches . + our results ( fig . [ sdf - gnf]b ) confirm the prediction of toner and tu : transverse superdiffusion holds at low - enough noise , while normal diffusion is observed in the disordered , high - noise phase . note that the systematic deviation appearing in our data at some large scale is induced by large fluctuations in the orientation of the global mean velocity during our numerical simulations ( not shown here ) . we take the opportunity of this discussion to come back to the superdiffusive behavior of particles observed in the transition region@xcite . there , sub - clusters emerge propagate ballistically and isotropically due to the absence of a well - established global order . particles trajectories consist in `` ballistic flights '' , occurring when a particle is caught in one of these coherently moving clusters , alternated with ordinary diffusion in disordered regions . the mean square displacement of particles exhibits the scaling @xmath186 @xcite . in view of our current understanding of the discontinuous nature of the transition , we now tend to believe that this isotropic superdiffusion is probably _ not _ asymptotic . we now turn our attention to the internal structure of the ordered regimes . as we noted in the previous section , these regimes do _ not _ consist of a single cluster of interacting particles moving coherently . even in the case where high density bands / sheets are present , these are in fact dynamical objects made of splitting and merging clusters . note that for the models considered here , clusters are unambiguously defined thanks to the strictly finite interaction range @xmath6 . as noticed first by aldana and huepe @xcite , clusters of size @xmath187 are distributed algebraically in the ordered region , i.e. @xmath188 . but a closer look reveals that the exponent @xmath189 characterizing the distribution of cluster sizes changes with the distance to the transition point . for noise intensities not too far from the threshold , when bands are observed , we find @xmath189 values larger than 2 , whereas @xmath190 in the bandless regimes present at low noise intensities ( fig . [ figcluster]a , b ) . thus , bands are truly complex , non - trivial structures emerging out of the transverse dynamics of clusters with a well defined mean size ( since @xmath191 ) . it is only in the bandless regime that one can speak , as aldana and huepe , of `` strong intermittency '' . we note in passing that the parameter values they considered correspond in fact to a case where bands are easily observed ( at larger sizes than those considered in @xcite ) . thus , clusters do have a well - defined mean size in their case . consequently , the probability distribution @xmath192 of the order parameter @xmath47 does _ not _ show the behavior reported in fig . 1 of @xcite as soon as the system size is large enough . whether in the band / sheet regime or not , @xmath192 shows essentially gaussian tails , is strongly peaked around its mean , and its variance decreases with system size ( fig . [ figcluster]c , d ) . although the picture of intermittent bursts between `` laminar''intervals proposed by aldana and huepe has thus to be abandoned , the anomalous density fluctuations reported in the previous section are probably tantamount to the strong intermittency of cluster dynamics in the bandless regime . again , these phenomena , reported also in the context of `` active '' nematics @xcite , deserve further investigation . ( black ) , 128 ( red ) , and 512 ( green ) from left to right ( @xmath12 , @xmath39 , @xmath40 , angular noise ) . ( c - d ) probability distribution functions of the order parameter @xmath47 ( arbitrary units ) for the same parameters and system sizes as in ( a - b ) . ( the most peaked distributions are for the largest size @xmath179 . ) left panels ( a , c ) : @xmath123 , bandless regime ; right panels ( b , d ) , regime with bands at @xmath160.,width=325 ] ( @xmath193 @xmath95 , @xmath40 ) . ( a ) two - point density correlation function @xmath194 ( coarse grained over a scale @xmath195 ) as a function of distance @xmath196 at different time steps : from left to right @xmath197 . noise amplitude is @xmath171 , data have been further averaged over @xmath198 different realizations . inset : log scales reveal the intermediate near - algebraic decay and the quasi - exponential cut - off . ( b ) lengthscale @xmath199 , estimated from the exponential cut - off positions , as a function of time . empty black circles : @xmath171 as in ( a ) ( i.e. regime in which bands are observed asymptotically ) . red full triangles : @xmath123 ( i.e. in the bandless regime ) . the dashed black line marks linear growth.,width=325 ] the ordered regimes presented above are the result of some transient evolution . in particular , the bands / sheets are the typical asymptotic structures appearing in _ finite _ domains with appropriate boundary conditions . in an infinite system , the phase ordering process is , on the other hand , infinite , and worth studying for its own sake . numerically , we have chosen to start from highly disordered initial conditions which have a homogeneous density and vanishing local order parameter . in practice , we quench a system `` thermalized '' at strong noise to a smaller , subcritical , @xmath37 value . typical snapshots show the emergence of structures whose typical scale seems to increase fast ( fig . [ coarse - snaps ] ) . during this domain growth , we monitor the two - point spatial correlation function of both the density and velocity fields . these fields are defined by a coarse - graining over a small lengthscale @xmath163 ( typically 4 ) . these correlation functions have an unusual shape ( fig . [ coarsening]a ) : after some rather fast initial decay , they display an algebraic behavior whose effective exponent decreases with time , and finally display a near - exponential cut - off . as a result , they can not be easily collapsed on a single curve using a simple , unique , rescaling lengthscale . nevertheless , using the late exponential cut - off , a correlation length @xmath199 can be extracted . such a lengthscale @xmath199 grows roughly linearly with time ( fig . [ coarsening]b ) . qualitatively similar results are obtained whether the noise strength is in the range where bands / sheets appear in finite boxes or not . we note that the above growth law is reminiscent of that of the so - called model h of the classification of halperin and hohenberg @xcite . since this model describes , in principle , the phase separation in a viscous binary fluid , the fast growth observed could thus be linked to the hydrodynamic modes expected in any continuous description of vicsek - like models @xcite . we now summarize our main results before discussing them at a somewhat more general level . we have provided ample evidence that the onset of collective motion in vicsek - style models is a discontinuous ( first - order ) phase transition , with all expected hallmarks , in agreement with @xcite . we have made the ( numerical ) effort of showing this in the limits of small and large velocity and/or density . we have shown that the ordered phase is divided in two regions : near the transition and down to rather low noise intensities , solitary structures spanning the directions transverse to the global collective velocity ( the bands or sheets ) appear , leading to an inhomogeneous density field . for weaker noise , on the other hand , no such structures appear , but strong , anomalous density fluctuations exist and particles undergoes superdiffusive motion transverse to the mean velocity direction . finally , we have reported a linear growth ( with time ) of ordered domains when a disordered configuration is quenched in the ordered phase . this fast growth can probably be linked to the expected emergence of long wavelength hydrodynamic modes in the ordered phase of active polar particles models . the high - density high - order traveling bands or sheets described here appear central to our main findings . they seem to be intimately linked to the discontinuous character of the transition which can , to some extent , be considered as the stability limit of these objects . in the range of noise values where they are observed , the anomalous density fluctuations present at lower noise intensities are suppressed . one may then wonder about the universality of these objects . simple variants of the vicsek - style models studied here ( e.g. restricting the interactions to binary ones involving only the nearest neighbor ) do exhibit bands and sheets @xcite . moreover , the continuous deterministic description derived by bertin _ et al . _ @xcite does possess localized , propagating solitary solutions rather similar to bands @xcite . although the stability of these solutions need to be further investigated , these results indicate that these objects are robust and that their existence is guaranteed beyond microscopic `` details '' . however , the emergence of regular , stable , bands and sheets is obviously conditioned to the shape and the boundary conditions of the domain in which the particles are allowed to move . in rectangular domains with at least one periodic direction , these objects can form , span across the whole domain , and move . but in , say , a circular domain with reflecting boundary conditions , they can not develop freely , being repeatedly `` frustrated '' . nevertheless , simulations performed in such a geometry indicate that the transition remains discontinuous , with the ordered phase consisting of one or several dense packets traveling along the circular boundary . note , though , that these packets intermittently emit elongated structures ( bands ) traveling towards the interior of the disk before colliding on the boundary . to sum up , bands appear as the `` natural '' objects in the transition region , but they may be prevented by the boundaries to develop into full - size straight objects . at any rate , time - series of the order parameter such as the one presented in fig . [ fig2 ] clearly show that the transition is discontinuous irrespective of the geometry and boundaries of the domain , and thus of whether bands / sheets can develop into stable regular structures or not : the sudden , abrupt , jumps from the disordered state to some ordered structure are tantamount to a nucleation phenomenon characteristic of a discontinuous transition . we would now like to offer the following speculative general picture : the key feature of the vicsek - like models studied here as well as of other models for active media made of self - propelled particles @xcite is the coupling between density and order . particles are forced to move , and , since they carry informations about the order , advection , density fluctuations and order are intimately linked . high - density means strong local order ( if noise is low enough ) because the many particles in a given neighborhood will adopt roughly the same orientation . the reverse is also true : in a highly - ordered region , particles will remain together for a long time and thus will sweep many other particles leading to a denser and denser group . , @xmath40 in a domain of size @xmath200 , ( same parameters as in fig . [ bandgnf]a ) . the local quantities were measured in boxes of linear size @xmath201 . here , the local order is represented by the angle between the orientation @xmath202 of the local order parameter and the global direction of motion @xmath203 . the black solid lines are running averages of the scatter plots . the red solid lines indicate the global density @xmath39 ( and thus marks the percolation threshold in a two - dimensional square lattice ) . ( a , b ) : @xmath123 and 0.2 : in the bandless regime , the ordered plateau starts below @xmath39 , i.e. ordered regions percolate . ( c ) approximately at the limit of existence of bands : the start of the plateau is near @xmath39 . ( d ) at higher noise amplitude in the presence of bands . , width=325 ] at a given noise level , one can thus relate , in the spirit of some local equilibrium hypothesis , local density to local order . in practice , such an `` equation of state '' approach can be justified by looking , e.g. , at a scatter plot of local order parameter vs local density . [ local - eq ] reveals that , in the ordered bandless regime , such a scatter plot is characterized by a plateau over a large range of local density values corresponding to order , followed , below some crossover density , by more disordered local patches . the regions in space where local density is below this crossover level do not percolate in the bandless regime , and order can be maintained very steadily in the whole domain ( this is corroborated by the fact that in spite of the large , anomalous density fluctuations , the order parameter field is , on the other hand , rather constant , see fig . [ bandgnf]c ) . the noise intensity at which bands emerge roughly corresponds to the value where the low - density disordered regions percolate . the reamining disconnected , dense patches then eventually self - organize into bands / sheets . the emergence of these elongated structures is rather natural : moving packets elongate spontaneously because they collect many particles ; superdiffusion in the directions perpendicular to the mean motion endow these nascent bands / sheets with some `` rigidity '' . at still stronger noise , the bands / sheets are destroyed and global order disappears . the above features are at the root of the approach by toner and tu @xcite . their predictions of strong density fluctuations , transverse superdiffusion , and peculiar sound propagation properties are correct as long as bands / sheets do not exist , i.e. for not too strong noise intensities . this is indeed in agreement with their assumption that the density field is statistically homogeneous in the frame moving at the global velocity ( albeit with strong fluctuations ) , which is only true in the bandless regime . the results presented here are almost entirely numerical . although they were obtained with care , they need to be ultimately backed up by more analytical results . a first step is the derivation of a continuous description in terms of a density and a velocity field ( or some combination of the two ) , which would allow to go beyond microscopic `` details '' . in that respect the deterministic equation derived by bertin _ from a boltzmann description in the dilute limit @xcite is encouraging . however , one may suspect that intrinsic fluctuations are crucial in the systems considered here if only because some of the effective noise terms will be multiplicative in the density . a mesoscopic , stochastic equation description is thus a priori preferable . this is especially true in view of the `` giant '' anomalous fluctuations present in the bandless ordered phase . these fluctuations clearly deserve further investigation , all the more so as they seem to be generic features of active particle models @xcite . ongoing work is devoted to both these general issues . most of this research has been funded by the european union via the fp6 starflag project . partial support from the french anr morphoscale project is acknowledged . we thank a. vulpiani and m. cencini for introducing us to the method outlined in ref . @xcite . , @xmath63 , @xmath69 ) . ( a ) : susceptibility _ _ vs _ time at reduced noise amplitude @xmath204 , @xmath205(dashed red line ) and @xmath206(plain black line ) . ( b ) : susceptibility _ vs. _ correlation at @xmath206 and @xmath207 , @xmath208 and @xmath209 from top to bottom . ( c ) : effective temperature _ vs _ noise amplitude . the vertical dashed line marks the transition point._______,width=325 ] in @xcite , vicsek _ et al . _ also studied the validity of the fluctuation - dissipation theorem and concluded , from numerical analysis , that it is violated . here we approach this question again , using the `` dynamical '' approach put forward in @xcite , rather than the equilibrium used in @xcite . the fluctuation - dissipation relation is expressed as : in both cases , at any rate , the key point is to investigate the effect of an external field on the ordering process . since here one can not rely on any hamiltonian structure , the external field remains somewhat arbitrary , as it can not be unambiguously defined as the conjugate variable of the order parameter . in @xcite , the field @xmath214 was directly confronted to the local average velocity , and the governing equation was replaced by : @xmath215 \;. \label{evfield}\ ] ] this manner of introducing @xmath214 leads to an effective intensity which depends on the local ordering : @xmath214 is comparatively stronger in disordered regions ( an important effect at the early stages of ordering ) than in ordered regions . this could in fact prevent the necessary linear regime to occur , even at very low field values ( see , for instance , fig . 6 of @xcite ) . a mean - field analysis has confirmed this view , showing a logarithmic variation of the response with the field intensity . to bypass this problem , we have preferred to use the following equation : @xmath216 \label{en2field}\ ] ] where the vectorial noise was chosen because , as shown above , it leads more easily to the asymptotic regime . the effective intensity of the field is now proportional to the local order . two ( scalar ) response functions can be defined in our problem : the longitudinal response @xmath217 along the field direction , and the transverse response @xmath218 . we consider the former . in practice , we quenched , at time @xmath219 , a strong - noise , highly - disordered system @xmath220 to a smaller noise value and started applying the constant , homogeneous field @xmath214 immediately . we then followed the response of the system by monitoring the growth of the order parameter . we measured the susceptibility @xmath221 which is nothing but the integrated response function : @xmath222 . in practice , we have : @xmath223 \;.\ ] ] in a well - behaved system , the susceptibility should be independent of the amplitude of the field , at least at small enough values ( `` linear '' regime ) . this is what we observed , as shown in fig . [ ffdt]a . correspondingly , the correlation function is defined as : @xmath224 the fluctuation - dissipation relation ( [ efdt1 ] ) can then be written in its integrated form : @xmath225 in fig . [ ffdt]b , we show that @xmath226 and @xmath212 are related linearly in time , confirming the validity of this relation and allowing an estimation of @xmath213 . this well - defined although not uniquely defined effective temperature varies as expected in parameter space . in particular , it increases systematically with the noise strength @xmath37 ( fig . [ ffdt]c ) , although this variation is not linear . note also , that , intriguingly , there is a small jump of @xmath213 at the noise value corresponding to the transition in this case ( @xmath227 ) .

the statistics of lagrangian pair dispersion in a homogeneous isotropic flow is investigated by means of direct numerical simulations . the focus is on deviations from richardson eddy - diffusivity model and in particular on the strong fluctuations experienced by tracers . evidence is obtained that the distribution of distances attains an almost self - similar regime characterized by a very weak intermittency . the timescale of convergence to this behavior is found to be given by the kinetic energy dissipation time measured at the scale of the initial separation . conversely the velocity differences between tracers are displaying a strongly anomalous behavior whose scaling properties are very close to that of lagrangian structure functions . these violent fluctuations are interpreted geometrically and are shown to be responsible for a long - term memory of the initial separation . despite this strong intermittency , it is found that the mixed moment defined by the ratio between the cube of the longitudinal velocity difference and the distance attains a statistically stationary regime on very short timescales . these results are brought together to address the question of violent events in the distribution of distances . it is found that distances much larger than the average are reached by pairs that have always separated faster since the initial time . they contribute a stretched exponential behavior in the tail of the inter - tracer distance probability distribution . the tail approaches a pure exponential at large times , contradicting richardson diffusive approach . at the same time , the distance distribution displays a time - dependent power - law behavior at very small values , which is interpreted in terms of fractal geometry . it is argued and demonstrated numerically that the exponent converges to one at large time , again in conflict with richardson s distribution . turbulent transport ; relative dispersion

1209.2296

it is known since taylor s seminal work @xcite that tracers transported by a turbulent flow approach a diffuse behavior on time scales much longer than the lagrangian correlation time of the flow . these ideas are now commonly used in applications , as for instance in air quality control , to model effective mixing properties in terms of an eddy diffusivity . such models give a good handle on long - term averages and are successfully used to determine , for instance , possible health hazards linked to a long exposure downstream a pollutant source . however they are unable to capture strong local fluctuations stemming from the complex structure of the turbulent flow . such events can not be directly predicted from the averaged concentration field as they relate to higher - order moments . accessing these fluctuations is crucial in order to quantify for instance the likeliness of finding a local concentration exceeding a high threshold . second - order statistics , such as the variance of a transported concentration field and more generally the spatial correlation of a passive scalar , are statistically related to the relative motion of tracers ( see , e.g. , @xcite ) . the problem consists then in investigating the time evolution of the separation @xmath0 between two lagrangian trajectories . in turbulence , the distance @xmath1 follows richardson obukhov superdiffusive law @xmath2 , where @xmath3 is the mean rate of kinetic energy dissipation . the long - term behavior is thus becoming independent of the initial separation @xmath4 , whence the designation of _ explosive _ pair separation . in his original arguments to derive this law , richardson @xcite assumed that pair separation is a diffusion process with a scale dependent diffusivity @xmath5 . then the transition probability density @xmath6 that two tracers are at a distance @xmath7 knowing that they were initially separated by @xmath4 satisfies the fokker planck equation @xmath8 . \label{eq : diffusion}\ ] ] the atmospheric measurements of richardson , later refined by obukhov @xcite in view of kolmogorov 1941 phenomenology , led to assume that for separations @xmath9 within the inertial range of turbulence @xmath10 . this implies that , at large times , @xmath11 and the transition probability reads @xmath12 , \label{eq : pdf_richardson}\ ] ] where @xmath13 is a positive constant . the considerations leading to the fokker planck equation ( [ eq : diffusion ] ) rely on the use of a central - limit theorem for @xmath14 and thus on the hypothesis that the velocity difference between the two tracers is correlated over timescales much smaller than those of interest . as noticed for instance in @xcite , such an hypothesis can hardly be invoked . it is indeed known in turbulence that eddies of size @xmath9 are correlated over a time of the order of their turnover time @xmath15 . hence for separations that grow like @xmath16 , one has @xmath17 , so that the dominant flow structures in the separation dynamics are in principle correlated over timescales of the order of the observation time . despite such shortcomings , the diffusive approach proposed by richardson and in particular the explicit form ( [ eq : pdf_richardson ] ) for the transition probability density have proven being relevant in some asymptotics @xcite . also , much work on relative dispersion has used it as a basis . for instance , improvements of ( [ eq : pdf_richardson ] ) were proposed using modified versions of the eddy diffusivity @xmath18 , adding a time dependence @xcite or , more recently , including finite reynolds number effects @xcite . all of these improvements strongly alter the functional form of the large-@xmath9 tail of the transition probability density . nevertheless , the physical mechanisms leading to the de - correlation of velocity differences and to models based on eddy diffusivity are still unclear and many questions remain open . for instance , the speed of convergence to the richardson obukhov law and the form of subleading terms are still not known . the first work dealing with the way pair separations converge to a superdiffusive behavior is due to batchelor @xcite . he argued that the explosive @xmath19 law is preceded by a ballistic phase during which the tracers keep their initial velocity and separate linearly in time , i.e. @xmath20 , up to a time @xmath21 , which is equal to the eddy turnover time associated to the initial separation . this first regime where velocity remains strongly correlated can clearly not be described by eddy - diffusivity approaches . various stochastic models have been designed to catch the two regimes . most of them are based on the observation that the acceleration difference between the two tracers is shortly correlated and do not assume that velocity differences get uncorrelated . the pair separation and the velocity difference can then be approximated as coupled markovian diffusive processes ( see @xcite for reviews ) . the usual path for designing such models requires imposing some constraints on the drift and the diffusion terms . thomson @xcite argued that they should satisfy the well - mixed condition : when averaging over uniformly separated pairs inside the whole inertial range , the statistics of velocity differences between the two tracers has to recover eulerian two - point statistics . definitively devising an admissible model requires an input from eulerian statistics @xcite . it is known in turbulence that the distribution of velocity differences is neither self - similar nor gaussian ( see , e.g. , @xcite ) . such models become thus so complicated that they can hardly be used to improve the understanding of the underlying phenomenology and , at the same time , they are not easily amenable for an analytical treatment . the goal of the present work is to give some new phenomenological understanding of the problem of turbulent relative dispersion . for this , we make use of large - scale direct numerical simulations and essentially focus our study on the role of violent events leading to strong fluctuations . to give grounds to possible improvements of models for relative dispersion , we introduce some new observables , measure their behavior , and interpret the results mostly in terms of geometrical arguments . the paper is organized as follows . we first report in sec . [ sec : averaged_separations ] numerical results on the time evolution of averaged separations and of higher - order statistics ; this is an extension of some of our recent work @xcite . in sec . [ sec : veldiff ] we focus on the behavior of velocity differences and discuss the question of intermittency and of a possible multiscaling for the time evolution of their moments . we also introduce a mixed moment of separation and velocity difference that attains quickly a scaling regime ( independent on time and initial separations ) . in sec . [ sec : extreme ] , we apply previous results and make use of geometrical considerations to investigate the very large and small fluctuations of the inter - trajectory distance . finally , sec . [ sec : conclusion ] encompasses concluding remarks and discuss open questions . we report in this section numerical results on the time evolution of the averaged squared distance between tracers . the aim is to provide a first insight on the numerical simulations that we have used and on the typical values of the timescales and lengthscales that are used throughout this paper . one of the major difficulties encountered when attacking numerically the problem of turbulent explosive separation is that it requires a huge timescale separation . indeed , to observe richardson obukhov @xmath19 superdiffusive regime , one needs to follow particle pairs on a time @xmath22 much longer than the eddy turnover time @xmath23 associated to their initial separation @xmath4 and much smaller than the time @xmath24 associated to the integral scale @xmath25 . also , it is in principle required that the initial distance @xmath26 belongs to the inertial range , so that its associated turnover time @xmath27 has to be much larger than the kolmogorov dissipative timescale @xmath28 . we should thus have @xmath29 . to give a concrete idea , if we require that these timescales are separated by at least a decade , this implies that one should have a taylor - based reynolds number @xmath30 . such a high level of turbulence is still far from the current state of direct numerical simulations and also from present - day experimental setups where accurate particle tracking techniques can be used ( see @xcite for recent reviews ) . for this reason , there is an important need for understanding the full process of convergence to the @xmath19 law in order to predict and detect for instance possible subleading corrections . this is the spirit in which we have tried to present this work . to investigate such issues , we have performed direct numerical simulations of the incompressible navier stokes equation in the three - dimensional @xmath31-periodic domain using a standard pseudo - fourier - spectral solver with a third - order runge kutta time marching . such a method is well adapted to incompressible homogeneous and isotropic turbulence at high reynolds numbers with an extended inertial range of scales . it has the advantages of combining a high degree of accuracy with very good performances on massive parallel supercomputers such as bluegene systems and large intel / amd clusters . we have used two sets of simulations whose parameters are summarized in table [ table ] ( more details on these simulations can be found in @xcite ) . to maintain a statistical steady state , the flow is forced by keeping constant the energy content of the two first shells of wavenumbers in fourier space . [ table ] ( a ) second - order eulerian structure function @xmath32 for the two values of the reynolds number investigated in this paper . the solid line represents kolmogorov 1941 scaling @xmath33 with @xmath34 . the dashed line corresponds to she - lvque anomalous scaling with @xmath35 . the vertical and horizontal colored dashed lines indicate the integral scale and the large - scale velocity , respectively . ( b ) logarithmic derivative @xmath36 . the two horizontal lines encode the two scalings shown in fig . ( a ) . ] the eulerian second - order structure function measured from these two simulations are represented in fig . [ fig : s2_2reynolds ] . as seen there , the largest one develops a significant scaling range where deviations from kolmogorov 1941 scaling start to be visible . for each value of the reynolds number , the flow is seeded with @xmath37 tracer particles whose motion is integrated using tri - linear interpolation . after a time sufficiently long to have converged to a statistical steady state , we start the analysis of the dispersion of tracer pairs . for this , we label at a fixed initial time ( that we fix here to be @xmath38 ) all couples whose distance @xmath39 is equal to @xmath40 for @xmath41 and equal to @xmath42 for @xmath43 with @xmath44 and @xmath45 . this bining was chosen such that each family indexed by @xmath26 contains a few hundreds of thousands of pairs . we then track forward in time all indexed pairs and perform statistics conditioned on their initial separation @xmath26 . for the sake of simplicity we denote by @xmath46 the lagrangian ensemble average conditioned on @xmath26 . time - evolution of the mean - squared distance for @xmath47 ( a ) and @xmath48 ( b ) for various initial separations @xmath26 as labeled . the horizontal and vertical solid lines represent the integral scale @xmath25 and its associated turnover time @xmath24 , respectively . the dashed line corresponds to the explosive richardson - obukhov law ( [ eq : explosive ] ) with @xmath49 . ] figure [ fig : r2_fn_time_diffsep ] shows for the two simulations the time evolution of the mean squared distance @xmath50 for various values of the initial separation @xmath26 . times and space are there represented in dissipative - scale units . after a transient ( which roughly corresponds to batchelor s ballistic regime ) , the mean - squared distance approaches the explosive richardson obukhov regime @xmath51 we observe for both values of the reynolds number a richardson obukhov constant @xmath52 . the low accuracy with which this constant is determined comes from the fact that , even at the higher resolution , the @xmath19 scaling is observed in a rather limited time range . this is even clearer from fig . [ fig : dr2_fn_time_compensate ] ( a ) , which shows the compensated mean squared increase of the distance @xmath53 , where @xmath54 designates the second - order eulerian structure function with absolute values . the choice of such a timescale was motivated in @xcite as that of deviations from batchelor s initial ballistic regime . surprisingly , the data shown in fig . [ fig : dr2_fn_time_compensate ] ( a ) corresponding to various initial separations @xmath26 far enough in the inertial range seem to collapse . ( a ) compensated mean - squared displacement @xmath55 as a function of @xmath56 with @xmath57 for various initial separations and @xmath47 ( @xmath58 ) and @xmath59 ( @xmath60 ) . the two curves show behaviors of the form @xmath61 , with @xmath62 , given by batchelor s ballistic regime ( black dotted line ) , and @xmath63 ( grey dashed line ) . ( b ) measured value of the constant @xmath64 in front of the subleading term as a function of the initial separation . it stabilizes to @xmath65 for @xmath66 ; this value is represented as a dashed line . , title="fig:",scaledwidth=50.0% ] ( a ) compensated mean - squared displacement @xmath55 as a function of @xmath56 with @xmath57 for various initial separations and @xmath47 ( @xmath58 ) and @xmath59 ( @xmath60 ) . the two curves show behaviors of the form @xmath61 , with @xmath62 , given by batchelor s ballistic regime ( black dotted line ) , and @xmath63 ( grey dashed line ) . ( b ) measured value of the constant @xmath64 in front of the subleading term as a function of the initial separation . it stabilizes to @xmath65 for @xmath66 ; this value is represented as a dashed line . , title="fig:",scaledwidth=50.0% ] this suggests that the timescale @xmath67 contains most of the dependence of pair dispersion upon the initial separation @xmath26 . also data indicate that the subdominant terms in ( [ eq : explosive ] ) are @xmath68 , leading to postulate @xmath69 , with a constant @xmath70 independent of @xmath26 when @xmath66 . the product of the constants @xmath71 and @xmath70 has been estimated numerically and results are shown in fig . [ fig : dr2_fn_time_compensate ] ( b ) . one can clearly see that @xmath65 when @xmath72 . one can also see from the figure that @xmath73 for @xmath74 . thus , for dissipative - range initial separations , the asymptotic @xmath19 behavior is attained from below . this can lead for such values of @xmath26 to an intermediate time range where the mean squared distance grows even faster than the explosive @xmath19 law , as for instance observed in @xcite . another remark that can be drawn from the data is that , independently of the reynolds number , the constant @xmath70 is equal to zero for @xmath75 . the first subleading terms are then @xmath76 , so that the convergence to the @xmath19 law is much faster for such an initial separation than for others . this observation could be useful for experimentalists to optimize their setup . however , such small values of @xmath26 are clearly not representative of the inertial - range behavior . ( a ) fourth - order moment @xmath77 and ( b ) sixth - order moment @xmath78 as function of @xmath56 for @xmath47 . both curves are normalized such that their expected long - time behavior is @xmath79 and @xmath80 , respectively . the black dashed lines represent such behaviors . ] we now turn to investigating the large - time behavior of higher - order moments of the separation . figure [ fig : r4_fn_time_diffsep ] shows the time evolution of @xmath81 ( a ) and of @xmath82 ( b ) . at times smaller than @xmath67 the separation grows ballistically , so that @xmath83 where @xmath84 denotes the velocity difference between the two tracers . the fact that we have chosen to rescale time by @xmath67 ( which depends on second - order statistics of the initial velocity difference ) implies that the moments do not collapse in this regime because of eulerian multiscaling . however the collapse occurs for @xmath85 where these two moments grow like @xmath86 and @xmath87 , respectively , with possible minute deviations . the measured power - laws give evidence that , at sufficiently long times , inter - tracer distances follow a scale - invariant law . also the observed collapses indicate that @xmath67 could be again the time of convergence to such a behavior . fourth ( a ) and sixth ( b ) order moments of @xmath88 as a function of its second - order moment for @xmath47 . the two gray dashed lines show a scale - invariant behavior , i.e. @xmath89 and @xmath90 , respectively . the two insets show the associated local slopes , that is the logarithmic derivatives @xmath91 , together with the normal scalings represented as dashed lines . ] the presence of a scale - invariant regime is also clear when making use of ideas borrowed from extended self - similarity and representing these two moments as a function of @xmath92 ( see fig . [ fig : r_no_intermittency ] ) . this time , for a fixed @xmath26 , the smallest separations correspond to the ballistic regime . there , we trivially have @xmath93 , which has a weak dependence on @xmath26 , because of an intermittent distribution of eulerian velocity increments , but does not depend on time . this normal scaling can be observed for @xmath94 in the insets of fig . [ fig : r_no_intermittency ] , which represent the logarithmic derivatives of the high - order moments with respect to the second order . at times of the order of @xmath67 , noticeable deviations to normal scaling can be observed . finally , at much larger scales , data corresponding to different values of the initial separation @xmath26 collapse but the curves start to bend down . one observes in the insets that the associated local slopes approach values clearly smaller than those corresponding to normal scaling . this gives evidence of a rather weak intermittency in the distribution of tracer separations . note that the presented measurements were performed for @xmath47 but the same behavior has been observed for @xmath59 . to our knowledge , the most convincing observation of an intermittent behavior in pair dispersion has been based on an exit - time analysis @xcite . however , the relation of such fixed - scale statistics to the usual fixed - time measurements we report here requires to consider pair separation velocities . as we will see in next section , the velocity difference between two tracers displays statistics that are much more intermittent than those for pair separation . this implies that there is no contradiction between an almost normal scaling for distances as a function of time and an anomalous behavior of exit times as a function of distance . probability density function of the distance @xmath9 at time @xmath95 ( a ) and @xmath96 ( b ) and for various values of the initial separation . we have here normalized it by @xmath97 and represented on a @xmath98 axis as a function of @xmath99 . with such a choice , richardson s diffusive density distribution ( [ eq : pdf_richardson ] ) appears as a straight line ( represented here as a black dashed line).,title="fig:",scaledwidth=50.0% ] probability density function of the distance @xmath9 at time @xmath95 ( a ) and @xmath96 ( b ) and for various values of the initial separation . we have here normalized it by @xmath97 and represented on a @xmath98 axis as a function of @xmath99 . with such a choice , richardson s diffusive density distribution ( [ eq : pdf_richardson ] ) appears as a straight line ( represented here as a black dashed line).,title="fig:",scaledwidth=50.0% ] to investigate further this weak intermittency in the separation distribution , we have represented in fig . [ fig : histo_r_4096 ] the probability density function ( pdf ) of the distance @xmath100 for various initial separation and at times where we expect to have almost converged to the explosive regime , namely at @xmath101 ( a ) and @xmath102 ( b ) . such measurements are compared to richardson s diffusive law ( [ eq : pdf_richardson ] ) . data suggest that a large part of the pdf core ( for @xmath103 . in this section , we are interested in the behavior of the velocity difference @xmath104 between two tracers as a function of time . initially , the statistics of @xmath105 are exactly given by the eulerian statistics of velocity increments at a separation @xmath26 . at large times , an explosive @xmath19 behavior for distances implies that @xmath106 . a naive picture would consist in interpolating between these two behaviors by assuming that @xmath107 where @xmath108 is a positive constant that can not be straightforwardly expressed as a function of @xmath71 . as argued in @xcite one of the two terms , which correspond to the ballistic or to the richardson obukhov regime , is dominant when @xmath22 is much smaller or much longer than @xmath109 . time behavior of the mean - squared velocity difference for @xmath47 . ( a ) @xmath110 , normalized by its initial value , as a function of @xmath56 ; the gray dashed line is a behavior of the form @xmath111 with @xmath112 . inset : same but zoomed at short times ; the black dotted line is the initial decrease associated to kinetic energy dissipation : @xmath113 . ( b ) time behavior of the velocity change @xmath114 ; the dashed curve is @xmath115 ; the dotted line has slope @xmath116 and corresponds to the initial abrupt variation of the velocity difference . inset : same but rescaling time by @xmath117.,title="fig:",scaledwidth=50.0% ] time behavior of the mean - squared velocity difference for @xmath47 . ( a ) @xmath110 , normalized by its initial value , as a function of @xmath56 ; the gray dashed line is a behavior of the form @xmath111 with @xmath112 . inset : same but zoomed at short times ; the black dotted line is the initial decrease associated to kinetic energy dissipation : @xmath113 . ( b ) time behavior of the velocity change @xmath114 ; the dashed curve is @xmath115 ; the dotted line has slope @xmath116 and corresponds to the initial abrupt variation of the velocity difference . inset : same but rescaling time by @xmath117.,title="fig:",scaledwidth=50.0% ] as seen from fig . [ fig : dv2_fn_time ] ( a ) such a form with @xmath118 ( represented as a dashed line ) seems to be a good first - order approximation . however , this picture is not completely satisfactory . when zooming closer to the initial times , one observes that the averaged pair kinetic energy @xmath119 first starts by decreasing for a time of the order of a few tens of @xmath67 . we indeed know that initially @xmath120 where @xmath121 denotes the initial difference of acceleration between the two tracers . this approximation leads to @xmath122see inset of fig . [ fig : dv2_fn_time ] ( a ) . the fact that this prediction ceases to describe the data after such a short timescale ( that is significantly smaller than @xmath67 instead of being of its order ) indicates that the subleading terms in ( [ eq : ballistic ] ) then become important . as a consequence , depending on the quantity we are interested in , batchelor s ballistic regime might end very quickly . after this subtle initial decrease , the averaged squared velocity difference amplitude tends asymptotically to a behavior @xmath123 . however this process occurs in a noticeably different manner to that of convergence of squared separation to richardson @xmath19 law : irrespective of the initial separation @xmath26 , the quantity latexmath:[$\langle behavior . in other terms , we observe that for a fixed value of @xmath56 , the pair kinetic energy @xmath125 is an increasing function of @xmath126 and seems to converge to ( [ eq : behav_vel ] ) only in the limit @xmath127 . at large times , the mean - squared velocity difference grows linearly . this behavior , together with the fact that acceleration differences are correlated over timescales of the order of the kolmogorov dissipative time @xmath117 , suggest that velocity differences have a diffusive behavior for @xmath128 ( see @xcite ) . if this was exact , it would lead to the behavior ( [ eq : behav_vel ] ) as the temporal increments of @xmath129 would be independent of its initial value . also , a purely diffusive behavior of @xmath129 would result in the fact that its mean - squared temporal increment @xmath130 is @xmath76 for all times @xmath128 . as seen in fig . [ fig : dv2_fn_time ] ( b ) , the diffusive regime is actually only asymptotic . also , data show that @xmath131 initially changes on times of the order of @xmath117 by a factor of the order of its typical initial value @xmath132^{1/2}$ ] . this abrupt evolution can be interpreted phenomenologically . with some finite probability , one of the two tracers is within a vortex filament at time @xmath67 . the typical energy content of this filament will contribute to the value of @xmath133 . however , after a time @xmath22 of the order of @xmath117 , the trajectory of this tracer will have turned around this filament , so that its velocity will have completely changed orientation ( without changing much its amplitude ) . this will result in @xmath134 , explaining the observed behavior . as a result of this sudden kinematic variations of velocity differences , latexmath:[$\langle manner to @xmath125 for times @xmath128 ( compare the dashed lines in fig . [ fig : dv2_fn_time ] ( a ) and ( b ) ) . let us also notice that the convergence to this behavior is again from below , irrespective of the initial separation @xmath26 . there is hence an abrupt change ( occurring on timescales of the order of @xmath117 ) that prevents from determining an effective initial velocity difference and thus from observing a clear diffusive behavior of @xmath131 . however , data suggest that the timescale of convergence to this behavior is , as for separations , of the order of @xmath67 . to understand further this question , we next turn to investigating the behavior of the longitudinal velocity difference between the tracers . we are here interested in the evolution of the longitudinal component @xmath136 of the velocity difference as a function of time . this quantity is important to characterize pair separation as @xmath137 . initially , the averaged longitudinal velocity vanishes , i.e. @xmath138 ; this is due to the statistical stationarity of the fluid flow . for times @xmath94 in the batchelor s ballistic regime , the pairs are keeping their initial velocity difference and one can easily check that @xmath139 where @xmath140 denotes the components of @xmath129 that are transverse to @xmath141 . it is thus clear that the average velocity at which tracer trajectories separate immediately becomes positive . figure [ fig : vpar_fn_time ] ( a ) , which represents the time evolution of @xmath142 , shows without doubt this initial linear growth . ( a ) time evolution of the averaged longitudinal velocity difference @xmath142 for various initial separations , as labeled , and for @xmath143 . ( b ) averaged time @xmath144 when trajectories with a given initial longitudinal velocity difference @xmath145 reach a minimal distance . ] this increase of the longitudinal velocity difference has an interesting geometrical interpretation . if all pairs were to maintain indefinitely their initial velocity differences @xmath105 , it is clear that they would eventually all separate at large times . indeed , they will reach a minimal distance at a finite time @xmath146 / |{\bm v}(0)|^2 = -r_0 v^\parallel(0 ) / approaching . after the time @xmath147 , the distance between particles increases and @xmath148 becomes positive . this leads to an increase of @xmath149 that comes from kinematic considerations and is obviously not due to any dynamics imposed by the turbulent flow . to estimate the typical value of the minimal distance at time @xmath147 , we have performed statistics on pair separation conditioned not only on the initial distance @xmath26 , but also on the initial longitudinal velocity difference by binning pairs with @xmath150 . from these statistics we have defined the averaged time latexmath:[$t_\ast = -r_0 v_0 / \langle initial longitudinal velocity difference @xmath145 are at a minimal distance . the data are shown in fig . [ fig : vpar_fn_time ] ( b ) for various values of @xmath26 and as a function of @xmath145 . we observe that for @xmath66 this time approximatively takes the form @xmath152 , where the function @xmath153 attains its maximum ( roughly equal to 0.05 ) at @xmath154 . this confirms the observation made in fig . [ fig : vpar_fn_time ] ( a ) that the initial growth of @xmath149 occurs on a time length of the order of a few hundredths of @xmath67 . in addition to the change in the mean longitudinal velocity difference discussed above , numerical measurements show that the full distribution of @xmath148 looses its symmetry and develops fatter tails when time increases . figures [ fig : skew_flat ] ( a ) and ( b ) represent the skewness @xmath155 and flatness @xmath156 of @xmath148 as a function of time and for the same initial separations as in fig . [ fig : vpar_fn_time ] ( a ) . these observables are frequently used in turbulence to quantify the shape of the velocity increment distribution . for two - particle lagrangian statistics , they are defined as @xmath157 ^ 3 \right\rangle}{\left\langle [ v^\parallel(t)-\langle v^\parallel(t ) \rangle]^2 \right\rangle^{3/2}}\quad\mbox{and}\quad \mathcal{f}(t ) = \frac{\left\langle [ v^\parallel(t)-\langle v^\parallel(t ) \rangle]^4 \right\rangle}{\left\langle [ v^\parallel(t)-\langle v^\parallel(t ) \rangle]^2 \right\rangle^{2}}. \label{def : skew_flat}\ ] ] time evolution of ( a ) the skewness @xmath158 and ( b ) the flatness @xmath159 defined in ( [ def : skew_flat ] ) for the same initial separations @xmath26 as in fig . [ fig : vpar_fn_time ] ( a ) . the dashed lines are the values corresponding to a gaussian distribution : @xmath160 and @xmath161 . the dotted line in ( a ) is the initial value of the skewness obtained when assuming kolmogorov 1941 scaling , namely @xmath162 . ] as already observed for instance in @xcite , these quantities strongly vary as a function of time and maintain a marked dependence upon the initial separation @xmath26 for rather long times . figure [ fig : skew_flat ] ( a ) shows that the skewness of @xmath163 starts from negative values ( to be in agreement with the eulerian 4/5 law ) and becomes positive at times larger than @xmath164 . this initial change of sign can also be interpreted geometrically in terms of the time @xmath147 when initially ballistically approaching pairs begin to move away . however , after this , the curves separate and each of them attains a maximum at times of the order of @xmath67 or slightly smaller . this maximal value of the skewness strongly depends on the initial separation : less is @xmath26 , higher it is . after this maximum , the skewness decreases without attaining an asymptotic regime that would be independent of @xmath26 . this could be either due to the fact that there is a persistent memory of @xmath26 in such quantities or to a contamination by finite reynolds number ( and finite size ) effects . the same kind of behavior is observed for the flatness @xmath156 of the distribution of @xmath148 as seen in fig . [ fig : skew_flat ] ( b ) . however , the increase of its maximal value when decreasing @xmath26 is even more pronounced . another noticeable difference is that the initial value of @xmath156 itself depends on @xmath26 and relates to the scale dependence of the eulerian flatness . time evolution for @xmath165 of the averaged distance ( a ) and of the squared velocity difference ( b ) conditioned on @xmath166 with @xmath167 ^ 2\rangle^{1/2 } = 0.1,0.4,0.8,1.2, ... ,2.8 $ ] . the red curves correspond to positive values of @xmath145 and the blue ones to negative values . in ( b ) the dashed - dotted line is @xmath76 ] the strong dependence on @xmath26 of the skewness and of the flatness can be interpreted phenomenologically in terms of the intermittent nature of velocity increments . a large positive value of @xmath155 corresponds to a large probability of having pairs separating faster than the average . such events will also be responsible for a large value of the flatness @xmath156 . it is clear that those particles that separate the faster are typically those which are the most separated . also , in a turbulent flow , the typical value of the velocity increases with scale , so that the particles which get quickly separated are likely to continue separating faster than the average . this is evidenced in fig . [ fig : r2_v2_condvelpar ] , which represents the mean squared distance and the averaged squared amplitude of the velocity difference conditioned on the initial value of the longitudinal velocity difference for @xmath169 . one observes that there is up to the latest time ( of the order of @xmath170 ) , a noticeable memory on the initial value of @xmath148 . besides this consideration , one also remarks in fig . [ fig : r2_v2_condvelpar ] ( b ) that the pairs having an initially large negative longitudinal velocity difference ( the blue curves ) dissipate much more kinetic energy than the others . however , this does not prevent them from separating at large times faster than the pairs having initially a smaller velocity difference . we now turn back to the explanation of the long - term dependence on @xmath26 of the skewness and flatness of @xmath148 . recall that , in a turbulent flow , violent velocity differences are more probable at small scales than at larger scales . this implies that pairs with a small initial separation are more likely to experience a large ( positive or negative ) initial velocity difference . this will make them separate faster than the average and thus experience even larger values of the velocity . the rapid and strong increase of fluctuations in their velocity differences is thus due to a kind of snowball effect . we will come back later to quantifying with more accuracy this phenomenon . all the considerations on the dependence of @xmath148 upon the initial separation are also visible in the probability distribution of the longitudinal velocity . figure [ fig : histo_vpar ] ( a ) shows the centered pdf of @xmath148 normalized to unit variance for various times and @xmath171 . the data clearly show that at times later than @xmath172 , there is a change in the sign of the skewness . also , one sees that the time dependence of the skewness and of the flatness comes from the right tails associated to large velocity differences , supporting the arguments discussed above . the left tails , which corresponds to approaching pairs , seem on the contrary to collapse . the right tail has a very rich behavior . it starts with broadening at times @xmath173 , in agreement with the increase of flatness . for @xmath174 , it then decreases and possibly goes back asymptotically to its initial form . if this was true , it would imply that the distribution of velocity differences keep in memory the initial separation at any later time . centered and normalized probability density functions of the longitudinal velocity difference : ( a ) for @xmath175 and various times and ( b ) for various initial separations and @xmath102 . in both cases , the black dashed line shows a gaussian distribution.,title="fig:",scaledwidth=50.0% ] centered and normalized probability density functions of the longitudinal velocity difference : ( a ) for @xmath175 and various times and ( b ) for various initial separations and @xmath102 . in both cases , the black dashed line shows a gaussian distribution.,title="fig:",scaledwidth=50.0% ] this is also evidenced from fig . [ fig : histo_vpar ] ( b ) , which shows the same pdfs for various separations and a time @xmath102 fixed . again we observe a rather good collapse of the tails associated to negative longitudinal velocity differences , but the right tails display very strong dependence on the initial separation . clearly , the behavior of this tail is a stretched exponential for @xmath176 and is faster than exponential for larger initial separations . the actual level of statistics do not allow us to relate systematically this behavior to that of the initial velocity difference distribution . fourth - order ( a ) and sixth - order ( b ) moments of the longitudinal velocity difference as a function of its second - order moment for various times and initial separations . the two dashed lines correspond to a scaling compatible with that of lagrangian structure functions proposed in @xcite , namely @xmath177 and @xmath178 . the insets show the logarithmic derivative @xmath179^p\rangle / \mathrm{d}\log \langle[v^\parallel(t)]^2\rangle$ ] for ( a ) @xmath180 and ( b ) @xmath181 as a function of @xmath56 ; there the bold dashed lines show the lagrangian multifractal scaling and the thin lines what is expected from a self - similar behavior . ] finally another way to address the question of intermittency of the velocity difference consists in finding how moments of its longitudinal component depend on time . for that we follow , as in the case of the moments of distances , an approach similar to that of extended self - similarity . figures [ fig : vpar_intermittency ] ( a ) and ( b ) show the fourth and sixth - order moments of @xmath163 as a function of its second - order moment . as evidenced in the insets , they display an anomalous behavior that differs from simple scaling . however , the collapse for various @xmath26 is much less evident than for the moments of the distance , except perhaps at sufficiently large times . one can there guess a power - law dependence of @xmath182^{p}\rangle$ ] as a function of @xmath182^{2}\rangle$ ] . surprisingly the power is compatible with the scaling exponent of the lagrangian structure functions that were obtained in @xcite by relating velocity increment along trajectories to she lvque multifractal spectrum for the eulerian field . the two dashed lines in fig . [ fig : vpar_intermittency ] ( a ) and ( b ) correspond to the predicted values @xmath177 and @xmath178 . confirming further this match would require much better statistics . as we have seen previously , the velocity difference between tracers displays very intermittent features and , as a consequence , does not converge to a behavior with temporal self - similarity , or does it only very slowly . the situation is very different when interested in mixed statistics between distances and longitudinal velocity differences . as seen in @xcite , the moment @xmath182 ^ 3/|{\bm r}(t)|\rangle$ ] , which is initially negative and equal to @xmath183 , tends very quickly to a positive constant see fig . [ fig : locdiss_fntime_r0_24eta ] ( a ) . ( a ) time evolution of the mixed moment @xmath184 ^ 3/|{\bm r}(t)|\rangle$ ] for different initial separations and the two values of the reynolds number : @xmath47 ( @xmath58 ) and @xmath59 ( @xmath60 ) . the black dashed line shows the asymptotic value @xmath185 . ( b ) same for @xmath186 and @xmath47 in black and for the same moment but conditioned on positive ( red curve ) and negative ( blue curve ) values of the initial longitudinal velocity difference @xmath187.,title="fig:",scaledwidth=50.0% ] ( a ) time evolution of the mixed moment @xmath184 ^ 3/|{\bm r}(t)|\rangle$ ] for different initial separations and the two values of the reynolds number : @xmath47 ( @xmath58 ) and @xmath59 ( @xmath60 ) . the black dashed line shows the asymptotic value @xmath185 . ( b ) same for @xmath186 and @xmath47 in black and for the same moment but conditioned on positive ( red curve ) and negative ( blue curve ) values of the initial longitudinal velocity difference @xmath187.,title="fig:",scaledwidth=50.0% ] the decrease at very large times comes from the contamination of the statistics by pairs that have reached a distance of the order of the integral scale . the asymptotic value @xmath188 seems to depend only weakly on the reynolds number . the collapse of the curves associated to different reynolds numbers and , for @xmath66 , to various initial separations indicate that the time of convergence is @xmath189 . figure [ fig : locdiss_fntime_r0_24eta ] ( b ) shows the same moment but conditioned on the sign of the initial longitudinal velocity difference . one observes that for initially separating pairs ( red curve ) , the convergence to the asymptotic value is on a time of the order of @xmath117 . conversely for tracers that initially approach each other ( blue curve ) , the convergence is less fast . we have seen in sec . [ subsec : geom_long ] that such pairs first attain their minimal distance at @xmath190 . then , at that time , @xmath184 ^ 3/|{\bm r}(t)| \,|\ , v^\parallel(0)<0\rangle$ ] changes sign and the convergence to @xmath191 occurs only later . such initially approaching pairs are leading the average so that the overall convergence is on timescales of the order of @xmath147 . the mixed moment @xmath192 ^ 3/|{\bm r}| \rangle$ ] , which is a kind of `` local dissipation '' along pairs of trajectories , is thus conserved by the lagrangian flow . ( a ) pdf of the `` local dissipation '' @xmath193 ^ 3/|{\bm r}(t)|$ ] for @xmath186 and various times ( as labeled ) . ( b ) same for various initial separations and a time @xmath194 fixed . @xmath56 goes here from 0.2 to 5 so that , in all cases , the mixed quantity @xmath195 ^ 3/|{\bm r}|$ ] has reached its asymptotic regime . in both figures the vertical dashed lines show the position of the average value @xmath196 . ] actually , it is not only the average of the `` local dissipation '' that converges to a constant but its full distribution seems to attain a stationary regime on times of the order of @xmath147 . figure [ fig : pdf_loc_diss ] ( a ) shows for a given initial separation , the convergence at large times of the pdf of @xmath193 ^ 3/|{\bm r}(t)|$ ] . one observes that the right and left tails converge on different timescales . the tail associated to large positive values ( separating pairs ) occurs on timescales of the order of @xmath117 ( which is in this case @xmath197 ) , while that for negative values ( approaching pairs ) converges slower . for the largest time ( @xmath198 ) one observes that deviations to the asymptotic distribution occur again at very large positive values . this is due to a contamination of such events by the large scales of the turbulent flow . this decrease is in agreement with the observed departure in fig . [ fig : locdiss_fntime_r0_24eta ] of the average from its asymptotic value at large times . figure [ fig : pdf_loc_diss ] ( b ) shows the pdfs of @xmath195 ^ 3/|{\bm r}|$ ] for different initial separations and at a fixed time sufficiently large to be ensured that all distributions have attained their asymptotic regime . one observes a robust collapse , much more pronounced than for both the distribution of separations and that of velocity differences . note that in fig . [ fig : pdf_loc_diss ] the distributions are raw and were not rescaled by any moment of @xmath195 ^ 3/|{\bm r}|$ ] . the asymptotic pdf is peaked around zero ( rather than its mean value ) and displays fat tails that , according to our data , are @xmath199 on both sides . to our knowledge , such a fast and manifest convergence of these mixed statistics has never been reported before . from a phenomenological viewpoint , one expects @xmath200 in the explosive richardson obukhov regime , so that the `` local dissipation '' @xmath195 ^ 3/|{\bm r}|$ ] should become constant at sufficiently large times . however we have observed here that the convergence of this quantity to its asymptotic value occurs much faster and in a much more definite manner than the convergence of the statistics of @xmath1 and @xmath201 to their respective asymptotic forms . this indicates that the statistical stationarity of the `` local dissipation '' is more likely to be a cause rather than a consequence of richardson obukhov explosive separation . at the moment we unfortunately lack a clear understanding of the physical mechanisms that are responsible for the observed behavior of @xmath195 ^ 3/|{\bm r}|$ ] and that could shed light on its relation to explosive separation in this section we turn back to the statistics of the distance @xmath100 between tracers . our goal is to explain the mechanisms leading to very large or very small values of this distance in the light of the various observations made in previous sections . considering the relative dispersion of tracers conditioned on their initial distance @xmath26 can be geometrically interpreted in terms of the time evolution of an initially spherical surface of radius @xmath26 that is centered on a reference trajectory . the transport of this surface by a turbulent flow is generally very complex . incompressibility implies that the volume of the sphere is conserved but the velocity field roughness will be responsible for strong distorsions of its surface . this is represented in fig . [ fig : evol_sphere ] that shows for @xmath59 at three various times the shape defined by the instantaneous position of 60 trajectories that are all initially at a distance @xmath202 from a reference tracer . quasi - lagrangian evolution of the set of tracers that are initially at a distance @xmath203 of a reference trajectory . ] we can qualitatively deduce from these snapshots that the large excursions of the inter - trajectory distance go together with strong pinches of the surface . we also see that the intense stretchings are occurring in a time - correlated manner : the surface angles that are visible at late time have formed at very early stages . as we will now see , the most - separated pairs have been so for long times and carry a reminiscent dependence on the initial separation . we first consider the events related to distances that are much larger than their average . it was argued in sec . [ subsec : higher - order ] that the large - value tail of the separation pdf is not well described by richardson s distribution . as seen in fig . [ fig : histo_r_4096 ] for @xmath102 , the tail is indeed broader than @xmath204 for @xmath205 and narrower otherwise . it seems to tend to a gaussian when either @xmath26 or @xmath22 are sufficiently large . we have checked that these behaviors are not due to a contamination by pairs that have reached the large scales of the turbulent flow , as the considered distances are still well below @xmath25 , except maybe for the largest initial separation @xmath169 . these observations suggest that for such extreme events , there is a long - term memory of the initial separation . to qualify further the history of pairs that are well separated at large times , we have performed the following analysis . fixing a final time @xmath206 sufficiently large to have reached the explosive richardson obukhov regime , we have carried out statistics conditioned on pairs that are far separated at @xmath207 , say such that their distance is latexmath:[$|{\bm r}(t_\mathrm{f})| \ge 2\ , \langle finite inertial subrange effects , we have restricted this analysis to @xmath143 and to initial separations @xmath209 , and we have chosen the largest compatible value of @xmath206 , namely @xmath210 . let us denote by @xmath211 the resulting conditional ensemble average , i.e. @xmath212 . figure [ fig : v_condr_fn_time ] ( a ) represents the relative increase @xmath213 of the mean - squared displacement . the various curves , which are associated to different initial separations @xmath26 , have a maximum at @xmath214 represented by a dashed line . the value of this maximum is becoming larger when @xmath26 decreases and approaches the dissipative scales . this is a signature of the dependence of the large - positive value tail upon @xmath26 . on the right - hand side of the maximum , the various curves tend back to @xmath215 . this indicates that separations that are large at a given intermediate time relax in average to an ordinary behavior at later times . this can be regarded as a consequence of the weakening of large - distance probability tails as a function of time . conditional statistics over pairs that are at a distance at least twice the average at time @xmath216 ( conditional average on @xmath217 is denoted by @xmath218 ) . ( a ) relative increase in the mean squared change of separation @xmath92 as a function of time . ( b ) averaged longitudinal velocity difference conditioned on the separation at time @xmath216 ( @xmath60 ) and without conditioning ( solid line ) . inset : initial value of the conditioned longitudinal velocity difference as a function of the initial separation ; the solide line is @xmath219 . ] the situation is rather different when interested in the left - hand side of the maximum . the curves do not converge to @xmath215 , and thus to an average behavior when @xmath220 . this means that pairs that are well separated at a given time are likely to have been so at any previous times . this asymmetry when going forward or backward in time could already be grasped in fig . [ fig : evol_sphere ] . the trajectories that are far away from the reference tracer at the latest time ( as , e.g. , those defining the left and top corners ) are visibly also well separated at the intermediate time . conversely the right corner at the intermediate time have later stopped separating faster than the average . figure [ fig : v_condr_fn_time ] ( b ) shows the time evolution of the averaged longitudinal velocity difference @xmath221 conditioned on having a separation twice the average at time @xmath210 . a first remark that can not be seen from the log - log plot is that the initial value of this average is strictly positive . the inset represents the variation of @xmath222 as a function of the initial separation . this implies that the pairs that are well separated at a late time are preferentially separating from the very beginning . the initial longitudinal velocity fluctuation that is necessary for the pairs to be far apart at time @xmath206 becomes weaker when @xmath26 increases . data suggest that @xmath223^{1/2 } \propto r_0^{-2/3}$ ] as seen in the inset . this initial separation makes the selected pairs reach an almost diffusive regime at time much shorter than the end of the average ballistic regime . for large - enough initial separations we indeed have @xmath224 for @xmath225 . the pair distance encountered a final acceleration right before the time of conditioning @xmath206 . after that , the longitudinal velocity difference relaxes slowly to the average regime . again here , as in the case of the average separation , we observe that the imposing of having a large distance at a large time selects pair histories . the main contribution to statistics is indeed given by couples encountering an initially violent separation and quickly reaching larger scales . such pairs converge rapidly to an explosive regime and continue to separate faster than the average for a long time . this is the snowball effect we have mentioned before . exponent @xmath226 of the right tail of the pdf of distances @xmath100 as a function of time and for various initial separations . the exponent was obtained by fitting @xmath227 to a power - law for @xmath228.,scaledwidth=60.0% ] in order to give a more quantitative handle on the far tail of the separation pdf , we have estimated and fitted its functional shape and studied its variations . for that , following the observations made in sec . [ subsec : higher - order ] , we have assumed that @xmath229 and measured how the exponent @xmath226 depends on both time and initial separation . figure [ fig : exp_rigthtail ] shows the time behavior of the exponent @xmath226 for various values of the initial separation . on the figure , the three dashed horizontal lines represent the tails of a gaussian ( @xmath230 ) , of an exponential ( @xmath231 ) and of richardson s distribution ( @xmath232 ) . at time @xmath38 , the distribution is peaked around @xmath26 and compactly supported , so that @xmath233 . the exponent then decreases , crosses @xmath231 ( so that the distribution becomes a stretched exponential ) for @xmath234 , and reaches a minimal value that depends on the initial separation @xmath26 . for the smallest @xmath26 , this minimal value is below richardson s prediction , as previously noticed . for larger separations , this minimum increases and when @xmath66 the curves seem to saturate to the value @xmath235 . also , we can not exclude that all curves converge to the exponential value when @xmath236 . the increase at the last stage can hardly be blamed on an integral - scale effect . we have indeed excluded here all pairs that are separated by a distance larger than @xmath237 . unscaled pdfs of the absolute separation between tracers for various times and the initial separations ( a ) @xmath175 and ( b ) @xmath186.,title="fig:",scaledwidth=50.0% ] unscaled pdfs of the absolute separation between tracers for various times and the initial separations ( a ) @xmath175 and ( b ) @xmath186.,title="fig:",scaledwidth=50.0% ] much more violent and intermittent events take place for particle pairs that separate much less than the average . as already stressed and observed in fig . [ fig : evol_sphere ] , the large excursions of inter - trajectory distances go together with strong pinches of separations . this is evidenced from fig . [ fig : histo_r_notscaled ] that represents the probability distribution of @xmath1 at various times and for two inertial - range values of the initial separation . one observes that at very short times , the pdf is peaked around @xmath26 . when time increases , its maximum , which roughly corresponds to the value of @xmath238 , shifts to larger values and the distribution broadens simultaneously at large and small values . this leads to the development for @xmath239 of an intermediate range of pairs whose separations lag behind the average evolution . in this subrange , the pdf behaves as a power law @xmath240 , where the exponent @xmath241 evolves as a function of time and @xmath26 . the power - law behavior is substantiated when measuring the cumulative probability @xmath242 = \int_0^r p(r^\prime)\,\mathrm{d}r^\prime \propto r^{\beta+1}$ ] of inter - trajectory distances . one expects from richardson arguments that @xmath243 , so that @xmath244 . two dimensional sketch of the time evolution of trajectories that are initially located at a distance @xmath26 of a reference tracer ( shown as a black dot at the center of the circle ) and then spread.,scaledwidth=75.0% ] the exponent @xmath241 can be interpreted in terms of fractal geometry . if all the trajectories were uniformly and independently distributed in space , the fraction of pairs at a distance less than @xmath9 would be @xmath245 , where @xmath246 is the space dimension . also one would expect this fraction to be @xmath247 if the trajectories were all confined on a surface , and @xmath248 if they were on a curve . in general , the exponent @xmath249/[\log r]$ ] measures the _ correlation dimension _ of a fractal set . @xmath250 behavior thus corresponds to the idea that , at sufficiently long times , trajectories forget about their initial separation and distribute homogeneously in space . as we have previously discussed , pair dispersion can be geometrically interpreted as the average lagrangian evolution of an initially spherical surface of radius @xmath26 centered on a reference trajectory . at time @xmath38 , it is clear that @xmath251 is a heaviside function centered on @xmath252 , so that @xmath253 . at later times the stretching and pinching of the sphere leads to a non - trivial behavior of @xmath251 for @xmath254 . this behavior is sketched for two dimensions in fig . [ fig : sketch_fractal ] . the exponent @xmath241 is thus measuring the fractality of the image of the sphere by the lagrangian flow . it relates to the fraction of this forward - in - time image that remains within a distance @xmath9 of the reference trajectory . at sufficiently large times ( right - most panel of fig . [ fig : sketch_fractal ] ) , one expects the set of trajectories to be mainly stretched by the flow eddies with the most relevant correlation time , that is those of size @xmath238 , and the small scale fluctuations to become less and less important . hence , at very large times , the only pairs that are still at a distance @xmath255 should be distributed on a surface ( curve in two dimensions ) , so that @xmath256 and @xmath257 . this approach contradicts the prediction of richardson that , somehow , postulates that the velocity difference correlation time does not depend on scale ( as it is assumed to be zero ) . ( a ) cumulative probability distribution of the distance @xmath258 $ ] as a function of @xmath9 for @xmath259 and at various times ; the gray dashed curve is the cumulative probability associated to richardson s distribution that displays a behavior @xmath260 at small distances . ( b ) time evolution of the left - tail power - law exponent @xmath241 for various values of the initial separation . the black solid line shows @xmath243 corresponding to richardson s distribution and the dashed line stands for the value @xmath261 that is argued in the text.,title="fig:",scaledwidth=50.0% ] ( a ) cumulative probability distribution of the distance @xmath258 $ ] as a function of @xmath9 for @xmath259 and at various times ; the gray dashed curve is the cumulative probability associated to richardson s distribution that displays a behavior @xmath260 at small distances . ( b ) time evolution of the left - tail power - law exponent @xmath241 for various values of the initial separation . the black solid line shows @xmath243 corresponding to richardson s distribution and the dashed line stands for the value @xmath261 that is argued in the text.,title="fig:",scaledwidth=50.0% ] to shed light on such ideas , we have estimated the exponent @xmath241 for various @xmath26 and at different times . for this purpose , we have used the cumulative probability @xmath251 ( shown in fig . [ fig : exponent ] ( a ) for @xmath259 ) to obtain @xmath262 through a fit of its local slope . the resulting measurements are displayed in fig . [ fig : exponent ] ( b ) as a function of time . up to the large error bars due to a lack of statistics , one observes that the timescale @xmath67 is again relevant to describe both the initial dynamics and the convergence to an asymptotic regime . at small times @xmath94 , the exponent @xmath241 takes large positive values , which , as explained above , are due to the initial conditioning @xmath263 . at time @xmath264 , the exponent becomes smaller than 3 . for @xmath85 , it asymptotically approaches the value @xmath243 , so that data give evidence in favor of the argument exposed above . interestingly , for small initial separations @xmath265 , the limit is reached from below and preceded by a minimum of @xmath241 . this means that for such intermediate times , the pinching ( or equivalently the stretching ) of the sphere is so strong that the intersection of the latter with spheres of radius @xmath9 defines a fractal object that is more concentrated than a surface . such a temporary singular behavior is certainly related to the presence of tails fatter than exponential at very large distances . in this paper we have confirmed and extended previous results of @xcite on the timescales of convergence of turbulent pair dispersion to an asymptotic regime . we have seen that low and medium - order moments of the particle separation approach an asymptotic regime on times of the order of @xmath266 , where @xmath133 denotes the ( absolute value ) structure function associated to the initial separation @xmath26 and @xmath3 is the mean rate of kinetic energy dissipation . this timescale has been shown to be relevant to describe also the initial kinematic change of sign of the longitudinal velocity difference @xmath148 and the tails of the distance distribution at small values . however , we have seen that @xmath67 is not relevant to describe the convergence of the velocity difference statistics to an asymptotic regime . we have observed that up to the largest time , the skewness , the flatness , and more generally the shape of the velocity difference distribution still depend on the initial separation , even when time is rescaled by @xmath67 . this leads to a behavior that is by far more intermittent than that of separations . also we obtained evidence that the far tail distribution of separation is also keeping a nontrivial memory on @xmath26 up to the largest times . throughout this paper , we made an important use of geometrical considerations to explain phenomenologically the statistical events leading to extreme fluctuations . in particular , we argued and obtained numerical evidence that , at sufficiently large times , the probability distribution @xmath267 of inter - tracer distances is @xmath268 for @xmath254 and decays as @xmath269 when @xmath270 . these two observations strongly contradicts richardson s eddy diffusivity approach , which seems nevertheless to give a good approximation of the core of the distribution . finally , we obtained a striking result concerning mixed distance and velocity statistics . we indeed found that the distribution of a `` local dissipation '' , defined as @xmath195 ^ 3/|{\bm r}|$ ] , attains relatively quickly a scaling regime that is independent of both @xmath26 and time . this behavior , which , to our knowledge , has never been observed before , gives very strong constraints for the development and validation of stochastic markovian models for turbulent relative dispersion . a central question that we raised concerns the physical mechanisms leading to this fast convergence of the `` local dissipation '' to a statistically stationary behavior . at the moment , an even phenomenological explanation is still missing . we have understood that kinematic considerations can be used to explain why @xmath192 ^ 3/|{\bm r}|\rangle$ ] , which is initially negative , becomes quickly positive . this is due to the fact that initially approaching trajectories will always eventually separate . however , there is no clear idea of why this average should converge to a time - independent value . possible ideas could consist in finding other mixed moments that are strictly ( and not asymptotically ) conserved by the two - point turbulent lagrangian flow . such considerations are kept for future work . finally , another possible extension of the current study is to apply the developed understanding of the geometry of relative dispersion to more complicated turbulent transport situations involving more than two trajectories . this is for instance the case for the forward - in - time dynamics of triangles or tetrahedrons studied in @xcite . the quality of today numerical data would be very useful to revisit such questions with an emphasize on extreme events . another situation that is much closer to applications is that of the transport by turbulence of pollutant patches . we have seen in the current study that relative dispersion relates to the distorsion of a sphere by the lagrangian flow . we related the small - scale behavior of separation statistics to the fractal dimension of this object . such geometrical considerations could be generalized to understand and quantify the large concentration fluctuations in the dispersion of a pollutant spot . we are grateful to l. biferale , m. cencini , g. falkovich , a. frishmann , g. krstulovic , a. lanotte , s.s . ray for useful discussions and remarks . access to the ibm bluegene / p computer jugene at the fz jlich was made available through the xxl - project hbo28 . the research leading to these results has received funding from dfg - for1048 and from the european research council under the european community s seventh framework program ( fp7/2007 - 2013 , grant agreement no .

using the multiwavelength photometric and spectroscopic data covering the chandra deep field south obtained within the great observatories origins deep survey , we investigate the rest - frame uv properties of galaxies to , including the evolution of the luminosity function , the luminosity density , star formation rate ( sfr ) and galaxy morphology . we find a significant brightening ( mag ) in the rest - frame 2800 characteristic magnitude ( ) over the redshift range and no evolution at higher redshifts . the rest - frame 2800 luminosity density shows an increase by a factor over the redshift range investigated . we estimate the star formation rate density to.2 from the 1500 and 2800 luminosities . when no correction for extinction is made , we find that the star formation rate derived from the 2800 luminosity density is almost factor two higher than that derived from the 1500 luminosities . attributing this difference to differential dust extinction , we find that an e(b v)=0.20 results in the same extinction corrected star formation rate from both 1500 and 2800 luminosities . the extinction corrected sfr is a factor ( ) higher than the uncorrected sfr derived from 1500 ( 2800 ) luminosity . we investigate the morphological composition of our sample by fitting srsic profiles to the acs galaxy images at a fixed rest - frame wavelength of 2800 at . we find that the fraction of apparently bulge - dominated galaxies ( srsic index ) increases from% at to% at . at the same time , we note that galaxies get bluer at increasing redshift . this suggests a scenario where an increased fraction of the star formation takes place in bulge - dominated systems at high redshift . this could be the evidence that the present day ellipticals are a result of assembly ( i.e. , mergers ) of galaxies at . finally , we find that galaxy sizes for a luminosity selected sample evolves as between redshifts and . this is consistent with previous measurements and suggests a similar evolution over the redshift range .

astro-ph0609016

over the last decade , a large number of studies have measured the evolution of cosmic star formation rate ( sfr ) in the range @xmath18 , using different diagnostics ( steidel et al . 1995 ; madau et al . 1996 ; chary & elbaz 2001 ; giavalisco et al . 2004a ; schiminovich et al . 2005 ; takeuchi et al . 2005 ; see also hopkins ( 2004 ) for a compilation of the sfr from a large number of multiwavelength surveys ) . these studies show a rapid increase in the sfr to @xmath19 , beyond which it flattens or turns over . while most studies are in broad agreement on the shape of the sfr , the absolute normalization is still uncertain , mainly due to unknown extinction in the uv to optical bands and systematic effects depending on star formation diagnostic , selection of the star - forming population and the adopted flux limit of the surveys used ( e.g. , hopkins & beacom 2006 ) . a further question in these investigations concerns the morphology of star forming galaxies and how it evolves with redshift . these studies require complete surveys of star forming galaxies with known redshifts , sfrs and morphologies . using the high spatial resolution and sensitivity of the advance camera for surveys ( acs ) on - board the hubble space telescope ( hst ) , a number of large , multi - waveband and deep surveys have now been completed , with the two deepest being the great observatories origins deep survey ( goods ; giavalisco et al . 2004b ) and the hubble ultra - deep field ( hudf ; beckwith , s. v. w. , et al . 2006 , in preparation ) . the combined wavelength coverage , spatial resolution and depth of these surveys allow accurate measurements of the sfr , morphology and size of these galaxies to faint flux levels ( @xmath20 ) and high redshifts ( @xmath21 ) . recently , the ultraviolet morphological properties of star forming galaxies in the goods and hudf were explored by a number of investigators . using wfpc2 observations of the goods parallel fields at f300w , de mello et al . ( 2006 ) studied rest - frame uv properties of galaxies to @xmath22 . they found that their sample included all major morphological types , with compact and peculiar morphologies becoming relatively more abundant at higher redshift ( @xmath23 ) . furthermore , lotz et al . ( 2006 ) studied the uv morphologies of star - forming galaxies in the goods - s at @xmath22 and @xmath24 and found no significant differences between the galaxy morphologies at these redshifts . conselice et al . ( 2004 ) used the @xmath8 acs observations of goods - s to identify populations of luminous diffuse objects and luminous asymmetric objects at @xmath25 which they argue are the progenitors of todays normal disk and elliptical galaxies . ravindranath et al . ( 2006 ) find that among their sample of bright lbgs at @xmath26 , about 40% have light profiles that can be approximated by an exponential profile as seen in disks , while 30% have close to r@xmath27 profiles as seen spheroids . however , they note that even these galaxies do show clumpy or faint asymmetric features characterestic of tidal interaction or minor merger . the method of gini coefficients used by lotz et al . ( 2006 ) is more sensitive to merger - like features . lotz et al . ( 2006 ) find that among their sample of lbgs at @xmath28 , 30% have relatively undisturbed spheroid - like morphologies , about 10 - 25% are major mergers . about 50% are like exponential disks or have minor mergers . except for small differences arising from the different methods that are employed , the results from ravindranath et al . ( 2006 ) are in good agreement with lotz et al . ( 2006 ) . finally , the evolution of galaxy sizes has also been investigated using the goods and hudf data sets by bouwens et al . ( 2004a , 2004b , 2005 ) and ferguson et al . ( 2004 ) . in this paper , we study uv properties of a complete sample of star - forming galaxies in the goods - s at rest - frame 1500 and 2800 wavelengths . the investigation includes a study of the evolution of the luminosity function , the luminosity density , the sfr , morphology , and galaxy size . in 2 we describe our data . this is followed by results on the luminosity functions and luminosity densities ( 3 ) , the sfr ( 4 ) , and morphology and size ( 5 ) . we conclude and summarize our results in 6 . throughout this paper we use @xmath29 , @xmath30 and @xmath31 km s@xmath32 mpc@xmath32 . magnitudes are in the ab system . the goods south data used here consist of deep wide - field @xmath8 acs observations in the f435w , f606w , f775w , and f850lp passbands ( hereafter @xmath33 and @xmath34bands ) combined with ground - based eso vlt / isaac near - ir @xmath35 , and @xmath36band observations . the limiting 10@xmath37 sensitivity for the acs observations are @xmath38 , @xmath39 , @xmath40 , and @xmath41 ( giavalisco et al . 2004b ) , while the isaac observations have limits @xmath42 , @xmath43 , and @xmath44 ( vandame , b. , et al . 2006 , in preparation ) . in this investigation we use a @xmath45-band selected catalog and adopt a limiting magnitude @xmath46 . we choose a limit @xmath1 mag brighter than the 10@xmath37 sensitivity to assure that we can reliably determine photometric redshifts and measure morphological parameters for our galaxy sample . the area of the field used here is covered by both acs and isaac and corresponds to @xmath11130 square arcmin . photometry is derived using sextractor ( bertin & arnouts 1996 ) . before deriving photometry , we convolve the space- and ground - based images to a common psf . however , when deriving morphological properties , we use original unconvolved acs images . besides the acs+isaac photometric catalog , we also produce a solely ground - based catalog including observations in @xmath47-band ( ctio , 4 m mosaic ) , @xmath48 , @xmath49 , @xmath50 and @xmath51bands ( eso , 2.2 m wfi ) , @xmath35 , and @xmath52band ( eso , ntt sofi ) and the above described isaac data . these observations are also centered on the goods south area , but the field is significantly wider , in total covering @xmath53 square arcmin , of which about one third is observed in near - ir . the 10@xmath37 sensitivities are @xmath54 , @xmath55 , @xmath56 , @xmath57 , @xmath58 , @xmath59 , @xmath60 , @xmath61 ( giavalisco et al . the limiting magnitude adopted for this @xmath62-band selected catalog is @xmath63 . note that we restrict our investigation to goods south and do not include goods north due to the shallower near - ir data of the latter , leading to less accurate photometric redshifts at @xmath64 and faint magnitudes . photometric redshifts are calculated using the template fitting method incorporating priors as described in dahlen et al . ( 2005 ) . for each object , we derive the photometric redshift , the redshift probability distribution and the best - fitting spectral type . the spectral types cover e , sbc , scd and i m templates ( coleman et al . 1980 , extended in the uv and near - ir by bolzonella et al . 2000 ) and two starburst templates ( kinney et al . 1996 ) . to test the accuracy of the photometric redshifts , we compare with a sample of 519 spectroscopic redshifts taken from the eso / goods - cdfs spectroscopy master catalogue , which is a compilation of a number of datasets with major contributions from le fvre et al . ( 2004 ) , vanzetta et al . ( 2005 ) and mignoli et al . we find an accuracy @xmath65=0.08 ( where @xmath66 ) , after excluding a small fraction ( @xmath113% ) outliers with @xmath67 . restricting the redshift range to that studied here , @xmath68 , results in a slightly lower outlier fraction ( @xmath112% ) . furthermore , the outliers have distributions in redshift and magnitudes similar to the full redshift sample . the distribution in spectral types somewhat biased towards later types for the outliers , but there are outliers of all sed types . therefore , we do expect to have a small fraction of outliers and galaxies with uncertain photometric redshifts . however , when deriving e.g. , luminosity functions ( hereafter lfs ) and luminosity densities ( lds ) , we use a method that incorporates the full photometric redshift probability distribution , instead of a single redshift , in order to minimize the effect of photometric redshift uncertainty . this method is further described in dahlen et al . ( 2005 ) . for a number of objects ( 312 of total 2976 ) , we replace the photometric redshifts with available spectroscopic redshifts . the spectral types of these objects are calculated using the photometric redshift technique after fixing the redshift to its spectroscopic value . one aim of this paper is to investigate the galaxy properties at rest - frame 2800 . to do this , for each acs band we identify a redshift range where the filter encompass the rest - frame 2800 band this is defined so that at least 25% of the filter s integrated transmission is short - ward of rest - frame 2800 ( lower redshift limit ) and at least 25% is long - ward ( upper redshift limit ) . resulting redshift ranges are given in table 1 . the redshifts that divide the volume of each bin into two equal halves are @xmath69 , @xmath70 , @xmath71 , and @xmath72 for @xmath73- , @xmath74- , @xmath75- and @xmath45-bands , respectively . in the following when we derive galaxy morphologies , we use measurements in different bands depending on redshift so that we always are probing the same rest - frame wavelength . by requiring that we observe the same rest - frame wavelength in all bins , we minimize the effects of photometric k - corrections when deriving rest - frame 2800 luminosities , as well as morphological k - corrections , which may otherwise bias morphological interpretations ( e.g. , papovich et al . 2005 ) . when deriving the lf , we also use the ground - based catalog in order to measure rest - frame 2800 at lower redshifts than possible with the acs bands . using the observed @xmath47-band , we probe the rest - frame 2800 at @xmath76 . to investigate how the uv derived sfr depends on the choice of rest - frame band , we also calculate the lf at rest - frame 1500 using the same galaxies as in the 2800 selected samples . in the two lowest redshift bins ( @xmath77 and @xmath78 ) , we do not have any filter that observe sufficiently close to the rest - frame 1500 and do therefore not include these bins . at @xmath79 , we use the @xmath73-band , which observes rest - frame @xmath112000 , and extrapolate to 1500 using the best - fitting spectral template derived from the photometric redshift fitting . in the two highest redshift bins , @xmath80 , and @xmath81 , we can more directly derive the rest - frame 1500 luminosity . we use the @xmath82-method ( schmidt 1968 ) to derive the lf . to account for the relatively large errors in the photometric redshifts , we incorporate the redshift probability distribution derived from the photometric redshift method when determining the lf . this procedure is described in detail in dahlen et al . ( 2005 ) . after deriving the lf , we fit the usual schechter function ( schechter 1976 ) to the data . at 2800 , we determine all the schechter function parameters ( @xmath83 , @xmath84 and @xmath85 ) in the redshift bin at ( @xmath78 ) , which is the bin where we reach the faintest limit in rest - frame 2800 absolute magnitude . in the remaining bins , we fix the faint - end slope to the value derived in this bin , @xmath86 . the resulting schechter function parameters are listed in table 1 . in figure [ fig1 ] , we show the rest - frame 2800 lfs in the five redshift bins . the best fitting schechter functions are shown with solid lines , while the result derived in the lowest redshift bin is shown with dotted lines . vertical dotted lines represent completeness limits . ( when fitting the schechter function parameters , we only use points brighter that these limits . ) the completeness limits are derived by calculating the absolute 2800 magnitude corresponding to the apparent magnitude limit ( @xmath87 ) for a range of galaxy seds from ellipticals to starbursts at the central redshift of each bin . we thereafter choose the brightest of these magnitudes as the limit , ensuring us that we are complete for all considered galaxy types . the figure clearly shows a strong evolution , where the characteristic magnitude gets brighter at higher redshifts . the total brightening in @xmath88 is @xmath1 mag between @xmath89 and @xmath90 . this is similar to the evolution in @xmath88 found by wolf et al . ( 2003 ) and gabasch et al . ( 2004 ) . between the two highest redshift bins , the evolution is consistent with being flat or slightly fading . for the 1500 lfs , we use the redshift bin at @xmath90 to determine the faint - end slope and find @xmath91 . we chose this bin since it directly observes the rest - frame 1500 ( in contrast to the @xmath92 bin which requires extrapolation ) . after fixing the faint - end slope , we derive the schechter parameters . results are given in table 1 and figure [ fig2 ] . similar to the 2800 results , the characteristic magnitude at 1500 shows a brightening with redshift from @xmath51.1 to @xmath51.7 , with no significant evolution at higher redshift . combined with results of bouwens et al . ( 2005 ) , who found that @xmath93 brightens by @xmath94 mag between @xmath21 and @xmath95 , we develop a scenario where the characteristic uv magnitude brightens , turns over , and finally fades when going from high to low redshift . such evolution is consistent with the hierarchical model ( e.g. , cole et al . 2004 ) where the initial brightening at rest - frame uv is due to infall of gas , mergers and merger induced star formation . when the merger rate , and the related star formation , decreases , galaxies start to fade as the reservoirs of cold gas are depleted . the observations suggest a peak in the uv characteristic magnitude at @xmath96 , which we note coincides with the proposed peak in the cosmic sfr , as well as the peak of the merger epoch for the brightest and most massive galaxies ( conselice et al . in contrast , we note that gabasch et al . ( 2004 ) find a monotonic brightening in the characteristic magnitude at both rest - frame 1500 and 2800 to redshift @xmath97 . to derive the uv ld over all magnitudes , we approximate the lf in each bin by the derived schechter function parameters . the ld is given by @xmath98 the resulting lds are listed in table 1 . in figure [ fig3 ] , we plot the 2800 lds as red dots . we note an increase in the ld by a factor @xmath4 between redshift @xmath89 and @xmath99 . at higher redshift , the evolution is consistent with being flat . for comparison , we also plot the lds from the canada france redshift survey ( cfrs ; lilly et al . 1996 ) , hdf - n ( connolly et al . 1997 ) and combo-17 ( wolf et al . there is in general an excellent agreement at overlapping redshifts between the goods and the combo-17 results , however , at @xmath89 , we find a somewhat lower value , suggesting a stronger evolution to @xmath51 . at @xmath131.5 , we find a ld that is a factor @xmath112 higher than the results given in connolly et al . we note , however , that part of the difference between results may be due to clustering variance , especially affecting surveys with small areas such as the hdf - n ( connolly et al . ) which only covers @xmath115 sq . the deviating point from lilly et al . at @xmath50.6 could also be due to a combination of statistical scatter and clustering variance . the rest - frame 1500 ld is plotted in figure [ fig4 ] as blue dots . there is a trend showing an increase in the ld between redshifts @xmath99 and @xmath100 . however , our highest redshift point again suggests a flattening in the evolution . in the figure , we also plot recent measurements from galex ( schiminovich et al . 2005 ; triangles and wyder et al . 2005 ; black dot ) . together , these measurements depict a scenario where the ld evolves rapidly between @xmath101 and @xmath99 , thereafter the evolution is somewhat less steep and at @xmath96 , the evolution flattens out , or may even turn over . since the uv luminosity is mainly produced by short - lived o and b stars , it is closely related to the ongoing star formation . using results from stellar synthesis codes and assumptions on the past star formation history , it is possible to derive a conversion factor relating the uv luminosity and the ongoing sfr . for example , madau et al . ( 1998 ) present relations between the 1500 and 2800 luminosities and the sfr assuming an exponentially declining star formation history , sfr@xmath102exp(t/@xmath103 ) . they find that the relation is quite insensitive to the past star formation history when varying @xmath103 in the interval 1 to 20 gyr . we have examined the results in madau et al . and confirm that at 1500 , the relation holds and is independent from the star formation history while at 2800 , the conversion factor varies by @xmath1130% depending on @xmath103 . the larger spread at 2800 is due to the fact that more long - lived a and f stars are significant contributors here in addition to o and b stars . to investigate the relation between luminosity density , ongoing star formation and star formation history averaged over an ensemble of galaxies , we use two different models . in the first , we assume a constant sfr(@xmath45 ) this should be approximately true at @xmath104 where we know from independent measurements ( e.g. , giavalisco et al . 2004a ) that the star formation rate is consistent with being flat . as our second model , we use the fit to a number of sfr measurements presented in giavalisco et al . and parameterized in strolger et al . ( 2004 ) , @xmath105 where @xmath106 is the age of the universe ( in gyr ) , @xmath107=13.47 gyr , @xmath108=0.021 , @xmath109=2.12 , @xmath110=1.69 , and @xmath111=0.207 . in both models , we assume an onset of the star formation at @xmath112 , the results are , however , not sensitive to the exact choice of this redshift . also , the conversion factor is only dependent on the shape of the sfr history and not the absolute normalization ( i.e. , parameter @xmath108 is arbitrary in the equation above ) . this also means that results are independent of any constant amount of dust extinction . input are also galexev stellar population synthesis models ( bruzual & charlot 2003 ) , giving the age luminosity evolution for a simple stellar population ( ssp ) at different uv luminosities . for these models , we assume solar metallicity and two different imfs , a standard salpeter imf ( salpeter 1955 ) and a chabrier imf ( chabrier 2003 ) . the relation between ongoing sfr and uv luminosity is finally derived by convolving the ssp results with the sfr history @xmath113 where @xmath114 is the flux at time @xmath115 after an instantaneous burst of the ssp and @xmath116 is the age of the universe at @xmath112 . the conversion factor can thereafter be used to calculate the ongoing sfr from the observed uv luminosity density using @xmath117 in figure [ fig5 ] , we show the resulting conversion factors between uv luminosity and ongoing sfr at rest - frame wavelengths 1500 and 2800 . note that we at this stage do not include any corrections for dust extinction . for the constant sfr scenario ( blue lines in figure [ fig5 ] ) , we find that the conversion factor is close to independent of redshift . this is expected since we in this case expect that the fraction of the uv luminosity coming from ongoing star formation remains constant . at 2800 , however , there is a slight increase in @xmath118 with time due to the build - up of old stars with residual 2800 luminosity . for the more realistic sfr history ( red lines in figure [ fig5 ] ) , we note a clear , albeit small , redshift dependence on the relation between uv luminosity and ongoing sfr . at 1500 , the conversion factor increases by @xmath119% between @xmath120 and @xmath121 ( similar for both imfs ) . the evolution is , as expected , stronger at 2800 , where we note an increase by @xmath116% over the the same redshift range . as a consistency check , we calculate what fraction of the observed 2800 luminosity density comes from galaxies that are best - fitted by an early - type sed , i.e. , representing non - star forming galaxies . we do this by fitting a schechter function to early - type population separately in the two lowest redshift bins . we find that @xmath122% of the luminosity density comes from this population , consistent with the theoretical predictions . at higher redshift , the statistics are too poor to allow a fit to the early - type population separately . we use the conversion factors derived from the realistic sfr history to derive the ongoing sfr for the goods dataset . results are shown in figure [ fig6 ] . for both imfs , we find that the sfr derived from the 2800 luminosity is a factor @xmath123 higher compared to the sfr derived from the 1500 luminosity . since we have accounted for the fraction of the 2800 luminosity that comes from a non - star forming population , the remaining discrepancy should have other causes . first , an imf with a different slope at high masses compared to the standard salpeter imf and chabrier imf , could change the relation between uv luminosities and sfr , and therefore explain some of the difference . second , it is expected that a large fraction of the uv luminosity is obscured by dust extinction . since the extinction is more severe at shorter wave - lengths , this could explain the difference in the sfrs derived from 1500 and 2800 luminosities . assuming a calzetti attenuation law ( calzetti et al . 2000 ) and a salpeter imf , we fit the amount of dust needed to bring the sfrs from 1500 and 2800 to agreement and find that a mean e(b v)=0.20@xmath124 . this is equivalent to an extinction corrected sfr a factor @xmath6 higher at 1500 and a factor @xmath7 higher at 2800 , compared to the uncorrected sfrs . the extinction correction we find here is higher compared to the median e(b v)=0.15 found by shapley et al . ( 2001 ) , who derive the extinction from optical and near ir photometry of a set of @xmath95 lyman break galaxies ( also assuming a calzetti attenuation law and a salpeter imf ) . however , the shapley et al . ( 2001 ) results are based on bruzual & charlot ( 1996 ) models . using the more recent models in bruzual & charlot ( 2003 ) results in a median e(b v)=0.20 for the same galaxy sample ( shapley , private communication ) , which is consistent with what found here . the correction at 1500 is somewhat larger than the factor @xmath125 derived from the galex data by schiminovich et al . our results suggest that the extinction corrected sfr is up to @xmath126 higher than previously derived extinction corrected rates based on uv luminosities , e.g. , giavalisco et al . ( 2004a ) . in figure [ fig7 ] , we show the extinction corrected sfrs derived from both the 1500 and 2800 luminosity densities . for the two lowest redshift bins , we derive the sfr from the 2800 luminosity density assuming the same extinction correction as derived from the results at @xmath127 . our results support a steep increase in the sfr between @xmath101 and @xmath99 . compared to the local measurement by wyder et al . ( 2005 ) , we find an increase in the sfr by at least a factor @xmath125 between @xmath101 and @xmath22 . at redshifts @xmath64 , there is a mild increase in the sfr , however , we can not exclude a flat rate over the range @xmath128 . in the figure , we also plot sfrs based on 1500 luminosity densities from schiminovich et al . ( 2005 ; triangles ) . we have used the corrections derived here to calculate the sfrs from the lds . we note that the agreement is good , and only for the lowest redshift point in the goods sample is the deviation outside the one sigma error bars . we use galfit ( peng et al . 2002 ) to determine the morphology at rest - frame 2800 for the objects in the @xmath8 acs @xmath129 bands . we use unconvolved images when deriving morphological catalogs . these are thereafter matched to the photometric catalogs ( based on convolved images ) . we do not include the ground - based @xmath47-band data in the morphological investigation ( i.e. , the @xmath77 redshift bin ) due to the lower resolution and broader psf in this band . the light distribution of each object is fit with a srsic ( 1968 ) radial profile @xmath130},\ ] ] where @xmath131 is the srsic index , @xmath132 is defined as @xmath132=(2@xmath1330.331 ) , @xmath134 is the effective radius containing half of the total galaxy light and @xmath135 is the surface brightness at @xmath134 . in the fitting we use a psf derived from stars in the field . a pure exponential profile has @xmath136 , while a de vaucouleurs profile has @xmath137 . we adopt a division between `` disk''-objects with @xmath138 , and `` bulge''-objects with @xmath10 , which is the same criterion as used by e.g. , barden et al . ( 2005 ) and ravindranath et al . bulge objects should mostly consist of ellipticals , while disk objects are foremost spirals and irregulars . we note , however , that at high redshifts and short rest - frame wavelengths , we do not expect the galaxy population to follow the hubble sequence as well as the case is locally . in order to derive srsic index and galaxy radius it is necessary that a major fraction of the galaxy luminosity comes from a main source for which a center can be well defined . when going to higher redshifts , and in particular in rest - frame uv , it is expected that galaxy morphology becomes more disturbed with the possibility of multiple peaks in the light distribution . this could affect the measured properties , making it difficult to interpret e.g. , srsic index and galaxy size . in order to investigate the galaxies included in this investigation , we visually inspected representative samples of galaxies in all redshift bins , including both `` disk''- and `` bulge''-objects . we find that `` bulge''-objects in all bins are well represented by symmetric and single peaked objects . at all redshifts , `` disk''-objects are , as expected , more irregular in shape . however , we find that the sample we select is clearly dominated by galaxies with a main luminous component for which a center can be defined . therefore , we trust that the srsic index and galaxy radius derived reflects the morphological properties of the galaxies . the reason for the relatively well behaved morphologies of these galaxies can be attributed to the selection , where we only include intrinsically bright objects . for these , we do not expect that star forming lumps or trails of tidal interactions affect the overall morphology more than marginally in most cases . for fainter objects , the situation should be different where secondary star forming regions can cause multiple peaks with similar luminosity . although we can calculate srsic index also for these objects , we do not include them in our investigation since the interpretation of these is more uncertain . in figure [ fig8 ] , we show representitative samples of `` disk''- and `` bulge''-objects in each redshift bin . we have chosen these randomly among galaxies brighter than @xmath139 ( described below ) . for the `` bulge''-objects , the selected galaxies represent 100% , 12% , 9% , and 24% of the full sample in the four redshift bins , respectively ( from low to high redshift ) . for `` disk''-objects , the corresponding fractions are 10% , 3% , 2% , and 17% , respectively . as discussed above , the figure shows that the `` bulge''-objects are more symmetric and mostly consist of a single central object while `` disk''-objects are more irregular in shape . to compare the same population in each band , we use an absolute magnitude cutoff that is related to the characteristic magnitude in each band ; @xmath140 where @xmath141 is derived from the completeness limit in the highest redshift bin , @xmath14219.75 . we find @xmath143 , i.e. , in each band we select galaxies with @xmath144 . the completeness in determining the srsic index for galaxies brighter than the magnitude limit is 80% , 94% , 94% and 93% in the @xmath73 , @xmath74 , @xmath75 and @xmath45-bands , respectively . there are two reasons for the incompleteness . first , since srsic indexes are derived using the unconvolved images , while photometry is derived using convolved , there are cases where objects do not match between catalogs . this is because multiple nearby objects in the unconvolved image may be merged into a single object in the convolved image . for these objects we do not assign a srsic index . about 75% of the objects without determined srsic index belong to this group . second , for the remaining 25% , the srsic fitting algorithm did not converge , mainly due to the faintness of the objects . besides deriving the _ morphological _ characteristics of the galaxy sample using the srsic index , we also derive the _ spectral _ type describing the overall shape of the galaxy sed . spectral types are given by the best fitting template sed derived from the photometric redshift fitting . we number spectral types 1 - 5 , with e=1 , sbc=2 , scd=3 , im=4 and starburst=5 ( for both starburst templates used ) . we also include intermediate types which are interpolations between subsequent templates . we divide the galaxies into two spectral types : early - type galaxies ( dominated by e spectrum , type @xmath145 1.5 ) and late - type galaxies ( type @xmath146 1.5 ) . using our full data set , we calculate the median srsic index for the early - type and late - type populations . we find that the early - type population has a high median srsic index , @xmath147 , indicating that galaxies with early - type spectra are bulge dominated . the median srsic index for the late - type population is @xmath148 , consistent with a disk - dominated ( spiral ) galaxy . if we assume that the srsic index in general follows the spectral type of the galaxy , we can use this assumption to correct for the incompleteness in derived srsic indexes described in 5.1 . we do this by placing galaxies with early - type spectra in the high srsic sample ( @xmath149 ) , while later type galaxies are placed in the low srsic sample ( @xmath150 ) . note , however , that there is a large dispersion in the correlation between galaxy spectral and morphological type , and we find that @xmath151 of the late - type galaxies have high srsic indexes ( @xmath10 ) , suggesting that they are bulge dominated , and that a similar fraction of the early - type galaxies have low srsic index . in figure [ fig9 ] , we show the distribution of srsic indexes in each redshift bin for galaxies brighter than the limit @xmath139 . the figure shows that the rest - frame 2800 morphology is dominated by low srsic indexes , i.e. , `` shallow '' disk dominated or irregular systems . this is natural since star formation is expected to occur in these systems . if we calculate the fraction of galaxies that are bulge dominated ( @xmath10 ) , we find that 8% , 15% , 24% and 31% of the galaxies belong to this group at @xmath78 , @xmath152 , @xmath153 , @xmath154 , respectively . this suggests that the fraction of bulge dominated systems ( in rest - frame 2800 ) increases with redshift . to investigate the effect on results due to incompleteness in determined srsic indexes , we assign a high srsic index ( @xmath10 ) to early - type galaxies and a low srsic index ( @xmath138 ) to late - type galaxies , as discussed in 5.2 . after making the correction , we find that 9% , 14% , 22% , and 29% of the galaxies have high srsic indexes , in the four bins , respectively . the trend in the evolution remains , suggesting that the incompleteness only marginally affects results . we have also examined if there is a bias in the derived srsic indexes as the mean apparent magnitude of the galaxies gets fainter at higher redshift . we investigate this by adding simulated galaxies with known radial profiles ( exponential and de vaucouleurs ) spanning a range of magnitudes and radii to the real acs @xmath45-band image using the iraf task mkobjects . we thereafter run galfit and determine the fraction of the galaxies for which we correctly recover the input profile . at magnitudes @xmath155 , we find that @xmath156% of both galaxy types have correctly determined profiles . at the faintest magnitudes included in this investigation @xmath157 , we find that 95% of the input exponential profiles are recovered as disk galaxies , while 88% of the de vaucouleurs profiles are correctly assigned a bulge profile . we therefore conclude that the magnitude bias when determining srsic index should not be severe . next , we investigate how the relation between morphological and spectral types evolves with redshift . deriving the median srsic index in the four redshift bins we find @xmath158=0.49 , 0.76 , 1.27 , and 1.34 at @xmath50.55 , @xmath51.14 , @xmath51.75 , and @xmath52.23 , respectively . this is consistent with the general trend found above showing that the fraction of high srsic index objects increases with redshift . calculating the median spectral type in the four bins , we find @xmath145type@xmath146=3.3 , 3.7 , 5.0 , and 5.0 , respectively . overall , this shows that the galaxy population producing the 2800 luminosity mainly consists of galaxies with late - type seds which get bluer at higher redshift . this is contrary to the overall correlation between types where later spectral types in general have lower srsic indexes . our result that the srsic index increases with redshift , while the galaxies at the same time get bluer , again suggests that a larger fraction of the star formation at high redshift occurs in more concentrated bulge - dominated objects compared to low redshifts . so far , we have investigated the fraction of the number of galaxies belonging to the different morphological types . we now estimate the fraction of the rest - frame 2800 luminosity emitted by disk- and bulge - dominated systems . we do this by summing the emitted flux in each redshift bin and thereafter determining what fraction comes from bulge systems . using the magnitude limit @xmath139 , we find fractions 0.11@xmath159 , 0.14@xmath159 , 0.21@xmath159 and 0.31@xmath160 , of the luminosity is emitted in bulge ( @xmath10 ) systems at @xmath78 , @xmath152 , @xmath153 , and @xmath154 , respectively . these results are plotted in figure [ fig10 ] . the trend we find is inconsistent with a flat non - evolution scenario at a @xmath161 level . note that we do not extrapolate the lf to faint magnitudes . if the faint - end slopes of the disk and bulge population differ , or evolve differentially with time , then conclusions could be different at the faint end . similar results , showing an increased fraction of the sfr occurring in spheroids / ellipticals to @xmath99 , are presented by menanteau et al . ( 2006 ) , who find that the fraction of the sfr occurring in ellipticals at rest - frame @xmath73 is @xmath162 at @xmath163 and @xmath164 at @xmath165 . at higher redshift ( @xmath166 ) , the fraction decreases in menanteau et al . , in contrast to the continuous increase suggested in this investigation . at lower redshifts , lauger et al . ( 2005 ) find a steep increase in the number of faint blue bulge - dominated galaxies between @xmath167 and @xmath92 in rest - frame @xmath48band . the emergence of this population could be related to the morphological evolution at higher redshifts seen in this investigation . conselice et al . ( 2005 ) investigate the relation between rest - frame @xmath73-band morphology and star formation in the hdf - n . they find that star formation is higher in more massive and early - type galaxies at @xmath127 compared to lower redshift . this is consistent with what is found in our investigation . our results are also in agreement with the result of menanteau et al . ( 2001 ) , who use @xmath8 data with @xmath168 mag from the hubble deep fields n and s and find that a significant fraction ( @xmath169 ) of intermediate redshift ( @xmath99 ) spheroids / ellipticals have blue colors . furthermore , lotz et al . ( 2006 ) find a fraction @xmath1130% spheroids at @xmath24 and @xmath1115% spheroids at @xmath22 . this is consistent with the trend found in this investigation showing a decrease in spheroid fraction at lower redshifts . the fractions found are also reasonably consistent with the numbers we find . note , however , that lotz et al . measure spheroid fraction at a different rest - frame ( fuv , @xmath170 ) and use a different method for defining spheroids , we could therefore expect some differences . most of the stellar mass in ellipticals is expected to have assembled by @xmath51 ( bundy et al . the increased fraction sfr in bulge systems at higher redshifts we observe could therefore be the sign of the build up of ellipticals at @xmath127 , and the subsequent decrease in sfr reflects a more passive evolution at @xmath171 . this is similar to the scenario menanteau et al . ( 2001 ) suggest where the blue galaxies they observe at @xmath51 are old elliptical systems undergoing recent star formation induced by mergers or inflow of material . however , we can not exclude that some part of the bulge systems we observe at high redshift may evolve to the bulges of todays population of old giant spirals . to summarize , we find an increasing fraction of galaxies at higher redshifts to be bulge - dominated in the rest - frame 2800 . this suggests that a significant amount of star formation occurs in these objects at high redshift . at least part of the increase in sfr should be due to the build up of todays ellipticals by infall or mergers at @xmath131 . simple scaling models for the expected redshift evolution of galaxy sizes in the hierarchical model is presented by e.g. , fall & efstathiou ( 1980 ) , mo et al . ( 1998 ) and bouwens & silk ( 2002 ) . the scale length of a spiral galaxy , @xmath172 , is assumed to be proportional to the virial radius , @xmath173 and therefore related to the virial mass via @xmath174 where @xmath175^{1/2}\ ] ] and @xmath176 is the circular velocity at @xmath177 . for fixed circular velocity , we therefore expect @xmath178 , while for fixed mass we expect @xmath179 . in the case of a sample with fixed rest - frame luminosity , the behavior depends on the evolution of the mass - to - light ratio and is expected to be intermediate between these relations ( bouwens & silk 2002 ; ferguson et al . 2004 ) . when investigating the size evolution , we want to compare galaxies within the same range of e.g. , absolute magnitudes or masses ( compared to the above investigation where we examine the brightest population in each bin ) . we investigate the completeness in each bin by adding simulated galaxies to the real images . the simulated galaxies span a wide range of absolute magnitude and size ( i.e. , surface brightness ) . in figure [ fig11 ] , we plot the half - light radius ( @xmath180 , as measured by sextractor ) at rest - frame 2800 of the observed galaxies as a function of absolute magnitude . we only include galaxies with disk morphology ( @xmath138 ) since equation ( 4 ) implies the formation of a spiral galaxy . over - plotted are the 50% ( dashed lines ) and 99% ( solid lines ) completeness limits derived from the simulations . the gray line shows the adopted magnitude limit , @xmath181 . in figure [ fig12 ] , we plot the distribution of radii for galaxies brighter than this limit . there are only a few objects brighter than our magnitude limit that fall within the area where we expect a completeness between 50% and 99% ( figure [ fig11 ] ) . this indicates that we are not severely affected by incompleteness and should not be missing more than a few objects at the most . as a further test , we have taken the galaxies detected @xmath74-band ( @xmath92 redshift bin ) and redshifted their properties ( luminosity , size ) , including effects of surface brightness dimming . we thereafter randomly distribute this population on the @xmath182 and @xmath34band images and run sextractor to derive the fraction of recovered galaxies . in the third redshift bin , @xmath90 , we recover 99% of the redshifted galaxies , with no dependence on the size of the galaxies . in the highest redshift bin , @xmath0 , the overall recovered fraction is 89% , with with a size dependence suggesting that we detect @xmath1175% of galaxies with log(r@xmath183)@xmath184 and @xmath1192% of galaxies with log(r@xmath183)@xmath185 . these results also suggest that we are not severely affected by incompleteness in our high redshift galaxy samples . the completeness in the highest redshift bin , derived from these simulations , is shown with the gray line in figure [ fig12 ] . note that we do not take into account any size - redshift relation ( i.e. , eq . [ 4 ] ) , when redshifting our objects from low to high redshift bins . if this relation exists , we will be even more complete . we use the results from our simulations to account for the incompleteness by weighting objects within the area between to 50% and 99% lines by the inverse of the completeness at the particular point in the absolute magnitude - radius diagram . we further note that only a few galaxies survive the selection in the lowest redshift bin . this is due to the general fading of the galaxy population at lower redshift and the small volume of the lowest bin . we keep this bin in our investigation , but note that due to the low statistics ( large statistical errors ) , our results regarding redshift evolution of the galaxy properties do not change whether we include this bin or not . as a consistency check on the selected galaxy samples , we calculate the median absolute magnitude and find @xmath186 in the three highest redshift bins , while the low redshift bin has a median magnitude @xmath187 . this reassures us that we are comparing samples with similar magnitudes . figure [ fig13 ] shows the size redshift relation for our magnitude selected galaxy sample . in each bin we plot the median radius in kpc for our adopted cosmology . over - plotted are the theoretical curves representing the cases where galaxy sizes evolve as @xmath178 and @xmath179 . following the approach in e.g. , bouwens et al . ( 2004b ) , we fit our data to the functional form @xmath188 , and find @xmath189 between @xmath190 ( including the @xmath12 point do not change the derived value of @xmath191 ) . if we use the @xmath178 and @xmath179 parameterizations and fit the curves to functional form @xmath188 over the redshift range @xmath70 to @xmath72 , we get @xmath192 and @xmath193 , respectively . this indicates that the evolution we find is somewhat steeper than expected for a mass selected sample . a possible explanation is that the stellar mass - to - light ratio decreases at higher redshift . such scenario is expected as the sfr in individual galaxies increases at higher redshift . a manifestation of this is the observed brightening of @xmath83 . compared to previous investigations , our results agree well with the evolution @xmath194 found by bouwens et al . ( 2005 ) between redshifts @xmath195 and @xmath196 for a large sample dropout galaxies . even though bouwens et measure the evolution at a shorter rest - frame wavelength ( @xmath197 ) and include all galaxy morphologies , these results suggest that the size - redshift correlation continues to @xmath198 . ferguson et al . ( 2004 ) investigate the evolution at rest - frame @xmath111500 and find an evolution , @xmath199 , over the redshift range @xmath200 , which is equivalent with @xmath201 . this is somewhat steeper than found here , with both investigations consistent with the @xmath199 relation . there are also investigations of the size redshift relation at optical wavelengths . in rest - frame @xmath48band , papovich et al . ( 2005 ) find @xmath202 between @xmath203 and @xmath99 . while in rest - frame v - band , trujillo et al . ( 2005 ) find an evolution @xmath204 at @xmath205 and @xmath206 over a larger redshift interval @xmath207 for galaxies with disk morphology ( @xmath208 ) . in combination , these results suggest a similar size - redshift evolution over the redshift range @xmath209 which is consistent with @xmath210 and , as noted by trujillo et al . ( 2005 ) , that the relation is similar at visual and uv rest - frame wavelengths . next we investigate the evolution of the size - redshift evolution at a fixed mass scale . we use the relation between mass and rest - frame @xmath74 magnitude and @xmath211 color given in bell et al . ( 2003 ) to determine masses . rest - frame magnitudes and colors are calculated using the observed bands that are closest to the redshifted rest - frame @xmath73 and @xmath74-bands , respectively , in combination with k - corrections calculated from the best - fitting template sed for each individual galaxy . more details on this procedure are given in dahlen et al . derived masses are calculated for a kroupa ( 2001 ) imf ( a salpeter ( 1955 ) imf gives masses @xmath11two times higher ( kauffman et al . 2003 ) ) . we stress that this is not directly a mass selected sample since it is based on photometry . derived masses are subjected to uncertainties and scatter in these relations . this may be especially severe for this investigation since we are using the very bluest galaxy population where the relation between mass in luminosity has a large dispersion . after calculating mass , we select galaxies with log(@xmath212)@xmath213 . to check that we are comparing galaxy samples with similar mass , we calculate the median mass in each redshift bin . we find masses log(@xmath212)=10.5 , 10,4 , 10.3 , and 10.3 , at redshifts @xmath214 , @xmath79 , @xmath80 , and @xmath81 , respectively . this shows that we are comparing samples with similar masses . in figure [ fig14 ] , we show the resulting size - redshift relation for the sample , where red dots represent the median value in each bin . fitting to the functional form @xmath188 , we find @xmath215 . this suggests a somewhat shallower evolution in the size - redshift relation for a mass selected sample compared to a luminosity selected sample , even though the difference is within errors . comparing the @xmath191 parameters between the luminosity and the mass selected samples indicates a decrease in stellar mass - to - light ratio at higher redshift . the evolution found is consistent with the simple scaling law @xmath179 ( which suggests @xmath216 over our redshift range ) . a shallower evolution is found in the rest - frame @xmath49band by trujillo et al . ( 2005 ) , who report @xmath217 between @xmath95 and @xmath101 . comparing with the @xmath49band evolution at fixed luminosity , this also indicates a decrease in the stellar mass - to - light ratio with redshift ( trujillo et al . 2004 , 2005 ) . in contrast , using disk galaxies ( @xmath138 ) in rest - frame @xmath74-band , barden et al . ( 2005 ) did not find any significant evolution in the mass - size relation over the redshift range @xmath218 . since this is derived over a different redshift range and rest - frame band , it is difficult to compare results . however , we note that a non - evolution scenario at @xmath190 is inconsistent with our results at a @xmath114@xmath37 level . therefore , further investigations are needed to determine the redshift - size evolution , in particular , it would be desirable to cover a larger redshift range using measurements in a single rest - frame band . we use the goods cdf - s optical and near - ir observations to derive the evolution of the rest - frame 1500 and 2800 luminosity functions , luminosity densities and star formation rates to @xmath0 . taking advantage of the high resolution @xmath8 acs imaging , we also derive the evolution the rest - frame uv morphological properties . we find : * there is a strong evolution in the uv characteristic magnitude with redshift . we find a brightening in @xmath83 by @xmath1 mag between @xmath89 and @xmath90 . at both 1500 and 2800 , we find no significant evolution at @xmath219 . * the rest - frame 2800 luminosity density increases by a factor @xmath4 over the redshift range @xmath220 , while at rest - frame 1500 the increase is a factor @xmath221 between @xmath92 and @xmath90 . the evolution flattens in both wave - lengths at @xmath219 . * we find that the uncorrected sfr derived from 2800 luminosity is a factor @xmath123 higher than the sfr derived from 1500 luminosity . assuming that the difference is due to differential dust extinction , we find that e(b v)=0.20@xmath124 ( using a calzetti attenuation law and a standard salpeter imf ) can explain the difference in the uncorrected sfrs . * the extinction corrected sfr is a factor @xmath6 ( @xmath7 ) higher than the uncorrected sfr calculated from 1500 ( 2800 ) luminosity densities . these corrections lead to a @xmath126 higher sfr compared to the sfrs derived from dropout samples which assume a smaller extinction correction ( e.g. , giavalisco et al . * the sfr we derive shows a steep increase out to @xmath222 . comparing with the local value from wyder et al . ( 2005 ) , we find that the sfr is a factor @xmath223 higher at @xmath22 compared to @xmath101 . we note , however , that our measurements are consistent with a flat rate at @xmath131.7 . * we find that the fraction of galaxies with high srsic index ( @xmath10 ) , indicating a bulge - like morphology , increases at higher redshift . at @xmath12 , we find that @xmath224 of the galaxies have bulge morphology , while the fraction is @xmath169 at @xmath0 . at the same time , the mean color of the galaxies gets bluer at higher redshift . this suggests that an increased fraction of the star formation takes place in objects with bulge - like morphology at high redshift . this could be the sign of the formation of todays elliptical population via mergers and infall at redshifts @xmath131 . at least part of the high redshift bulge systems may , however , evolve into the bulges of todays population of old spirals . * investigating the luminosity - size redshift evolution of disk galaxies at rest - frame 2800 over the redshift range @xmath225 , we find that galaxies become larger at lower redshift . our results are consistent with trends found at both higher and lower redshift and suggests a size evolution @xmath226 between redshifts @xmath21 and @xmath101 . this is somewhat steeper then the simple scaling law @xmath179 , expected in the hierarchical model for a sample selected by mass . this suggests a stellar mass - to - light ratio that decreases at higher redshift . * we find a somewhat smaller evolution @xmath227 when selecting galaxies by mass ( where mass is indirectly derived from the galaxies rest - frame colors and absolute magnitudes ) . this milder evolution supports the change in stellar mass - to - light ratio with redshift . compared to measurements in the literature , we find a steeper evolution in the size evolution for the mass selected sample . however , investigations are made at different rest - frame bands and redshift ranges , and are therefore difficult to directly compare . barden , m. , et al . 2005 , preprint , a - ph/0502416 bell , e. f. , mcintosh , d.h . , katz , n. , & weinberg , m. d. 2003 , apjs , 149 , 289 bertin , e. , & arnouts , s. 1996 , a&as , 117 , 393 bolzonella , m. , miralles , j. -m . , & pell , r. 2000 , a&a , 363 , 476 bouwens , r. & silk , j. 2002 , apj , 568 , 522 bouwens , r. j. , illingworth , g. d. , blakeslee , j. p. , broadhurst , t. j. , & franx , m. 2004a , apj , 611 , l1 bouwens , r. j. , et al . 2004b , apj , 606 , l25 bouwens , r. j. , illingworth , g. d. , blakeslee , j. p. , & franx , m. 2005 , apj , submitted ( preprint a - ph/0509641 ) bruzual , g. & charlot , s. 2003 , mnras , 344 , 1000 bundy , k. , ellis , r. s. , & conselice , c. j. 2005 , apj , 625 , 621 calzetti , d. , armus , l. , bohlin , r. c. , kinney , a. l. , koornneef , j. , & storchi - bergmann , t. 2000 , apj , 533 , 682 chabrier , g. 2003 , pasp , 115 , 763 chary , r. , & elbaz , d. 2001 , apj , 556 , 562 cole , s. , lacey , c. g. , baugh , c. m. , & frenk , c. s. 2000 , mnras , 319 , 168 coleman , g. d. , wu , c .- c . , & weedman , d. w. 1980 , apjs , 43 , 393 connolly , a.j . , szalay , a.s . , dickinson , m.e . , subbarao , m.u . , & brunner , r.j . 1997 , apj , 486 , l11 conselice , c. j. , bershady , m. a. , dickinson , m. , & papovich , c. 2003 , aj , 126 , 1183 conselice , c. j. , et al . 2004 , apj , 600 , l139 conselice , c. j. , blackburne , j. a. , & papovich , c. 2005 , apj , 620 , 564 dahlen , t. , mobasher . b. , somerville , r. s. , moustakas l. a. , dickinson , m. , ferguson , h. c. , & giavalisco , m. 2005 , apj , 631 , 126 de mello , d. f. , wadadekar , y. , dahlen , t. , casertano , s. , gardner , j. p. 2006 , aj , 131 , 216 fall , s. m. , & efstathiou , g. 1980 , mnras , 193 , 189 ferguson , h. c. , et al . 2004 , apj , 600 , l107 gabasch , a. , et al . 2004 , a&a , 421 , 41 giavalisco , m. , et al . , 2004a , apj , 600 , l103 giavalisco , m. , et al . , 2004b , apj , 600 , l93 hopkins , a. m. 2004 , apj , 615 , 209 hopkins , a. m. & beacom , j. f. 2006 , apj , submitted ( preprint a - ph/0601463 ) kauffmann , g. , et al . 2003 , mnras , 341 , 33 kinney , a. l. , calzetti , d. , bohlin , r. c. , mcquade , k. , storchi - bergmann , t. , & schmitt , h. r. 1996 , apj , 467 , 38 kroupa , p. 2001 , mnras , 322,231 lauger , s. , ilbert , o. , buat , v. , tresse , l. , burgarella , d. , arnouts , s. , le fvre , o. , & vvds collaboration . 2005 , in edp - sciences , conference series , sf2a-2005 : semaine de lastrophysique francaise , ed . f. casoli , t. contini , j.m . hameury & l. pagani , 667 le fvre , o. , et al . 2004 , a&a , 428 , 1043 lilly , s.j . , le fvre , o. , hammer , f. , & crampton d. 1996 , aj , 460 , l1 lotz , j. m. , madau , p. , giavalisco , m. , primack , j. , & ferguson , h. c. 2006 , apj , 636 , 592 madau , p. , ferguson , h. c. , dickinson , m. , giavalisco , m. , steidel , c. c. , & fruchter , a. 1996 , mnras , 283 , 1388 madau , p. , pozzetti , l. , & dickinson , m. 1998 , apj , 498 , 106 menanteau f. , abraham , r. g. , & ellis , r. s. 2001 , mnras , 322 , 1 menanteau f. , ford , h. c. , motta , v. , benitez , n. , martel , a. , blakeslee , j. p. , & infante , l. 2006 , aj , 131 , 208 mignoli , m. , et al . 2005 , a&a , 437 , 883 mo , h. j. , mao , s. , & white , s. d. m. 1998 , mnras , 295 , 319 papovich , c. , dickinson , m. , giavalisco , m. , conselice c. j. , & ferguson , h. c. 2005 , apj , 631 , 101 peng , c. y. , ho , l. c. , impey , c. d. , & rix , h. , -w . 2002 , aj , 124 , 266 ravindranath , s. , et al . 2006 , apj , in press ( preprint a - ph/0606696 ) salpeter , e. e. 1955 , apj , 121 , 161 schechter , p. 1976 , apj , 203 , 297 schiminovich d. , et al . 2005 , apj , 619 , l47 schmidt , m. 1968 , apj , 151 , 393 srsic , j. l. , 1968 , atlas de galaxias australes ( crdoba : obs . astron . crdoba ) shapley , a. e. , steidel , c. c. , adelberger , k. l. , dickinson , m. , giavalisco , m. , & pettini , m. 2001 , apj , 562 , 95 steidel , c. c. , pettini , m. , & hamilton , d. 1995 , aj , 110 , 2519 strolger , l. g. , et al . 2004 , apj , 613 , 200 takeuchi , t. t. , buat , v. , & burgarella , d. 2005 , a&a , 440 , l17 trujillo , i. , et al . 2004 , apj , 604 , 521 trujillo , i. , et al . 2005 , apj , submitted ( preprint a - ph/0504225 ) vanzella , e. , et al . 2005 , a&a , 434 , 53 wolf , c. , meisenheimer , k. , rix , h .- w . , borch , a. , dye , s. , & kleinheinrich , m. 2003 , a&a , 401 , 73 wyder , t. k. , et al . 2005 , apj , 619 , l15 lccccclcc 0.29 - 0.37 & 0.33 & @xmath228 & -&21.8@xmath229 & 25.84@xmath230 & @xmath231 + 0.46 - 0.63 & 0.55 & @xmath232 & @xmath233 & 62.3@xmath234 & 26.29@xmath235 & @xmath236 + 0.92 - 1.33 & 1.14 & @xmath237 & -&39.9@xmath238 & 26.40@xmath239 & @xmath240 + 1.62 - 1.88 & 1.75 & @xmath241 & -&40.8@xmath242 & 26.56@xmath243 & @xmath244 + 2.08 - 2.37 & 2.23 & @xmath245 & -&33.8@xmath246 & 26.43@xmath247 & @xmath248 + & & & & & & + & & & & & & + 0.92 - 1.33 & 1.14 & @xmath249 & -&29.6@xmath250 & 26.19@xmath230 & @xmath251 + 1.62 - 1.88 & 1.75 & @xmath252 & @xmath253 & 31.1@xmath254 & 26.46@xmath255 & @xmath256 + 2.08 - 2.37 & 2.23 & @xmath257 & -&33.2@xmath258 & 26.34@xmath239 & @xmath259 [ table1 ]

we investigate the effects of abundance and age on the mid - uv spectra and mg strengths of stellar populations using simple population synthesis models . these models are used to constrain the star formation history of four nearby elliptical galaxies and spiral bulges . the mid - uv ( 1800 - 3200 ) light of evolved stellar populations ( 1 gyr ) is dominated by the main sequence turn - off , unlike the optical light which is dominated by the red giant branch . because the main sequence turn - off is sensitive to age and metallicity in ways different from the rgb , a detailed investigation of the mid - uv features of elliptical galaxies may help break the age - metallicity degeneracy that plagues optical techniques . also , a better understanding of this wavelength region is useful for the studies of 0.5 z 1.5 galaxies for which the rest frame mid - uv is redshifted into the visible . we create simple , single age ( 3 - 20 gyr ) , single metallicity ( z 0.0004 - 0.05 ) spectral energy distributions ( seds ) extending into the uv using the kurucz model stellar fluxes . comparison to standard stars mid - uv spectra reveals that the kurucz model fluxes accurately model a blend feature of fei and mgi at 2538 ( bl2538 ) and the slope of the continuum between 2600 and 3100 ( ) . we find that our simple single age , single metallicity seds agree well with these mid - uv features of globular clusters . however , the majority of the galaxies do not agree with the bl2538 , , and mg values given by these simple models . the mid - uv features appear to require both an old metal - rich and an old metal - poor ( z 0.001 ) population . the implied metal - poor population is less than 10% of the total mass for all the galaxies but dominates the seds shortward of 3000 . intermediate age ( 1 - 5 gyr ) populations are not required to match the uv for any of the galaxies , but are not ruled out . despite being limited by the quality of the model stellar fluxes , our study has yielded two promising mid - uv spectral diagnostics ( bl2538 and ) and suggests unique and complex star formation histories for elliptical galaxies .

astro-ph9911233

the integrated light of elliptical galaxies contains many clues to their star formation histories . currently , the most popular approach to deciphering the broad band colors and spectral features of elliptical galaxies and star clusters is evolutionary population synthesis ( charlot & bruzual 1991 ; bressan , chiosi , & fagotto 1994 ; worthey 1994 ) . this technique models the integrated light for a stellar population of a given age and metallicity by summing up the light from individual stars along theoretical isochrones . the resulting spectral energy distribution ( sed ) is dependent on the libraries of stellar evolutionary tracks and stellar fluxes , the chosen star formation rate and initial mass function ( imf ) , and in the case of multi - metallicity populations , the chemical enrichment history . charlot , worthey , & bressan ( 1996 ) examined the differences between several recent models and found that they are seriously limited by the dearth of reliable libraries for cool and non - solar composition stars and by the uncertain properties of post main sequence stars . nevertheless , population synthesis is appealing because of its ability to prescribe a fundamental recipe for a galaxy . attempts to unravel the history of star formation in elliptical galaxies via spectral synthesis are plagued by the `` age - metallicity degeneracy '' ( worthey 1994 and references therein ) . the effective temperature of the red giant branch ( rgb ) decreases with increasing age and metallicity in such a way that the rgb of an old solar metallicity population looks nearly identical to that of a more metal - rich , slightly younger population . because the rgb dominates both the integrated line strength and broad band color longward of 4000 , a degeneracy of these properties in age and abundance results for stellar populations older than 1 gyr . the most promising attempts to disentangle age and metallicity from integrated spectra have involved use of the balmer absorption lines @xcite and mid - uv colors ( dorman , oconnell , & rood 1995 ) . all of these methods rely on spectral diagnostics for which the main sequence begins to dominate over the rgb . the contribution of the main sequence turn - off to the integrated light is greatest in the mid - uv region ( @xmath5 2500 - 3500 ) ; the contribution of the far - uv bright hb and post hb stars to the integrated light is less than 50% at 2500 ( worthey , dorman , & jones 1996 ) . @xcite suggest that the spectral break at 2900 is correlated with spectral type of the turn - off population but weakly influenced by metallicity . also , the continuum of mid - uv is affected by the metallicity of the stellar population . @xcite found that the color ( 2500-v ) was correlated with metallicity and fanelli et al . ( 1990 , 1992 ) found @xmath6 ( 2600-v ) 10 times more sensitive to [ fe / h ] than @xmath6 ( b - v ) . because the spectral type of the main sequence turn - off stars dominates the absorption features of the mid - uv and the continuum is highly sensitive to chemical content , a detailed investigation of the integrated mid - uv light of stellar populations may help break the age - metallicity degeneracy . one observational result population synthesis models have sought to explain is the correlation between the `` uv upturn '' shortward of 2000 and the optical mg@xmath0 absorption line strength in elliptical galaxies @xcite . the brightest galaxies in uv also have the strongest mg@xmath0 absorption ( burstein et al . 1988 ) , suggesting that these uv strong galaxies are metal - rich and possibly enhanced in @xmath7 element abundances . in the most metal - rich galaxies , the primary source of far - uv emission appears to be extreme horizontal branch ( ehb ) stars and their descendants , the agb - manqu stars . while the bulk of the evidence favors a metal - rich origin for the ehb stars @xcite , models with a distribution of metallicities including a metal - poor tail @xcite can not be excluded . in any case , the models for hot populations are strongly dependent on mass loss on the red giant branch and on the helium abundance . small changes in these two parameters can give wildly different uv upturns at fixed age and metallicity . for the purposes of this paper , the uv upturn is simply an underlying continuum that must be subtracted from the galaxy spectra prior to analysis of the spectral features . the star formation history of nearby ellipticals , particularly m32 , has been in the subject of much debate . in 1980 , @xcite proposed that a 5 gyr stellar sub - population is responsible for the relatively blue @xmath8 color of m32 . the idea that m32 contains a large intermediate age population was supported by detailed comparison of m32 and 47 tuc ( an old `` metal - rich '' globular cluster ) near uv spectra ( 3800 - 4400 ) @xcite . the study of h@xmath9 and h@xmath10 absorption lines , which are dependent on the main sequence turn - off stars in old stellar populations , suggest intermediate ages for many ellipticals @xcite . however , recent hst observations of m32 s red giant branch implies that at least half of m32 s stars are older than 8 gyr @xcite . the metallicity distribution of m32 and other ellipticals has also been investigated . @xcite found evidence for a hot metal - poor population in an otherwise metal - rich population in the mid - uv spectra of m32 . the width of m32 s rgb requires that the stars span a substantial range in metallicity @xcite . the estimated metallicity distribution has a low metallicity tail extending down to [ fe / h ] = -1.5 but is strongly skewed to solar metallicity and is much narrower than expected from a closed box chemical enrichment scenario . comparisons to synthetic spectra suggest that other galaxies , including ngc 1404 , ngc 4649 , and the bulge of m31 , also show a deficit of low metallicity stars relative to closed box models @xcite . the ages of distant elliptical galaxies has been the subject of some debate as well . @xcite determined an age of @xmath5 3.5 gyr ( assuming a solar metallicity ) for lbds 53w091 ( z @xmath5 1.55 ) by comparing the rest frame mid - uv spectral features to solar metallicity f and g stars and population synthesis models . a galaxy this old at a redshift of 1.55 would rule out a @xmath11 , @xmath12 , @xmath13 universe . a later comparison of this galaxy s spectra to improved f star models gave an age of 1 - 2 gyr ( assuming z @xmath14 @xmath15 ) , relaxing the cosmological constraints ( heap et al . because galaxies with 0.5 @xmath2 z @xmath2 1.5 have their mid - uv spectra redshifted into the visible range , a better understanding of the mid - uv features of old stellar populations will be useful for dating distant galaxies . using simple population synthesis models , we examine the mid - uv features and mg@xmath0 strengths of four nearby elliptical galaxies and spiral bulges . in order to study the spectral properties of the underlying main sequence and rgb branch stars , we exclude the uncertain hot hb and post hb stars from our models isochrones and attempt to subtract out the far - uv component from the galaxies spectra . specifically we use the blend feature at 2538 ( bl2538 ) and mg@xmath0 line indices and introduce a new tool for examining the mid - uv continua , the slope between 2600 and 3100 ( @xmath4 ) , to determine the galactic properties . the investigation of the mid - uv dependence on age and abundance via population synthesis is severely limited by inaccuracies in our model stellar fluxes . however , bl2538 and @xmath4 appear to be adequately modeled by the kurucz fluxes . we find for our simple population models that populations older than 3 gyr separate out in the s2850 v. bl2538 and s2850 v. mg@xmath0 planes . although globular cluster spectra agree with simple , single metallicity , single age populations , a small but mid - uv bright component of low metallicity ( z @xmath2 0.001 ) stars may explain the spectral features of ngc 1399 . m32 s uv features can not be explained purely by a significant population of intermediate age ( 1 - 3 gyr ) solar metallicity stars ; a small number of metal - poor stars are suggested by the mid - uv in this case as well . because metal - poor stars are so much brighter than metal - rich stars in the mid - uv , we can not ignore the effects of the metallicity distribution on our population synthesis models . the rest of the paper is arranged as follows : first , in section 2 we test our library of stellar fluxes ( lejeune , cuisnier , & buser 1997 ) for inaccuracies and determine the influence of these inaccuracies on our model spectral energy distributions ( seds ) . in section 3 , we explain our population synthesis models , their mid - uv and optical properties , and compare them to the mid - uv features of several composite globular cluster iue spectra ( bonatto , bica , & alloin 1995 ) . in section 4 , we present the fos spectra of ngc 1399 and ngc 3610 and the iue spectra of ngc 1399 , m32 and the bulge of m31 . we compare our models to these galaxies . in section 5 we examine the effects of a metal - poor sub - population on the integrated spectra of an otherwise metal - rich population . in order to examine the effect of abundance on the integrated light of old stellar populations , kurucz model stellar fluxes @xcite with [ fe / h ] = -1.5 , -1.0 , -0.5 , 0.0 , and + 0.5 were used for our population synthesis models . due to the paucity of non - solar abundance stars in empirical libraries , the kurucz models offer the advantage of the studying the effects of a full range of metallicity on the model stellar populations . however , at cooler temperatures the model fluxes do suffer from serious limitations . @xcite compared late g and k solar chemical composition giants and found that , while the optical and infrared observations matched the kurucz fluxes , significant discrepancies existed in the uv . they concluded that the models failed to account for chromospheric emission and thus underestimated the flux at wavelengths less than 3100 . also , the optical mg@xmath0 index of cool , metal - rich giants may not be well modeled by the kurucz stellar fluxes . when the kurucz 1992 models were compared to the improved kurucz 1993 models , chavez , malgnini , & morossi ( 1995 ) found that for @xmath16 @xmath2 5000k and [ fe / h ] @xmath14 0.0 , the newer models predicted a mg@xmath0 values up to 0.58 magnitudes stronger than the 1992 models . the kurucz 1993 fluxes corrected an error in the fei continuum opacity for cool stars , had more optical depth points , and included detailed modeling of scattering effects in the source function not accounted for in the 1992 fluxes . clearly , while the kurucz models are powerful tools for the modeling of stellar populations , a detailed understanding of their limitations is necessary to interpret the results of our model stellar populations correctly . here we quantify the accuracy of the kurucz model stellar fluxes in the mid - uv and the optical . we test the mid - uv spectra and optical features mg@xmath0 and h@xmath9 of the solar abundance kurucz 1992 models against the composite iue stellar main sequence spectral energy distributions complied by @xcite and standard stars in the @xcite catalog . the importance of the cool giants on the mid - uv and optical portions of our model seds is determined . we have used the un - calibrated kurucz model fluxes presented by @xcite as our stellar library . lejeune et al . ( 1997 ) have calculated correction functions for each of the kurucz models in order to yield synthetic broad - band colors which match the empirical color - temperature relations . however , they note that the dwarf stellar fluxes have a negligible optical color correction for @xmath16 @xmath14 3500 k , and that the color correction is significant for the giants only for @xmath16 @xmath2 4000 k. because we use the non - color corrected fluxes , we first tested the un - calibrated solar metallicity model colors against the colors of the stars in the gunn - stryker catalog @xcite . the broad band colors u - b and b - v of the kurucz fluxes match both the giant and dwarf stars in the gunn - stryker catalog to within @xmath18 0.2 magnitudes ( figure 1a & b ) . the coolest red giant colors are least well fit by the kurucz fluxes , with a discrepancy of @xmath5 0.2 magnitudes . fanelli et al . ( 1990 , 1992 ) examined the mid - uv spectral morphology of 218 normal stars observed by iue and found that the spectral lines in the uv are sensitive to and change significantly with spectral types ranging from o to k. we measured the spectral indices fe2402 , bl2538 , fe2609 , mg2800 , mg2852 , mgwide , and fe3000 as defined by @xcite ( table 1 and figure 2 ) for the kurucz model fluxes corresponding to solar , main - sequence ( ms ) stars and compared them to the indices of the iue stellar spectra measured by @xcite . unfortunately , few non - solar metallicity stars were observed , and so our discussion is limited to the comparison of solar abundance stars and the corresponding kurucz models . it is important to note that the `` continuum '' band passes for these indices contain very little true continuum . due to the strong line blanketing in the uv , we are forced to use pseudo - continuum band passes around the absorption features to compute indices and our results will depend on the accuracy of the pseudo - continuum as well as that of the absorption features of the stellar models . in figure 3a - g , we plot the b - v color against the mid - uv line indices for both the iue composite spectra ( triangles ) and kurucz fluxes ( connected crosses ) . the b - v colors for the kurucz models were measured using iraf task synphot.calcphot . only the bl2538 feature is well matched ( to better than 0.05 magnitudes ) by the full temperature range of kurucz main sequence model fluxes ( figure 3b ) . ( when the individual iue spectra are compared to the kurucz fluxes , the stellar bl2538 values show a great deal of scatter about the kurucz values ( dorman , oconnell , & rood 1997 ) . this may be due to noise in the individual iue spectra or metallicity and gravity variations . the difference in bl2538 between a g0v and g0iii star is @xmath5 0.4 magnitudes ; the difference between a g0v and g0iv star is much less , @xmath5 0.03 magnitudes . we examine the composite iue spectra of stars identified as luminosity class v because these stars will dominate the mid - uv light . ) in general , the kurucz model fluxes for b - v @xmath2 0.1 ( @xmath16 @xmath14 8000 k ) match the observed line strengths of iue data , but overestimate the line strengths for cooler stars ( b - v @xmath14 0.2 , @xmath16 @xmath2 7000 k ) . in particular , the @xmath16 = 6000 k , log g=4.5 solar kurucz model is too strong by 0.6 magnitudes for fe2404 , 0.4 for fe2609 , 0.4 for mg2800 , 0.4 for mg2852 , 0.3 for mgwide , and 0.15 too weak for fe3000 . these discrepancies may be due in part to the inevitably incomplete atomic line lists . they may also be the result of departures from lte , chromospheric emission filling in the mid - uv absorption lines for cooler dwarfs , and opacity due to tri - atomic molecules , none of which are modeled by the kurucz stellar fluxes . the continuum of the mid - uv range is strongly sensitive to the metallicity of the stellar population @xcite age may also influence the mid - uv continuum : the spectrum of an old stellar population at 2500 is dominated by flux from main sequence turn - off stars , and thus may be used to constrain this turn - off population more closely than the k and m giant dominated optical region of the spectrum . again , it is important to realize that the vast number of atomic absorption lines in the uv suppress most of the true continuum and what we observe is a pseudo - continuum . to examine the dependence of this region s pseudo - continuum on age and metallicity , we have measured the slope of the pseudo - continuum ( @xmath19 magnitude/1000 ) in the 2600 to 3100 range of our model simple populations ( section 3 ) . the slope of the pseudo - continuum between 2600 and 3100 ( hereafter @xmath4 ) was estimated using the iraf v2.11.1 task onedspec.continuum with high rejection set at 3 @xmath20 and low rejection set at 1 @xmath20 in order to exclude the effects of the magnesium features at 2800 and 2852 on the spectral shape . again , to determine possible inaccuracies in the model fluxes , we tested @xmath4 of the kurucz models against @xmath4 of the iue observations . in figure 3h , the @xmath4 for the main sequence iue spectra and kurucz models are plotted against @xmath8 for the range of slopes observed in the galaxies . the kurucz models pseudo - continua closely match that of the observations ( to better than 0.5 @xmath19 magnitude/ 1000 ) , except for the coolest stars ( @xmath21 1.1 ) . ( note : the color - corrected lejeune et al . 1997 version of the stellar models do not reproduce s2850 ; the shifts in b - v are small but the s2850 feature is shallower by up to 2 mag/ 1000 . ) we also examine the robustness of the kurucz models line strengths at optical wavelengths for solar metallicity giants . the mg@xmath0 feature is primarily a measurement of the galaxy s metallicity , but is also strongly linked to age . the h@xmath9 feature at 4860 is sensitive to main sequence turn - off temperature , and hence may be an indicator of age @xcite . observed elliptical galaxy h@xmath9 and h@xmath10 absorption strengths suggest intermediate ages ( @xmath2 8 gyr ) . we seek to determine the relationship between the mid - uv properties and these optical line strengths for elliptical galaxies ( section 4 ) . in figure 4a - d , we compared the mg@xmath0 and h@xmath9 features of the kurucz solar ms dwarfs and giants models to a number of standard ms dwarfs and giants taken from the gunn - stryker atlas . the kurucz models have an h@xmath9 equivalent width @xmath5 1 - 2 less than the standard stars , so we do not use this optical feature for our analysis . although mg@xmath0 is well modeled by the kurucz models with @xmath16 @xmath14 4500 k , cooler giants increasingly overestimate the feature by 0.2 magnitudes or more ( figure 4a ) . the coolest dwarf models also overestimate mg@xmath0 strength . given that @xcite found that the cool metal - rich giant kurucz models mg@xmath0 values were strongly sensitive to small changes in the scattering parameter , we will also not use the mg@xmath0 values estimated by the kurucz models . instead we use the empirical fitting formula given by @xcite ( the dashed line in figure 4a , b ) . @xcite determined the dependence of @xmath5 eighty stars mg@xmath0 on effective temperature , gravity , [ fe / h ] , and [ mg / h ] . our model seds were created using the technique of `` isochrone synthesis '' , whereby a stellar population is created by summing up stellar fluxes along isochrones for a given age and metallicity ; in section 3 we describe this process in greater detail . comparing the isochrones of different ages and metallicity @xcite , we see that old , metal - rich populations possess a significant number of k and m giants . since the kurucz stellar fluxes poorly model these stars in the uv , we examined the importance of the cool giants to the total integrated light of our oldest , most metal - rich ( 20 gyr , z=0.05 ) model population . we have removed the contribution from giant stars with @xmath16 @xmath2 4500 k from a 20 gyr , z=0.05 sed and compared it to the un - modified sed ( figure 5 ) . we see that the giants contribute very little light to the mid - uv ( 1800 - 3200 ) , and become significant for wavelengths @xmath14 4000 . thus we are confident that the mid - uv continuum of our synthetic populations is not significantly affected by the known inaccuracies of cool giants . in summary , after close investigation of the mid - uv and optical line strengths of the main sequence and giant solar metallicity kurucz models , we are left with only the blend feature of fei and mgi at 2538 in the mid - uv ( @xmath22 ) and the slope of the continua between 2600 and 3100 ( @xmath4 ) as features reproduced well by the kurucz models . we are satisfied that the mid - uv is not significantly influenced by the discrepancies in the cool giant stellar models . however , if further detailed investigation of the narrow band absorption features in the mid - uv is to be done , new model stellar fluxes with more accurate mid - uv and optical line strengths need to be constructed . it is important to note that our tests of the bl2538 and @xmath4 features have only shown that the relative fluxes of the pseudo - continuum over small wavelength regions are reasonably matched by the kurucz models . for bl2538 , we have only examined the relative flux of the local pseudo - continuum to the blend absorbtion feature , and for @xmath4 , we have tested the average slope of the pseudo - continuum over a 500 range . a 10 % error in the mid - uv pseudo - continuum would introduce a 0.1 magnitude error in bl2538 and a 0.2 magnitude/ 1000 error in @xmath4 . problems with the mid - uv pseudo - continuum over larger wavelength regions and its normalization relative to the optical pseudo - continuum may still exist . single metallicity , single age simple stellar populations were constructed using the `` isochrone synthesis '' technique , which sums up model stellar fluxes using a given set of isochrones , imf and star formation rate . we used kurucz model fluxes for stars with 3500 k @xmath2 @xmath16 @xmath2 40000 k , the padova isochrones @xcite , a salpeter imf , and an exponentially decaying star formation rate with @xmath23 = @xmath24 . spectral energy distributions were created for z=0.0004 , 0.001 , 0.008 , 0.02 , and 0.05 and ages 3 , 7 , 10 , 12 , 15 , and 20 gyrs . further details of the isochrone synthesis code are given in @xcite . the padova isochrones only contain `` normal '' p - agb stars and do not model post - early agb ( pe - agb ) , agb - manqu or `` extreme blue horizontal branch '' stars ( hb stars with zahb @xmath16 @xmath14 16,000 k ) which are all possible contributors to the uv upturn found in elliptical galaxies @xcite . while the existence of ehb stars and their descendents in elliptical galaxies is reasonably secure observationally , theoretical models of how the ehb is populated are uncertain . the number of ehb stars depends on parameters such as the metallicity , age , helium abundance of the stellar population , and on the details of mass loss at the tip of the red giant branch . while it is possible to choose plausible values for these parameters and construct synthetic spectra , the dependence on the poorly constrained rgb mass loss rate means that the uv upturn is not yet useful for constraining the age and chemical content of the stellar population . for galaxies with a strong uv upturn , the mid - uv portion of the spectrum contains flux both from ehb stars ( @xmath16 @xmath5 25,000 k ) and from cooler hb and ms turn - off stars . because we are interested in the ms turn - off stars , we have chosen to subtract the hot stellar component from both the models and the data ( section 4 ) . the isochrones were modified to eliminate hb and p - agb stars with @xmath16 @xmath14 9,000 k ( figure 6 ) . with these modified tracks , spectral energy distributions were created for z=0.0004 , 0.001 , 0.008 , 0.02 , and 0.05 and ages 3 , 7 , 10 , 15 , and 20 gyrs . in section 2 , we evaluated the accuracy of the kurucz models and found that the blend feature at 2538 and the slope @xmath4 were well modeled . we measured bl2538 and @xmath4 for the single age ( 3 - 20 gyr ) , single metallicity ( z=0.0004 - 0.05 ) models . mg@xmath0 was calculated using the empirical formula from @xcite : @xmath25 - 0.16\cdot [ fe / h]^{2 } \nonumber \\ + 0.981\cdot [ mg / fe]\},\end{aligned}\ ] ] where @xmath26 = 5040/@xmath16 ( plotted as a dashed line in figure 4a , b ) . because the kurucz model stellar fluxes do not model the effect of enhanced @xmath7 element abundance on the mid - uv spectrum , we assumed [ mg / fe ] = 0 . an enhanced [ mg / fe ] has a strong effect on mg@xmath0 at the higher metallicities ; a 15 gyr , [ fe / h]=0.0 , [ mg / fe]= + 0.5 population has a mg@xmath0 feature 0.15 magnitudes stronger than a similar [ mg / fe]=0.0 population . in figures 7 - 9 we plot the relations between @xmath22 , @xmath4 , and mg@xmath27 as a function of age and metallicity . the models generally occupy a narrow locus on these figures , indicating an age - metallicity degeneracy for simple stellar populations . however , young metal - rich populations do not overlap with old metal - poor populations due to the curvature in the relations shown in figures 7 and 9 . this curvature also means that populations with a spread in metallicity will not follow the simple stellar population locus . we will return to this in section 5 . the tests of the kurucz model fluxes in section 2 were valid only for solar metallicity . in this section we compare the features predicted by the simple stellar population models to the observed features in globular clusters to test the synthesis model s validity at low metallicity and old ages . @xcite published mid - uv equivalent widths ( ew ) and continuum point relative flux values normalized to the flux at 2646 for composite globular cluster iue spectra ( table 2 ) . we chose the seven composite spectra for globular clusters with ages @xmath14 10 gyr and metallicities z= 0.0002 to 0.006 to test the validity of the slopes and mid - uv bl2538 features of our simple models . we are not able to estimate the hot stellar component of the integrated globular cluster spectra because the full spectra are not given . therefore we computed our model seds with the original isochrones ( including the hot hb and post hb stars ) and compared these to the composite globular cluster features ( figure 10 ) . the presence of the hot stellar component has little effect on the low metallicity models , but significantly changes the bl2538 feature for the z @xmath14 0.02 models . @xcite measure the equivalent width between 2506 - 2574 . this is approximately 40 wider than the bl2538 index , and encompasses sii absorption lines at 2507 , 2514 , 2516 , thus translating this ew into an index may cause a slight overestimation of the strength of the bl2538 index . fitting the slope of the globular cluster spectra was relatively straight - forward . the slope of a line fit to the log of the flux at the continuum points 2646 , 2959 , and 3122 was measured . in order to convert the equivalent widths to indices , the flux of the continuum for the ew bandpass at 2556 ( @xmath28 ) was interpolated between the continuum flux at 2466 and 2646 . then the `` area '' of the absorption feature was determined by multiplying the ew by this interpolated continuum flux @xmath28 . then the depth of the absorption feature was estimated by dividing this area by the width of the ew bandpass ( 68 ) . finally the flux of the absorption feature at 2556 ( @xmath29 ) was calculated by subtracting the depth of the absorption feature from @xmath28 and the bl2538 index value was estimated by -2.5 log ( @xmath29/@xmath28 ) . table 2 gives the ew and continuum values measured by @xcite , and the calculated indices and slopes . the calculated indices and slopes have large uncertainties , since these values are only rough estimates from the data . nevertheless , plotting the globular cluster points ( triangles ) on figure 10 , we find good agreement with our simple seds , except for 47 tuc . the difference between the globular clusters ( except for 47 tuc ) and the models is @xmath2 0.05 in @xmath22 and @xmath2 0.35 in @xmath4 . when the ew of our z=0.0004 , 0.001 , and 0.008 , 15 gyr model seds were measured , they agree well with the globular cluster ew although the model z=0.008 ew is slightly less than the clusters ews . the globular clusters slopes and bl2538 strengths increase with metallicity , as predicted by our single age , single z models . note that the cluster slopes are slightly lower than the model predictions ( although within the uncertainties ) , which may be due an underestimation of the slope of the clusters because of uncertainties in the hot hb and post hb stellar component . the padova isochrones do not possess extreme horizontal branches , which may affect the mid - uv light of some of the globular clusters . note the difference between the the `` blue '' z=0.0007 cluster and the `` red '' z=0.0006 cluster in bl2538 . again , this may be the result of the blue component of stars washing out the bl2538 feature . the effects of interstellar extinction on the @xmath4 and bl2538 were found to be less than the uncertainty of the model seds and galaxy data . galactic extinction curves with a@xmath30 = 0.0 , 0.05 , 0.10 , 0.25 , 0.50 , and 0.75 @xcite were used to redden the seds of solar metallicity models of ages 1 , 3 , 7 , 10 , 15 , and 20 gyr and the seds of a 7 gyr population with z= 0.0004 , 0.001 , 0.008 , 0.02 , and 0.05 . in all cases , @xmath19 bl2538 @xmath2 0.05 magnitudes . for a@xmath30 @xmath2 0.50 , @xmath19 @xmath4 @xmath2 0.4 and was 0.6 for a@xmath30 = 0.75 . thus , for galactic reddening with a@xmath30 @xmath2 0.50 , the reddening effects on the slope and the blend feature are less than the error bars on the galaxy data points ( figure 7 ) . ngc 1399 and ngc 3610 were observed with the hst faint object spectrograph on 1993 17 january and 1992 30 november , respectively ( figure 11 ) . gratings and exposure times are shown in table 3 . all observations used a 1.0 diameter circular aperture . the observations were with the aberrated hst optics , prior to installation of costar . in this paper we focus on the g270h data . the target acquisition procedure involved measuring the flux a 4x4 grid for spaced at 0.25 `` intervals , and choosing the brightest spot in the grid . this procedure thus centered the galaxy to an accuracy of 0.25 '' in the aperture , and also resulted in the observations with the red channel and the blue channel being slightly offset from each other ( in retrospect a 5x5 grid would have been better ) . the fos data were reduced following standard pipeline procedures . this involves correcting for geomagnetic - induced image motion ; correcting for particle - induced dark current based on the geomagnetic position of the spacecraft ; correcting for scattered light via subtraction of a constant count - rate determined from diodes that are unilluminated by the selected grating ; flat - field correction ; wavelength calibration ; and flux calibration based on observations spectrophotometric standards . a full correction for grating - scattered light is in principle much more complex than the simple subtraction of a constant we have adopted here ( bushouse , rosa , & mueller 1995 ) . in particular , due to the high sensitivity of the fos detectors to long - wavelength photons , the fos spectra contain a component of light scattered from wavelengths longward of 3000 . we have used bspec @xcite to estimate the uncertainty introduced by ignoring the wavelength dependence of the scattered light correction . for ngc1399 we use a composite spectrum constructed from the the hut spectrum , the iue spectrum of ngc4649 , and an optical spectrum from the las campanas observatory ( kimble , davidsen , & sandage 1989 ) , normalized to agree in the wavelength regions where they overlap . for this model , the average scattered light contribution to the measured flux from 1700 - 3270 is less than 1% . we have obtained uv spectra for ngc 1399 , m32 and the bulge of m31 , extracted from the iue database by d. calzetti ( figure 11 ) . all the spectra have been corrected for redshift obtained from ned using the iraf v 2.11.1 task onedspec.dopcor . literature values for mg@xmath0 ( burstein et al . 1988 ; worthey , faber , & gonzalez 1992 ) may overestimate the galaxy s mean mg@xmath0 index strength because the spectra are often taken with a narrow aperture centered on the inner - most ( and probably most metal - rich ) regions of the galaxy . we measure the mg@xmath0 values from the optical spectra of @xcite which were taken with an aperture matched to the 10 @xmath31 20 iue aperture . for ngc 1399 , 20 corresponds to @xmath5 2 kpc . however , for m 31 and m 32 , 20 corresponds to @xmath5 80 pc , so for these galaxies even with the wide aperture slit we are observing only the very central regions . in the case of ngc 3610 , wide aperture optical spectra were not available and the mg@xmath0 value taken with the lick 1.4@xmath31 4 aperture @xcite was compared to our models . if we assume ngc 3610 has a metallicity gradient typical of other elliptical galaxies @xcite , no appreciable age gradient , and an age @xmath5 10 gyr , at a 10 radius , ngc 3610 should have a slope 0.5 magnitudes/1000 less steep , bl2538 0.05 magnitudes weaker , and an mg@xmath0 0.05 magnitudes weaker than measured in a 1 aperture . our goal is to examine the main sequence and cooler stellar sub - population of these galaxies , and not the hot horizontal and post hb populations . we must therefore determine which galaxies possess significant uv upturns and attempt to subtract away this hot component of the spectra . we estimated the strength of the uv upturn by measuring the @xmath32 color of the galaxies with the iraf task onedspec.sband . the 1550 and 2500 bands were centered at 1550 and 2500 respectively and were both 600 wide . our calculations of @xmath32 agreed with previously published values for the galaxies . m32 had the reddest @xmath32 color ( @xmath5 1.5 ) . the rest of the galaxies have significant uv upturns indicative of a hot stellar population . @xcite found that the far - uv flux of m31 shows a drop - off blue - ward at 1000 , which requires that the @xmath16 be @xmath5 25,000 k for the uv emitting stars . thus , for the `` normal '' galaxies with no recent star formation , the uv upturn can be roughly approximated by a @xmath16 @xmath5 20000 - 30000 k star . between 1350 and 3200 , the spectral shape changes very little with temperature for stars hotter than 20,000 k. for the galaxies with 1550 - 2500 colors @xmath2 0.5 magnitudes , we fit a 20,000k log g=5 kurucz model to the galaxies flux between 1350 and 1650 and subtracted the stellar spectrum from the galaxies spectra ( figure 12 ) . to test whether this method did a good job of eliminating the light from the hot component , models of a simple single metallicity population where created using the original isochrones with hot hb and p - agb stars . then the 20,000 k stellar model was fit and subtracted from this model sed . the resulting spectra had an identical slope and bl2538 strength as the model created using the modified isochrones ( section 3.1 ) . thus if a galaxy s uv upturn is due to a @xmath5 20,000 k hot stellar component , the subtracted galaxy spectra should have @xmath4 and bl2538 which reflect the metallicity and ages of the cooler stars . if a significant population of `` blue stragglers '' with @xmath33 exists , the resulting spectra will still have contamination from this non - msto population . however , like the hot hb and post hb population , the blue straggler population in galaxies is not well constrained . spinrad et al . ( 1997 ) estimate that if elliptical galaxies have blue straggler populations similar to that of old metal - poor clusters , blue stragglers may contribute as much as 20 - 50 % of flux at 2600 . on the other hand , for the higher metallicity cluster 47 tuc , the blue straggler contribution to the mid - uv flux has been estimated to be 7% @xcite . figures 7 - 9 show the measured spectral indices for the four galaxies in our sample . while the sample is small and the error bars are large , there are at least hints of some general trends . in particular , as mg@xmath27 increases , the @xmath22 feature tends to get weaker and the slope @xmath4 tends to get shallower . this is clearly evident in the comparison of ngc 1399 ( mg@xmath27 = 0.312 ) and ngc 3610 ( mg@xmath27 = 0.271 ) in figure 11 . it is also clear in figure 7 - 9 that the galaxies do not follow the behavior expected for simple stellar populations . the galaxies are generally displaced from the model locus and the trends are nearly orthogonal to those expected for simple variations of age and metallicity . our understanding of chemical evolution dictates that galaxies , unlike star clusters , have complex chemical make - ups , and so the comparison of single metallicity seds to real galaxies is limited in its usefulness . color and mg@xmath0 line strength gradients with galactic radius @xcite and the width of the rgb in m32 @xcite are evidence of stellar populations of varying metallicities within ellipticals . however , the details of the metallicity distribution in ellipticals are not well understood . comparing the mg@xmath0 and fe line strengths for synthetic populations and observed galaxies , @xcite finds high average metallicities , small metallicity dispersions and old ages for ellipticals . @xcite find that for m31 and m32 and @xcite find that for ngc1404 and ngc4649 only @xmath2 10% by mass of stars are metal poor ( z @xmath2 0.008 ) , which suggests that the ` g dwarf problem ' is not limited to the disk of the galaxy . because simple , single abundance seds did not match the mid - uv features of the majority of the galaxies in our sample , we looked to see how a spread in metallicities affects the light of a single stellar population . we consider a very simple bi - modal metallicity scenario for old stellar populations . two different bi - modal metallicity models are created by combining the spectra of the metal - poor z=0.0004 12 gyr population with the metal - rich spectra z=0.05 or z=0.02 12 gyr populations . the models are constructed so that 0.5% , 1% , 5% , 10% , and 15% of the total galaxy mass originates from the metal - poor z=0.0004 sub - population . these bi - modal metallicity models are then measured for bl2538 , mg@xmath0 , and @xmath4 . figure 13 shows the combined 10% z=0.0004 mass and 90% z=0.05 mass 12 gyr model ( solid line ) and each of its sub - population seds ( short dashed line for z=0.0004 , long dashed line for z=0.05 ) . while the metal - rich component dominates the visible light , the metal - poor component dominates the mid - uv . ( table 5 ) in figures 14a , b , c we show the single metallicity models for a given age and the results of the bi - modal metallicity populations . the z=0.05 / z=0.0004 models are represented by the pentagons where the model with the highest slope and smallest pentagon has 0.5% of its mass from the metal - poor sub - population and the model with the lowest slope and largest pentagon has 15 % of its mass from the metal - poor sub - population . the z=0.02 / z=0.0004 models are represented by stars , again with the smallest star representing the 0.5% metal - poor mass model and the largest star representing the 15% metal - poor mass model . comparing the bi - modal models with the galaxies , we see that the positions of most of the galaxies in the @xmath4 v. bl2538 plane may be explained by a significant contribution of a metal - poor stellar population to the flux at 2600 ( figure 14a ) . on the other hand , contamination by `` blue stragglers '' or problems with the hot component subtraction may also result in weaker mid - uv features . but it is important to note that a small percentage of metal - poor stars can contribute most of the light at 2600 . while our results do not predict the number of low metallicity stars needed to `` solve '' the g - dwarf problem , it is not unreasonable to expect a few low metallicity stars in ellipticals however , except for ngc 1399 , for which the bi - modal metallicity tracks fit particularly well in all three planes , the mg@xmath0 strengths are slightly too strong to agree with most of the galaxies . the combination of bl2538 , @xmath4 , and mg@xmath0 provides a unique tool for investigating elliptical galaxy evolution . the uv features are most responsive to changes in the effective temperature of the turn - off population , while mg@xmath27 to first order measures the effective temperature of the giant branch . the positions of the galaxies in the line - strength vs. line - strength diagrams suggest : ( a ) these galaxies are not well represented by simple ( single - age , single - metallicity ) populations , and ( b ) that the star - formation and chemical enrichment histories were unique and complex . the addition of a small metal - poor component to a dominant old , metal - rich component helps to reconcile the models to the data . part of the difficulty in understanding the observations probably stems from the heterogeneous nature of the sample of galaxies in our sample , which to a certain extent represent the extremes of phenomena seen in elliptical galaxies , rather than the norm . it is thus worth considering each galaxy in turn , and examining the uv results in the context of other information about the galaxy . _ m32 : _ m 32 , a compact e1 satellite of m 31 , is the closest example of an elliptical galaxy . m 32 is unusual in that it has a very weak mg@xmath0 feature and no uv upturn . the stellar population in this galaxy has been the subject of intense study , both via integrated light and via the resolved stellar population . based on its blue ( b - v ) color , and on spectral synthesis , @xcite has suggested that it contains a @xmath5 5 gyr subpopulation . the population synthesis study of @xcite found that m 32 spectrum is dominated by a 13 - 15 gyr , z @xmath2 @xmath15 population , and has evidence for an additional 5 gyr stellar population . the mid - uv region of m 32 s spectrum is not well fit by the closed box enrichment model and is more closely matched when the contribution from z @xmath2 0.008 stars is removed . in @xcite and later in @xcite , the infall of gas is suggested as a mechanism for increasing the galaxy s metallicity quickly and preventing the formation of a significant number of metal - poor stars . hst photometry a few arcminutes from the m 32 s nucleus @xcite seems to have confirmed the dominance of old solar metallicity stars and the deficiency of metal - poor stars . the observed width of the red giant branch requires that at least half of the stars in m 32 be older than 8 gyr . grillmair et al.s study also requires that the stars span a wide metallicity range but are strongly skewed to solar abundances . based on the comparison of iue mid - uv observations of m 32 s nucleus with the globular cluster 47 tuc , @xcite found the mid - uv spectrum of m 32 was best fit by a combination of a solar metallicity msto , a solar giant branch , and a hot a star component consisting of metal - poor stars and possibly some p - agb stars . the mid - uv spectral breaks 2609/2660 and 2828/2923 and absorption features mg2800 and mg2852 are weaker in m 32 than in 47 tuc , but the c26 - 30 color ( similar to our @xmath4 ) is redder in m 32 . while rose & deng ( 1999 ) were able to reproduce the spectral breaks and absorption lines with a solar msto population with a @xmath5 25 % metal - poor contribution of the mid - uv light , the simulated c26 - 30 color was @xmath5 0.1 magnitude bluer than m 32 . since this color is dominated by the msto whereas the other features are influenced more by the hotter population , a slightly higher metallicity msto may reconcile the simulation with m 32 . from the simple single age , single metallicity seds ( figures 7 - 9 ) , we see the mid - uv and mg@xmath0 are not consistent with a single metallicity . the slope of m32 points to a dominant population with a metallicity z @xmath14 0.02 , the mid - uv bl2538 value implies a lower metallicity population of z @xmath5 0.001 - 0.008 , and mg@xmath0 gives a z @xmath5 0.008 - 0.02 population . invoking a bi - modal abundance model improves the situation slightly . the 12 gyr , 0.5 % z=0.0004/ 99.5 % z=0.05 sed matches the uv features , but gives a mg@xmath0 value stronger that what is observed ( figure 14a - c ) . we have also modeled the mid - uv features for composite z=0.05 populations with both a 12 gyr and 3 gyr population ( figure 15 ) . while the @xmath4 of m 32 is matched by 5% 3 gyr/ 95 % 12 gyr z=0.05 composite population , the @xmath22 of the composite model is too strong . solar metallicity composite 3 gyr/12 gyr populations would also have @xmath22 features stronger than m 32 and shallower slopes . figure 16 shows a direct comparison between the m 32 spectra , the best fitting metal - poor component model ( 0.5% z=0.0004 / 99.5% z=0.05 12 gyr ) , and the best fitting young component model ( 5% 3 gyr/ 95% 12 gyr z=0.05 ) . the slopes are identical for m 32 and the two models , but the @xmath22 of the multi - metallicity model agrees more closely with m 32 . a more complex model with a distribution of metallicities and ages may reproduce both mg@xmath0 and the mid - uv . _ invoking an intermediate 3 - 5 gyr metal - rich sub - population is not sufficient to explain m 32 s uv features ; the presence of a small number of metal - poor stars is implied . _ the number of low abundance stars predicted is very small ( @xmath5 0.5% ) . this agrees with @xcite and @xcite result of a deficiency of m 32 metal - poor stars as compared to a closed box chemical evolution model . however , the existence of a 2 - 5 gyr metal - rich population is not ruled out . the steep slope of m 32 s spectrum seems to require a z @xmath1 @xmath15 msto population ( figures 14a , 15 ) . it is important to note that our mid - uv m 32 spectrum is of only the inner - most 80 pc , hence we may be observing the most metal - rich region of the galaxy , whereas the hst color magnitude diagram @xcite which gives a peak abundance @xmath5 @xmath15 ) is of a region @xmath5 500 pc from the nucleus . _ _ bulge of m31:__m 31 has one of the reddest and strongest absorption - featured bulges known . its mg@xmath0 index in the innermost regions is comparable to that found for ellipticals and @xmath7 element enhancement has been suggested @xcite . however , its h@xmath9 strength @xcite and the reported presence of luminous agb stars suggest an age younger than our galactic halo of @xmath5 8 - 10 gyr @xcite . absorption line gradients suggest that the mean stellar population becomes older and more metal poor with increasing radius @xcite . like m 32 , the spectral features can not be fit by a single metallicity model and are more consistent with an additional low abundance population . in figures 7 - 9 , we find that both mg@xmath0 and @xmath4 are consistent with high metallicity ( z @xmath14 0.02 ) dominant population , while bl2538 gives a lower metallicity ( z @xmath5 0.001 ) . the 12 gyr , 0.5 % z=0.0004 / 99.5 % z=0.05 model matches m 31 in the mid - uv , but produces a slightly stronger mg@xmath0 than observed ( figure 14 a - c ) . the mid - uv features of the bulge of m 31 suggest a small number of metal - poor stars ( @xmath2 1% by mass ) , while its mg@xmath0 may point to the contribution of a metal - rich , intermediate age population . _ ngc 3610 : _ this galaxy has many properties of a 3 - 7 gyr merger remnant : ngc 3610 has the richest fine structure measured for an elliptical ; it has an unusually blue color ; and it possesses a weak mg@xmath0 feature and an enhanced h@xmath9 absorption @xcite . there is not a significant amount of dust or cold gas in the galaxy , nor is there any evidence for ongoing star formation . hst observations have detected a number of red globular clusters in ngc 3610 , consistent with an age of 4 gyr @xcite . comparison with the single age , single metallicity seds shows that ngc 3610 s slope and mg@xmath0 are consistent with high metallicity ( z @xmath14 0.02 ) dominant population , while the bl2538 feature implies a lower metallicity ( z @xmath5 0.001 ) . the 12 gyr , 1% z=0.0004 / 99% z=0.05 model is consistent with the mid - uv features , but predicts a slightly stronger mg@xmath0 value than observed in ngc 3610 . here again , we detect the signature of a small number of metal - poor stars ( @xmath2 1% by mass ) in the mid - uv light , and while the mg@xmath0 feature may be influenced by a population of metal - rich 3 -7 gyr stars . _ ngc 1399 : _ this ge in has one of the strongest mg@xmath0 values and uv upturns known . like other ges , ngc 1399 is thought to be old and very metal - rich . ngc 1399 possesses @xmath5 5800 globular clusters ; recent keck spectroscopy @xcite of @xmath5 20 of these clusters give a mean [ fe / h ] @xmath5 -0.8 dex and found no very metal - poor clusters ( although the sample may have been biased against blue clusters ) . surprisingly , both mid - uv features point to a low metallicity population when compared to the simple models , while the mg@xmath0 implies a high metallicity dominant population . the 12 gyr model with 5 - 10% z=0.0004 / z=0.05 agrees with the mid - uv features and mg@xmath0 of the iue data . the fos data , which observes only the center - most region of ngc 1399 , imply an even higher metallicity and/or a lower metallicity population . using the @xcite calibration ( equation 1 ) , a z @xmath34 , [ mg / fe ] = 0.0 , 10 - 15 gyr model is sufficient to explain the mean ngc 1399 mg@xmath0 value obtained by the wide 10 @xmath31 20 aperture ( section 4.2 , table 4 ) . an @xmath7 enhanced ( [ mg / fe ] @xmath1 0.0 ) model is not required to explain ngc 1399 s strong mg@xmath0 , but not ruled out . _ ngc 1399 appears to have the greatest percentage of low abundance stars ( @xmath5 10 % ) , indicating a broader metallicity distribution than less massive ellipticals . _ the strength of mg@xmath0 may rule out the existence of an intermediate age population . we have attempted to set additional constraints on the evolution of elliptical galaxies and spiral bulges by analyzing their mid - uv spectra . through comparisons iue observations of individual stars to the kurucz - model fluxes , two spectral indices ( bl2538 and @xmath4 ) are identified that appear to be promising diagnostics of the main - sequence turn - off population ( given the current generation of model stellar fluxes ) . single metallicity , single age model populations are constructed and found to agree well with the mid - uv features of globular clusters . however most of the of the galaxies do not fall on the single - age , single - metallicity tracks . contamination by blue stragglers may be partially responsible . another explanation is that the flux from metal - poor stars , invisible in the optical region of the spectrum , begins to dominate the flux in the mid - uv . in order to determine the effect of a small percentage ( by mass ) of metal - poor stars on the sed of an otherwise metal - rich population , simple single age , bi - modal metallicity models were created . for all the galaxies , a small ( @xmath2 10 % by mass ) number of low metallicity stars are implied by the mid - uv features . metal - poor stars are of course expected from chemical - evolution arguments , but evidence for a metallicity spread has been hard to find in the optical spectra . the metallicity distribution function can no longer be ignored if we hope match simultaneously the mid - uv and the optical spectra . it is difficult to draw strong conclusions about the recent star - formation history of the galaxies in our sample from the mid - uv . ngc 1399 , one of the strongest mg@xmath0 and uv upturn galaxies known , is consistent with a multi metallicity 12 gyr model in both the mid - uv and mg@xmath0 strength and may possess the greatest percentage of low metallicity stars observed . the strength of mg@xmath0 suggests that ngc 1399 has few , if any , stars younger than 10 gyr . however , for the remaining galaxies , especially m 32 , a more sophisticated model invoking both old metal - poor and intermediate age metal - rich stars may be needed to reconcile the mg@xmath0 and mid - uv features . on the other hand , variations the blue straggler populations and @xmath7 element abundance could also account for the disparities between the optical and mid - uv spectra . because they probe the main - sequence turn - off population , mid - uv spectral diagnostics can in principle provide a a powerful diagnostic of the star - formation history and chemical evolution of elliptical galaxies . our study presents a rather discouraging , but nevertheless realistic , picture of the current status of the models and the data in this wavelength regime . the mid - uv features we have identified suggest that significant progress can be made with additional high - quality spectra of elliptical galaxies , even if the model stellar fluxes do not improve . however , the promise of this wavelength regime for untangling long - standing issues of galaxy evolution should also motivate attempts to improve the model stellar fluxes , which will ultimately put such analyses on firmer ground , and allow the use of additional spectral features . we wish to thank the referee for his many constructive comments and suggestions . we also thank daniela calzetti and anne kinney for allowing us to use their extracted iue and optical spectra , and michael gregg for useful discussions about m 32 . this paper is based in part on observations with the nasa / esa hubble space telescope obtained at the space telescope science institute . support for this work was provided by nasa through grant number go-03647.01 - 91a from the space telescope science institute , which is operated by the association of universities for research in astronomy , incorporated , under nasa contract nas5 - 26555 . babul , a. & ferguson , h. c. 1996 , , 458 , 100 bertelli , g. , bressan , a. , chiosi , c. , fagotto , f. , & nasi , e. 1994 , a&as , 106 , 275 bonatto , c. , bica , a. , & alloin , d. 1995 , a&as , 112 , 71 borges , a. c. , idiart , t. p. , de freitas pacheco , j. a. , & thevenin , f. 1995 , , 110 , 2408 bressan , a. , chiosi , c. , & fagotto , f. 1994 , , 94 , 63 bressan , a. , chiosi , c. , & tantalo , r. 1996 , , 311 , 425 brown , t. m. , ferguson , h. c. , davidsen , a. f. , & dorman , b. 1997 , , 482 , 685 burstein , d. , bertola , f. , buson , l. m. , faber , s. m. , & lauer , t. r. 1988 , , 328 , 440 bushouse , h. , rosa , m. , & mueller , t. 1995 , in astronomical data analysis software and systems iv , ed . r. a. shaw , h.e . payne , & j. j. e. hayes ( asp conference series ) , 345 carollo , c. m. , danziger , j. j. , & buson , l. 1993 , , 265 , 553 charlot , s. , & bruzual , g. a. 1991 , , 367,126 charlot , s. , worthey , g. , & bressan , a. 1996 , , 457 , 625 chavez , m. , malagnini , m. l. , & morossi , c. 1995 , , 440 , 210 davidge , t. j. 1997 , , 113 , 985 davies , r. l. , sadler , e. m. , & peletier , r. f. 1993 , , 262 , 650 dorman , b. , oconnell , r. w. , & rood , r. t. 1995 , , 442 , 105 dorman , b. , oconnell , r. w. , & rood , r. t. 1997 , , 191 , 4903 faber , s. m. , friel , e. d. , burstein , d. , & gaskell , c. m. 1985 , , 57 , 711 fanelli , m. n. , oconnell , r. w. , burstein , d. , & wu , c. c. 1990 , , 364 , 272 fanelli , m. n. , oconnell , r. w. , burstein , d. , & wu , c. c. 1992 , , 82 , 197 ferguson , h. c. , & davidsen , a. f. 1993 , , 408 , 92 gonzalez , j. j. 1993 , phd . thesis , uc santa cruz greggio , l. 1997 , , 285 , 151 grillmair , c. j. et al . 1996 , , 112 , 1975 gunn , j. e. , & stryker , l. l. 1983 , , 52 , 121 heap , s. r. , et al . 1998 , , 492 , 131 kimble , r. a. , davidsen , a. f. , & sandage , a. 1989 , , 157 , 237 kinney , a. , calzetti , d. , bohlin , r. c. , mcquade , k. , storchi - bergmann , t. , & schmitt , h. r. 1996 , , 467 , 38 kissler - patig , m. , brodie , j. p. , schroder , l. l. , forbes , d. a. , grillmair , c. j. , & huchra , j. p. 1998 , , 115 , 105 lejeune , t. , cuisnier , f. , & buser , r. 1997 , a&as , 125 , 229 morossi , c. , franchini , m. , malagnini , m. l.,kurucz , r. l. , & buser , r. 1993 , , 277 , 173 oconnell , r. w. 1980 , , 236 , 430 park , j. h. , & lee , y. w. 1997 , , 476 , 28 pei , y. c. 1992 , , 395 , 130 rich , r. m. & mighell , k. j. 1995 , , 439 , 145 rosa , m. 1994 , fos isr 127 rose , j. a. 1994 , , 107 , 206 rose , j. a. , & deng , s. 1999 , , 117 , 2213 schweizer , f. & seitzer , p. 1992 , , 104 , 1039 schweizer , f. , seitzer , p. , faber , s. m. , burstein , d. , ore , c. m. d. , & gonzalez , j. j. 1990 , , 364 , 33 spinrad , h. , dey , a. , stern , d. , dunlop , j. , peacock , j. , jimenez , r. , & windhorst , r. 1997 , , 484 , 581 tantalo , r. , chiosi , c. , bressan , a. , & fagotto , f. 1996 , , 311 , 361 whitmore , b. c. , miller , b. w. , schweizer , f. , & fall , s. m. 1997 , , 114 , 1797 worthey , g. 1994 , , 95 , 107 worthey , g. , dorman , b. , & jones , l. 1996 , , 112 , 948 worthey , g. , faber , s. m. , & gonzalez , j. j. 1992 , , 398 , 69 ccc fe2402 & 2382 - 2422 & 2285 - 2325 + & & 2432 - 2458 + bl2538 & 2520 - 2556 & 2432 - 2458 + & & 2562 - 2588 + fe2609 & 2596 - 2622 & 2562 - 2588 + & & 2647 - 2673 + mg2800 & 2784 - 2814&2762 - 2782 + & & 2818 - 2838 + mg2852 & 2839 - 2865 & 2818 - 2838 + & & 2906 - 2936 + mgwide & 2670 - 2870 & 2470 - 2670 + & & 2930 - 3130 + fe3000 & 2965 - 3025 & 2906 - 2936 + & & 3031 - 3051 + h@xmath9 & 4848 - 4877&4828 - 4848 + & & 4877 - 4892 + mg@xmath0 & 5154 - 5197 & 4895 - 4958 + & & 5301 - 5366 + cccccccc g2b & 0.006 & 28.6 & 0.53 & 1.55 & 2.07 & 0.593 & -1.65 + g47tuc & 0.004 & 34.2 & 0.46 & 1.46 & 2.06 & 0.759 & -1.60 + g3b & 0.0011 & 20.7 & 0.63 & 1.27 & 1.33 & 0.396 & -0.68 + g3r & 0.0010 & 23.4 & 0.66 & 1.24 & 1.42 & 0.496 & -0.80 + g4b & 0.0007 & 12.7 & 0.80 & 1.12 & 1.24 & 0.224 & -0.48 + g4r & 0.0006 & 21.6 & 0.68 & 1.23 & 1.30 & 0.416 & -0.63 + g5 & 0.0002 & 9.1 & 0.85 & 1.18 & 1.27 & 0.156 & -0.55 + cccccc m 32 & 0.45 & - & -2.92 & - & 0.194 + m 31 & 0.34 & 0.49 & -2.29 & -2.73 & 0.286 + ngc 1399 & 0.07 & 0.30 & -1.02 & -1.65 & 0.312 + ngc 3610 & 0.30 & 0.38 & -2.59 & -2.73 & 0.271 + ngc 1399 & 0.08 & 0.23 & -0.80 & -2.04 & 0.357 +

various recent studies proved that cosmological models with a significant contribution from cold dark matter isocurvature perturbations are still compatible with most recent data on cosmic microwave background anisotropies and on the shape of the galaxy power spectrum , provided that one allows for a very blue spectrum of primordial entropy fluctuations ( ) . however , such models predict an excess of matter fluctuations on small scales , typically below . we show that the proper inclusion of high - resolution high signal - to - noise lyman- forest data excludes most of these models . the upper bound on the isocurvature fraction , defined at the pivot scale mpc , is pushed down to , while ( 95% confidence limits ) . we also study the bounds on curvaton models characterized by maximal correlation between curvature and isocurvature modes , and a unique spectral tilt for both . we find that ( 95% c.l . ) in that case . for double inflation models with two massive inflatons coupled only gravitationally , the mass ratio should obey ( 95% c.l . ) .

astro-ph0509209

with the most recent measurements of the cosmic microwave background ( cmb ) anisotropies and large scale structures ( lss ) of the universe as well as various other astronomical observations , it is now possible to have a clear and consistent picture of the history and content of the universe since nucleosynthesis . in particular , it is well established that the cosmological perturbations which gave rise to the cmb anisotropies and the lss of the universe were inflationary , with a close to scale - invariant harrison - zeldovich spectrum . moreover , the cmb and lss data allow to test the paradigm of adiabaticity of the cosmological perturbations and hence the precise nature of the mechanism which has generated them . the simplest realizations of the inflationary paradigm predict an approximately scale invariant spectrum of adiabatic ( ad ) and gaussian curvature fluctuations , whose amplitude remains constant outside the horizon , and therefore allows cosmologists to probe directly the physics of inflation from current cmb and lss observations . however , this is not the only possibility . models of inflation with more than one field generically predict that , together with the adiabatic component , there should also be entropy , or isocurvature perturbations @xcite , associated with fluctuations in number density between different components of the plasma before photon decoupling , with a possible statistical correlation between the adiabatic and isocurvature modes @xcite . baryon isocurvature ( bi ) perturbations and cold dark matter isocurvature ( cdi ) perturbations were proposed long ago @xcite as an alternative to adiabatic perturbations . these bi and cdi modes are qualitatively similar , since they are related by a simple rescaling factor @xmath10 , or @xmath11 for the cross - correlation : thus , by studying the case of mixed ad + cdi modes , one implicitly includes the case of ad + bi , for which the allowed isocurvature fraction is larger roughly by the above factor evaluated near the maximum likelihood model . a few years ago , two other modes , neutrino isocurvature density ( nid ) and velocity ( niv ) perturbations , have been added to the list @xcite . moreover , isocurvature perturbations have been advocated in order to explain the high redshift of reionization claimed by the wmap team @xcite . note , however , that in the case all fields thermalize at reheating , no isocurvature mode will survive @xcite . the simplest assumption for generating observable cdi perturbations is that one of the inflaton fields remains uncoupled from the rest of the plasma between inflation and its decay into cdm particles . since baryons and neutrinos are usually assumed to be in thermal equilibrium in the early universe , it is more difficult to build realistic models for the generation of bi , nid and niv modes than for cdi - but some possibilities still exist , based on non - zero conserved quantities and chemical potentials ( see e.g. @xcite ) . moreover , it is well known that entropy perturbations seed curvature perturbations outside the horizon @xcite , so that it is possible that a significant component of the observed adiabatic mode could be maximally correlated with an isocurvature mode . such models are generically called _ curvaton models _ @xcite , and are now widely studied as an alternative to the standard paradigm . furthermore , isocurvature modes typically induce non - gaussian signatures in the spectrum of primordial perturbations @xcite . in the last few years , various models with a correlated mixture of adiabatic and isocurvature perturbations have been tested by several authors , with different combinations of data sets and theoretical priors . a crucial difference between these analyses lies in the assumptions concerning the scale - dependence of the various modes . some groups assumed for simplicity that the adiabatic and isocurvature mode shared exactly the same scale - dependence @xcite , but enriched the analysis by considering the full mixtures of several modes at a time ( ad , cdi , nid , niv ) . other groups concentrated on the ( correlated ) mixture of two modes only ( ad+cdi in @xcite , ad+nid and ad+niv in @xcite ) , with a different power law for the three components ( adiabatic , isocurvature and cross - correlation ) , as expected in the general case . finally , an intermediate approach consists in studying the mixture of two modes with a scale - independent mixing angle , i.e. , only two tilts @xcite . in addition to these references , some groups studied the case of the curvaton scenario , which requires some specific analyses @xcite since it involves a maximal correlation / anti - correlation and a unique spectral index for the adiabatic and isocurvature modes . furthermore , two groups have quantified the need for isocurvature modes through a bayesian evidence computation on the basis of current cmb and galaxy power spectrum data , reaching somewhat different conclusions due to a different choice of priors @xcite . in this work , we are particularly interested in mixed models with ad+cdi modes and three different tilts , for which it was shown in refs . @xcite and @xcite that a significant fraction of isocurvature perturbations is still allowed . this sounds surprising at first sight , since the isocurvature mode is known for suppressing small - scale cmb anisotropies . this is true indeed for a scale - invariant spectrum of primordial isocurvature fluctuations , but not in general : a significant isocurvature contribution with a very blue tilt ( @xmath12 ) can contribute to cmb anistropies even on small scales , and can be compatible to some extent with the cmb temperature and temperature - polarization spectra , in spite of the small shift induced in the scale of the acoustic peaks . these models predict generically an excess of matter fluctuations on small scales . using the shape and amplitude of the _ linear _ power spectrum derived from galaxy surveys at wavenumbers @xmath13mpc , one can exclude such an excess for wavelengths @xmath14 larger than @xmath15 mpc . the main goal of this work is to push the constraints down by making use of lyman-@xmath2 forest data , which probe large - scale structure at redshift @xmath16 and on scales @xmath17 , in the mildly non - linear regime . therefore , in any comparison between lyman-@xmath2 observations and linear theoretical predictions , it is necessary to take into account the non - linear evolution with n - body or hydrodynamical simulations . usually , these simulations are carried under the assumption of adiabaticity . however , it is not difficult to generalize them to the case of mixed adiabatic plus isocurvature models . during matter domination , the perturbations seeded by each of the two modes are indistinguishable : the only difference lies in their scale - dependence , but not in their nature or time evolution . so , a given mixed model is entirely specified by a single matter transfer function , defined for instance soon after the time of equality . therefore , the lyman-@xmath2 forest data can be safely applied to non - adiabatic models provided that one takes into account the fact that the matter transfer function has more freedom than in the purely adiabatic case . in the following analysis , we will carefully take this point into account . we will use here the linear matter power spectrum inferred from two large samples of quasar ( qso ) absorption spectra @xcite using state of the art hydrodynamical simulations @xcite combined with cosmic microwave background data from the wmap satellite @xcite ; as well as from the small - scale temperature anisotropy probed by vsa @xcite , cbi @xcite and acbar @xcite ; from the matter power spectrum measured by the 2-degree - field galaxy redshift survey ( 2dfgrs ) @xcite and the sloan digital sky survey ( sdss ) @xcite ; and finally from the recent type ia supernova ( sn ) compilation of ref . we note that the cosmological parameters recovered from the data sets used in this paper are in good agreement with subsequent studies made by the sdss collaboration using a different data set and a very different theoretical modelling ( @xcite ) . this demonstrates that the analysis of the lyman-@xmath2 forest qso spectra is robust and that many systematic uncertainties involved in the measurement are now better understood than a few years ago . the plan of the paper is as follows . in section ii we describe the notations we used for the isocurvature sector . in section iii we introduce the lyman-@xmath2 data that we are employing . in section iv we discuss the general bounds on our full ad+cdi parameter space from lyman-@xmath2 , cmb , lss and sn data using a bayesian likelihood analysis . we also check explicitly with a hydrodynamical simulation the robustness of our lyman-@xmath2 data - fitting procedure , and we address the subtle issue of the role of parametrizations and priors on the isocurvature bounds and in the interpretations of our results . we also discuss the specific curvaton models with maximal anticorrelation and equal tilts for both adiabatic and isocurvature modes , as well as bounds on double inflation models . in section v we draw our conclusions . for the theoretical analysis , we will use the notation and some of the approximations of ref . @xcite . during inflation , more than one scalar field could evolve sufficiently slowly that their quantum fluctuations perturbed the metric on scales larger than the hubble scale during inflation . these perturbations will later give rise to one adiabatic mode and several isocurvature modes . we will restrict ourselves here to the situation where there are only two fields , @xmath18 and @xmath19 , and thus only one isocurvature and one adiabatic mode . introducing more fields would complicate the inflationary model and even then , it would be rather unlikely that more than one isocurvature mode contributes to the observed cosmological perturbations . therefore , the two - point correlation function or power spectra of both adiabatic and isocurvature perturbations , as well as their cross - correlation , can be parametrized with three power laws , i.e. three amplitudes and three spectral indices , @xmath20 \delta_{{\cal s}}^2(k)\!&\equiv&\ ! { k^3\over2\pi^2}\langle{{\cal s}}_{\rm rad}^2\rangle = { k_0 ^ 3\over2\pi^2 } b^2\left({k\over k_0}\right)^{{n_{\rm iso}}-1}\,,\\[2 mm ] \delta_{{{\cal r}}{{\cal s}}}^2(k)\!&\equiv&\!{k^3\over2\pi^2 } \nonumber \langle{{\cal r}}_{\rm rad}{{\cal s}}_{\rm rad}\rangle \ , \nonumber\\[2 mm ] & = & { k_0 ^ 3\over2\pi^2 } a\,b\,\cos\delta_{k_0 } \left({k\over k_0}\right)^{{n_{\rm cor}}+{{1\over2}}({n_{\rm ad}}+{n_{\rm iso}})-1}\,.\nonumber\end{aligned}\ ] ] here , @xmath21 stands for the curvature perturbation , and @xmath22 for the cdi perturbation . both are evaluated during radiation domination and on super - hubble scales . we also introduced an arbitrary pivot scale @xmath23 , at which the amplitude parameters are defined through @xmath24 and @xmath25 . in addition to the fact that curvature and entropy perturbations are generally correlated at the end of inflation , some extra correlation can be generated later by the partial conversion of isocurvature into adiabatic perturbations . the correlation angle @xmath26 is in general a function of @xmath27 , and in the above definitions , we approximated @xmath28 by a power law with amplitude @xmath29 and tilt @xmath30 . so , we assumed implicitly that the inequality @xmath31 holds over all relevant scales . we will enforce this condition in the following analysis . the angular power spectrum of temperature and polarization anisotropies seen in the cmb today can be obtained from the radiation transfer functions for adiabatic and isocurvature perturbations , @xmath32 and @xmath33 , computed from the initial conditions @xmath34 and @xmath35 , respectively , and convolved with the initial power spectra , @xmath36 ^ 2\ , \left({k\over k_0}\right)^{{n_{\rm ad}}-1}\,,\\[2 mm ] c_l^{\rm iso}\!&\equiv&\,\int { dk\over k}\ , \left[\theta_l^{\rm iso}(k)\right]^2\ , \left({k\over k_0}\right)^{{n_{\rm iso}}-1}\,,\nonumber\\[2 mm ] c_l^{\rm cor}\!&\equiv&\,\int { dk\over k}\ , \theta_l^{\rm ad}(k)\,\theta_l^{\rm iso}(k)\ , \left({k\over k_0}\right)^{{n_{\rm cor}}+{{1\over2}}({n_{\rm ad}}+{n_{\rm iso}})-1}\,,\nonumber\end{aligned}\ ] ] then , the total angular power spectrum reads @xmath37 in many works ( see for instance @xcite ) , the following parametrization is employed : @xmath38\,,\ ] ] where @xmath39 represents the entropy to curvature perturbation ratio during the radiation era at @xmath40 . we will use here a slightly different notation , used before by other groups @xcite : @xmath41\,,\ ] ] where @xmath42 represents the isocurvature fraction at @xmath23 , and runs from purely adiabatic ( @xmath43 ) to purely isocurvature ( @xmath44 ) , while @xmath45 defines the correlation coefficient at @xmath23 , with @xmath46 corresponding to maximally correlated(anticorrelated ) modes . there is an obvious relation between both parametrizations : @xmath47 this notation has the advantage that the full parameter space of @xmath48 is contained within an ellipse . the north and south rims correspond to fully correlated ( @xmath49 ) and fully anticorrelated ( @xmath50 ) perturbations , with the equator corresponding to uncorrelated perturbations ( @xmath51 ) . the east and west correspond to purely isocurvature and purely adiabatic perturbations , respectively . any other point within the ellipse is an arbitrary admixture of adiabatic and isocurvature modes . we should emphasize that the three amplitude parameters @xmath52 , @xmath2 and @xmath45 are defined at @xmath53 , and that comparing bounds from various papers is straightforward only when the pivot scale is the same . for instance , in the simple case where @xmath54 , @xmath2 is independent of @xmath23 , but this is not the case for @xmath45 : if @xmath55 , points within the @xmath48 ellipse are shifted vertically toward the edges of the ellipse when one increases @xmath23 and shifted toward the horizontal @xmath56 line when one decreases @xmath23 . when @xmath57 , both @xmath2 and @xmath45 depend on the pivot scale . in addition , by changing the prior on the amplitudes , a shift in the pivot scale affects the @xmath58 likelihood quite dramatically @xcite . throughout this paper , we will use @xmath59 mpc@xmath5 , which is the most frequent choice in the literature . this value corresponds roughly to the multipole number @xmath60 . therefore , the ratio @xmath61 is roughly independent of the tilt values . for cosmological parameters close to the best - fit @xmath62cdm model , one finds @xmath63 . the smallness of this number comes from the fact that @xmath33 is strongly suppressed with respect to @xmath64 for large wavenumbers . indeed , the metric perturbations induced by isocurvature perturbations remain small during radiation domination : so , for small scales entering early inside the hubble radius , the amplitude of the photon acoustic oscillations is also small ( as can be seen via its transfer function ) . as a consequence , even if during radiation domination one has @xmath65 ( corresponding to @xmath66 or @xmath67 ) the isocurvature mode contributes only to 1% of the observed anisotropy near @xmath68 . of course , if @xmath58 is very different from @xmath69 , there could still be a large isocurvature contribution at either larger or smaller scales . since in the following we will focus on the constraints induced on mixed ad+cdi models by the lyman-@xmath2 data , let us give a few details on the shape of the linear matter power spectrum @xmath70\,.&\end{aligned}\ ] ] here @xmath71 and @xmath72 are computed from the initial conditions @xmath34 and @xmath35 respectively , exactly like @xmath64 and @xmath33 , and the cross - correlated term is simply given by @xmath73^{{{1\over2}}},\ ] ] where the minus sign comes from the fact that with our definition of @xmath74 , a positive correlation @xmath75 in the early universe implies a reduction of the matter power spectrum today , and vice - versa . in the limit @xmath76 , where @xmath77 corresponds to modes crossing the hubble length at the time of equality , it is well known ( see e.g. @xcite ) that the power spectra obey , to first approximation , @xmath78 which shows that for @xmath79 the isocurvature contribution to the small - scale power spectrum is generically much redder than the adiabatic one . the relative amplitude depends on the cosmological parameters . in the vicinity of the concordance @xmath62cdm model , one finds @xmath80 for cdi . so , like for cmb anisotropies , we see that even when @xmath65 in the early universe ( i.e. @xmath81 or @xmath82 ) , the isocurvature contribution to the currently observed power spectrum is only of the per cent order , at least near the pivot scale . however , for large @xmath58 , the contribution may be large on small scales . indeed , a large portion of the parameter region allowed by previous studies corresponds to a significant isocurvature fraction @xmath83 and to a very blue tilt @xmath84 . in this case , the matter power spectrum is affected or even dominated by the non - adiabatic contribution on small scales ( typically for wavenumbers @xmath85mpc ) . we illustrate this behavior on fig . [ fig1 ] for a particular set of ad+cdi models with two different values of the isocurvature tilt , @xmath86 or @xmath87 , and many possible values of @xmath88 . the impact of the non - adiabatic contribution consists either in a smooth change of the effective slope on small scales , or more radically in a sharp feature ( a pronounced break or a dip ) . the second situation can occur on relevant scales for large positive @xmath45 , and of course large enough values of @xmath2 and @xmath58 . it is well established by analytical calculation and hydrodynamical simulations that the lyman-@xmath2 forest blueward of the lyman-@xmath2 emission line in qso spectra is produced by the inhomogeneous distribution of a warm ( @xmath89 k ) and photoionized intergalactic medium ( igm ) along the line of sight . the opacity fluctuations in the spectra arise from fluctuations in the matter density and trace the gravitational clustering of the matter distribution in the quasi - linear regime @xcite . the lyman-@xmath2 forest has thus been used extensively as a probe of the matter power spectrum on comoving scales of @xmath90mpc @xcite . the lyman-@xmath2 optical depth in velocity space @xmath91 ( km / s ) is related to the neutral hydrogen distribution in real space as ( see e.g. ref . @xcite ) : @xmath92dy \ ; , \label{eqtau2}\ ] ] where @xmath93 @xmath94 is the hydrogen ly@xmath2 cross - section , @xmath95 is the real - space coordinate ( in km s@xmath5 ) , @xmath96 is the standard voigt profile normalized in real - space , @xmath97 is the velocity dispersion in units of @xmath98 , @xmath99 the hubble parameter , @xmath100 is the local density of neutral hydrogen and @xmath101 is the peculiar velocity along the line - of - sight . the density of neutral hydrogen can be obtained by solving the photoionization equilibrium equation ( see e.g. @xcite ) . the neutral hydrogen in the igm responsible for the lyman-@xmath2 forest absorptions is highly ionized due to the metagalactic ultraviolet ( uv ) background radiation produced by stars and qsos at high redshift . this optically thin gas in photoionization equilibrium produces a lyman-@xmath2 optical depth of order unity . the balance between the photoionization heating by the uv background and adiabatic cooling by the expansion of the universe drives most of the gas with @xmath102 , which dominates the lyman-@xmath2 opacity , onto a power - law density relation @xmath103 , where the parameters @xmath104 and @xmath105 depend on the reionization history and spectral shape of the uv background and @xmath106 is the local gas overdensity ( @xmath107 ) . the relevant physical processes can be readily modelled in hydrodynamical simulations . the physics of a photoionized igm that traces the dark matter distribution is , however , sufficiently simple that considerable insight can be gained from analytical modeling of the igm opacity based on the so called fluctuating gunn peterson approximation neglecting the effect of peculiar velocities and the thermal broadening @xcite . the fluctuating gunn peterson approximation makes use of the power - law temperature density relation and describes the relation between lyman-@xmath2 opacity and gas density ( see @xcite ) along a given line of sight as follows , @xmath108 where @xmath109 in the range @xmath110 , @xmath111 the hi photoionization rate , @xmath112 km / s / mpc the hubble parameter at redshift zero . for a quantitative analysis , however , full hydrodynamical simulations , which properly simulate the non - linear evolution of the igm and its thermal state , are needed . equations ( [ eqtau2 ] ) and ( [ eqtau ] ) show how the observed flux @xmath113 depends on the underlying local gas density @xmath114 , which in turn is simply related to the dark matter density , at least at large scales where the baryonic pressure can be neglected @xcite . statistical properties of the flux distribution , such as the flux power spectrum , are thus closely related to the statistical properties of the underlying matter density field . the power spectrum of the observed flux in high - resolution lyman-@xmath2 forest data provides meaningful constraints on the dark matter power spectrum on scales of @xmath115 , roughly corresponding to scales of @xmath90mpc ( somewhat dependent on the cosmological model ) . at larger scales the errors due to uncertainties in fitting a continuum ( i.e. in removing the long wavelength dependence of the spectrum emitted by each qso ) become very large while at smaller scales the contribution of metal absorption systems becomes dominant ( see e.g. @xcite ) . in this paper , we will use the dark matter power spectrum that viel , haehnelt & springel @xcite ( vhs ) inferred from the flux power spectra of the croft et al . @xcite ( c02 ) sample and the luqas sample of high - resolution lyman-@xmath2 forest data @xcite . the c02 sample consists of 30 keck high resolution hires spectra and 23 keck low resolution lris spectra and has a median redshift of @xmath116 . the luqas sample contains 27 spectra taken with the uves spectrograph and has a median redshift of @xmath117 . the resolution of the spectra is 6 km / s , 8 km / s and 130 km / s for the uves , hires and lris spectra , respectively . the s / n per resolution element is typically 30 - 50 . damped and sub - damped lyman-@xmath2 systems have been removed from the luqas sample and their impact on the flux power spectrum has been quantified by @xcite . estimates for the errors introduced by continuum fitting , the presence of metal lines in the forest region and strong absorptions systems have also been made @xcite . vhs have used numerical simulation to calibrate the relation between flux power spectrum and linear dark matter power spectrum with a method proposed by c02 and improved by @xcite and vhs . a set of hydrodynamical simulations for a coarse grid of the relevant parameters is used to find a model that provides a reasonable but not exact fit to the observed flux power spectrum . then , it is assumed that the differences between the model and the observed linear power spectrum depend linearly on the matter power spectrum . the hydrodynamical simulations are used to determine a bias function between flux and matter power spectrum : @xmath118 , on the range of scales of interest . in this way the linear matter power spectrum can be recovered with reasonable computational resources . this method has been found to be robust provided the systematic uncertainties are properly taken into account @xcite . running hydrodynamical simulations for a fine grid of all the relevant parameters is unfortunately computationally prohibitive ( see discussion in @xcite on a possible attempt to overcome this problem ) . we have seen in section [ mps ] that the isocurvature mode contribution can create distortions in the small - scale linear matter power spectrum . of course , this extra freedom was not taken into account in the definition of the grid of models in vhs . in principle , we should run simulations for a new grid with extra parameters ( @xmath2 , @xmath45 , @xmath58 , @xmath30 ) . alternatively , we can carry a tentative analysis with the same function @xmath119 and the same error bars as in the pure adiabatic case , and check the validity of our results _ a posteriori_. the idea is simply to select a marginally excluded model with the largest possible deviation from adiabaticity in the matter power spectrum . for this model , we run a new hydrodynamical simulation and we compare @xmath120 with the function @xmath121 used in the analysis . in case of good agreement , the results will be validated . we expect this agreement to be fairly good on large scales , but deviations should appear on small scales , because of the different non - linear evolution . the use of state - of - the - art hydrodynamical simulations is a significant improvement compared to previous studies which used numerical simulation of dark matter only @xcite . we use the parallel treesph code gadget - ii @xcite in its treepm mode which speeds up the calculation of long - range gravitational forces considerably . the simulations are performed with periodic boundary conditions with an equal number of dark matter and gas particles . radiative cooling and heating processes are followed using an implementation similar to @xcite for a primordial mix of hydrogen and helium . the uv background is given by @xcite . to maximise the speed of the simulation a simplified criterion of star formation has been applied : all the gas at overdensities larger than 1000 times the mean overdensity is turned into stars @xcite . the simulations were run on cosmos , a 152 gb shared memory altix 3700 with 152 cpus hosted at the department of applied mathematics and theoretical physics ( cambridge ) . there is a number of systematic uncertainties and statistical errors which affect the inferred power spectrum and an extensive discussion can be found in @xcite . vhs estimated the uncertainty of the overall _ rms _ fluctuation amplitude of matter fluctuation to be 14.5 % with a wide range of different factors contributing . we present here a brief summary . the effective optical depth , @xmath122 which is essential for the calibration procedure has to be determined separately from the absorption spectra . as discussed in vhs , there is a considerable spread in the measurement of the effective optical depth in the literature . determinations from low - resolution low s / n spectra give systematically higher values than high - resolution high s / n spectra . however , there is little doubt that the lower values from high - resolution high s / n spectra are appropriate and the range suggested in vhs leads to a 8% uncertainty in the _ rms _ fluctuation amplitude of the matter density field ( see table 5 in vhs ) . other uncertainties are the slope and normalization of the temperature - density relation of the absorbing gas which is usually parametrised as @xmath103 . @xmath104 and @xmath105 together contribute up to 5% to the error of the inferred fluctuation amplitude . vhs further estimated that uncertainties due to the c02 method ( due to fitting the observed flux power spectrum with a bias function which is extracted at a slightly different redshift than the observations ) contribute about 5% . they further assigned a 5 % uncertainty to the somewhat uncertain effect of galactic winds and finally an 8% uncertainty due the numerical simulations ( codes used by different groups give somewhat different results ) . summed in quadrature , all these errors led to the estimate of the overall uncertainty of 14.5% in the _ rms _ fluctuation amplitude of the matter density field . for our analysis we use the inferred dm power spectrum in the range @xmath115 as given in table 4 of vhs . ( note that , as in @xcite we have reduced the power spectrum values by 7% to mimick a temperature - density relation with @xmath123 , the middle of the plausible range for @xmath105 @xcite ) . unfortunately at smaller scales the systematic errors become prohibitively large mainly due to the large contribution of metal absorption lines to the flux power spectrum ( see fig . 3 of ref . @xcite ) and due to the much larger sensitivity of the flux power spectrum to the thermal state of the gas at these scales . likelihood for the ad+cdi model , using all our data set . the first eleven parameters are independent , while the last four are related parameters ( with non - flat priors ) . , height=340 ] .[tab_lya_cdi ] @xmath124 confidence limits for the ad+cdi model , using all our data set , for the eleven basis parameters with flat priors , and below , for related parameters . [ cols="^,^",options="header " , ] two - dimensional likelihood for the amplitude of the isocurvature mode and of the cross - correlation component , near the pivot scale . we adopted a flat prior within the ellipse ( which appears here as a circle ) in which these parameters are defined . , height=302 ] any ad+cdi model is described by the usual six parameters of the @xmath62cdm model , plus four parameters for the isocurvature sector ( two amplitudes and two tilts ) . like in most of the literature , we define the amplitudes parameters at the pivot scale @xmath125mpc@xmath5 . for the isocurvature fraction , we could decide to impose a flat prior on @xmath126 , or @xmath2 , or any function of them ; different choices are not equivalent , in general . we will come back to the dependence of the final result on the choice of priors in section [ sec : priors ] . meanwhile , we chose a specific set of parameters which appear linearly in the expression of the observable power spectra , @xmath2 and @xmath127 , and that we believe are physically relevant . as already mentioned , these two parameter are defined within an ellipse , in which we assume a flat prior . furthermore , we must take into account the inequality @xmath128 which should hold at least over the scales probed by the data , i.e. typically between @xmath129mpc@xmath5 and @xmath130mpc@xmath5 . this is achieved by introducing a new parameter @xmath131 , with a flat prior within the range @xmath132 . in summary , our basis parameters with flat priors consists of : * the baryon density , @xmath133 , * the cold dark matter density , @xmath134 , * the ratio @xmath135 of the sound horizon to the angular diameter distance multiplied by 100 , * the optical depth to reionization , @xmath136 , * the adiabatic tilt , @xmath69 , * the isocurvature tilt , @xmath137 , * the parameter related to the tilt of the cross - correlation angle , @xmath138 $ ] , * the overall normalization , @xmath139 $ ] , * the isocurvature fraction , @xmath2 , * the cross - correlation amplitude , @xmath140 . in addition , there are three independent parameters related to observations : the lyman-@xmath2 calibration parameter @xmath141 defined in @xcite , on which we impose the same gaussian prior @xmath142 ; and the two bias parameters associated to the 2df and sdss data with flat priors . our full parameter space is therefore 13-dimensional . we compute the marginalized bayesian likelihood of each parameter with a monte carlo markov chain method , using the public code cosmomc @xcite . the results are displayed in fig . [ fig_lya_cdi ] and table [ tab_lya_cdi ] ( after marginalization over the 2df and sdss bias parameters ) . the data favors purely adiabatic models , but remains compatible with an isocurvature fraction @xmath143 at the 2@xmath144 ( 95% ) confidence level ( cl ) , with a tilt @xmath145 ( @xmath146 cl ) . the one - dimensional likelihoods for @xmath2 , @xmath140 must be interpreted with care : the fact that these parameters are defined within an ellipse implies that there is more parameter space available near @xmath147 and @xmath148 . more interesting are the two - dimensional likelihood contours for @xmath149 displayed in fig . [ fig_lya_cdi_2d ] , since in this representation the prior is really flat inside the ellipse . from this figure , it is clear that the data prefers an uncorrelated isocurvature contribution . the flatness of the @xmath150 likelihood shows that the data give no indication on the tilt of the cross - correlation angle . likelihood of the isocurvature - related parameters , for the three combinations of data sets described in section [ impact ] : `` lyman-@xmath2 '' ( red ) , `` 2df bias prior '' ( green ) and `` none '' ( blue ) . _ ( left ) _ marginalized 1@xmath144 and 2@xmath144 confidence levels in the @xmath151^{1/2})$ ] space . _ ( right ) _ marginalized probability distribution for @xmath58.,height=158 ] favored ranges for the matter power spectrum @xmath152 in the three runs `` lyman-@xmath2 '' ( dark ) , `` 2df bias prior '' ( medium ) and `` none '' ( light ) , compared with our lyman-@xmath2 data , from the luqas quasar spectra _ ( left ) _ and from the re - analyzed croft et al . ( right)_. the bands represent the envelope of all the matter power spectra in the markov chains ( after eliminating models with the worse likelihood ) . each power spectrum has been computed at the median redshift of the data and re - expressed in units of km / s . in addition to the statistical errors , the data points share an overall effective calibration error , whose standard deviation is displayed in the top right corners . for the run including the lyman-@xmath2 data , each power spectrum has been divided by the value of the calibration parameter . the red dashed curves show the particular power spectrum discussed in section [ posteriori].,height=158 ] the lyman-@xmath2 forest provides a powerful indication on both the amplitude and the shape of the matter power spectrum for @xmath153 s / km , i.e. roughly larger than @xmath154mpc . in order to illustrate the importance of this data set in our results , we repeat the same analysis without lyman-@xmath2 data . in this case , there are two options : we can either use the 2df and sdss galaxy power spectrum data as a constraint only on the _ shape _ of the matter power spectrum , as already done in the previous analysis of section [ sec : results ] ; or introduce a bias prior derived e.g. from the third and fourth - order galaxy correlation function of the 2df catalogue @xcite , in order to keep an information on the amplitude of the matter power spectrum . for these three cases , that we call `` lyman-@xmath2 '' , `` 2dfbias prior '' and `` none '' , the 2@xmath144 upper bound on @xmath2 are respectively equal to 0.4 , 0.5 and 0.5 . the likelihoods for the most interesting parameters are displayed in fig . [ lya_vs_bias_cdi ] . as expected , the lyman-@xmath2 data set is significanty more powerful than the 2df bias prior for cutting out models with large @xmath2 , and even more clearly , with large @xmath58 or large anticorrelation , as can be seen in fig . it is important to note that without these data , all results depend on our arbitrary prior @xmath155 : values far beyond this upper bound could still be compatible with the data , as also found in ref . @xcite when using the same pivot scale . in the presence of the lyman-@xmath2 data , we get a robust upper bound on @xmath58 , and none of our priors play a role in the final results , with the exception of the well - motivated @xmath150 prior . the impact of the lyman-@xmath2 data can be understood visually from fig . [ fig_lya_pk ] . after running each case , we consider the collection of all matter power spectra in our markhov chains ( except models with a bad posterior likelihood @xmath156 ) . the gray bands in fig . [ fig_lya_pk ] correspond to the envelope of all these @xmath152 s , compared to the lyman-@xmath2 data points . as expected , when the lyman-@xmath2 is not used , the band gets very wide above the wavenumber @xmath157mpc@xmath158s / km ( note that for models with @xmath159 , the small - scale power spectrum is asymptotically flat ) . the role of the bias prior is marginal : it simply favors models with the lowest global normalization , but without affecting the isocurvature fraction and tilt . using the lyman-@xmath2 data , we can exclude any break in the power spectrum on scales @xmath160mpc@xmath161s / km . this results in much stronger constraints for the parameters @xmath162 , as can be seen from fig . 4 . the ratio @xmath163 at @xmath164 , computed from the hydrodynamical simulations as the ratio of the flux power spectrum ( averaged over 1000 line - of - sights ) over the input linear matter power spectrum . the solid blue curve shows the result for an adiabatic @xmath62cdm close to the best - fit model , while the dashed red curve was obtained from the `` most extreme ad+cdi mixed model '' defined in section [ posteriori ] and here labelled as @xmath62cdi . the green band shows the region in which the lyman-@xmath2 data is used in the present analysis.,height=340 ] we apply the strategy described in section [ hydro ] in order to check the validity of our lyman-@xmath2 data fitting procedure . we take the large number of samples contained in our markov chains , and eliminate all models with a likelihood smaller than @xmath165 ( in terms of effective @xmath166 , this corresponds to @xmath167 ) . we then select the model with the largest value of @xmath2 , which represents the strongest deviation from the purely adiabatic model . the corresponding matter power spectrum is plotted in fig . [ fig_lya_pk ] and has a break around @xmath168mpc@xmath169s / km . above this wavenumber , the slope of the power spectrum is given by eq . ( [ shape_pk_iso ] ) with @xmath170 . for this `` extreme '' model , we perform a hydrodynamical simulation as described in section [ hydro ] , and compare the bias function @xmath119 with that assumed throughout the analysis . as shown in fig.[bias ] , in the range @xmath171km / s probed by the data , the difference between the two functions is very small with respect to the statistical errors on the data . we conclude that in the present context , our lyman-@xmath2 data fitting procedure is robust , and does not introduce an error in the 1@xmath144 or 2@xmath144 bounds derived for each parameter of the ad+cdi mixed model . two - dimensional 68% and 95% confidence limits for the cdi mode amplitude and cross - correlation angle ( evaluated at the pivot scale ) . the two plots show the results of two independent runs with different parameter basis and priors . on the left ( solid red curves ) , the parameters are @xmath172 , with a flat prior within the ellipse . on the right ( solid black lines ) , the parameters are @xmath173 , related to the previous parameters through eqs.([correspondance ] ) , with a flat prior within the square . the dashed curves show for comparison the likelihood contours obtained for one parameter set , assuming a flat prior on the _ other _ parameter set . , height=158 ] the fact of choosing a top - hat prior in the @xmath172 parameter space is rather arbitrary . other groups prefer to take top - hat priors on @xmath174 , defined in eq.([correspondance ] ) , and @xmath175 . due to the non - linear transformation between the two basis , they are clearly not equivalent in terms of priors ( see the discussion of this point in @xcite , in the context of bayesian evidence calculation for adiabatic versus mixed models ) . we checked this issue explicitly with an independent run based on the @xmath173 basis . the results are summarized in fig . [ lya_cdi_op_vs_fiso ] . as expected from the jacobian , the @xmath173 option gives more weight to models with a small isocurvature fraction . for instance , the run with a flat prior on @xmath174 gives a 1@xmath144 bound @xmath176 , while that with a flat prior on @xmath2 gives @xmath177 . however , at the 2@xmath144 level , the relative difference is small ( @xmath178 versus @xmath179 ) because the jacobian is asymptotically flat . in principle , in bayesian analysis , the choice of parameter basis and priors should reflect one s knowledge on the model before comparision with the data . however , in the absence of a unique underlying physical model motivating the presence of isocurvature modes , different scientists might put foward different choices of prior . this intrinsic freedom in bayesian analyses should always be kept in mind when quoting bounds , especially for parameters which represent physical ingredients not strictly needed by the data , which is the case here for the isocurvature sector parameters ( for other parameters such that the data picks up a narrow allowed region , a change of priors wo nt affect the bounds very much ) . however , even for the isocurvature parameters discussed here , it is reassuring to see from our analysis that the 2@xmath144 contours obtained from the the two runs and compared in fig . [ lya_cdi_op_vs_fiso ] are roughly in agreement . likelihood for @xmath174 in the curvaton model . ] in this section we derive bounds on the specific case of curvaton models . the curvaton hypothesis is an ingenuous way to generate the observed curvature perturbation from a field ( the curvaton ) different from that which drives inflation ( the inflaton ) @xcite . in practice there is not much difference in the phenomenological signatures left in the cmb and lss compared to an ordinary inflationary model . however , there are a few cases in which it is possible to leave a residual " isocurvature component , together with the dominant curvature contribution . more specifically , in the curvaton models in which the curvaton field is responsible for the cdm component of matter , there are various possibilities depending on the time of creation of cdm versus the decay of the curvaton field . in all these cases , the curvature and isocurvature perturbations are related to the gauge invariant bardeen variable @xmath180 as @xmath181 & & { { \cal r}}=-\zeta\,.\end{aligned}\ ] ] let us classify here the different cases : 1 ) when cdm - creation occurs before the curvaton decays and the fraction @xmath182 of the total energy density in the curvaton field at the time of its decay is negligible . then @xmath183 , which corresponds to @xmath184 ( @xmath185 ) , and @xmath49 ( maximally correlated ) , with @xmath186 ; 2 ) when cdm - creation occurs before the curvaton decays but the fraction @xmath182 at decay is important . this case requires specific model input and in principle can have any value of @xmath174 and @xmath45 , while @xmath186 ; 3 ) when cdm - creation occurs at the decay of the curvaton and the fraction @xmath187 . in this case , @xmath188 and thus @xmath189 , which corresponds to @xmath190 , i.e. @xmath50 ( maximally anticorrelated ) and @xmath186 ; 4 ) when cdm - creation occurs after the curvaton decay . then there is only one thermal fluid in equilibrium , @xmath191 , and there is no way to generate an isocurvature perturbation , @xmath192 . since case 1 ) is already excluded at many sigma , and case 2 ) is essentially identical ( except for @xmath186 ) to our generic analysis , we will concentrate on case 3 ) of a maximally anticorrelated mixture of isocurvature and adiabatic modes with equal tilts and @xmath193 . our results are summarized in fig . [ fig_curv ] , which shows the likelihood distribution for the generic curvaton model . we have used @xmath194 , @xmath193 , and @xmath195 , which is equivalent to @xmath196 and @xmath126 positive or negative : @xmath197 corresponds to @xmath196 , or positive correlation between @xmath198 and @xmath199 , i.e. suppression of power in @xmath152 and in the large - scale cmb temperature spectrum ; while @xmath200 corresponds to the opposite anti - correlated case . we find @xmath201 at the 2@xmath144-level , which implies @xmath202 at the same cl . in our opinion , such a stringent constraint on the fraction of energy density in the curvaton at decay calls for a tremendous finetuning ( there is no physical reason to expect that the curvaton should decay precisely when it is starting to dominate the total energy density of the universe , within 2% ) , which makes the curvaton hypothesis in its most attractive scenario very unlikely . another chance to generate an observable isocurvature signature is through the possible presence of two scalar fields driving inflation @xcite . the simplest case at hand is that of two massive fields coupled only gravitationally : @xmath203 where @xmath204 and @xmath205 are the masses of the heavy and light fields respectively . + we assume slow - roll conditions during inflation , and use the number of e - folds till the end of inflation @xmath206 to parametrize the fields as : @xmath207 using the field and friedmann equations , we can solve for the rate of expansion during inflation : @xmath208 where @xmath209 , and find the number of e - folds as a function of @xmath135 : @xmath210 the perturbed einstein equations can be solved for long wavelength modes in the longitudinal gauge . assuming that the heavy field decays into cdm whereas the ligth field produces other species , we find the magnitudes of the curvature and entropy perturbation at horizon crossing . during radiation domination and for super - hubble modes , this gives : @xmath211 where @xmath212 are gaussian random fields associated with the quantum fluctuations of the fields , and the subindex @xmath27 implies the value of the corresponding quantity at horizon crossing during inflation . one typically expects @xmath213 . it can be seen from ( [ primordial_p ] ) that the correlation power spectrum has no scale dependence , and thus , for this model @xmath214 , while the adiabatic and isocurvature tilts have expressions @xmath215 { n_{\rm iso}}&= & 1 - { ( r^2 - 1)(r^2\tan^4\theta-1)(1+\tan^2\theta)\over s_k(1+r^2\tan^2\theta)^2(1+r^4\tan^2\theta)}\,,\end{aligned}\ ] ] whose values , for @xmath216 , are typically @xmath217 and @xmath58 in the range @xmath218 $ ] for @xmath219 $ ] . since @xmath220 at 95% c.l . , models with large values of @xmath221 are ruled out . it was shown in @xcite that a relationship beteween @xmath2 and @xmath45 can be found . it can be simply expressed as a straight line in our parameter space : @xmath222 on the other hand , for these models , the parameters @xmath2 and @xmath45 have minimum and maximum values respectively , which only depend on the ratio @xmath221 and the number of e - folds @xmath223 , @xmath224 \beta_{\rm max } & = & { r^2 - 1\over r^2 + 1}\,,\\[2 mm ] \left.2\beta\sqrt{\alpha(1-\alpha)}\right|_{\rm max } & = & { 2s_k(r^2 - 1)\over s_k^2+(r^2 + 1)^2}\,.\end{aligned}\ ] ] using the results of section [ sec : results ] , we find that the inclusion of the lyman-@xmath2 data significantly improves the previous bound on r to @xmath225 at 95% c.l . this bound comes mainly from a combination of bounds on @xmath140 and @xmath58 . we did not find necessary to generate a @xmath214 sampling for this model . in our results , the parameter @xmath150 has a flat distribution and thus is unconstrained . we therefore expect similar results when fixing it to zero . in addition to cmb , lss and snia data , we used some recent lyman-@xmath2 forest data to further constrain the bounds on possible cdm - isocurvature primordial fluctuations . we find that the systematics induced @xmath226 in particular , those associated with the recovery of the linear dark matter power spectrum from the flux power spectrum @xmath226 are greatly compensated by the valuable information on the small - scale matter power spectrum provided by the lyman-@xmath2 data . before summarizing our results , it is worth mentioning that when we omit the lyman-@xmath2 forest data our bounds agree very well with those of ref . the authors of @xcite work with a pivot scale @xmath227mpc@xmath5 , but they also show how their results are modified when they take @xmath228mpc@xmath5 like in the present paper : in that case the agreement with us is particularly good . the comparison of our results with the wmap analysis from ref . @xcite is more puzzling : using or not some lyman-@xmath2 data , they always find much stronger bounds on @xmath174 than us . it is true that we have one more free parameter @xmath150 , and that we do not introduce a prior on the 2df bias ; however , even when we fix @xmath229 and introduce such a prior , our @xmath174 bound remains much weaker . so far , private communications with the wmap team did not allow us to understand the origin of the discrepancy . using all our data set , we find at the 95% confidence level , an isocurvature fraction @xmath230 , a cross - correlation amplitude @xmath231 , and an isocurvature tilt @xmath232 . the tilt of the correlation angle remains unconstrained . if we switch to the basis used for instance in ref . @xcite we find @xmath233 at 95% c.l . in the case of a curvaton scenario where cdm - creation occurs at the decay of the curvaton a case in which the adiabatic and isocurvature modes are maximally anti - correlated , @xmath50 , and @xmath194 we find @xmath234 , still at the 95% confidence level this requires that the fraction @xmath182 of the total density in the curvaton field at that time be fine - tuned between 0.98 and one . finally , if we assume a double - inflation model with two massive inflatons coupled only gravitationally , such that the heaviest field decays into cdm , while the lightest one into standard model particles , we find that the mass ratio should obey @xmath9 ( 95% c.l . ) . the simulations were done at the uk national cosmology supercomputer center funded by pparc , hefce and silicon graphics / cray research . mv thanks pparc for financial support . mb thanks the group at sussex university for their warm hospitality and acknowledges support by the european community programme human potential under contract no . hpmt - ct-2000 - 00096 . this work was supported in part by a cicyt project fpa2003 - 04597 , as well as a spanish - french collaborative grant between cicyt and in2p3 . we whish to thank a. liddle , h. peiris and l. verde for useful discussions . a. d. linde , phys . b * 158 * , 375 ( 1985 ) ; l. a. kofman and a. d. linde , nucl . b * 282 * , 555 ( 1987 ) ; s. mollerach , phys . lett . b * 242 * , 158 ( 1990 ) ; a. d. linde and v. mukhanov , phys . d * 56 * , 535 ( 1997 ) ; m. kawasaki , n. sugiyama and t. yanagida , phys . d * 54 * , 2442 ( 1996 ) ; p. j. e. peebles , astrophys . j. * 510 * , 523 ( 1999 ) . d. polarski and a. a. starobinsky , phys . d * 50 * , 6123 ( 1994 ) ; m. sasaki and e. d. stewart , prog . phys . * 95 * , 71 ( 1996 ) ; m. sasaki and t. tanaka , prog . phys . * 99 * , 763 ( 1998 ) . j. garca - bellido and d. wands , phys . d * 53 * , 5437 ( 1996 ) ; * 52 * , 6739 ( 1995 ) . c. gordon , d. wands , b. a. bassett and r. maartens , phys . d * 63 * , 023506 ( 2001 ) ; n. bartolo , s. matarrese and a. riotto , phys . d * 64 * , 123504 ( 2001 ) . d. wands , n. bartolo , s. matarrese and a. riotto , phys . d * 66 * , 043520 ( 2002 ) . f. finelli and r. h. brandenberger , phys . d * 62 * , 083502 ( 2000 ) ; f. di marco , f. finelli and r. brandenberger , phys . d * 67 * , 063512 ( 2003 ) . d. langlois , phys . d * 59 * , 123512 ( 1999 ) ; d. langlois and a. riazuelo , phys . d * 62 * , 043504 ( 2000 ) . g. efstathiou and j. r. bond , mon . not . r. astron . a * 218 * , 103 ( 1986 ) ; * 227 * , 33 ( 1987 ) ; p. j. e. peebles , nature * 327 * , 210 ( 1987 ) . h. kodama and m. sasaki , int . j. mod . phys . a * 1 * , 265 ( 1986 ) ; * 2 * , 491 ( 1987 ) ; s. mollerach , phys . rev . d * 42 * , 313 ( 1990 ) . m. bucher , k. moodley and n. turok , phys . d * 62 * , 083508 ( 2000 ) ; phys . rev . * 87 * , 191301 ( 2001 ) . s. weinberg , phys . d * 70 * , 083522 ( 2004 ) . d. h. lyth , c. ungarelli and d. wands , phys . d * 67 * , 023503 ( 2003 ) . c. gordon and k. a. malik , phys . d * 69 * , 063508 ( 2004 ) . k. enqvist and m. s. sloth , nucl . b * 626 * , 395 ( 2002 ) . d. h. lyth and d. wands , phys . b * 524 * , 5 ( 2002 ) ; t. moroi and t. takahashi , phys . b * 522 * , 215 ( 2001 ) [ erratum - ibid . b * 539 * , 303 ( 2002 ) ] ; phys . d * 66 * , 063501 ( 2002 ) . n. bartolo , e. komatsu , s. matarrese and a. riotto , phys . rept . * 402 * , 103 ( 2004 ) . r. trotta , a. riazuelo and r. durrer , phys . lett . * 87 * , 231301 ( 2001 ) ; phys . d * 67 * , 063520 ( 2003 ) . m. bucher , j. dunkley , p. g. ferreira , k. moodley and c. skordis , phys . lett . * 93 * , 081301 ( 2004 ) ; k. moodley , m. bucher , j. dunkley , p. g. ferreira and c. skordis , phys . d * 70 * , 103520 ( 2004 ) ; j. dunkley , m. bucher , p. g. ferreira , k. moodley and c. skordis , arxiv : astro - ph/0507473 . j. valiviita and v. muhonen , phys . * 91 * , 131302 ( 2003 ) . m. beltran , j. garcia - bellido , j. lesgourgues and a. riazuelo , phys . d * 70 * , 103530 ( 2004 ) . h. kurki - suonio , v. muhonen and j. valiviita , phys . d * 71 * , 063005 ( 2005 ) . l. amendola , c. gordon , d. wands and m. sasaki , phys . rev . * 88 * , 211302 ( 2002 ) . h. v. peiris _ et al . _ , astrophys . j. suppl . * 148 * , 213 ( 2003 ) . p. crotty , j. garca - bellido , j. lesgourgues and a. riazuelo , phys . lett . * 91 * , 171301 ( 2003 ) . f. ferrer , s. rasanen and j. valiviita , jcap * 0410 * , 010 ( 2004 ) . d. parkinson , s. tsujikawa , b. a. bassett and l. amendola , arxiv : astro - ph/0409071 . c. gordon and a. lewis , phys . d * 67 * , 123513 ( 2003 ) ; new astron . rev . * 47 * , 793 ( 2003 ) . g. lazarides , r. r. de austri and r. trotta , phys . d * 70 * , 123527 ( 2004 ) . m. beltran , j. garcia - bellido , j. lesgourgues , a. r. liddle and a. slosar , phys . d * 71 * , 063532 ( 2005 ) . r. trotta , arxiv : astro - ph/0504022 . t. s. kim , m. viel , m. g. haehnelt , r. f. carswell and s. cristiani , mon . not . soc . * 347 * , 355 ( 2004 ) . r. a. c. croft _ et al . _ , astrophys . j. * 581 * , 20 ( 2002 ) . m. viel , m. g. haehnelt and v. springel , mon . not . soc . * 354 * , 684 ( 2004 ) . c. l. bennett _ et al . _ [ wmap collaboration ] , astrophys . j. suppl . * 148 * , 1 ( 2003 ) ; d. n. spergel _ et al . _ , astrophys . j. suppl . * 148 * , 175 ( 2003 ) ; r. rebolo _ et al . _ [ vsa collaboration ] , mon . not . 353 * , 747 ( 2004 ) ; c. dickinson _ et al . _ , mon . not . soc . * 353 * , 732 ( 2004 ) . t. j. pearson _ et al . _ [ cbi collaboration ] , astrophys . j. * 591 * , 556 ( 2003 ) ; j. l. sievers _ et al . _ , astrophys . j. * 591 * , 599 ( 2003 ) ; a. c. s. readhead _ et al . _ , astrophys . j. * 609 * , 498 ( 2004 ) . c. l. kuo _ et al . _ [ acbar collaboration ] , astrophys . j. * 600 * , 32 ( 2004 ) ; j. h. goldstein _ et al . _ , astrophys . j. * 599 * , 773 ( 2003 ) . j. a. peacock _ et al . _ [ 2dfgrs collaboration ] , nature * 410 * , 169 ( 2001 ) ; w. j. percival _ et al . _ , mon . not . r. astron . a * 327 * , 1297 ( 2001 ) ; * 337 * , 1068 ( 2002 ) . m. tegmark _ et al . _ [ sdss collaboration ] , astrophys . j. * 606 * , 702 ( 2004 ) . a. g. riess _ et al . _ [ supernova search team collaboration ] , astrophys . j. * 607 * , 665 ( 2004 ) . u. seljak _ et al . _ , phys . d * 71 * , 103515 ( 2005 ) . r. stompor , a. j. banday and k. m. gorski , astrophys . j. * 463 * , 8 ( 1996 ) ; p. j. e. peebles , astrophys . j. * 510 * , 531 ( 1999 ) ; e. pierpaoli , j. garca - bellido and s. borgani , jhep * 9910 * , 015 ( 1999 ) ; m. kawasaki and f. takahashi , phys . b * 516 * , 388 ( 2001 ) ; k. enqvist , h. kurki - suonio and j. vliviita , phys . rev . d * 62 * , 103003 ( 2000 ) ; * 65 * , 043002 ( 2002 ) . a. r. liddle and d. h. lyth , `` cosmological inflation and large - scale structure '' , cambridge university press ( 2000 ) ; e. bertschinger , arxiv : astro - ph/0101009 ; h. bi , astrophys . j. * 405 * , 479 ( 1993 ) ; m. viel , s. matarrese , h. j. mo , m. g. haehnelt and t. theuns , mon . not . soc . * 329 * , 848 ( 2002 ) ; m. zaldarriaga , r. scoccimarro and l. hui , astrophys . j. * 590 * , 1 ( 2003 ) . p. mcdonald _ et al . _ , arxiv : astro - ph/0407377 . l. hui , n. y. gnedin and y. zhang , astrophys . j. * 486 * , 599 ( 1997 ) . n. katz , d. h. weinberg and l. hernquist , astrophys . j. suppl . * 105 * , 19 ( 1996 ) . j. e. gunn and b. a. peterson , astrophys . j. * 142 * , 1633 ( 1965 ) ; j. n. bahcall and e. e. salpeter , astrophys . j. * 142 * , 1677 ( 1965 ) . l. verde _ et al . _ , astrophys . j. suppl . * 148 * , 195 ( 2003 ) . a. cuoco , j. lesgourgues , g. mangano and s. pastor , phys . d * 71 * , 123501 ( 2005 ) . j. silk and m. s. turner , phys . d * 35 * , 419 ( 1987 ) ; d. polarski and a. a. starobinsky , nucl . * b385 * , 623 ( 1992 ) ;

we present accurate relative ages for a sample of galactic globular clusters . the ages have been obtained by measuring the difference between the horizontal branch and the turnoff in two , internally photometrically homogeneous databases . the mutual consistency of the two data sets has been assessed by comparing the ages of globular clusters in common between the two databases . we have also investigated the consistency of our relative age determination within the recent stellar model framework . all clusters with }<-1.7 $ ] are found to be old , and coeval , with the possible exception of two objects , which are marginally younger . the age dispersion for the metal poor clusters is gyr ( rms ) , consistent with a null age dispersion . intermediate metallicity clusters (}<-0.8 $ ] ) are on average gyr younger than the metal poor ones , with an age dispersion of gyr ( rms ) , and a total age range of gyr . about of the intermediate metallicity clusters are coeval with the oldest clusters . all the clusters with}>-0.8 $ ] are gyr younger than the most metal poor ones , with a relatively small age dispersion , though the metal rich sample is still too small to allow firmer conclusions . there is no correlation of the cluster age with the galactocentric distance . we briefly discuss the implication of these observational results for the formation history of the galaxy .

astro-ph0503594

galactic globular clusters ( gcs ) play a key role in stellar astrophysics and cosmology . stars in a given gc are to a good approximation coeval , were born with the same initial chemical composition , and are located approximately at the same distance from the observer . these characteristics make them valuable benchmarks for testing stellar evolution theories through comparisons of theoretical isochrones and luminosity functions with their observational counterparts ( see , e.g. @xcite ) . in addition , due to the uniform age and initial chemical composition of stars belonging to a given cluster , the estimation of the cluster age is relatively straightforward , as shown in the pioneering work of @xcite , @xcite . given that gcs are the oldest systems populating our galaxy ( and external galaxies ) for which an age estimate is feasible , their ages provide a lower limit to the age of the universe , whilst their relative ages ( and possible correlations with metallicity , abundance patterns , and position ) disclose fundamental information about galactic formation mechanisms and timescales . recent work on the age distribution of sizeable samples of gcs @xcite has led to a broadly consistent scenario in which metal poor clusters up to some intermediate metallicities are largely coeval , whereas at higher @xmath2}$ ] values an age spread exists , with possibly an age - metallicity relation . the age distribution with respect to the position within the galaxy shows that an age spread is present at all distances , without any correlation with the distance from the galactic centre @xcite . unfortunately , absolute age measurements are still affected by a number of uncertainties , in particular by the large remaining errors on the gc distances and reddening @xcite . relative ages can be obtained with much higher accuracy by measuring the position of the main sequence turn - off ( to ) which is the most - used age indicator relative to some other feature in the colour magnitude diagram ( cmd ) whose location is little or not at all dependent on age ( stetson , vandenberg , & bolte 1996 , sarajedini , chaboyer , & demarque 1997 ) . however , any observed age indicator must be compared with theoretical predictions in order to derive an age . uncertainties in both the input physics @xcite , and the transformation from the theoretical to the observational plane @xcite , constitute a significant source of errors in the final age estimate , particularly when we want to compare ages of clusters with different metal content . in the present work , in order to minimise the uncertainties coming from the models , we used an updated set of theoretical tracks . more specifically , the theoretical isochrones employed in our analysis are the @xmath10-enhanced counterparts of the @xcite model and isochrone library , and have already been discussed briefly in @xcite . the reliability and accuracy of this theoretical framework has already been tested by comparison with various empirical data sets by @xcite and by @xcite . in the measurement of accurate relative ages , the photometric homogeneity of the data , and the homogeneity in the methods used to measure the age indicator parameters , are of fundamental importance , as we showed in @xcite , where , for the first time , we have been able to measure relative ages on a truly photometrically homogeneous database for a large sample ( 34 ) of gcs . in this paper we intend to extend the work of @xcite , by adding a new , larger , photometrically homogeneous catalogue of cmds , i.e. the hst snapshot catalogue by @xcite . for methodological consistency , we have also re - measured all the parameters relevant for an age estimate in the cmds of @xcite used by @xcite . the age - dating method used in our analysis is based on the magnitude difference @xmath11 ( in the @xmath12 band for the hst data , and in the analogous @xmath13-johnson band for the goundbased ones ) between the to and the horizontal branch ( hb ) , i.e. the so called vertical method @xcite . the to brightness is the age indicator ( the to gets fainter for increasing age ) whereas the hb level is unaffected in the age range typical of gcs ; this implies that older clusters display a larger value of @xmath11 . for ages around @xmath14 gyr the parameter @xmath11 scales approximately as @xmath15 . since it is a differential quantity , the comparison between observed and predicted @xmath11 values is unaffected by uncertainties in cluster reddening and distance modulus estimates , superadiabatic convection treatment in the stellar models as well as uncertainties in the colour - effective temperature relation adopted for transferring the models from the theoretical plane to the observational one . another advantage of using the @xmath11 parameter as age indicator is that it is weakly sensitive to uncertainties in the cluster @xmath2}$ ] , because at a fixed age the to magnitude scales with @xmath2}$ ] almost like the hb level . the plan of this paper is as follows . section 2 describes the observational data , the measurement of the @xmath11 from the observed cmds , and the estimate of the associated errors . section 3 is devoted to the estimate of the age distribution of our cluster sample . a discussion of the main results is in section 4 . conclusions are in section 5 . the present investigation is based on two databases of gc cmds : * the hst snapshot database of @xcite ; * the groundbased database of @xcite . each of the two databases is internally photometrically homogeneous . all the cmds from both catalogues can be found at the padova globular cluster group web page + ( http://dipastro.pd.astro.it/globulars ) . here , for the first time , we present an age estimate based on the hst snapshot catalogue . this is a catalogue of cmds for @xmath16 galactic gcs , observed in their central regions with the wfpc2/hst camera in the photometric bands @xmath17 and @xmath12 ( similar to the @xmath18 , @xmath13 in the johnson - cousins photometric system ) . for details about the observations , the reduction , and the photometric measurements please refer to @xcite . the groundbased database has already been used for relative age measurements by @xcite , who selected the best cmds from the two , photometrically homogeneous data sets collected , respectively , at the eso / dutch telescope ( for the southern sky gcs ) by @xcite , and at the isaac newton group jacobus kapteyn telescope ( for the northern gcs ) by @xcite . for details about the observations , the reduction , and the photometric measurements please refer to @xcite . forty one out of @xmath16 clusters from the hst snapshot catalogue , and @xmath19 out of @xmath20 clusters from the groundbased catalogues , have a cmd which can be used for an age measurement . the remaining objects were excluded for several reasons : too shallow photometry , large field contamination , small number of member stars . sixteen clusters are in common between the two catalogues , which allowed us to check the consistency of the ages derived from the two databases . combining the two catalogues , we have been able to estimate ages for @xmath0 gcs , representing more than @xmath21% of the galactic gc population . these gcs cover a metallicity interval @xmath22}<-0.3 $ ] , and are located from the very central part of the galaxy out to @xmath23 kpc , therefore extending the metallicity and galactocentric distance coverage of the original @xcite paper . in our attempt to be as homogeneous as possible , and consistent with our previous investigations based on the hst snapshot catalogue , we have adopted the metallicities listed in @xcite calibrated over both the ( * ? ? ? * hereafter cg ) , as extended by @xcite , and ( * ? ? ? * hereafter zw ) metallicity scales . in this paper , we will present the results obtained adopting both metallicity scales . for those clusters that do not appear in the @xcite catalogue , we adopted the original zw value when available . otherwise , we calculated the zw metallicity from ( * ? ? ? * in the revised version of 2003 ) by fitting a straight line to the relation between the two scales . for clusters not in the @xcite compilation , @xmath2}$ ] values in the cg system were calculated from zw values using the following equation @xcite : @xmath2}_{cg } = 0.61 + 3.04 { \rm [ fe / h]}_{zw } + 1.981 { \rm [ fe / h]}_{zw}^2 + 0.532 { \rm [ fe / h]}_{zw}^3 $ ] . all the other cluster parameters used in this paper have been extracted from ( * ? ? ? * in the revised version of 2003 ) . the relevant age indicator parameters described below have been extracted from both databases . it is worth emphasising the fact that the hst snapshot cmds are in the @xmath17 , @xmath12 wfpc2 flight system , while the groundbased cmds are in the @xmath13 , @xmath24 johnson - cousins photometric system . therefore , the vertical parameters obtained from the two databases can not be directly compared . first , they must be compared with the theoretical models , transformed into the appropriate observational plane , for the age determinations . only these final ages can be compared . in order to apply the vertical method , we need to measure the magnitude of the to point and the zero age horizontal branch ( zahb ) level . for the to measurement , we proceeded in the following way . first , the cmd of each clusters was cleaned by selecting stars based on the daophot psf fitting parameters ( chi , sha ) and their photometric errors . second , we extracted the fiducial main sequence ( ms ) for each cluster by taking the median of the colour distributions obtained in magnitude boxes containing a fixed number of stars , ranging from @xmath25 to @xmath26 , depending on the total number of ms stars . a preliminary to was then defined as the bluest point of the ms fiducial lines . for each cluster , we extracted a number of different ms fiducial lines by using different numbers of stars per box , and for each fiducial ms we associated a preliminary to to its bluest point . finally , we adopted as to magnitude the mean value of all the preliminary tos . the error on the mean of the preliminary to magnitudes gave a first estimate of the error associated to the adopted to magnitude . the to magnitude error was then better quantified following the procedure explained in the next section . for the zahb magnitudes of the hst snapshot clusters , we adopted the values recently published by @xcite . they present distance moduli and reddening estimates for @xmath27 galactic globular clusters , based on the same photometric catalogue we are using here . the zahb magnitudes were estimated starting from the rr lyrae level , for low and intermediate metallicity clusters , and from the fainter envelope of the red hb , for metal rich ones . for a detailed description of the adopted method , please refer to @xcite . here it suffices to say briefly that : * the zahb level for _ metal rich clusters _ ( @xmath2}\geq-1.0 $ ] ) was set as the magnitude of the lower envelope of the hb minus @xmath28 times the photometric error ( resulting from artificial star experiments ) at that magnitude ; * for _ low and intermediate metallicity clusters _ ( @xmath2}<-1.0 $ ] ) , a template cluster with similar metallicity and with a sizeable population of rr lyrae stars was selected from the literature . five different templates were used to cover the entire metallicity range . the cmd of the template cluster was shifted in colour and magnitude until its hb overlapped the hb of the cluster whose zahb level was to be determined . this procedure allowed to obtain the mean apparent magnitude of the cluster rr lyrae stars which was then transformed into the apparent zahb magnitude by using the relation @xcite : @xmath29}.\ ] ] for methodological consistency , we repeated the measures of the to and zahb magnitudes also for the groundbased cmds . the to measurement method by @xcite is very similar to what we did in the present paper , while the approach to obtaining the zahb magnitudes is completely different . the new zahb magnitudes for the groundbased cmds have been measured exactly as described above for the hst data , and in more detail by @xcite . for the groundbased low and intermediate metallicity clusters we used the same templates as for the hst ones . however , the reference cmds are different , as the groundbased catalogue is in the @xmath13 , @xmath24 photometric system . table [ ground ] lists the reference clusters , the metallicity range of the comparison clusters , and the source of the photometry for the reference cmds . figures [ fig0a ] and [ fig0b ] show two examples of our results . the cmds of two clusters are shown , one from the groundbased catalogue ( ngc 6218 ) and one from the the snapshot database ( ngc 362 ) . in fig . [ fig0a ] , the brighter part of the cmd of one of the templates used for the zahb measurements is also shown . the template is ngc 5904 . ngc 6218 has a quite scarcely populated hb , nevertheless by applying our technique we were able to estimate the zahb magnitude . we would like to emphasise the fact that ngc 6218 is also contained in the snapshot catalogue and zahb measurements were made on the hst cmd and published by @xcite . those measurements are in excellent agreement with those obtained from the groundbased cmd , considering the different but similar bandpasses . in order to estimate the uncertainty in the adopted to magnitude , we used the same method as in @xcite . we built about one hundred synthetic cmds for each cluster , using the isochrones by @xcite . the synthetic cmds were constructed by adopting for each cluster the corresponding metallicity , an age in the range between @xmath14 and @xmath30 gyr , the photometric errors ( as estimated from artificial star experiments ) , and the total number of stars in the observed cmd , and varying only the initial random number generator seed . the same procedure used to determine the to in the observed cmd ( see the explanation in the previous section ) was then applied to the synthetic diagrams . we estimated the error in the to colour and magnitude in each observed cmd as the standard deviation of all the tos measured in the corresponding synthetic diagrams . this procedure does not take into account the effects due to differential reddening . therefore the error estimated for clusters whose cmds show this effect ( namely ngc 4372 , ngc 5927 , ngc 6273 , and ngc 6544 ) is likely to be underestimated . the error on the mean of the preliminary to magnitudes for ngc 6273 ( the worst case ) is approximately a factor 2 larger than the error calculated from the synthetic diagrams . nevertheless , we decided to adopt this technique for its reliability and for consistency with @xcite . regarding the errors on the zahb magnitude of the intermediate and metal poor clusters , they are dominated by the error in matching the hb of the template cluster with that of the object cluster , particularly for clusters with only a blue hb . as in @xcite , we repeated the fit many times , and estimated the error as the maximum difference between the single shift values and the final adopted one . this error estimate has been divided by @xmath28 , in order to have the classical 1-@xmath31 value for a gaussian distribution of the errors ( in order to make it comparable with the errors of the to magnitudes ) . for the metal rich clusters , the main source of error is the uncertainty in the position of the lower envelope , particularly in clusters with a small number of stars . in estimating errors in the vertical parameter , we considered the errors on the zahb and to magnitudes to be independent , and thus summed in quadrature the two contributions . table [ tab1 ] lists the most relevant cluster parameters , including those measured in this paper , for the hst snapshot clusters . 1 gives the cluster name ; col . 2 and 3 give the metallicity in the zw and cg scale , respectively ; col . 4 gives the galactocentric distance . columns 5 , 6 of table [ tab1 ] give the to magnitudes and errors , respectively , in the @xmath12 band , while the @xmath12 zahb magnitudes and errors are in col . 7 , 8 . finally , col . 9 , 10 list the vertical parameter value , in the @xmath12 band , for the snapshot clusters . table [ tab2 ] provides the same quantities as table [ tab1 ] , but for the groundbased clusters . the magnitudes of table [ tab2 ] are therefore in the @xmath13-band . as already mentioned , we have used the same database @xcite used by @xcite , but for consistency reasons we decided to re - measure the vertical parameter . figure [ confr.gbn-gba ] shows the comparison between the to and zahb magnitudes , and between the vertical parameters of the present study and those listed by @xcite for the groundbased clusters . while the to values are in very good agreement , there is some hint of a trend with the metallicity ( though it is mostly within the average measurement errors ) for the differences in zahb magnitudes in the intermediate and low metallicity range . this was expected because of the different method adopted in the two studies and the different definition of the zahb level . @xcite used the same ( empirical ) template for all the intermediate and metal poor clusters . we think that the method used in the present paper is more reliable , as it is able to better account for the differences in the hb at varying the cluster metallicities . to measure the cluster relative ages , we compared our observed vertical parameters with the theoretical counterpart obtained from the set of @xmath10-enhanced isochrones of @xcite , available in both the wfpc2/hst photometric bands , and the johnson - cousins system . for each isochrone , the theoretical to was determined as the bluest point of the ms , and the zahb level was taken as the magnitude of the zahb model at @xmath32 . we performed a @xmath33-dimensional bicubic spline to set a finer grid in age and metallicity , and thus to give an estimate of the age for each cluster in the sample . figures [ c04aesnap ] and [ c04aegb ] show the theoretical @xmath34 and @xmath11 values as a function of @xmath2}$ ] for various ages , together with the observational points from the hst snapshot and groundbased observations , respectively . the upper panel of each figure adopts the zw metallicity scale , whereas the lower panel refers to the cg scale . the two data sets are presented separately due to the different photometric systems . both figures [ c04aesnap ] and [ c04aegb ] show that the mean age of the metal poor clusters is about @xmath5 gyr older than the mean age of the clusters with @xmath2}_{\rm zw}>-1.7 $ ] ( @xmath2}_{\rm cg}>-1.4 $ ] ) . at intermediate metallicities ( @xmath4}_{\rm zw } < -0.8 $ ] ) , the age spread is increased by a group of a few clusters showing a younger age . this will be discussed further in the next section . keeping in mind that the purpose of this work is to measure relative ages , and following @xcite , we calculated for each cluster a * normalised age * , defined as the ratio between the cluster age and the mean age of the group of metal poor clusters . more specifically , in case of the zw metallicity , we used as normalising factor the mean age of the clusters with @xmath2}_{\rm zw}<-1.7 $ ] , i.e. @xmath35 gyr . for the cg metallicity scale , the normalising age is equal to @xmath36 gyr , corresponding to the mean age of the clusters with @xmath2}_{\rm cg}<-1.4 $ ] . columns 1114 of tables [ tab1 ] and [ tab2 ] give the estimated normalised ages and errors for the hst snapshot and the groundbased sample , respectively . in both tables , col . 11 , 12 give the normalised ages , and the corresponding errors adopting the zw metallicity scale , and col . 13 , 14 the normalised ages and the corresponding errors for the cg metallicity scale . as it can be inferred from the values in tables [ tab1 ] and [ tab2 ] , and as it will be discussed in section 4 , the group of clusters at intermediate metallicity shows a large age dispersion . we have verified that the age differences derived from the vertical parameter are indeed correct . in particular , we focussed our attention on a few clusters with nearly the same metallicity , within @xmath37 dex , at @xmath2}_{\rm zw}\sim-1.3 $ ] . some of these clusters have been the subject of a long and lively discussion in the literature in the last 10 years or so ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) in the attempt to establish whether they are coeval or not . according to the vertical parameter , ngc 362 , ngc 1261 , ngc 1851 , ngc 2808 , ngc 5904 have all the same age , within the uncertainties ( 5 - 8% ) , and are 10 - 15% younger than another group of older ( and coeval among them ) clusters which includes ngc 288 , ngc 5946 , ngc 6218 , ngc 6121 , ngc 6266 , ngc 6362 , ngc 6717 . in order to check this observational evidence , in figures [ comp_ml1 ] and [ comp_ml2 ] we directly compare the hb population and the fiducial points reproducing the location of the main sequence and sub - giant branch ( sgb ) of the cmds of four couples of clusters selected among the above mentioned objects . the hb stars and the fiducial points have been shifted in magnitude in order to have the to point at magnitude 0 . one of the two clusters has been shifted in colour by the amount required to take the tos at the same colour . figure [ comp_ml1 ] shows this comparison for two couples of nearly coeval clusters ( ngc 362 - ngc 1261 and ngc 362 - ngc 1851 ) : all the sequences match remarkably well , and it is hard to infer any age difference among these objects . these clusters were also found to be nearly coeval by @xcite and @xcite . figure [ comp_ml2 ] compares two of these young clusters ( ngc 362 and ngc 1851 ) with ngc 6121 , which results older : despite the to and sgb regions match well , the hb levels significantly differ , as expected from our vertical parameters . the difference in magnitude between the hb levels is indicative of the different ages . the homogeneity of the observational database is a crucial ingredient for accurate relative age measurement . in this paper , we use two catalogues of cmds coming from rather different observational systems , with similar , but not identical bandpasses . this means that we can not directly compare the parameters obtained from the empirical cmds . even if the models used for the age determination have been appropriately transformed into the two observational systems before comparison with the observed parameters , it is important to verify that the ages coming from the two data sets are mutually consistent . for this reason , we took advantage of the @xmath1 clusters in common between the hst snapshot and the groundbased data set . figure [ snap - gbn ] shows the difference between the ages from the snapshot and the groundbased catalogues for the common clusters . the average difference is consistent with zero , and there is no trend with the cluster metal content . figure [ snap - gbn ] tells us that we can merge the two age catalogues , and use them together in analysing the relative ages of the galactic globular clusters . comparison with stellar models is critical for any cluster age determination . in principle , differences in the input physics ( i.e. equation of state , opacities ) input parameters ( i.e. the helium enrichment ratio @xmath38 ) and transformations from the theoretical to the observational plane , may have a significant impact on the final ages and the inferred star formation history of the galactic gc system . to assess the consistency of our relative age determination , at least within the recent theoretical framework , we compared in fig . [ confteo ] our normalised ages for the groundbased sample with the values estimated using the two isochrone sets already employed by @xcite , i.e. the straniero , chieffi , & limongi ( 1997 ) and @xcite isochrones calculated by interpolating the evolutionary tracks by @xcite . apart from the heavy element mixture , these two sets of isochrones differ from our reference models mainly with respect to the adopted equation of state , bolometric corrections , and helium abundance at a given metallicity . about this latter point , the models by @xcite use @xmath39 , whereas @xcite employ @xmath40 , and @xcite make use of @xmath41 . it is worth noticing that the initial he abundance ( @xmath42 ) adopted in the stellar computation has a strong effect on both the core h - burning lifetime and on the hb luminosity level . more in details , for a fixed age , when the value of @xmath42 increases , the to of the isochrone becomes fainter , whereas the hb luminosity level increases . as a consequence the value of the @xmath11 parameter increases . so for a fixed age , the adopted helium - enrichment ratio , affects the trend of the @xmath11 parameter with @xmath43 . the net effect on the age estimate is to make a cluster of a given metallicity younger when he increases . therefore , an increase of the helium - enrichment ratio enhances , if present , any age - metallicity relationship ( in the sense of making the more metal rich clusters younger in comparison with the more metal poor ones ) . given that @xcite models include a larger helium enrichment ratio with respect to @xcite , we expect that their use produces younger relative ages for the more metal rich clusters . the different helium - enrichment ratio is also the main reason for the different behaviours of the @xmath11 parameter with the metallicity at a fixed age . in addition , the @xcite models account for helium diffusion , which is neglected in the other isochrone sets . when deriving the theoretical @xmath11age@xmath2}$ ] calibration from the @xcite scaled solar isochrones , we used the property that the @xmath11 values at a given age are to a good approximation the same for both a scaled solar metal distribution and an @xmath10-enhanced one typical of gcs ( see , e.g. @xcite ) with the same metal mass fraction @xmath43 , at least for metal poor objects . bearing in mind that the relation between global metallicity @xmath44}$ ] and @xmath2}$ ] is different for @xmath10-element enhanced mixtures , the @xmath11age@xmath44}$ ] calibration obtained in this way was then transformed to a @xmath45age@xmath2}$ ] relationship using @xmath44}={\rm [ fe / h]}+0.3 $ ] , which is appropriate for @xmath46}=+0.4 $ ] . the comparison of the normalised ages discloses very good agreement for @xmath2}_{\rm zw}<-1.9 $ ] . moving towards higher @xmath2}$ ] our ages become increasingly younger up to @xmath2}_{\rm zw}\sim-1.1 $ ] . the maximum difference with the @xcite and @xcite models is @xmath47% and @xmath48% , respectively , of the same order of magnitude as the error bars due to the observational uncertainties . for more metal rich clusters the difference decreases to @xmath47% for the @xcite models , whereas the difference between our ages and those obtained using the @xcite models seems to be reversed , with the @xcite ages younger than ours by @xmath47% for the three clusters at @xmath2}\sim-0.6 $ ] . we have also compared our normalised ages with the results obtained using the @xcite models . in this case , there is generally a better agreement between the two age sets , with some differences ( always smaller than @xmath49% ) at the two extremes of the metallicity interval covered by our data set . despite these discrepancies , for the purposes of this paper , the comparisons in fig . [ confteo ] show that there is a substantial agreement among the ages obtained with recent stellar evolution models : the overall picture of the age distribution discussed in the next section would not change if we were to consider the normalised ages obtained from the @xcite , @xcite or @xcite models . it is the size of the trends of the average normalised ages with metallicity that would marginally change , although an agreement at the level of @xmath50% ( comparable with the observational errors ) among rather different model sets is encouraging . by contrast the trend of the age dispersion with metallicity is more solid , as it mainly depends on the dispersion of the observed parameters , and much less ( and in any case much less than the observational errors ) on the adopted models . figures [ figmna2_zw ] and [ figmna2_cg ] plot the normalised ages as a function of the metallicity , and of the distance from the galactic centre for the two adopted metallicity scales . different symbols represent the different metallicity ranges listed in the figure caption . open symbols represent the hst snapshot data , whereas filled symbols are for the groundbased clusters . for the clusters in common between the two catalogues , we plot the age obtained from the vertical parameter measured on the hst snapshot cmds . the dashed line represents the zero point : clusters with a normalised age equal to @xmath51 , have an age of @xmath35 gyr and @xmath36 gyr for the zw and cg metallicity scales , respectively . the overall trend already displayed by figures [ c04aesnap ] and [ c04aegb ] is more easily visible in figures [ figmna2_zw ] and [ figmna2_cg ] : the most metal poor clusters ( @xmath2}_{\rm zw}<-1.7 $ ] , @xmath2}_{\rm cg}<-1.4 $ ] ) are all older , and coeval , with an age dispersion smaller than @xmath3 gyr , which is compatible with a null age dispersion , if we take into account the measurement errors . the clusters with @xmath4}_{\rm zw}<-0.8 $ ] ( @xmath52}_{\rm cg}<-0.8 $ ] ) are , on average , @xmath5 gyr and @xmath53 gyr respectively , if the models of @xcite and @xcite are adopted . ] younger , and show a larger age dispersion ( @xmath6 gyr ) , with an age total range of @xmath54 gyr . interestingly enough , @xmath55% ( @xmath49 out of @xmath56 ) of the clusters in this metallicity interval seems to be coeval with the most metal poor ones . the age dispersion is independent of the adopted stellar evolution model . more metal rich clusters have on average the same age as the intermediate metallicity ones , but apparently with a smaller age dispersion , though the small number of objects and the differences in the ages from the different model sets ( cf . fig [ confteo ] ) do not allow any firm conclusion for this group of clusters . there is no evidence for a dependence of age on the distance from the galactic centre , though all the clusters with @xmath57 kpc are old ( but these are all metal poor clusters ) . in the following we consider a few peculiar objects . on average , we note that the metal poor clusters are the oldest clusters . however , among them , two clusters ( ngc 4590 and ngc 7078 ) seem to be about @xmath58% younger . interestingly enough , @xcite listed these two gcs among a group of low - metallicity clusters that display a planar alignment in the outer halo , probably the consequence of an accretion event from a galactic satellite system . on the other hand , in the same paper also ngc 7099 and ngc 5024 are indicated as members of the same group , while in our study they seem to be coeval with the remaining metal poor clusters . there are a few intermediate metallicity clusters that show a much younger age , namely ngc 362 ( the youngest of the whole sample , with an age @xmath59% younger than the zero relative age level ) , ngc 1261 , ngc 2808 , ngc 3201 and ngc 1851 . all these clusters were already known to be younger objects @xcite . figure 1 in @xcite displays the position of some galactic clusters in galactocentric coordinates indicating the position of two planes passing in the vicinity of some satellite galaxies of the milky way . ngc 1261 , ngc 2808 and ngc 1851 are located very close to the fornax - leo - sculptor plane . the galactocentric coordinates of ngc 362 and ngc 3201 can be calculated starting from the galactic coordinates ( @xmath60 ) , and the distance from the galactic centre ( * ? ? ? * in the revised version of 2003 ) : both clusters are found to be very close to the same plane . none of the clusters suspected of being members of the sagittarius stream @xcite , and present in our catalogue ( ngc 288 , ngc 4147 , ngc 5053 , ngc 5466 , ngc 5634 , ngc 7089 ) , seems to be significantly younger than the average of the other clusters with similar metallicity , though it is well established that other clusters of the same stream are definitely young , e.g. pal 12 @xcite , ruprecht 106 @xcite , terzan 7 @xcite . we presented a catalogue of homogeneous , relative ages obtained from the so called vertical parameter for a sample of galactic gcs . the analysis of the relative age distribution gives some hints on the origin of the gc system of our galaxy . in particular , we have shown that the low metallicity clusters ( @xmath2}\leq-1.7 $ ] ) are the oldest gcs . the low metallicity clusters show a very small age dispersion ( @xmath61 gyr ) , compatible with a null age dispersion , if we account for the measurement errors . among the low - metallicity clusters , there are a couple of objects which are marginally younger . these objects are part of a suspected galactic stream , which might indicate accretion from outside the galaxy , or debris from the tidal disruption of a galactic satellite . the clusters at intermediate metallicities ( @xmath62}_{\rm zw}\le-0.9 $ ] ) are up to @xmath28 gyr younger than the more metal poor ones , with an age dispersion of @xmath63 gyr . a small sample of objects ( @xmath55% ) seems to be coeval with the most metal poor gcs . it must be explicitly noted that , while the age dispersion seems to be a solid result , the dependence of the _ mean _ age on metallicity and its value are model dependent . the empirical facts outlined above imply that most of the gcs with @xmath2}_{\rm zw}\ge-1.7 $ ] were born between @xmath64 and @xmath51 gyr after a first generation of more metal poor clusters . a few of the intermediate metallicity clusters are up to @xmath28 gyr younger than the oldest clusters in the same metallicity range and this last result is not model dependent . interestingly enough , all these young clusters seem to be related to some galactic streams , i.e. linked to some tidal disruption event ( of an external object or a smaller satellite ) , though these streams contain also some of the oldest known gcs . finally , all the metal rich clusters are at least @xmath51 gyr younger than the most metal poor ones , with a relatively small age dispersion . the metal rich sample is still too poor , however , to enable firmer conclusions to be reached . we do not see any correlation of the cluster ages with the galactocentric distance . however , all the clusters with galactocentric distances greater than @xmath65 kpc are old , with a very small age dispersion . these clusters are also in the metal poor sample . in concluding this paper , it is worth adding a final remark . other parameters , as helium variations , @xmath46}$ ] differences , cluster to cluster deep mixing ( due to internal rotation ) affect the vertical separation between the to and the zahb . we have already commented the effect of he - enrichment as a function of metallicity in section 3.3 . here , we note that there is no evidence of significant cluster to cluster he variations , as shown by salaris et al . ( 2004 ) using our snapshot database . however , recent results on the massive clusters @xmath66 centauri ( bedin et al . 2004 , piotto et al . 2005 ) and ngc 2808 ( dantona & caloi 2004 ) indicates that within the same cluster , there could be two generations of stars , with the second generation being helium enriched . if this occurrence is confirmed , and if the same scenario is present also in less massive clusters , the ages derived from the vertical parameter could be affected . as for the other parameters , we plotted our relative ages as a function of @xmath46}$ ] for the clusters for which this parameter has been measured , and we did not find any correlation . the effect of deep mixing variations is at the moment impossible to assess from the observational point of view . in conclusion , the main parameter able to explain our measured vertical parameter variations , at least inside the metallicity bins of figures [ figmna2_zw ] and [ figmna2_cg ] is still age . we warmly thank a.r . walker and e. sandquist for providing us their data for some clusters . we also thank douglas c. heggie for a careful reading of the manuscript , and the anonymous referee for useful comments . gp , sc acknowledge partial support from the ministero dellistruzione , universit e ricerca ( _ miur _ , prin2002 , prin2003 ) , and from the agenzia spaziale italiana ( _ asi _ ) . bellazzini , m. , ferraro , f. r. , & ibata , r. 2003 , , 125 , 188 bedin , l. r. , piotto , g. , anderson , j. , cassisi , s. , king , i. r. , momany , y. , & carraro , g. 2004 , , 605 , l125 bergbusch , p. a. , & vandenberg , d. a. 2001 , , 556 , 322 buonanno , r. , corsi , c. e. , pulone , l. , fusi pecci , f. , & bellazzini , m. 1998 , , 333 , 505 buonanno , r. , corsi , c. e. , pecci , f. f. , richer , h. b. , & fahlman , g. g. 1995a , , 109 , 650 buonanno , r. , corsi , c. e. , pulone , l. , pecci , f. f. , richer , h. b. , & fahlman , g. c. 1995b , , 109 , 663 buonanno , r. , buscema , g. , fusi pecci , f. , richer , h. b. , & fahlman , g. g. 1990 , , 100 , 1811 carretta , e. , cohen , j. g. , gratton , r. g. , & behr , b. b. 2001 , , 122 , 1469 carretta , e. & gratton , r. g. 1997 , , 121 , 95 cassisi , s. & salaris , m. 1997 , , 285 , 593 cassisi , s. , salaris , m. , castelli , f. , & pietrinferni , a. 2004 , , astro - ph/0408111 chaboyer , b. , demarque , p. , kernan , p. j. , & krauss , l. m. 1998 , , 494 , 96 dantona , f. , & caloi , v. 2004 , , 611 , 871 fusi pecci , f. , bellazzini , m. , cacciari , c. , & ferraro , f. r. 1995 , , 110 , 1664 girardi , l. , bressan , a. , bertelli , g. , & chiosi , c. 2000 , , 141 , 371 gratton , r. g. , bragaglia , a. , carretta , e. , clementini , g. , desidera , s. , grundahl , f. , & lucatello , s. 2003 , , 408 , 529 harris , w. e. 1996 , , 112 , 1487 hoyle , f. & schwarzschild , m. 1955 , , 2 , 1 pietrinferni , a. , cassisi , s. , salaris , m. , & castelli , f. 2004 , , 612 , 168 piotto , g. , et al . 2002 , , 391 , 945 piotto , g. , et al . 2005 , apj , astro - ph/0412016 recio - blanco , a. , et al . 2005 , , astro - ph/0408462 renzini , a. & fusi pecci , f. 1988 , , 26 , 199 riello , m. , et al . 2003 , , 410 , 553 rosenberg , a. , aparicio , a. , saviane , i. , & piotto , g. 2000a , , 145 , 451 rosenberg , a. , piotto , g. , saviane , i. , & aparicio , a. 2000b , , 144 , 5 rosenberg , a. , saviane , i. , piotto , g. , & aparicio , a. 1999 , , 118 , 2306 rosenberg , a. , saviane , i. , piotto , g. , & held , e. v. 1998 , , 339 , 61 rutledge , g. a. , hesser , j. e. , stetson , p. b. , mateo , m. , simard , l. , bolte , m. , friel , e. d. , & copin , y. 1997 , , 109 , 883 salaris , m. , riello , m. , cassisi , s. , & piotto , g. 2004 , , 420 , 911 salaris , m. & weiss , a. 2002 , , 388 , 492 salaris , m. & weiss , a. 1998 , , 335 , 943 salaris , m. , chieffi , a. , & straniero , o. 1993 , , 414 , 580 sandage , a. r. & schwarzschild , m. 1952 , , 116 , 463 sandquist , e. l. , bolte , m. , stetson , p. b. , & hesser , j. e. 1996 , , 470 , 910 sarajedini , a. , chaboyer , b. , & demarque , p. 1997 , , 109 , 1321 stetson , p. b. , vandenberg , d. a. , & bolte , m. 1996 , , 108 , 560 straniero , o. , chieffi , a. , & limongi , m. 1997 , , 490 , 425 vandenberg , d. a. 2000 , , 129 , 315 vandenberg , d. a. , swenson , f. j. , rogers , f. j. , iglesias , c. a. , & alexander , d. r. 2000 , , 532 , 430 walker , a. r. 1994 , , 108 , 555 walker , a. r. 1998 , , 116 , 220 yoon , s. & lee , y. 2002 , science , 297 , 578 zinn , r. & west , m. j. 1984 , , 55 , 45 , but adopting the cg metallicity scale . circles refer to clusters with [ fe / h]@xmath71 ; triangles to clusters with @xmath72[fe / h]@xmath73 , and diamonds to clusters with [ fe / h]@xmath74.[figmna2_cg],scaledwidth=100.0% ] ccl i d & metallicity interval & cmd source + ngc 4590 & @xmath75[fe / h]@xmath76 & @xcite + ngc 5272 & @xmath77[fe / h ] @xmath78 & @xcite + ngc 5904 & @xmath79[fe / h]@xmath80 & sandquist ( priv . comm . ) + ngc 1851 & @xmath81[fe / h]@xmath82 & @xcite + ngc 6362 & @xmath83[fe / h]@xmath84 & walker ( priv . comm . ) + cccccccccccccc ic 4499 & -1.50 & -1.29 & 15.7 & 21.26 & 0.11 & 17.75 & 0.03 & 3.51 & 0.11 & 1.02 & 0.11 & 1.01 & 0.12 + ngc 104 & -0.71 & -0.78 & 7.4 & 17.63 & 0.06 & 14.10 & 0.03 & 3.53 & 0.06 & 0.97 & 0.07 & 0.99 & 0.07 + ngc 362 & -1.33 & -1.09 & 9.3 & 18.75 & 0.07 & 15.50 & 0.03 & 3.25 & 0.08 & 0.74 & 0.06 & 0.74 & 0.06 + ngc 1261 & -1.32 & -1.08 & 18.2 & 20.11 & 0.05 & 16.85 & 0.03 & 3.26 & 0.06 & 0.75 & 0.05 & 0.75 & 0.05 + ngc 1851 & -1.23 & -1.03 & 16.7 & 19.49 & 0.03 & 16.15 & 0.03 & 3.34 & 0.05 & 0.80 & 0.05 & 0.81 & 0.04 + ngc 1904 & -1.67 & -1.37 & 18.8 & 19.61 & 0.08 & 16.25 & 0.03 & 3.36 & 0.09 & 0.90 & 0.08 & 0.87 & 0.08 + ngc 2808 & -1.36 & -1.11 & 11.0 & 19.63 & 0.09 & 16.35 & 0.03 & 3.28 & 0.10 & 0.77 & 0.08 & 0.76 & 0.08 + ngc 3201 & -1.53 & -1.24 & 9.0 & 18.08 & 0.04 & 14.85 & 0.03 & 3.23 & 0.05 & 0.77 & 0.04 & 0.73 & 0.04 + ngc 4147 & -1.77 & -1.50 & 21.3 & 20.46 & 0.09 & 16.98 & 0.03 & 3.48 & 0.09 & 1.03 & 0.08 & 1.01 & 0.09 + ngc 4372 * & -2.03 & -1.88 & 7.1 & 18.94 & 0.05 & 15.58 & 0.03 & 3.36 & 0.06 & 0.98 & 0.05 & 0.98 & 0.05 + ngc 4590 & -2.11 & -2.00 & 10.1 & 19.04 & 0.06 & 15.73 & 0.03 & 3.31 & 0.07 & 0.92 & 0.06 & 0.93 & 0.06 + ngc 4833 & -1.92 & -1.71 & 6.9 & 19.14 & 0.05 & 15.70 & 0.03 & 3.44 & 0.06 & 1.01 & 0.05 & 1.00 & 0.06 + ngc 5024 & -2.04 & -1.86 & 18.8 & 20.27 & 0.08 & 16.83 & 0.03 & 3.44 & 0.08 & 1.02 & 0.07 & 1.03 & 0.07 + ngc 5694 & -1.93 & -1.74 & 29.1 & 22.01 & 0.08 & 18.53 & 0.03 & 3.48 & 0.08 & 1.05 & 0.07 & 1.05 & 0.07 + ngc 5824 & -1.85 & -1.60 & 25.8 & 21.98 & 0.03 & 18.53 & 0.03 & 3.45 & 0.04 & 1.01 & 0.04 & 1.00 & 0.04 + ngc 5904 & -1.38 & -1.12 & 6.2 & 18.49 & 0.07 & 15.15 & 0.03 & 3.34 & 0.08 & 0.83 & 0.07 & 0.81 & 0.07 + ngc 5927 * & -0.32 & -0.64 & 4.5 & 20.25 & 0.05 & 16.79 & 0.03 & 3.46 & 0.06 & 0.94 & 0.06 & 0.92 & 0.06 + ngc 5946 & -1.39 & -1.22 & 7.4 & 21.01 & 0.15 & 17.60 & 0.03 & 3.41 & 0.15 & 0.90 & 0.14 & 0.89 & 0.15 + ngc 5986 & -1.65 & -1.35 & 4.8 & 20.08 & 0.05 & 16.70 & 0.03 & 3.38 & 0.06 & 0.91 & 0.05 & 0.88 & 0.06 + ngc 6171 & -1.09 & -0.95 & 3.3 & 19.23 & 0.07 & 15.70 & 0.03 & 3.53 & 0.07 & 0.98 & 0.07 & 0.99 & 0.08 + ngc 6218 & -1.40 & -1.14 & 4.5 & 18.25 & 0.07 & 14.80 & 0.03 & 3.45 & 0.08 & 0.94 & 0.08 & 0.92 & 0.08 + ngc 6235 & -1.46 & -1.18 & 2.9 & 20.41 & 0.14 & 17.00 & 0.03 & 3.41 & 0.15 & 0.91 & 0.14 & 0.89 & 0.15 + ngc 6266 & -1.23 & -1.02 & 1.7 & 19.76 & 0.05 & 16.30 & 0.03 & 3.46 & 0.06 & 0.92 & 0.06 & 0.92 & 0.06 + ngc 6273 * & -1.80 & -1.53 & 1.6 & 19.95 & 0.04 & 16.55 & 0.03 & 3.40 & 0.05 & 0.96 & 0.03 & 0.93 & 0.04 + ngc 6284 & -1.24 & -1.13 & 6.9 & 20.90 & 0.04 & 17.50 & 0.03 & 3.40 & 0.04 & 0.86 & 0.04 & 0.87 & 0.05 + ngc 6287 & -2.07 & -1.91 & 1.7 & 20.59 & 0.11 & 17.13 & 0.03 & 3.46 & 0.12 & 1.05 & 0.11 & 1.06 & 0.11 + ngc 6342 & -0.66 & -0.69 & 1.7 & 20.55 & 0.05 & 17.07 & 0.03 & 3.48 & 0.06 & 0.92 & 0.07 & 0.94 & 0.07 + ngc 6362 & -1.18 & -0.99 & 5.3 & 18.88 & 0.03 & 15.42 & 0.03 & 3.46 & 0.04 & 0.91 & 0.04 & 0.92 & 0.04 + ngc 6544 * & -1.48 & -1.20 & 5.4 & 18.58 & 0.09 & 15.25 & 0.03 & 3.33 & 0.09 & 0.84 & 0.08 & 0.81 & 0.09 + ngc 6584 & -1.51 & -1.30 & 7.0 & 19.94 & 0.10 & 16.60 & 0.03 & 3.34 & 0.10 & 0.85 & 0.09 & 0.83 & 0.09 + ngc 6637 & -0.72 & -0.78 & 1.6 & 19.50 & 0.09 & 16.04 & 0.03 & 3.46 & 0.09 & 0.89 & 0.10 & 0.91 & 0.10 + ngc 6652 & -0.99 & -0.97 & 2.4 & 19.52 & 0.06 & 16.06 & 0.03 & 3.46 & 0.07 & 0.89 & 0.07 & 0.92 & 0.07 + ngc 6681 & -1.64 & -1.35 & 2.1 & 19.15 & 0.10 & 15.75 & 0.03 & 3.40 & 0.10 & 0.93 & 0.09 & 0.91 & 0.10 + ngc 6717 & -1.33 & -1.09 & 2.3 & 19.24 & 0.07 & 15.80 & 0.03 & 3.44 & 0.08 & 0.92 & 0.08 & 0.91 & 0.08 + ngc 6723 & -1.12 & -0.96 & 2.6 & 19.06 & 0.11 & 15.55 & 0.03 & 3.51 & 0.11 & 0.96 & 0.11 & 0.97 & 0.12 + ngc 6838 & -0.58 & -0.73 & 6.7 & 18.01 & 0.05 & 14.54 & 0.03 & 3.47 & 0.06 & 0.91 & 0.06 & 0.92 & 0.06 + ngc 6934 & -1.54 & -1.32 & 14.3 & 20.28 & 0.06 & 16.95 & 0.03 & 3.33 & 0.07 & 0.85 & 0.06 & 0.83 & 0.06 + ngc 6981 & -1.50 & -1.21 & 12.9 & 20.22 & 0.08 & 16.90 & 0.03 & 3.32 & 0.08 & 0.83 & 0.07 & 0.80 & 0.08 + ngc 7078 & -2.13 & -2.02 & 10.4 & 19.21 & 0.09 & 15.88 & 0.04 & 3.33 & 0.10 & 0.94 & 0.08 & 0.95 & 0.08 + ngc 7089 & -1.61 & -1.31 & 10.4 & 19.44 & 0.10 & 16.03 & 0.03 & 3.41 & 0.11 & 0.94 & 0.10 & 0.91 & 0.11 + ngc 7099 & -2.05 & -1.92 & 7.1 & 18.73 & 0.05 & 15.23 & 0.03 & 3.50 & 0.05 & 1.08 & 0.06 & 1.10 & 0.06 + cccccccccccccc ngc 104 & -0.71 & -0.78 & 7.4 & 17.65 & 0.08 & 14.10 & 0.03 & 3.55 & 0.08 & 1.00 & 0.10 & 1.02 & 0.10 + ngc 288 & -1.40 & -1.14 & 12.0 & 18.92 & 0.04 & 15.48 & 0.03 & 3.45 & 0.05 & 0.94 & 0.05 & 0.92 & 0.05 + ngc 362 & -1.33 & -1.09 & 9.3 & 18.84 & 0.09 & 15.57 & 0.03 & 3.27 & 0.09 & 0.76 & 0.08 & 0.76 & 0.07 + ngc 1261 & -1.32 & -1.08 & 18.2 & 19.98 & 0.06 & 16.74 & 0.03 & 3.24 & 0.07 & 0.74 & 0.05 & 0.74 & 0.05 + ngc 1851 & -1.23 & -1.03 & 16.7 & 19.58 & 0.07 & 16.23 & 0.03 & 3.35 & 0.07 & 0.81 & 0.07 & 0.82 & 0.07 + ngc 1904 & -1.67 & -1.37 & 18.8 & 19.62 & 0.09 & 16.25 & 0.03 & 3.37 & 0.10 & 0.91 & 0.09 & 0.88 & 0.09 + ngc 2808 & -1.36 & -1.11 & 11.0 & 19.61 & 0.07 & 16.36 & 0.03 & 3.25 & 0.07 & 0.75 & 0.06 & 0.74 & 0.05 + ngc 3201 & -1.53 & -1.24 & 9.0 & 18.11 & 0.05 & 14.83 & 0.03 & 3.28 & 0.06 & 0.80 & 0.05 & 0.77 & 0.05 + ngc 4590 & -2.11 & -2.00 & 10.1 & 19.05 & 0.07 & 15.74 & 0.03 & 3.31 & 0.07 & 0.93 & 0.07 & 0.94 & 0.07 + ngc 5053 & -2.10 & -1.98 & 16.8 & 20.00 & 0.06 & 16.64 & 0.03 & 3.37 & 0.07 & 0.99 & 0.05 & 1.01 & 0.05 + ngc 5272 & -1.66 & -1.34 & 12.2 & 19.00 & 0.04 & 15.76 & 0.03 & 3.24 & 0.05 & 0.80 & 0.04 & 0.76 & 0.04 + ngc 5466 & -2.22 & -2.20 & 16.2 & 19.89 & 0.07 & 16.50 & 0.03 & 3.39 & 0.07 & 1.01 & 0.06 & 1.03 & 0.06 + ngc 5897 & -1.93 & -1.73 & 7.3 & 19.77 & 0.07 & 16.32 & 0.03 & 3.46 & 0.08 & 1.04 & 0.07 & 1.04 & 0.07 + ngc 5904 & -1.38 & -1.12 & 6.2 & 18.47 & 0.03 & 15.12 & 0.03 & 3.34 & 0.04 & 0.83 & 0.04 & 0.81 & 0.04 + ngc 6093 & -1.75 & -1.47 & 3.8 & 19.77 & 0.07 & 16.35 & 0.03 & 3.42 & 0.08 & 0.97 & 0.07 & 0.95 & 0.08 + ngc 6121 & -1.27 & -1.05 & 5.9 & 16.87 & 0.03 & 13.43 & 0.03 & 3.44 & 0.04 & 0.91 & 0.04 & 0.91 & 0.04 + ngc 6171 & -1.09 & -0.95 & 3.3 & 19.17 & 0.06 & 15.68 & 0.03 & 3.48 & 0.07 & 0.93 & 0.07 & 0.95 & 0.07 + ngc 6205 & -1.63 & -1.33 & 8.7 & 18.53 & 0.06 & 15.01 & 0.03 & 3.52 & 0.07 & 1.05 & 0.07 & 1.02 & 0.07 + ngc 6218 & -1.40 & -1.14 & 4.5 & 18.31 & 0.07 & 14.84 & 0.05 & 3.48 & 0.09 & 0.97 & 0.08 & 0.95 & 0.09 + ngc 6254 & -1.55 & -1.25 & 4.6 & 18.48 & 0.05 & 15.11 & 0.05 & 3.37 & 0.07 & 0.89 & 0.06 & 0.86 & 0.07 + ngc 6341 & -2.24 & -2.16 & 9.6 & 18.64 & 0.06 & 15.18 & 0.03 & 3.47 & 0.07 & 1.08 & 0.06 & 1.10 & 0.06 + ngc 6362 & -1.18 & -0.99 & 5.3 & 18.86 & 0.08 & 15.43 & 0.03 & 3.43 & 0.08 & 0.88 & 0.09 & 0.89 & 0.09 + ngc 6366 & -0.58 & -0.73 & 5.0 & 19.14 & 0.06 & 15.70 & 0.03 & 3.44 & 0.07 & 0.89 & 0.07 & 0.90 & 0.07 + ngc 6397 & -1.94 & -1.76 & 6.0 & 16.44 & 0.04 & 13.03 & 0.03 & 3.41 & 0.05 & 1.00 & 0.03 & 0.99 & 0.05 + ngc 6681 & -1.64 & -1.35 & 2.1 & 19.21 & 0.09 & 15.82 & 0.05 & 3.39 & 0.10 & 0.92 & 0.10 & 0.90 & 0.10 + ngc 6723 & -1.12 & -0.96 & 2.6 & 19.07 & 0.09 & 15.53 & 0.03 & 3.54 & 0.09 & 0.99 & 0.10 & 1.01 & 0.10 + ngc 6752 & -1.54 & -1.24 & 5.2 & 17.45 & 0.08 & 13.91 & 0.05 & 3.53 & 0.09 & 1.05 & 0.10 & 1.03 & 0.10 + ngc 6809 & -1.80 & -1.54 & 3.9 & 17.93 & 0.12 & 14.44 & 0.03 & 3.49 & 0.12 & 1.05 & 0.12 & 1.04 & 0.13 + ngc 6838 & -0.58 & -0.73 & 6.7 & 18.01 & 0.06 & 14.52 & 0.03 & 3.49 & 0.07 & 0.95 & 0.08 & 0.95 & 0.08 + ngc 7078 & -2.13 & -2.02 & 10.4 & 19.24 & 0.06 & 15.89 & 0.03 & 3.35 & 0.07 & 0.97 & 0.06 & 0.98 & 0.06 +

this paper is concerned with dynamic resource allocation in a cellular wireless network with slow fading for support of data traffic having heterogeneous transmission delay requirements . the multiple - input single - output ( miso ) fading broadcast channel ( bc ) is of interest where the base station ( bs ) employs multiple transmit antennas to realize simultaneous downlink transmission at the same frequency to multiple mobile users each having a single receive antenna . an information - theoretic approach is taken for characterizing capacity limits of the fading miso - bc under various transmission delay considerations . first , this paper studies transmit optimization at the bs when some users have delay - tolerant `` packet '' data and the others have delay - sensitive `` circuit '' data for transmission at the same time . based on the convex optimization framework , an online resource allocation algorithm is derived that is amenable to efficient cross - layer implementation of both physical ( phy ) -layer multi - antenna transmission and media - access - control ( mac ) -layer multiuser rate scheduling . secondly , this paper investigates the fundamental throughput - delay tradeoff for transmission over the fading miso - bc . by comparing the network throughput under completely relaxed versus strictly zero transmission delay constraint , this paper characterizes the limiting loss in sum capacity due to the vanishing delay tolerance , termed the _ delay penalty _ , under some prescribed user fairness for transmit rate allocation . broadcast channel ( bc ) , fading channel , multi - antenna , throughput - delay tradeoff , dynamic resource allocation , cross - layer optimization , convex optimization . [ section ] [ section ] [ section ] [ section ] [ section ] [ section ]

0804.3261

in mobile wireless networks , communications typically take place over time - varying channels . when this time - variation or fading is `` fast '' such that the channel state information ( csi ) is hardly obtainable at the transmitter , a classical approach for mitigating impairments of fading to transmission reliability is to apply _ diversity techniques _ , such as coded diversity , antenna diversity , and path diversity . on the other hand , when the fading channel changes sufficiently `` slowly '' such that the transmitter is able to acquire the csi , a general approach to compensate for the fading is _ dynamic resource allocation _ , whereby transmit resources such as power , bit - rate , antenna - beam and bandwidth are dynamically allocated based upon the fading distribution . effective implementation of dynamic resource allocation usually requires joint optimization of both physical ( phy ) -layer transmission and media - access - control ( mac ) -layer rate scheduling in the classical communication protocol stack , and thus demands for a new _ cross - layer _ design methodology . one challenging issue to be addressed for dynamic resource allocation in tomorrow s wireless networks is how to meet with user s heterogeneous transmission quality - of - service ( qos ) requirements . among others , the demand for wireless high - speed connectivity for both delay - tolerant `` packet '' data and delay - sensitive `` circuit '' data is expected to rise significantly in the next decade . therefore , study on both spectral and power efficient transmission schemes for support of _ heterogeneous delay - constrained data traffic _ becomes an important area for research . on the other hand , because tolerance for a larger transmission delay incurred to data traffic allows for more flexible transmit power and rate adaptation over time and thereby leads to a larger transmission throughput in the long term , there is in general a fundamental _ throughput - delay tradeoff _ associated with dynamic resource allocation over fading channels . characterization of such fundamental tradeoff is another important research problem because it reveals the ultimate gain achievable by dynamic resource allocation under realistic transmission delay requirements . this paper is aimed to provide concrete answers to the aforementioned problems by considering the fading broadcast channel ( bc ) that models the downlink transmission in a typical wireless cellular network . an information - theoretic approach is taken in this paper to address some fundamental limits of dynamic resource allocation for the fading bc under various transmission delay considerations . in particular , the fading multiple - input single - output ( miso ) bc is considered where multi - antennas are equipped at the transmitter of the base station ( bs ) , and single antenna at the receiver of each mobile user . because of multi - antennas at the transmitter , spatial multiplexing can be used at the bs to support simultaneous transmission to mobile users at the same frequency , named space - division - multiple - access ( sdma ) . a slow - fading environment is assumed , and for simplicity , the block - fading ( bf ) channel model is adopted . it is further assumed that the bs has perfect user csi at its transmitter , and is thus able to perform a centralized dynamic resource allocation based upon multiuser channel conditions . this paper s main contributions are summarized as follows : * this paper studies optimal dynamic resource allocation for the fading miso - bc when both no - delay - constrained ( ndc ) packet data and delay - constrained ( dc ) circuit data are required for transmission at the same time . a cross - layer optimization approach is taken for jointly optimizing capacity - achieving multi - antenna transmission at the phy - layer and fairness - ensured multiuser rate scheduling at the mac - layer . a convex optimization framework is formulated for minimizing the average transmit power at the bs subject to both ndc and dc user rate constraints . a _ two - layer lagrange - duality method _ is shown to be the key for solving this problem . based on this method , a novel online resource allocation algorithm that is amenable to efficient cross - layer implementation is derived , and its convergence behavior is validated . * this paper investigates the fundamental throughput - delay tradeoff for the fading miso - bc under optimal dynamic resource allocation . by taking the difference between the maximum sum - rate of users under ndc and dc transmission subject to the constraint that the rate portion allocated to each user needs to be regulated by the same prescribed _ rate - profile _ , the paper presents a novel characterization for the limiting loss in sum capacity due to the vanishing delay tolerance , termed the _ delay penalty _ , for the fading miso - bc . thereby , the delay penalty provides the answer to the following interesting question : comparing no delay constraint versus zero - delay constraint for all users in the network , how much is the maximum percentage of throughput gain achievable for _ all _ users by optimal dynamic resource allocation ? the capacity region under ndc or dc transmission for a fading single - input single - output ( siso ) bc has been characterized in @xcite , @xcite , and for a fading siso multiple - access channel ( mac ) in @xcite , @xcite . a similar scenario like in this paper with mixed ndc and dc transmission has also been considered in @xcite for the single - user multiple - input multiple - output ( mimo ) fading channel , and in @xcite for the fading siso - bc . the comparison of achievable rates between ndc and dc transmission has been considered in @xcite for the fading mimo - bc . however , none of the above prior work has considered transmit optimization with mixed ndc and dc data traffic for the fading miso - bc , which is addressed in this paper . on the other hand , throughput - delay and power - delay tradeoffs for communications over fading channels by exploiting the combined csi and data buffer occupancy at the transmitter have been intensively studied in the literature for both single - user and multiuser transmission ( e.g. , @xcite and references therein ) . in contrast to prior work , this paper studies the throughput - delay tradeoff from a new perspective by characterizing the fundamental delay penalty in the network throughput owing to stringent ( zero ) transmission delay constraint imposed by all users . the concept of rate - profile , or its equivalent definitions for specifying some certain fairness in user rate allocation have also been considered for the siso multiuser channel in @xcite , @xcite , and for the mimo multiuser channel in @xcite , @xcite . however , to the author s best knowledge , application of rate - profile for characterizing the delay penalty in a multiuser fading channel is a novelty of this paper . there has been recently a great deal of study on the real - time resource allocation algorithm , named proportional fair scheduling ( pfs ) ( e.g. , @xcite-@xcite and references therein ) , which maximizes the network throughput by exploiting the multiuser channel variation and at the same time , maintains certain fairness among users in rate allocation . however , pfs is unable to guarantee any prescribed user rate demand . in this paper , a novel online scheduling algorithm is proposed to ensure that all ndc and dc user rate demands are satisfied with the minimum transmit power consumption at the bs . the remainder of this paper is organized as follows . section [ sec : system model ] illustrates the fading miso - bc model and provides a summary of known information - theoretic results for it . section [ sec : mixed traffic ] addresses the optimal cross - layer dynamic resource allocation problem for support of simultaneous transmission of heterogeneous delay - constrained traffic . section [ sec : tradeoffs ] characterizes the fundamental throughput - delay tradeoff for the fading miso - bc . section [ sec : simulation results ] provides the simulation results . finally , section [ sec : conclusion ] concludes the paper . _ notation _ : this paper uses upper case boldface letters to denote matrices and lower case boldface letters to indicate vectors . for a square matrix @xmath0 , @xmath1 and @xmath2 are its determinant and inverse , respectively . for any general matrix @xmath3 , @xmath4 denotes its conjugate transpose . @xmath5 and @xmath6 indicate the identity matrix and the vector with all zero elements , respectively . @xmath7 denotes the euclidean norm of a vector @xmath8 . @xmath9 $ ] denotes statistical expectation over the random variable @xmath10 . @xmath11 denotes the @xmath12-dimensional real euclidean space and @xmath13 is its non - negative orthant . @xmath14 is the space of @xmath15 matrices with complex number entries . the distribution of a circularly - symmetric complex gaussian ( cscg ) vector with the mean vector @xmath8 and the covariance matrix @xmath16 is denoted by @xmath17 , and @xmath18 means `` distributed as '' . @xmath19 denotes the non - negative part of a real number @xmath20 . a miso - bc channel with @xmath21 mobile users each having a single antenna and a fixed bs having @xmath12 antennas is considered , as shown in fig . [ fig : system model ] . because of multi - antennas at the transmitter , the bs is able to employ sdma to transmit to multiple users simultaneously at the same bandwidth . it is assumed that the transmission to all users is synchronously divided into consecutive blocks , and the fading occurs from block to block but remains static within a block of symbols , i.e. , a block - fading ( bf ) model . furthermore , it is assumed that the fading process is stationary and ergodic . let @xmath10 be the random variable representing the fading state . at fading state @xmath10 , the miso - bc can be considered as a discrete - time channel represented by @xmath22=\left[\begin{array}{c } { \mbox{\boldmath{$ h $ } } } _ 1(n ) \\ { \mbox{\boldmath{$ h $ } } } _ k(n)\end{array}\right ] { \mbox{\boldmath{$ x $ } } } ( n)+\left [ \begin{array}{c } z_1(n ) \\ \vdots \\ z_k(n ) \end{array } \right],\end{aligned}\ ] ] where @xmath23 , @xmath24 and @xmath25 denote the received signal , the @xmath26 downlink channel vector , and the receiver noise for user @xmath27 , respectively , and @xmath28 denotes the transmitted signal vector from the bs . it is assumed that @xmath29 . the transmitted signal @xmath30 can be further expressed as @xmath31 where @xmath32 and @xmath33 represent the precoding vector and the transmitted codeword symbol for user @xmath27 , respectively , at fading state @xmath10 . it is assumed that each user employs the optimal gaussian code - book with normalized codeword symbols , i.e. , @xmath34 , and the rate of code - book for user @xmath27 at fading state @xmath10 is denoted as @xmath35 . the allocated transmit power to user @xmath27 at fading state @xmath10 is denoted by @xmath36 , and it can be easily verified that @xmath37 . the total transmit power from the bs at fading state @xmath10 is then expressed as @xmath38 . assuming full knowledge of the fading distribution , the bs is able to adapt the transmission power @xmath36 and rate @xmath35 ( could be both zero for some fading state @xmath10 ) allocated to user @xmath27 in order to exploit multiuser channel variations over time . assuming a long - term power constraint ( ltpc ) @xmath39 over different fading states , the average transmit power at the bs needs to satisfy @xmath40\leq p^{*}$ ] . supposing that @xmath41 is given , the achievable rates @xmath42 of users need to be contained in the corresponding capacity region of the miso - bc at fading state @xmath10 , denoted by @xmath43 . characterization of @xmath44 will become useful later in this paper when the issue on how to dynamically allocate transmit power and user rates at different fading states is addressed . in many cases , it is more convenient to apply the celebrated duality result between the gaussian bc and mac @xcite to transform the capacity region characterization for the original bc to that for its dual mac . assuming that in the dual simo - mac of the original miso - bc considered in this paper , each user employs the optimal gaussian code - book of rate @xmath45 and has a transmit power @xmath46 , @xmath47 , at fading state @xmath10 , by @xcite the capacity region of the dual simo - mac at fading state @xmath10 can be expressed as @xmath48 where @xmath49 $ ] . the duality result @xcite then states that an achievable rate region for the original miso - bc at fading state @xmath10 with total transmit power @xmath41 can be expressed as @xmath50 it was later shown in @xcite that the above achievable rate region @xmath51 is indeed the capacity region @xmath52 for the gaussian bc . by applying the above results , it follows that any rate - tuple @xmath42 that is achievable in the fading miso - bc by the user power - tuple @xmath53 is also achievable as @xmath54 , @xmath55 , in the dual fading simo - mac by the corresponding user power - tuple @xmath56 provided that @xmath57 . note that the power allocation @xmath36 for user @xmath27 in the original bc is not necessarily equal to @xmath46 in the dual mac . the transforms between @xmath56 and @xmath53 as well as the corresponding precoding vectors @xmath58 for the same set of achievable rates @xmath42 and @xmath54 can be found in @xcite , and are thus omitted in this paper for brevity . this section studies optimal dynamic resource allocation algorithms for the bf miso - bc to support simultaneous transmission of data traffic with heterogeneous transmit rate and delay constraints . first , section [ subsec : problem formulation ] provides the problem formulation . then , section [ subsec : proposed solution ] presents the solution based on the lagrange - duality method of convex optimization . at last , section [ subsec : implementation ] derives an online algorithm that is suitable for real - time implementation of the proposed solution . the following rule for transmission scheduling at the bs is considered . as illustrated in fig . [ fig : traffic model ] , each user s data arising from some higher layer application is first placed into a dedicated buffer . periodically , the bs removes some of the data from each user s buffer , jointly encodes them into a block of symbols , and then broadcasts the encoded block to all users through the miso - bc . for simplicity , it is assumed that all user s data arrive to their dedicated buffers synchronously at the beginning of each scheduling period . the data arrival processes of users are assumed to be stationary and ergodic , mutually independent , and also independent of their channel realizations . this paper considers two types of data traffic with very different delay requirements : one is the delay - tolerant packet data and the other is the delay - sensitive circuit data , for which the following assumptions are made : * for a user with packet data application , the data arrival process is not necessarily continuous in time and the amount of arrived data in each scheduling period may be variable . all data are stored in a buffer of a sufficiently large size such that data dropping due to buffer overflow does not occur . in order for the scheduler to optimally exploit the channel dynamics , the allocated transmit rate can be variable during each scheduling period . it is assumed that there is always a sufficient amount of backlogged data in the buffer for transmission . the scheduler needs to ensure that the transmit rate averaged over scheduling periods in the long run is no smaller than the average data arrival rate . however , the exact amount of delay incurred to transmitted data in the buffer is not guaranteed . * for a user with circuit data application , the data arrival process is continuous with a constant - rate during each scheduling period . the arrived data is stored in the buffer for only one scheduling period and then transmitted . therefore , the amount of delay incurred to transmitted data is minimal . however , the scheduler needs to ensure a constant - rate transmission independent of channel condition . let the users with packet data applications be represented by the set @xmath59 where ndc refers to no - delay - constrained , and the users with circuit data applications represented by @xmath60 where dc refers to delay - constrained . in this paper , we consider optimal dynamic resource allocation to minimize the average transmit power at the bs over different fading states subject to the constraint that all ndc and dc user rate demands are satisfied . recall that @xmath35 denotes the rate assigned to user @xmath27 by the scheduler at fading state @xmath10 . for a ndc user , it is required that the average transmit rate @xmath61 $ ] over fading states needs to be no smaller than its average data arrival rate @xmath62 . in contrast , for a dc user , the transmit rate @xmath35 at any fading state @xmath10 needs to satisfy its constant data arrival rate @xmath62 . by considering the dual simo - mac in section [ sec : system model ] with @xmath63 , optimal allocation of transmit rates and powers at different @xmath10 can be obtained by solving the following optimization problem ( * p1 * ) : @xmath64 \\ \mathtt { subject \ to } & & \mathbb{e}_n[r_k(n ) ] \geq r_k^ { * } , \ \forall k\in \mathcal{u}_{\rm ndc } \label{eq : vr constraint } \\ & & r_k(n ) \geq r_k^ { * } , \ \forall n , \ \forall k\in \mathcal{u}_{\rm dc } \label{eq : cr constraint } \\ & & { \mbox{\boldmath{$ r $ } } } ( n)\in\mathcal{c}^{\rm mac}_n\left(\{q_k(n)\ } , \{{\mbox{\boldmath{$ h $ } } } _ k^{\dag}(n)\}\right ) , \ \forall n \label{eq : capacity constraint } \\ & & q_k(n)\geq 0 \ \forall n , k . \label{eq : power constraint}\end{aligned}\ ] ] for the above problem , the objective function and all the constraints except ( [ eq : capacity constraint ] ) are affine . it can also be verified from ( [ eq : capacity region dual mac ] ) that @xmath65 is a convex set with any given positive @xmath56 . therefore , problem p1 is a convex optimization problem @xcite and thus can be solved using efficient convex optimization techniques , as will be shown next . the lagrange - duality method is usually applied when a convex optimization problem can be more conveniently solved in its dual domain than in its original form . in this paper , we also apply this method for solving problem p1 . the first step for the lagrange - duality method is to introduce dual variables associated with some constraints of the original problem . for problem p1 that has multiple constraints , there are also various ways to introduce dual variables that might result in different dual problems . for the following proposed solution , dual variables are chosen with an aim to facilitate implementing it in the real time , as will be explained later in section [ subsec : implementation ] . as a first step , a set of dual variables @xmath66 , @xmath67 , are introduced for the ndc users with respect to ( w.r.t . ) their average - rate constraints in ( [ eq : vr constraint ] ) . the lagrangian of problem p1 can be then expressed as @xmath68-\sum_{k\in\mathcal{u}_{\rm ndc}}\mu_k\left(\mathbb{e}_n[r_k(n)]- r_k^{*}\right).\end{aligned}\ ] ] denote the set of @xmath56 and @xmath54 specified by the remaining constraints in ( [ eq : cr constraint ] ) , ( [ eq : capacity constraint ] ) and ( [ eq : power constraint ] ) as @xmath69 , the lagrange dual function is expressed as @xmath70 the dual problem of the original ( primal ) problem can be then expressed as @xmath71 because the primal problem is convex and also satisfies the slater s condition @xcite , are achievable . ] the duality gap between the optimal value of the primal problem and that of the dual problem becomes zero . this suggests that the problem at hand can be equivalently solved in its dual domain by first minimizing the lagrangian @xmath72 to obtain the dual function @xmath73 for some given @xmath66 , and then maximizing @xmath73 over @xmath66 . considering first the minimization problem in ( [ eq : lagrange dual power min ] ) to obtain @xmath73 for some given @xmath66 . it is interesting to observe that this problem can be solved by considering parallel subproblems each corresponding to one fading state @xmath10 . from ( [ eq : lagrange power min ] ) , the subproblem for fading state @xmath10 can be written as ( * p2 * ) @xmath74 hence , the dual function @xmath73 can be obtained by solving subproblems all having the identical structure , a technique usually referred to as the _ lagrange - dual decomposition_. for solving problem p2 for each @xmath10 , a new set of positive dual variables @xmath75 , are introduced for dc users w.r.t . their constant - rate constraints in ( [ eq : cr constraint sub ] ) . the lagrangian of problem p2 can be then expressed as @xmath76 note that for brevity , the index @xmath10 is dropped in @xmath46 , @xmath45 and @xmath77 in ( [ eq : lagrange power min sub ] ) since it is applicable for all @xmath10 . the corresponding dual function can be then defined as @xmath78 where @xmath79 denotes the set of @xmath80 and @xmath81 specified by the remaining constraints ( [ eq : capacity constraint sub ] ) and ( [ eq : power constraint sub ] ) at fading state @xmath10 . the associated dual problem is then defined as @xmath82 similar like p1 , problem p2 can also be solved by first minimizing @xmath83 to obtain the dual function @xmath84 for a given set of @xmath85 , and then maximizing @xmath86 over @xmath85 . from ( [ eq : lagrange power min sub ] ) , the minimization problem in ( [ eq : lagrange dual power min sub ] ) can be expressed as @xmath87 let @xmath88 , if @xmath89 , and @xmath90 , if @xmath91 , and @xmath92 be a permutation over @xmath93 such that @xmath94 , and let @xmath95 . thanks to the polymatroid structure of @xmath65 @xcite , the above problem can be simplified as ( * p3 * ) @xmath96 problem p3 is convex because all the constraints are affine and the objective function is convex w.r.t . @xmath80 . hence , this problem can be solved , e.g. , by the interior - point method @xcite . in appendix [ appendix : algorithm ] , an alternative method based on the block - coordinate decent principle @xcite by iteratively optimizing @xmath97 with all the other @xmath98 as fixed is presented . this method can be considered as a generalization of the algorithm described in @xcite , where all @xmath99 , are equal . so far , solutions have been presented for the minimization problem in ( [ eq : lagrange dual power min ] ) to obtain @xmath73 for some given @xmath66 and that in ( [ eq : lagrange dual power min sub ] ) to obtain each @xmath84 for some given @xmath85 . next , the remaining issue on how to update @xmath66 to maximize @xmath73 for the dual problem in ( [ eq : dual problem ] ) is addressed . similar techniques can also be used for updating @xmath85 to maximize @xmath84 for each dual problem in ( [ eq : dual problem sub ] ) . from ( [ eq : lagrange power min ] ) and ( [ eq : lagrange dual power min ] ) , it is observed that the dual function @xmath73 , though affine w.r.t . @xmath66 , is not directly differentiable w.r.t . @xmath66 . hence , standard method like newton method can not be employed to update @xmath66 for maximizing @xmath73 . an appropriate choice here may be the _ sub - gradient - based _ method @xcite that is capable of handling non - differentiable functions . this method is an iterative algorithm and at each iteration , it requires a sub - gradient at the corresponding value of @xmath66 to update @xmath66 for the next iteration . suppose that after solving problem p2 for some given @xmath66 at all fading states of @xmath10 , the obtained rates and powers are denoted by @xmath100 and @xmath101 , respectively , @xmath102 . the following lemma then provides a suitable sub - gradient for @xmath66 : [ lemma : sub - gradient ] if @xmath103 , then the vector @xmath104 defined as @xmath105 $ ] for @xmath106 is a sub - gradient of @xmath107 at @xmath66 . since for any set of @xmath108 s , @xmath109 , it follows that @xmath110\right).\end{aligned}\ ] ] hence , it is clear that the optimal dual solution @xmath111 that maximizes @xmath73 should not lie in the half space represented by @xmath112 . as a result , any new update of @xmath66 for the next iteration , denoted by @xmath113 , should satisfy @xmath114 . by applying lemma [ lemma : sub - gradient ] , a simple rule for updating @xmath66 is as follows : is the ellipsoid method @xcite , which at each iteration removes the half space specified by @xmath115 for searching @xmath111 . ] @xmath116\right)\right\}^+ , \ \ k\in\mathcal{u}_{\rm ndc},\end{aligned}\ ] ] where @xmath117 is a small positive step size . similarly , at each fading state @xmath10 , @xmath118 can be also updated towards maximizing @xmath119 as @xmath120 where @xmath100 , @xmath121 , is the solution obtained after solving problem p3 at fading state @xmath10 . to summarize , the proposed solution is implemented by a _ two - layer _ lagrange - duality method . at the first layer , the algorithm searches iteratively for @xmath111 with which the average - rate constraints of all ndc users are satisfied . at each iteration , an update for @xmath66 is generated and then passed to the second layer where the algorithm starts a parallel search for @xmath122 , each for a fading state @xmath10 , such that the constant - rate constraints of all dc users are satisfied at all fading states . the resultant @xmath100 of ndc users is then passed back to the first layer for another update of @xmath66 . the overall algorithm is summarized in table [ table : minpower algorithm ] . the complexity of this algorithm can be derived as follows . supposing that the ellipsoid method is used to iteratively update dual variables , the required number of iterations for convergence has @xmath123 , where @xmath124 is the size of the problem . let @xmath125 and @xmath126 denote the size of @xmath59 and @xmath60 , respectively , where @xmath127 . therefore , the ellipsoid method will need @xmath128 iterations for obtaining @xmath111 and @xmath129 iterations for obtaining @xmath122 for all fading states , assuming the number of fading states @xmath10 is finite and is equal to @xmath130 . the complexity for solving problem p3 is @xmath131 by , e.g. , the interior - point method . hence , the total complexity of the algorithm is @xmath132 . at last , take note that the proposed algorithm jointly optimizes transmit powers and rates together with decoding orders ( determined by the magnitudes of @xmath133 and @xmath122 @xcite ) of users at all fading states for the dual fading simo - mac . by the bc - mac duality result @xcite , the optimal transmit powers , rates , precoding vectors as well as encoding orders of users for the original fading miso - bc can be obtained . one important issue yet to be addressed for implementing the proposed solution for problem p1 is how to relax its assumption on perfect knowledge of the distribution of fading state @xmath10 . notice that this knowledge is necessary for computing the achievable average rates @xmath134\}$ ] , @xmath102 , which are needed for updating the dual variables @xmath66 for ndc users in ( [ eq : mu update ] ) . in practice , although it is reasonable to assume each user s fading channel is stationary and ergodic , the space of fading states is usually continuous and infinite and hence it is infeasible for the bs to initially acquire the channel distribution information for all users at all fading states . even though this information is available for off - line implementation of the proposed solution , the computational complexity becomes unbounded as the number of fading states goes to infinity . therefore , in this paper a modified `` online '' algorithm is developed that is able to adaptively update @xmath66 towards @xmath111 as transmission proceeds over time . let @xmath135 denote the transmission block index , @xmath136 . the key for the online algorithm is to approximate the statistical average @xmath137 $ ] , @xmath106 , at time @xmath135 by a time average of transmitted rates up to time @xmath138 , denoted by @xmath139 $ ] , where @xmath140 $ ] is obtained as @xmath141=(1-\epsilon)\bar{r}_k[t-1]+\epsilon r_k[t],\ ] ] where @xmath142 $ ] is the transmitted rate at time @xmath135 , and @xmath143 , @xmath144 , is a parameter that controls the convergence speed of @xmath140\rightarrow \mathbb{e}_n[r'_k(n)]$ ] as @xmath145 . by replacing @xmath137 $ ] by @xmath139 $ ] in ( [ eq : mu update ] ) , @xmath146\}$ ] at time @xmath135 can be updated accordingly such that as @xmath147 it converges to @xmath111 because @xmath140\rightarrow r_k^ * , \forall k\in\mathcal{u}_{\rm ndc}$ ] . this modified online algorithm is summarized in table [ table : online algorithm ] . * cross - layer implementation : * the proposed online algorithm based on the lagrange - duality method is amenable to cross - layer implementation of both phy - layer transmission and mac - layer multiuser rate scheduling . one challenging issue for cross - layer optimization is on how to select useful and succinct information for different layers to exchange and share so as to optimize their individual operations . the lagrange - duality method provides a new design paradigm for efficient cross - layer information exchange . on the one hand , since the mac - layer has the knowledge of user rate demands @xmath148 as well as their delay requirements ( ndc or dc ) , it can update dual variables @xmath146\}$ ] and @xmath149\}$ ] accordingly ( see table [ table : online algorithm ] ) , and then pass them to the phy - layer for computing the desirable transmission rates of users @xmath150\}$ ] to meet with each user s specific rate demand . on the other hand , the phy - layer is able to provide the mac - layer the updated @xmath150\}$ ] that optimize the phy - layer transmission by solving problem p3 given @xmath146\}$ ] and @xmath149\}$ ] . therefore , by exchanging dual variables and transmission rates between phy- and mac - layers , cross - layer dynamic resource allocation can be efficiently implemented . * comparison with pfs : * it is interesting to draw a comparison between the proposed online algorithm and the well - known pfs algorithm . pfs is designed for real - time multiuser rate scheduling in a mobile wireless network to ensure some certain fairness for user rate allocation while maximizing the network throughput . pfs applies to packet data transmission and hence is equivalent to the transmission scenario considered in this paper when only ndc users are present . at each time @xmath135 , transmit rates @xmath150\}$ ] assigned to users by pfs maximize the weighted sum - rate of users @xmath151r_k[t]$ ] , where the weights are given by @xmath152=\frac{1}{\bar{r}_k[t-1]}$ ] and @xmath139 $ ] is the estimated average rate for user @xmath27 up to time @xmath138 the same as expressed in ( [ eq : rate update ] ) . using this rule , it has been shown ( e.g. , @xcite-@xcite and references therein ) that as @xmath147 , @xmath140\rightarrow r_k^*$ ] where @xmath62 is the average achievable rate for user @xmath27 in the long term . furthermore , pfs maximizes @xmath153 over the expected capacity region ( please refer to definition [ def : capacity region ] in section [ sec : tradeoffs ] ) and , hence , @xmath148 can be considered as the unique intersection of the surface specified by @xmath154 and the boundary of the expected capacity region . because of the @xmath155 function , the rate assignments among users by pfs are regulated in a balanced manner such that no user can be allocated an overwhelmingly larger rate than the others even if it has a superior channel condition . however , the achievable rates @xmath148 by pfs are not guaranteed to satisfy any desired rate demand of users . in contrast , the proposed online algorithm ensures that each ndc user s average - rate demand is satisfied by applying a different rule ( see table [ table : online algorithm ] ) for updating user weights @xmath146\}$ ] for the resource allocation problem ( see problem p3 ) to be solved at each @xmath135 . for a single - user bf channel , the expected capacity @xcite , @xcite and the delay - limited capacity @xcite can be considered as the fading channel capacity limits under two extreme cases of delay constraint . the former is always larger than the latter and their difference , termed the _ delay penalty _ , then characterizes a fundamental throughput - delay tradeoff for dynamic resource allocation over a single - user fading channel . the delay penalty may or may not be significantly large for a single - user fading channel . for example , for a siso - bf channel , the delay penalty can be substantial because the delay - limited capacity is indeed zero if the fading channel is not `` invertible '' with a finite average transmit power @xcite . however , when multi - antennas are employed at the transmitter and/or receiver , the delay - limited capacity of the mimo - bf channel can be very close to the channel expected capacity @xcite , i.e. , a negligible delay penalty . this result can be explained by the `` channel - hardening '' effect @xcite for random mimo channels , i.e. , the mutual information of independent mimo channels becomes less variant because of antenna - induced space diversity , and hence the value of power and rate adaptation over time vanishes as the number of antennas becomes large . this result indicates that from an information - theoretic viewpoint , mimo channels are highly suitable for transmission of real - time and delay - constrained data traffic . characterization of the delay penalty in a multiuser fading channel is more challenging . the capacity definitions for the single - user fading channel can be extended to the multiuser channel as the _ expected capacity region _ under no delay constraint for all users , and the _ delay - limited capacity region _ under zero - delay constraint for all users . therefore , the delay penalty can be measured by directly comparing these two capacity regions . since capacity region contains the achievable rates of all users , it lies in a @xmath21-dimensional space where @xmath21 is the number of users in the network . as a result , characterization of capacity region becomes inconvenient as @xmath21 becomes large . in order to overcome this difficulty , prior research work usually adopts the maximum sum - rate of users over the capacity region , termed the _ sum capacity _ , as a simplified measure for the network throughput . applying the sum capacity to the corresponding capacity region , the _ expected throughput _ and the _ delay - limited throughput _ can be defined accordingly . however , the conventional sum capacity does not consider the rate allocation among users . as a consequence , each user s allocated rate portion in the expected throughput may be different compared to that in the delay - limited throughput . hence , a similar measure for the delay penalty like in the single - user case by taking the difference between the expected and delay - limited throughput looks problematic at a first glance in the multiuser case . this section presents a novel characterization of the fundamental throughput - delay tradeoff for the fading miso - bc . instead of considering mixed ndc and dc transmission like section [ sec : mixed traffic ] , it is assumed here that there are only ndc or dc users present , and comparison of the network throughput under these two extreme cases of delay constraint is of interest . because it is hard to compare directly the expected and the delay - limited capacity region as the number of users becomes large , the sum capacity is also considered for simplicity . however , unlike the conventional sum capacity that does not guarantee the amount of rate allocation among users , the expected and delay - limited throughput in this paper are defined under a new constraint that regulates each user s allocated rate portion based upon a prescribed _ rate - profile_. then , the delay penalty is characterized by the difference between the expected and delay - limited throughput _ under the same rate - profile_. the above concepts are more explicitly defined as follows : [ def : capacity region ] the expected capacity region for a fading miso - bc expressed in ( [ eq : mimo bc ] ) under a ltpc @xmath39 can be defined as @xmath156 , \forall k , { \mbox{\boldmath{$ r $ } } } ( n)\in\mathcal{c}^{\rm bc}_n(p(n),\{{\mbox{\boldmath{$ h $ } } } _ k(n)\ } ) , \forall n , \mathbb{e}_n[p(n)]\leq p^*\right\}.\end{aligned}\ ] ] similarly , the delay - limited capacity region is defined as @xmath157\leq p^*\right\}.\end{aligned}\ ] ] let @xmath62 denote @xmath27-th user s rate demand ( average - rate for a ndc user or constant - rate for a dc user ) @xmath47 , the rate profile is defined as a vector @xmath158 , where @xmath159 . [ def : throughput ] the expected throughput @xmath160 ( delay - limited throughput @xmath161 ) associated with a prescribed rate - profile @xmath162 under a ltpc @xmath39 is defined as the maximum sum - rate of users over the expected ( delay - limited ) capacity region under the constraint that the average ( constant ) transmit rate of each user @xmath163 must satisfy @xmath164 . [ def : delay penalty ] for some given delay profile @xmath162 and ltpc @xmath39 , the delay penalty @xmath165 is equal to @xmath166 . the proposed definition for delay penalty is illustrated in fig . [ fig : delay penalty ] for a 2-user case . from definition [ def : delay penalty ] , it is noted that characterization of @xmath165 requires that of both expected and delay - limited throughput . in the next , we present the algorithm for characterizing @xmath160 for some given @xmath39 and @xmath162 . similar algorithm can also be developed for characterizing @xmath161 and is thus omitted here for brevity . according to definition [ def : throughput ] and ( [ eq : expected capacity region ] ) and using the bc - mac duality result in section [ sec : system model ] , the expected throughput @xmath160 can be obtained by considering the following optimization problem ( * p4 * ) : @xmath167 \geq c_e\alpha_k , \\ & & { \mbox{\boldmath{$ r $ } } } ( n)\in\mathcal{c}^{\rm mac}_n\left(\{q_k(n)\ } , \{{\mbox{\boldmath{$ h $ } } } _ k^{\dag}(n)\}\right ) , \ \forall n \\ & & q_k(n)\geq 0 \ \forall n , k . \\ & & \mathbb{e}_n\left[\sum_{k=1}^k q_k(n)\right]\leq p^*\end{aligned}\ ] ] similar like problem p1 , it can be verified that the above problem is also convex , and hence can be solved using convex optimization techniques . here , instead of solving problem p4 from a scratch , the proposed solution transforms this problem into a special form of problem p1 and hence the same algorithm for problem p1 can be applied . first , considering the following transmit power minimization problem ( * p5 * ) for support of any arbitrary set of average - rate demands @xmath168 that satisfy a given rate - profile constraint @xmath162 , i.e. , @xmath169 , @xmath170 : @xmath64 \\ \mathtt { subject \ to } & & \mathbb{e}_n[r_k(n ) ] \geq r_{\rm sum}^*\alpha_k , \ \forall k , \\ & & { \mbox{\boldmath{$ r $ } } } ( n)\in\mathcal{c}^{\rm mac}_n\left(\{q_k(n)\ } , \{{\mbox{\boldmath{$ h $ } } } _ k^{\dag}(n)\}\right ) , \ \forall n \\ & & q_k(n)\geq 0 \ \forall n , k.\end{aligned}\ ] ] note that @xmath171 . it is observed that the above problem is a special case of problem p1 if all users have ndc transmission , i.e. , those constant - rate constraints for dc users in ( [ eq : cr constraint ] ) are removed . hence , problem p5 can also be solved by the proposed algorithm in table [ table : minpower algorithm ] . let @xmath172 denote the minimal transmit power obtained after solving problem p5 . notice that @xmath160 is a non - decreasing function of @xmath39 with some given @xmath162 because the expected capacity region @xmath173 corresponding to a larger @xmath39 always contains that with a smaller @xmath39 . hence , if @xmath174 , it can be inferred that @xmath160 must be larger than the assumed @xmath175 . otherwise , @xmath176 . by using this property , @xmath160 can be easily obtained by a bisection search @xcite . * fairness penalty : * there is an interesting relationship between the conventional sum capacity over the expected capacity region , and the expected throughput as a function of delay - profile , as shown in fig . [ fig : sum rate capacity ] for a 2-user case . the sum capacity is obtained by maximizing the sum - rate of users over the expected capacity region so as to maximally exploit the multiuser diversity gain in the achievable network throughput . however , it does not guarantee the resultant rate allocation among users . let the resultant rate portion allocated to each user in the sum capacity be specified by a rate - profile vector @xmath177 . in contrast , the expected throughput maximizes the sum - rate of users under any arbitrary rate - profile vector @xmath178 . due to this hard fairness constraint , the expected throughput in general is smaller than the sum capacity if @xmath178 is different from @xmath177 , and their difference can be used as a measure of the _ fairness penalty _ , which is denoted by @xmath179 . similarly , the fairness penalty can also be defined in the delay - limited case . * asymptotic results : * consider the miso - bc with asymptotically large number of users @xmath21 but fixed number of transmit antennas @xmath12 at the bs . it is assumed that the network is homogeneous where all users have mutually independent but identically distributed channels , and have identical rate demands , i.e. , @xmath180 . under these assumptions , in @xcite it has been shown that as @xmath181 the expected throughput @xmath182 under any finite power constraint @xmath39 scales like @xmath183 . in the following theorem , we provide this asymptotic result for the delay - limited throughput : [ theorem : dlc throughput ] under the assumption of symmetric fading and symmetric user rate demand , the delay - limited throughput @xmath184 for a fading miso - bc under a ltpc @xmath39 is upper - bounded by @xmath185 as @xmath181 , where @xmath186 is a constant depending solely on the channel distribution . please refer to appendix [ appendix : proof theorem ] . the above results suggest very different behaviors of the achievable network throughput with ndc versus dc transmission as @xmath21 becomes large . on the one hand , for the expected throughput , transmission delay is not an issue and hence the optimal strategy is to select only a subset of users with the best joint channel realizations for transmission at one time . the expected throughput thus scales linearly with @xmath12 and in double - logarithm with @xmath21 , for which the former is due to spatial multiplexing gain and the latter arises from the multiuser - diversity gain . on the other hand , in the delay - limited case , a constant - rate transmission needs to be ensured for all users . consequently , as the number of users increases , though more degrees of freedom are available for optimizing transmit parameters such as precoding vectors and encoding order of users , the delay - limited throughput is eventually saturated . the above comparison demonstrates that for a sdma - based network with a large user population , transmission delay can be a critical factor that prevents from achieving the maximum asymptotic throughput . however , notice that this may not be the case for the network having similar @xmath12 and @xmath21 , as will be verified later by the simulation results . in this section , simulation results are presented for evaluating the performances of dynamic resource allocation for the fading miso - bc under various transmission delay considerations . since the network throughput is contingent on transmit delay requirements as well as many other factors such as number of transmit antennas at the bs , number of mobile users , user channel conditions and rate requirements , and the transmit power constraint at the bs , various combinations of these factors are considered in the following simulations with an aim to demonstrate how transmission delay interplays with other factors in determining the achievable network throughput . the simulation results are presented in the following subsections . note that in the following simulation , user channel vectors @xmath187 are independently generated from the population of cscg vectors , and if not stated otherwise , it is assumed that @xmath188 . first , consider a miso - bc with @xmath189 with two users having ndc transmission and the other two having dc transmission . for convenience , it is assumed that the target average transmit rates for the two users with ndc transmission are both equal to @xmath190 , and the target constant rates for the two users with dc transmission both equal to @xmath191 . let @xmath192 denote the _ loading factor _ representing the ratio of the total amount of ndc traffic to that of the sum of ndc and dc traffic , i.e. , @xmath193 . if the proposed algorithm in table [ table : minpower algorithm ] that achieves optimal dynamic resource allocation is used , the required average transmit power at the bs is expected to be the minimum for satisfying both ndc and dc user rate demands for any given @xmath192 . for purpose of comparison , two suboptimal transmission schemes are also considered : * _ time - division - multiple - access ( tdma ) : _ a simple transmission scheme is to divide each transmission period into @xmath21 consecutive equal - duration time - blocks , each dedicated for transmission of one user s data traffic . if coherent precoding is applied at each fading state @xmath10 , i.e. , the precoder for user @xmath27 , defined as @xmath194 , is equal to @xmath195 , it can be shown that the miso - bc is decomposable into @xmath21 single - user siso channels . depending on each user s delay requirement , conventional water - filling power control @xcite and channel - inversion power control @xcite can be applied over different fading states for users with ndc and dc transmission , respectively , to achieve the minimum average transmit power under the given user rate demand . * _ zero - forcing ( zf ) -based sdma : _ another possible transmission scheme is also based on the sdma principle , i.e. , supporting all user s transmission simultaneously by multi - antenna spatial multiplexing at the bs . however , instead of using the dirty - paper - coding ( dpc)-based non - linear precoding assumed in this paper , a simple zf - based linear precoding is employed at the bs . the zf - based precoder @xmath196 for user @xmath27 at each fading state @xmath10 maximizes the user s own equivalent channel gain @xmath197 subject to the constraint that its associated co - channel interference must be completely removed , i.e. , @xmath198 . like tdma , zf - based precoding also decomposes the miso - bc into @xmath21 ( assuming @xmath199 ) single - user siso channels , and thereby optimal single - user power control schemes can be applied . in fig . [ fig : mixed traffic high power ] and fig . [ fig : mixed traffic low power ] , these three schemes are compared for two cases of network throughput , one corresponding to the sum - rate of users @xmath200 bits / complex dimension , and one corresponding to 2 bits / complex dimension . the required average transmit power over 500 randomly generated channel realizations is plotted versus the loading factor @xmath192 . first , in fig . [ fig : mixed traffic high power ] it is observed that in the case of high throughput ( bandwidth - limited ) , zf - based sdma outperforms tdma because of its larger spectral efficiency by spatial multiplexing . however , in the case of low throughput ( power - limited ) , in fig . [ fig : mixed traffic low power ] it is observed that tdma achieves better power efficiency than zf - based sdma . this is because coherent precoding in tdma provides more diversity and array gains , which become more dominant over spatial multiplexing gains at the power - limited regime . secondly , it is observed that in both cases of high and low throughput , the proposed scheme always outperforms both tdma and zf - based sdma given any loading factor @xmath192 . this is because the proposed scheme optimally balances the achievable spatial multiplexing , array , and diversity gains for the fading miso - bc . thirdly , it is observed that for all schemes , the required transmit power associated with some @xmath192 , @xmath201 , is always larger than that with @xmath202 ( e.g. , comparing @xmath203 and @xmath204 ) , i.e. , given the same portion of data traffic in the total traffic , ndc traffic has a better power efficiency than dc traffic . this is because ndc traffic allows for more flexible dynamic resource allocation than dc traffic and thus leads to a better power efficiency . the convergence of the online algorithm in table [ table : online algorithm ] is validated by simulations . for simplicity , it is assumed that only ndc users are present in the network . a miso - bc with @xmath205 and @xmath206 is considered . the target average - rates for user-1 and user-2 are 3 and 1 bits / complex dimension , respectively . the online algorithm is implemented for 3000 consecutive transmissions with randomly generated channel realizations . initially , the dual variables are set as @xmath207=\mu_2[0]=1 $ ] , and the estimated average transmit rates as @xmath208=\bar{r}_2[0]=0 $ ] . the updates for @xmath209\}$ ] and @xmath146\}$ ] as @xmath135 proceeds are shown in fig . [ fig : rbar online ] and fig . [ fig : mu online ] , respectively , for @xmath210 . it is observed that the online algorithm converges to the optimal dual variables and target average - rates for both users after a couple of hundreds of iterations . simulations results ( not shown in this paper ) for other values of @xmath117 and @xmath143 indicate that in general a larger step size leads to a faster algorithm convergence but also results in more frequent oscillations . [ fig : comp dp ] compares the network throughput under two extreme cases of delay constraint considered in this paper , namely , the expected throughput and the delay - limited throughput . it is assumed that @xmath211 , and two types of networks are considered : one is a 2-user network with user rate profile @xmath212 $ ] ; and the other is a 4-user network with @xmath213 $ ] . notice that the ratio of rate demand between any of the first two users and any of the last two users in the second case is the same as that between the two users in the first case , which is equal to @xmath214 . first , it is observed that for both networks , the delay penalties are only moderate for all considered transmit power values . the delay penalty increases as @xmath21 becomes larger than @xmath12 , but only slightly . small delay penalties in both cases can be explained by extending the multi - antenna channel hardening effect @xcite in the single - user case to the fading miso - bc , i.e. , as the number of degrees of fading increases with @xmath12 for some constant @xmath21 , not only the mutual information associated with each user s channel , but also the whole capacity region of the bc becomes less variant over different fading states . therefore , the sum - rate of users for any given rate profile also changes less dramatically , and as a consequence , imposing a set of strict constant - rate constraints at each fading state ( equivalent to a fixed rate - profile ) does not incur a large throughput loss for the fading miso - bc . secondly , it is observed that the expected throughput for a 4-user network outperforms that for a 2-user network given that both networks have the similar allocated rate portion among users . this throughput gain can be explained by the well - known multiuser diversity effect @xcite for ndc transmission . as the number of degrees of fading increases with @xmath21 for some constant @xmath12 , the bs can select relatively fixed number of users ( around @xmath12 ) with the best joint channel realizations from a larger number of total users for transmission at each time . thereby , the network throughput is boosted provided that each user has sufficient delay tolerance . on the other hand , it is observed that , maybe more surprisingly , the delay - limited throughput for the 4-user network also outperforms that for the 2-user network under the similar allocated rate portion among users . notice that in this case , all user s rates need to be constant at each fading state , and consequently , adaptive rate allocation for achieving the expected throughput does not apply here . this throughput gain in the dc case can also be explained by a more generally applied multiuser diversity effect . even though a set of constant - rates of users need to be satisfied at each fading state in the dc case , a larger number of users provides the bs more flexibility in jointly optimizing allocation of transmit resources among users based on their channel realizations . this multiuser diversity gain becomes more substantial as @xmath21 increases because a new user can bring along additional @xmath12 degrees of fading . however , it is also important to take note that the multiuser diversity gain in the delay - limited case does have certain limitation , e.g. , as @xmath21 becomes overwhelmingly larger than @xmath12 , the delay - limited throughput eventually gets saturated ( see theorem [ theorem : dlc throughput ] ) . at last , the tradeoff between the achievable network throughput and the fairness for user rate allocation is demonstrated . the ndc transmission is considered and hence the expected throughput is of interest . a network with 2 users is considered and it is assumed that @xmath211 , and the average ltpc at the bs is fixed as 10 . two fading channel models are considered : one has symmetric fading where both user channel vectors @xmath215 are assumed to be distributed as @xmath216 ; and the other has asymmetric fading where @xmath217 , and @xmath218 . notice that the asymmetric - fading case may correspond to a near - far situation in the cellular network where user-1 is closer to the bs and hence has an average channel gain of approximately @xmath219 db compared to user-2 . let @xmath220 denote the ratio of average - rate demand between user-1 and user-2 , i.e. , for the corresponding rate profile @xmath162 , @xmath221 . in fig . [ fig : comp fp ] , the expected throughput is shown as a function of @xmath220 for both symmetric- and asymmetric - fading cases . it is observed that in the symmetric - fading case , a strict fairness constraint for equal rate allocation among users , i.e. , @xmath222 , also corresponds to the maximum expected throughput or the sum capacity . in contrast , for the case of asymmetric fading , the maximum expected throughput is achieved when @xmath223 , i.e. , user-1 is allocated 70% of the expected throughput because of its superior channel condition . however , in the latter case , a strict fairness constraint with @xmath222 yields a throughput loss of only 0.3 bits / complex dimension . this small fairness penalty can be explained by observing that the expected throughput for the fading miso - bc under both symmetric and asymmetric fading is quite insensitive to @xmath220 at a very large range of its values , indicating that by optimizing resource allocation , transmission with very heterogeneous rate requirements may incur only a negligible network throughput loss . this paper investigates the capacity limits and the associated optimal dynamic resource allocation schemes for the fading miso - bc under various transmission delay and fairness considerations . first , this paper studies transmit optimization with mixed delay - constrained and delay - tolerant data traffic . the proposed online resource allocation algorithm is based on a two - layer lagrange - duality method , and is amenable to real - time cross - layer implementation . secondly , this paper characterizes some key fundamental tradeoffs between network throughput , transmission delay and user fairness in rate allocation , and draws some novel insights pertinent to multiuser diversity and channel hardening effects for the fading miso - bc . this paper shows that when there are similar numbers of users and transmit antennas at the bs , the delay penalty and the fairness penalty in the achievable network throughput may be only moderate , suggesting that employing multi - antennas at the bs is an effective means for delivering data traffic with heterogeneous delay and rate requirements . the results obtained in this paper can also be extended to the uplink transmission in a cellular network by considering the fading mac under individual user power constraint instead of the sum - power constraint in this paper as a consequence of the bc - mac duality . hence , another important factor needs to be taken into account by dynamic resource allocation for the fading mac is the fairness in user transmit power consumption . the concept of rate profile in this paper can also be applied to define a similar power profile for regulating the power consumption between users for the mac @xcite . furthermore , although the developed results in this paper are under the assumption of capacity - achieving transmission using dpc - based non - linear precoding at the bs , they are readily extendible to other suboptimal transmission methods such as linear precoding provided that the achievable rate region by these methods is still a convex set and , hence , like in this paper similar convex optimization techniques can be applied . in problem p3 , the constraints are separable and the objective function is convex . hence , this problem can be solved iteratively by the block - coordinate decent method @xcite . at each iteration , this method minimizes the objective function w.r.t . one @xmath97 while holding all the other @xmath97 s constant . more specifically , the method minimizes ( [ eq : objective func ] ) w.r.t . @xmath224 with constant @xmath225 , and then @xmath226 with constant @xmath227 , to @xmath228 with constant @xmath229 , and after that the above routine is repeated . because after each iteration the objective function only decreases , the convergence to the global minimum of the objective function is ensured . each iteration of the above algorithm is described as follows . without loss of generality , assuming that in ( [ eq : objective func ] ) , @xmath230 . considering any arbitrary iteration for minimizing ( [ eq : objective func ] ) w.r.t . @xmath231 , @xmath232 , with all the other @xmath97 s constant , problem p3 can be rewritten as @xmath233 by introducing the dual variable @xmath234 , @xmath235 , associated with the constraint @xmath236 , the karush - kuhn - tacker ( kkt ) optimality conditions @xcite state that the optimal @xmath237 and dual variable @xmath238 for this problem must satisfy @xmath239 let @xmath240 denote the function on the left - hand - side of ( [ eq : kkt1 ] ) . from ( [ eq : kkt2 ] ) , it is inferred that @xmath241 only if @xmath242 . from ( [ eq : kkt1 ] ) and by taking note that @xmath240 is a non - increasing function of @xmath243 for @xmath244 , it follows that @xmath241 occurs only if @xmath245 . thus , it follows that @xmath246 where @xmath247 is the unique root for @xmath248 . @xmath247 can be easily found by a bisection search @xcite over @xmath249 $ ] , where the above upper - bound for @xmath247 is obtained by the following inequalities and equality : @xmath250 first , we obtain an upper - bound for the delay - limited throughput @xmath184 by assuming that there is no co - channel interference between users , as opposed to the successive interference pre - subtraction by dpc . under this assumption , the fading miso - bc is decomposed into @xmath21 parallel single - user fading channels . let @xmath251 denote the average transmit power assigned to user @xmath27 , @xmath47 . the maximum constant - rate achievable over all fading states ( or the so - called delay - limited capacity @xcite ) for user @xmath27 can be expressed as @xcite @xmath252 where @xmath253 $ ] . because of the assumed symmetric fading ( hence , @xmath254 ) and symmetric rate demand ( hence , @xmath255 ) , it follows that @xmath256 , achieves the maximum average sum - rate of users . hence , the delay - limited throughput is upper - bounded by @xmath257 which holds for any @xmath258 . by taking the limit of the right - hand - side of ( [ eq : sum dlc ] ) as @xmath181 , the proof of theorem [ theorem : dlc throughput ] is completed . d. tse and s. hanly,``multi - access fading channels - part i : polymatroid structure , optimal resource allocation and throughput capacities , '' _ ieee trans . inf . theory _ , 44 , no . 7 , pp . 2796 - 2815 , nov . 1998 . y. c. liang , r. zhang , and j. m. cioffi , `` transmit optimization for mimo - ofdm with delay - constrained and no - delay - constrained traffics , '' _ ieee trans . signal proc . 54 , no . 8 , pp . 3190 - 3199 , 2006 . h. viswanathan , s. venkatesan , and h. c. huang , `` downlink capacity evaluation of cellular networks with known interference cancelation , '' _ ieee j. sel . areas commun . 802 - 811 , jun . 2003 . z. shen , j. g. andrews , and b. l. evans , `` adaptive resource allocation in multiuser ofdm systems with proportional rate constraints '' , _ ieee trans . wireless commun . 4 , no . 6 , pp . 2726 - 2737 , nov . m. mohseni , r. zhang , and j. cioffi , `` optimized transmission of fading multiple - access and broadcast channels with multiple antennas , '' _ ieee j. sel . areas commun . 24 , no . 8 , pp . 1627 - 1639 , aug . 2006 . g. caire , r. r. m@xmath259ller , and r. knopp , `` hard fairness versus proportional fairness in wireless communications : the single - cell case '' , _ ieee trans . theory , _ vol . 4 , pp . 1366 - 1385 , apr . s. vishwanath , n. jindal , and a. goldsmith , `` duality , achievable rates , and sum - rate capacity of gaussian mimo broadcast channels , '' _ ieee trans . inf . theory _ , 10 , pp . 2658 - 2668 , oct .

the nonequilibrium dynamics of anharmonic chains is studied by imposing an initial domain - wall state , in which the two half lattices are prepared in equilibrium with distinct parameters . we analyse the riemann problem for the corresponding euler equations and , in specific cases , compare with molecular dynamics . additionally , the fluctuations of time - integrated currents are investigated . in analogy with the kpz equation , their typical fluctuations should be of size and have a tracy - widom gue distributed amplitude . the proper extension to anharmonic chains is explained and tested through molecular dynamics . our results are calibrated against the stochastic leroux lattice gas .

1607.05205

the cold atom community has revived the study of the approach to thermal equilibrium for large isolated quantum systems . we refer to @xcite and references therein on previous work . most accessible , both numerically and experimentally , are lattice systems in one dimension . one dimension is peculiar , since there are models with an extensive number of locally conserved fields . examples of such quantum integrable systems are the xxz spin chain and the continuum lieb - liniger @xmath1-bose gas . obviously the pathway to equilibrium will depend crucially on whether the system is integrable or not @xcite . but in addition there is also the dependence on initial conditions which is potentially overwhelming . one could prepare the system already in thermal equilibrium and study the response to small initial perturbations @xcite . these are the much investigated time response and correlation functions in equilibrium . the initial state could be translation invariant , to some extent thereby suppressing the mostly slow spatial variations @xcite . recently initial domain - wall states have become very popular @xcite . such a state is obtained by joining two distinct thermal states at a single point ( and at two points in case of periodic boundary conditions ) . domain - wall states will be the main focus of our contribution . browsing the introductions to the papers mentioned above , one might have the impression that the approach to equilibrium for classical systems in one dimension is a well - covered topic . we study here fermi - pasta - ulam type anharmonic chains with domain - wall initial conditions and are not aware of any previous systematic study . the structure of equilibrium time - correlations for such chains has been elucidated only recently @xcite . in particular one now understands the link to anomalous transport which is most directly observed when coupling the chain to thermal reservoirs at distinct temperatures , see @xcite . as in the quantum world , there are integrable chains , in our context the most famous one being the toda chain . but the kam theorem signals in addition the possibility that , as a function of the energy , the structure may change from integrable to chaotic . this energy threshold is fascinating from the perspective of nonlinear dynamics and has attracted considerable attention @xcite . we hope that a better understanding of classical models also serves as an incentive to look for related phenomena in quantum systems . the parameters of the initial domain - wall state will be chosen such that in the accessible part of phase space the chain dynamics is sufficiently chaotic . then one would expect that the conserved fields as computed from the chain dynamics are approximated by the respective solution of the macroscopic euler equations , for times limited by diffusive effects . how well such expectations work out will have to be studied . the euler equations are based on the notion of local thermodynamic equilibrium . the microscopic local conservation laws are deduced from the chain dynamics and are then averaged in the stipulated local equilibrium state so to arrive at a closed set of equations for the conserved fields . in particular , to reach non - trivial predictions , the thermal average of the microscopic currents is not allowed to vanish . this is ensured if the interaction potential depends only on positional differences , as @xmath2 , implying momentum conservation . upon adding an on - site potential , @xmath3 , momentum conservation would be broken , all euler currents would vanish , and the evolution of the initial step profile is determined by diffusive effects only . the euler equations are a system of @xmath4 hyperbolic conservation laws , @xmath5 for our case of anharmonic chains . they are of the generic form @xmath6 @xmath7 , @xmath8 , with given current functions @xmath9 . in the mathematical literature the domain - wall initial data are known as riemann problem for eq . , which means @xmath10 for a wide class of current functions , there is a unique entropy solution to with initial conditions , see the exposition @xcite , sections 48 . this solution scales ballistically as @xmath11 where @xmath12 is bounded and continuous except for isolated jumps , possibly . there is a well developed theory of how to compute @xmath12 , at least in principle @xcite , sections 13 . in our case the current functions are determined through the microscopic particle model , and hence of a very particular form . thus our task is twofold . firstly we have to investigate the solution to the riemann problem . secondly such predictions should be compared with numerical simulations of the dynamics . let us return for a moment to the distinction between integrable and non - integrable systems , both starting from a domain - wall initial state . as supported by a variety of studies on quantum integrable models @xcite , one expects that still holds in the integrable case . thus at first sight there seems to be little difference . of course , the macroscopic profiles are computed by using completely different methods for the two cases . but the real distinguishing feature is the appearance of shocks . an ideal gas with step - initial conditions shows ballistic spreading but no shocks . the entropy solution for the euler equations is a mathematical shorthand for the limit of small dissipation , which is meaningful only if the underlying dynamics is sufficiently chaotic . merely invoking the conservation laws , the euler equations admit stable and unstable shocks . such unphysical solutions to are removed by requiring a positive entropy production at the shock , as will be illustrated in the examples below . in a local region away from the shock , the local state is ( to very good approximation ) in thermal equilibrium . our study adds current fluctuations as an item , which can no longer be based on the euler equations . most simple - mindedly , one would consider the fluctuations of the time - integrated current across the origin . for anharmonic chains the current is a three - vector . in most cases one would find gaussian fluctuations of size @xmath13 , thus not so interesting from a theoretical perspective . a more global picture emerges by considering the current integrated along the ray @xmath14 for some prescribed velocity @xmath15 . the ray is chosen to lie in the interior of a rarefaction wave . in addition , one has to consider a computable but particular linear combination of the three currents . as will be discussed , then the integrated current fluctuations are of size @xmath0 , smaller as for all other linear combinations , and the statistics is given by the tracy - widom distribution known from random matrix theory . our paper consists of three , at first sight somewhat unrelated parts . we start with a stochastic model with two conserved fields , as always with domain - wall initial conditions . for our particular system the validity of the euler equations has been established mathematically @xcite . thus the model is used to explain the method by which one obtains the solution of the riemann problem and as a numerical benchmark . we proceed to anharmonic chains , first with a general interaction potential . analytically and numerically we then consider two specific choices for the potential , which generate sufficiently chaotic dynamics ( as known from previous studies ) . in the third part we discuss the fluctuations of time - integrated currents . a prototypical stochastic lattice gas is the totally asymmetric simple exclusion process ( tasep ) . particles are located on @xmath16 , at most one particle per site . independently , after an exponentially distributed waiting time , a particle hops one step to the right , provided the target site is empty . clearly , the particle number is the only conserved field . to move towards several conservation laws , we look for a minimal extension of the tasep to a model with two conserved fields . in the literature a standard generalization is known as leroux stochastic lattice gas . ( this name goes back to fritz and tth @xcite , who prove the hydrodynamic limit globally in time . apparently , leroux first wrote down this particular system of conservation laws @xcite . ) the leroux lattice gas has two types of particles with label @xmath17 . subject to the exclusion rule , the 1 particles jump to the right and the @xmath18 particles to the left , both according to the tasep rule . furthermore , a neighboring pair @xmath19 is exchanged to @xmath20 with rate @xmath21 , which leads to the simplification that the stationary measures are bernoulli . a generalization of leroux is the arndt - heinzel - rittenberg ( ahr ) model @xcite . more formally , we introduce occupation variables @xmath22 , @xmath23 . the only allowed exchanges are [ cols="^,^ " , ] note that in our convention the labels of the components are interchanged in comparison to @xcite . clearly , the only conserved fields are the two particle numbers . the invariant bernoulli measures are parametrized by the average densities @xmath24 and @xmath25 . the hydrodynamic equations simplify when written in terms of the average number of holes and the average velocity , i.e. , @xmath26 with @xmath27 we refer to @xmath28 as states , more appropriately , but also more lengthy , as steady state parameters , resp . as thermodynamic states in case of anharmonic chains . the single - site probabilities of the steady states are @xmath29 averages will be denoted by @xmath30 , the subscripts being omitted when obvious from the context . since the steady states are of product form , their average current is easily computed . thus , on a large space - time scale the conserved fields are governed by the entropy solution of @xmath31 with the current vector @xmath32 to discuss the solution to the riemann problem , we follow fairly closely the conventions of ref . @xcite . one rewrites in semilinear form as @xmath33 where @xmath34 the eigenvalues of @xmath35 are @xmath36 and the corresponding right and left eigenvectors , @xmath37 , @xmath38 , are given by @xmath39 here @xmath40 and @xmath41 are positive normalization constants . for the riemann problem their explicit form is not needed . setting @xmath42 , one obtains for the change of @xmath43 along the vector fields @xmath44 , @xmath45 and strictly positive for @xmath46 . the rarefaction curves , @xmath47 , are obtained by solving the cauchy problem in @xmath48-space , @xmath49 for @xmath50 , with @xmath44 the right eigenvectors of @xmath35 , see . the normalization has been absorbed into the @xmath51-parameter . the integral curves are then determined by @xmath52 it follows that @xmath53 which is negative for @xmath54 and positive for @xmath55 . the solution of is @xmath56 as visualized in fig . [ fig : leroux_rarefaction_shock ] . maximally , @xmath54 starts at @xmath57 and ends at @xmath58 , whereas @xmath55 starts at @xmath59 and ends at @xmath60 . the local eigenvalue is @xmath61 to convert the solution from @xmath48-space to position space , we set @xmath62 . the solution is self - similar and we may assume @xmath63 . then @xmath64 the boundary speeds of @xmath54 are @xmath65 and of @xmath55 @xmath66 eq . as a function of @xmath67 describes solutions of the euler equation . the two solutions with @xmath50 are mirror images of each other . in state space for the leroux system according to eqs . and . traversal in arrow direction corresponds to a rarefaction wave and increasing @xmath43 , and traversal in opposite arrow direction to a shock curve.,scaledwidth=50.0% ] shock curves are determined by the rankine - hugoniot jump condition @xmath68 hence in our case @xmath69 according to the first equation the shock speed is @xmath70 we eliminate @xmath71 with the result @xmath72 if @xmath73 , the two solutions for @xmath74 are @xmath75 with @xmath50 and @xmath76 the sound speed on the left side of the shock . in particular , one recovers @xmath77 for @xmath78 . in fact , the rarefaction curves coincide with the solution to the rankine - hugoniot equations . this can be seen by defining @xmath79 implicitly via @xmath80 then @xmath81 inserting into one arrives at the shock speed @xmath82 the coincidence of rarefaction and shock curves is the defining property of the temple class @xcite . the leroux system is a further member of this class . in our context the interest results from a maximally simple underlying particle dynamics . the _ lax admissibility condition _ states that characteristics must move `` towards '' the shock : @xmath83 since @xmath71 is the mean value of the sound speeds , the condition simplifies to @xmath84 . together with @xmath85 , this is equivalent to @xmath86 for @xmath87 and @xmath88 for @xmath89 . the stable , physically admissible part of the rankine - hugoniot curve are the shock curves @xmath90 and @xmath91 as displayed in fig [ fig : leroux_rarefaction_shock ] . the construction of the general solution is illustrated in fig . [ fig : leroux_path ] . starting from the asymptotic left value @xmath92 , one first follows either the rarefaction curve @xmath55 or the stable part of the shock curve @xmath91 ( shown as linear orange - red line in fig . [ fig : leroux_rarefaction_shock ] ) up to a unique intermediate state @xmath93 . then @xmath93 is connected by either a rarefaction curve @xmath54 or shock curve @xmath90 ( blue - green in fig . [ fig : leroux_rarefaction_shock ] ) to the asymptotic right value @xmath94 . rarefaction curves correspond to traversal in arrow direction in fig . [ fig : leroux_rarefaction_shock ] , equivalently increasing eigenvalue @xmath43 . this procedure splits the parameter domain into four distinct pieces according to rarefaction - rarefaction , shock - rarefaction , rarefaction - shock , and shock - shock . the two domain boundaries correspond to either a single rarefaction or a single shock , with no intermediate value @xmath93 . in fig . [ fig : leroux_rarefaction_shock_xt ] we show the space - time plot corresponding to the case shock - rarefaction of fig . [ fig : leroux_rarefaction_shock_path ] . we perform monte carlo simulations of the leroux model for @xmath95 sites with periodic boundary conditions . to obtain an initial domain wall state , we sample the single - site probability distribution on the left half @xmath96 using parameters @xmath97 , and on the right half @xmath98 using parameters @xmath99 . the dynamics is simulated by random exchanges at exponentially distributed waiting times up to @xmath100 . this procedure is realized @xmath101 times to compute average profiles , as shown below . + + first , we illustrate the special case of a single rarefaction wave @xmath54 connecting @xmath102 to @xmath103 , which are chosen maximally as @xmath104 and @xmath105 with @xmath106 . this is the particular choice in @xcite , and corresponds to the extremal points of the green - blue line segment in fig . [ fig : leroux_rarefaction_shock ] . besides the @xmath107 riemann problem centered at @xmath108 and generating rarefaction @xmath54 , the periodic boundary condition translates to an additional @xmath109 riemann problem centered at @xmath110 . since the solutions of the rarefaction and shock curves coincide except for orientation , one concludes that the solution of the @xmath109 riemann problem is a single shock curve @xmath90 . in other words , one traverses the green - blue line segment in fig . [ fig : leroux_rarefaction_shock ] in opposite direction . the numerical monte carlo profiles , shown as orange dots in fig . [ fig : leroux_profiles ] , agree very well with the theoretical prediction ( solid black lines ) . in particular , note the sharp jump at the shock . this shock curve is located at @xmath111 and moves to the right , with shock speed @xmath112 . next , we perform molecular dynamics simulations corresponding to the general case in fig . [ fig : leroux_path ] . specifically , the left @xmath113 , the intermediate @xmath114 and the right @xmath115 . fig . [ fig : leroux_profiles_general ] shows the molecular dynamics profiles , with the shock curve @xmath91 indicated in red and the rarefaction wave @xmath54 in blue . one observes that the mc profile of @xmath91 is less sharp than the shock in fig . [ fig : leroux_profiles ] , presumably due to the higher shock speed . the thermodynamic entropy for the probability distribution of is @xmath116 using the standard physics convention for the sign . the definition in @xcite uses a convex function which has the opposite sign . the corresponding entropy flux @xmath117 has to satisfy @xmath118 with @xmath119 and @xmath120 . up to a constant the solution for @xmath117 is @xmath121 note that @xmath122 is an odd function in @xmath74 . [ fig : leroux_entropy ] visualizes both entropy and entropy flux . the entropy inequality admissibility condition states that @xmath123 in the sense of distributions . for continuously differentiable solutions due to one has @xmath124 and @xmath125 in particular , no entropy is produced at a rarefaction wave . on the other side , at shocks one can follow the same steps as for the rankine - hugoniot jump condition to derive from that @xmath126 for any continuously differentiable test function @xmath127 with compact support . the integration in proceeds along the shock curve , and @xmath128 , @xmath129 are the values on the left side of the shock . inserting the shock speed and the shock solution into and , one obtains ( with @xmath130 ) @xmath131\big ) + \frac{1}{b } \left(1 - b^2 - ( 1-\rho)(1-\rho_0)\right ) \frac{1}{2 } \log\!\left[\frac{1 - ( \rho / b)^2}{1 - ( \rho_0/b)^2}\right ] \\ - \left(1 - \rho - \rho_0 + \frac{\rho\,\rho_0}{b^2}\right ) \left(\mathrm{arctanh}\big(\frac{\rho_0}{b}\big ) - \mathrm{arctanh}\big(\frac{\rho}{b}\big)\right ) \bigg).\end{gathered}\ ] ] note that this expression is invariant under the interchange @xmath132 and simultaneously @xmath133 , as expected from symmetry of the shock curves moving to the right and left . as required , the solution of the riemann problem satisfies @xmath134 . with the leroux lattice gas as guiding example , we turn to our central theme which is the time evolution for domain - wall initial conditions of a system of anharmonically coupled mechanical point particles . the @xmath135-th particle has mass @xmath136 , position @xmath137 , momentum @xmath138 , and is coupled to its neighbors @xmath139 and @xmath140 through the potential @xmath141 . then newton s equations of motion are given by @xmath142 we read this equation as a discretized wave equation with @xmath143 the displacement of the wave field at lattice site @xmath135 . the hamiltonian of the chain is @xmath144 for the harmonic potential , @xmath145 , eq . reduces to a discrete linear wave equation . in the theoretical analysis we use @xmath23 . numerically , @xmath146 $ ] with periodic boundary conditions . in contrast to the leroux lattice gas , the dynamics is deterministic . the initial conditions are however random , specifically to have a domain - wall state . as before , we expect the euler equations to provide an accurate description on a macroscopic scale , provided the times are not too long . the validity of the euler equations is based on maintaining local stationarity away from shocks . for a stochastic system the local approach to stationarity is in a certain sense build into the dynamics . for mechanical systems one relies on sufficiently strong dynamical chaos . thereby integrable systems , as the harmonic and toda chain are ruled out . in the leroux lattice gas the jumps are asymmetric . the steady states are non - equilibrium and the dynamics does not satisfy the principle of detailed balance . in contrast , for the anharmonic chain the domain - wall state is manufactured by joining two thermal equilibrium states . because of momentum conservation , the thermal average defining the euler currents does not vanish . if one broke this conservation law by adding in an on - site potential , then the euler currents would vanish identically , no riemann problem ensues , and the first macroscopic time - scale is diffusive . it is convenient to introduce the stretch @xmath147 . then the equations of motion turn into @xmath148 from which one concludes that stretch and momentum are conserved . the respective currents are @xmath149 and @xmath150 . in addition , we define the local energy @xmath151 which changes in time as @xmath152 as anticipated , the energy is locally conserved . its current equals @xmath153 . more precisely than before , in our context non - integrable means that there are no further locally conserved fields . unfortunately , this property is difficult to check . besides the harmonic chain , the only known integrable system is the toda chain with @xmath154 . [ the calogero - moser chain has a two - sided decaying potential , which is not allowed in our context . ] the thermodynamic fields conjugate to stretch , momentum , and energy are the pressure @xmath155 , the mean momentum , @xmath156 , and the inverse temperature , @xmath157 , respectively . from we conclude that in thermal equilibrium the @xmath138 s and @xmath158 s are independent . the probability density function for @xmath138 is the shifted maxwellian @xmath159 and the one for @xmath158 is given by @xmath160 to obtain a finite spatial partition function @xmath161 , we require that @xmath162 is bounded from below and has at least a one - sided linearly growing lower bound as @xmath163 . then @xmath164 for @xmath155 in a suitably chosen interval . equilibrium averages are denoted by @xmath165 , the subscripts being omitted if obvious from the context . to assemble a domain - wall initial state , we set @xmath166 for @xmath167 and @xmath168 for @xmath169 in and . by construction these initial data are in thermal equilibrium except for the jump at the origin . a little bit more of thermodynamics will be needed . the average stretch is given by @xmath170 the average momentum is @xmath171 , and the average _ internal _ energy is @xmath172 the average total energy is then @xmath173 . later on we will need @xmath174 and @xmath175 as the inverse to eqs . and . by convexity of the respective thermodynamic potential this inverse is uniquely defined . the euler equations are obtained by assuming that the conserved fields are slowly varying on the scale of the lattice . they thus become functions of space - time @xmath176 . we now drop the prefix `` average '' and call the space - time fields simply stretch , momentum , and energy . since all particles have the same mass , we will follow the standard convention in using velocity instead of momentum as hydrodynamic field . if local equilibrium is propagated , then the macroscopic euler currents are the thermal average of the microscopic currents , explicitly , @xmath177 through eq . the currents become functions of the local fields , where the pressure is evaluated as @xmath178 then the euler equations read @xmath179 we combine the conserved fields as 3-vector @xmath180 . then the euler equations take the canonical form @xmath181 since the _ total _ energy @xmath182 ( instead of internal energy @xmath183 ) is locally conserved , @xmath182 has to be used when applying the theory of hyperbolic conservation laws . but for the riemann problem , it will turn out to be more concise to use the internal @xmath183 as parameter . in fact , eq . is identical to the euler equations of a one - dimensional fluid in lagrangian coordinates . its riemann problem has been studied in great detail starting with the pioneering work of bethe @xcite . the interest in one - dimensional fluids also served as a strong motivation to develop a mathematical theory of hyperbolic conservation laws with several components @xcite . for the physics perspective we refer to the excellent review by menikoff and plohr @xcite . the tutorial by bressan @xcite provides the necessary mathematical background . so it appears that we only have to point at the relevant literature . from the physics side the main goal is to understand the qualitative link between the equation of state and the solution to the riemann problem , in particular in case the equation of state allows for a phase transition . on the other hand , we plan to quantitatively compare the microscopic dynamics with the solution of the riemann problem and thus need shock and rarefaction profiles in a fairly explicit form . compared to one - dimensional fluids , anharmonic chains have the advantage that the grand - canonical potential is given in terms of a single one - dimensional integral . static correlations of the conserved fields vanish except for coinciding points . no phase transition is possible . before turning to our concrete examples , we have to recall a few general properties . the linearization matrix is given by @xmath184 the eigenvalues of @xmath35 are @xmath185 , with the adiabatic sound speed @xmath186 , @xmath187 , defined by @xmath188 the right eigenvectors of @xmath35 corresponding to the eigenvalues @xmath189 and @xmath190 , @xmath50 , read @xmath191 and the left eigenvectors of @xmath35 are @xmath192 by construction @xmath193 for @xmath194 . for the solution of the riemann problem , the positive normalization constants are not needed . as explained in appendix [ appendix : general_anharm ] , they are fixed by requiring that normal modes are orthonormal with respect to the equilibrium measure . since in equilibrium the @xmath138 s and @xmath158 s are independent , we can use the generalisation of to probability densities in order to obtain the physical entropy , @xmath195 , per volume . inserting from , yields @xmath196 with @xmath197 , @xmath198 . denoting by @xmath199 the gradient in state space , one obtains @xmath200 and for a smooth solution @xmath201 anharmonic chains are special in the sense that the entropy current vanishes . entropy may be produced at shock discontinuities , but is not propagated . this behavior should be contrasted with the leroux lattice gas , which has a non - zero entropy current , compare with . the rarefaction curves are obtained by solving the following cauchy problem in state space @xmath202 @xmath203 , where @xmath204 are the right eigenvectors of @xmath35 , see @xcite for details , [ [ eigenvalue-0 ] ] eigenvalue @xmath189 : + + + + + + + + + + + + + + + + + + + + + + + + + + + this is a contact discontinuity , which can be thought of as a rarefaction wave in the limit of zero extension with a non - zero jump . . for @xmath205 reads @xmath206 with @xmath155 evaluated at @xmath207 . the normalization constant has been absorbed into @xmath51 . since for the velocity @xmath208 and @xmath209 , one obtains the closed system @xmath210 which is of hamiltonian form with @xmath174 as hamiltonian function . across a contact discontinuity both pressure and velocity are conserved . by and the entropy changes as @xmath211 thus , the change of entropy across the contact discontinuity equals @xmath212 where @xmath213 and @xmath214 . [ [ eigenvalue - sigma - c ] ] eigenvalue @xmath190 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + eq . for @xmath50 reads @xmath215 the first equation of is solved by @xmath216 . for the internal energy it follows that @xmath217 however , instead of the energy equation it is more convenient to use the conservation of entropy @xmath218 inserting @xmath219 into , the velocity is then determined by @xmath220 the rarefaction curves can be obtained without actually solving differential equations . the gradient of @xmath190 along trajectories of the vector field @xmath44 , for @xmath50 , is @xmath221 a common simplifying assumption for hyperbolic conservation laws is genuine nonlinearity , i.e. , @xmath222 for any @xmath223 , @xmath183 . on general grounds , there is no reason why genuine nonlinearity should hold . a point in case are hard - point particles with alternating masses , discussed in sect . [ sec : hard_point_particles ] below . but genuine nonlinearity does not hold for a square - well interaction potential , see sect . [ sec : square_well ] . according to the rankine - hugoniot jump condition , we search for nontrivial solutions of @xmath224 using the shorthand notation @xmath225 , more explicitly eq . reads @xmath226 according to eq . , the shock speed is @xmath227 eqs . , and the relation @xmath228 lead to @xmath229 and inserting eq . , one arrives at @xmath230 multiplying eqs . and leads to the condition @xmath231 there is no general procedure to solve the rankine - hugoniot equations . also the issue of stability can be discussed only once the solution is of a more explicit form . a widely studied anharmonic chain is the fermi - pasta - ulam lattice with potential @xmath232 in the historical notation @xcite . as follows from and , @xmath233 and @xmath234 are given by simple integrals . but one still has to invert the pair of functions . to our knowledge , the corresponding riemann problem has never been studied . to simplify one looks for the factorized ideal gas ansatz @xmath235 , which holds if the potential takes only the values @xmath189 , @xmath236 . this leaves the choice @xmath237 for @xmath238 and @xmath239 for @xmath240 and @xmath241 . the parameter @xmath242 describes a hard core at which particles are specularly reflected from each other . the limiting hard - point , @xmath243 , is also allowed . there is an inward collision at separation @xmath244 . physically one can imagine that neighboring particles are connected by a massless string of maximal length @xmath244 . from the simulation perspective such a potential has the advantage that no differential evolution equation has to be solved . one can simply proceed from collision to collision . at a collision , the momenta are exchanged . thus @xmath245 with general @xmath246 is conserved . the dynamics is integrable . the standard resolution is to prescribe alternating masses , @xmath247 for even labels and @xmath248 for odd labels . thereby the collisions become nontrivial and seem to generate sufficient chaos . now the unit cell contains two particles . but , as discussed in @xcite , the euler equations still retain their form upon substituting for @xmath136 the average mass @xmath249 , again denoted by @xmath136 . we specialize the results from sect . [ sec : riemanngeneral ] to the case of a hard - point potential , @xmath250 for @xmath251 and @xmath252 for @xmath253 . then the inverse temperature is @xmath254 , the pressure is given by @xmath255 and the sound speed is obtained as @xmath256 the right eigenvectors of @xmath35 are @xmath257 and the left eigenvectors @xmath258 [ [ eigenvalue-0 - 1 ] ] eigenvalue @xmath189 : + + + + + + + + + + + + + + + + + + + + + + + + + + + for hard - point particles eq . reads @xmath259 with initial state @xmath260 . note that @xmath261 , since @xmath262 , and the solution obeys @xmath263 in particular the pressure is conserved , as required . [ [ eigenvalue - sigma - c-1 ] ] eigenvalue @xmath190 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for hard - point particles , eqs . and lead to @xmath264 and @xmath265 in particular , @xmath266 the gradient of @xmath190 along trajectories of the vector field @xmath44 , eq . , becomes then @xmath267 i.e. , genuine nonlinearity holds . for hard - point particles the condition reads @xmath268 implying that @xmath269 for @xmath270 . inserted into eq . leads to @xmath271 with @xmath50 . the shock speed is then @xmath272 with @xmath273 the sound speed of the initial state . the _ lax admissibility condition _ states that characteristics must run `` towards '' the shock , @xmath274 inserting the relation gives @xmath275 such that the lax admissibility condition becomes @xmath276 for @xmath87 , this is equivalent to @xmath277 , and for @xmath89 equivalent to @xmath278 . [ fig : hp_energy_stretch ] displays the internal energy in dependence of the stretch , both for the rarefaction and shock curves , denoting by @xmath279 the @xmath280-th rarefaction curve and by @xmath281 the @xmath280-th shock curve . analogously , fig . [ fig : hp_velocity_stretch ] displays the change of velocity in dependence of the stretch . these figures should be compared with fig . [ fig : leroux_rarefaction_shock ] . considering only rarefactions and shocks , the solution to both riemann problems are qualitatively the same . the contact discontinuity is merely an independent additional feature . the entropy for the hard - point particles is @xmath282 up to a constant shift by @xmath283 . for a jump along a shock curve with speed @xmath71 , the entropy admissibility condition @xmath284 becomes @xmath285 compare to eq . ( 47 ) in @xcite with opposite sign . we follow the physics convention of a concave entropy function , while @xcite prefers a convex function . in other words , at a shock a region with higher entropy invades a region with lower entropy . fig . [ fig : entropy_shock ] schematically visualizes the time evolution of the entropy across a shock with speed @xmath286 . for a molecular dynamics simulation we use a box of size @xmath287 with periodic boundary conditions , for which we adopt the domain @xmath288 $ ] . imposing domain - wall initial conditions as @xmath289 at the origin implies that somewhere else one has the reversed initial condition @xmath290 . we call this the periodic riemann problem . the riemann problem centered at @xmath189 is of key interest . but numerically we automatically realize two distinct riemann problems . the longest time of simulation is limited by collisions between the two solution branches . a conventional choice for the masses is @xmath291 , corresponding to alternating masses with @xmath292 and @xmath293 . as in @xcite , we prescribed @xmath294 and determine the entries of @xmath295 such that the rarefaction waves and shocks as shown in fig . [ fig : hardpoint_finite_vol_summary ] arise . in @xcite the goal was to have a wide rarefaction wave . here we explain in detail how the solution of the periodic riemann problem is constructed . with periodic boundary conditions and domain - wall initial conditions . the largest time corresponds to @xmath296.,scaledwidth=50.0% ] for conciseness , we denote the quotients of the stretches by @xmath297 , @xmath298 , @xmath299 and @xmath300 . considering the right rarefaction @xmath54 , condition for the internal energy reads @xmath301 , and for the velocity @xmath302 we can still choose @xmath303 , which we set to @xmath304 in fig . [ fig : hardpoint_finite_vol_summary ] . accordingly , @xmath305 , @xmath306 and @xmath307 . it remains to show that @xmath295 and @xmath308 can be chosen in accordance with fig . [ fig : hardpoint_finite_vol_summary ] . the contact discontinuity @xmath309 at the origin implies that the velocity is conserved , i.e. , @xmath310 . furthermore @xmath311 . considering the shock curve @xmath91 , according to , @xmath312 since @xmath313 as defined above , and according to , @xmath314 finally , for the shock curve @xmath90 implies that @xmath315 and that @xmath316 in summary , it holds that @xmath317 , i.e. , @xmath318 similarly , @xmath319 which enforces that @xmath320 the analogous procedure for the velocity leads to @xmath321 solving , , and for @xmath322 , @xmath323 and @xmath324 with the help of a computer algebra program leads to rather lengthy expressions in terms of roots of certain polynomials , which we do not write down explicitly . instead , we report the numerical values for @xmath304 , namely @xmath325 , @xmath326 and @xmath327 . note that @xmath322 and @xmath323 are close to @xmath328 . the numerical values for @xmath308 are @xmath329 , and for @xmath295 the values read @xmath330 . the shock @xmath91 is hardly visible , while @xmath331 and @xmath332 are well - developed . the rounding of the edges is presumably a finite size effect . a similar feature is noted in fig . [ fig : leroux_profiles_general ] for the leroux lattice gas . an alternative explanation would be rounding because of diffusion . but then this should show also in fig . [ fig : leroux_profiles ] . to decide larger size systems would have to be simulated . + + fig . [ fig : hp_profiles ] shows stretch , velocity , and internal energy profiles obtained by molecular dynamics simulations , after averaging over @xmath101 simulation runs with initial states chosen according to @xmath295 and @xmath93 in fig . [ fig : hardpoint_finite_vol_summary ] . a shock generated through rapid compression is studied in @xcite . hard - point particles are an example of a genuinely nonlinear hyperbolic conservation law . as discussed in @xcite , in general , one - dimensional fluids do not have such a property and the structure of solutions to the riemann problem is considerably richer than for hard - points . a similar observation is well known for stochastic dynamics . for a single component genuine nonlinearity corresponds to a convex flux function , which imposes an additional constraint . a two - component stochastic system is studied in @xcite . for anharmonic chains we still want to comply with the factorized ideal gas law @xmath235 , we consider the hard - core potential with core diameter @xmath243 and inward reflection at @xmath244 . this defines the _ square - well interaction potential _ ( see also @xcite ) @xmath333 since the potential is zero within the well , @xmath334 and the pressure factorizes as @xmath335 where @xmath336 is the inverse function of @xmath337 . the unit length can be chosen such that @xmath338 , which we adopt in the following . then is rewritten as @xmath339 hard - point particles are obtained in the limiting case @xmath340 , which corresponds to setting @xmath341 . @xmath342 is visualized in fig . [ fig : squarewell_h ] . , defining the pressure @xmath343.,scaledwidth=50.0% ] note that derivatives and indefinite integrals can be expressed through @xmath336 itself , for example @xmath344 and @xmath345 it is more transparent to keep for a while a general @xmath336 , specializing to the square - well potential at the end . the only constraint is @xmath346 , ensuring thermodynamic stability . wendroff @xcite discusses a two component model , which in essence corresponds to such a choice upon dropping the contact discontinuity . inserting into results in the linearized currents @xmath347 and the square of the sound speed @xmath348 the right eigenvectors of @xmath35 are @xmath349 the gradient of @xmath350 along trajectories of the vector field @xmath351 for @xmath50 , i.e. , for the special case , is @xmath352 for the hard - point case , @xmath341 , this simplifies to . note that the right hand side of is independent of @xmath353 , which is achieved by an appropriate choice of the sign of @xmath351 . the square - well potential turns out to violate genuine nonlinearity , since the expression changes sign at @xmath354 . by a suitable choice of @xmath336 , presumably one can generate shocks and rarefactions of the same richness as in @xcite . compared to the genuinely nonlinear case the main novel feature is to have a shock at the borderline of a rarefaction wave . for the square - well potential only the case of a left bordering shock is realized . in principle , the shock could also switch to the opposite side , but it can not lie in the interior of the rarefaction wave . [ [ eigenvalue-0 - 2 ] ] eigenvalue @xmath189 : + + + + + + + + + + + + + + + + + + + + + + + + + + + according to eq . @xmath355 with the pressure @xmath174 conserved . thus from , it follows that @xmath356 [ [ eigenvalue - sigma - c-2 ] ] eigenvalue @xmath190 : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + according to eqs . , , the stretch obeys @xmath216 and the internal energy @xmath357 . together with , one obtains @xmath358,\ ] ] which only depends on @xmath51 via @xmath359 . inserting this relation into the differential equation @xmath360 , the sound speed depending on @xmath359 and @xmath219 via , leads to @xmath361 \sqrt{2 h(r(\tau))^2 - h'(r(\tau ) ) } \ , r'(\tau)\ ] ] and integrates to @xmath362 \sqrt{2 h(s)^2 - h'(s)}.\ ] ] for @xmath155 as in , eq . leads to @xmath363 for @xmath364 if @xmath365 and @xmath366 if @xmath367 . the condition leads to @xmath368 for the hard - point particles with @xmath341 , eq . simplifies to the expression . inserting into results in the shock speed @xmath369 for the square - well potential fig . [ fig : sqw_energy_stretch ] visualizes an integral curve of the internal energy in dependence of the stretch , both for the rarefaction and shock curves . analogously , fig . [ fig : sqw_velocity_stretch ] visualizes the velocity in dependence of the stretch . for the square - well potential with @xmath370 , @xmath371 , @xmath372 . the green curve shows the shock speed in dependence of the stretch ; the condition only holds within the thick curve segment.,scaledwidth=45.0% ] the lax admissibility condition @xmath373 can be tested numerically for the square - well @xmath342 . [ fig : lax_condition ] shows an example with @xmath371 where the condition is satisfied within an interval @xmath374 . for larger @xmath223 , the sound speed ( evaluated along the shock solution ) becomes larger than @xmath375 . thus a shock wave is followed immediately by a rarefaction wave , as shown in fig . [ fig : squarewell_shock_rarefaction ] with @xmath376 . the state to the left of the shock curve is @xmath377 , and the transition from shock to rarefaction wave appears at @xmath378 with @xmath379 . the state @xmath93 is connected by a rarefaction wave to @xmath380 . specifically for the square - well interaction potential , the entropy equals @xmath381 up to a constant shift by @xmath283 . as expected , for the hard - point case @xmath341 , this expression simplifies to . + + + fig . [ fig : squarewell_profiles ] shows stretch , velocity , internal and total energy profiles obtained by molecular dynamics simulations with square - well interaction potential , after averaging over @xmath101 simulation runs and initial states chosen according to @xmath295 and @xmath308 in fig . [ fig : squarewell_shock_rarefaction ] . the alternating masses are set as @xmath382 and @xmath383 , such that the average mass @xmath370 . for stretch and velocity the shock followed by a rarefaction wave is well reproduced . for the total energy the shock is not so strong and hence hardly visible . the internal energy is not conserved . our prediction is based on local equilibrium which apparently is not so accurate close to the shock . the entire system shows still further shocks and rarefaction waves due to periodic boundary conditions . the maximum time @xmath384 in fig . [ fig : squarewell_profiles ] is chosen prior to their collision with the structure shown in fig . [ fig : squarewell_shock_rarefaction ] . in a famous contribution johansson @xcite considered the tasep with @xmath385 step initial conditions and particles hopping only to the left . he proved that the time - integrated current along a given ray @xmath386 with @xmath387 behaves for large @xmath388 as @xmath389 following standard conventions , the time scale is denoted by @xmath390 , in our particular case @xmath391 . @xmath392 is the overall sign of the amplitude . the amplitude itself , @xmath393 , is a tracy - widom gue distributed random variable , which was originally obtained as the distribution of the largest eigenvalue of a gue random matrix @xcite . in formulas , @xmath394 with the airy kernel @xmath395 @xmath396 denoting the standard airy function . @xmath397 , @xmath398 , and @xmath399 are computed , model - dependent parameters , while the exponent @xmath400 and @xmath401 are universal . using distinct methods , later the result was extended to a general initial step @xcite and also to the asep @xcite . the euler equation for the tasep reads @xmath402 @xmath403 the particle density . to have tracy - widom fluctuations the solution to the @xmath404 riemann problem for has to develop a rarefaction wave and the ray of integration must lie in the interior of the wave . in fact , the rarefaction profile happens to be linear , as for the leroux lattice gas . in general the profile will be nonlinear . still , provided @xmath15 is properly chosen , asymptotically the fluctuations of the time - integrated current are expected to have the same probability law as in . we regard this observation as a strong indication that also the leroux lattice gas , even more ambitiously anharmonic chains , has tracy - widom statistics for the time - integrated current . there is one immediate difficulty with such a conjecture . the current is a vector . so which linear combination has a statistics governed by @xmath393 ? we will first study the fluctuations of time - integrated currents abstractly and then specialize to the leroux lattice gas and anharmonic chains with square - well type potential including the hard - point limit . to define the time - integrated current , in general , let us start from a conservation law of the form @xmath405 thus , as a property special for one dimension , the vector field @xmath406 is curl - free and hence admits a potential , @xmath407 , up to a constant which we fix by @xmath408 . @xmath407 is then the current integrated from @xmath409 to @xmath176 along an arbitrary integration path . numerically a convenient choice , to be used later on , is @xmath410 assuming @xmath411 , @xmath412 . for a system with @xmath4 components , the same definition applies to each component separately and we set @xmath413 . clearly , the same argument works also for a spatial lattice with the @xmath414-integration replaced by a lattice sum . we now consider an @xmath4 component hyperbolic conservation law in the form @xmath415 the linearization matrix @xmath35 has eigenvalues @xmath416 , left eigenvectors , @xmath417 , and right eigenvectors , @xmath418 , @xmath419 . the eigenvalues are assumed to be non - degenerate . we consider the @xmath420 riemann problem such that its solution contains a rarefaction wave across which @xmath421 increases ( or decreases ) smoothly for @xmath422 . the current is integrated along the ray @xmath386 , which has to lie inside the rarefaction wave , i.e. @xmath423 . the rarefaction wave is associated with a particular eigenvalue , whose label is denoted by @xmath353 and regarded as fixed in the following . the quantity of interest is the distribution of integrated current @xmath424 . along @xmath386 the fields take the value @xmath425 and @xmath426 . hence , averaging in local equilibrium to leading order in @xmath388 , @xmath427 to access fluctuations we consider a point on @xmath386 , field value @xmath425 , and want to study small fluctuations with shape function @xmath428 which varies on a scale small compared to the variation of the rarefaction wave . thus we have to linearize relative to a homogeneous background @xmath425 . the resulting time evolution is given by @xmath429 where @xmath430 denotes the scalar product for @xmath4-vectors and @xmath431 . since @xmath426 , only the term with @xmath432 propagates along the ray @xmath433 , while all other components separate from it linearly in time . hence only @xmath434 can build up anomalous fluctuations . in case of a single component , the quadratic term is responsible for the @xmath0 fluctuations . for several components , one first has to transform to normal modes . the strength of the self - coupling is denoted by @xmath435 . the slope of the rarefaction profile does not vanish identically , hence classified to be in the kpz universality class of curved profiles . for it the time scale of the anomalous fluctuations is set by @xmath436 . thus we conjecture that for large @xmath388 @xmath437 any linear combination of currents other than in eq . encounters to some part almost independent contributions . hence , if @xmath438 is not parallel to @xmath439 , the standard central limit theorem should apply in the form @xmath440 where @xmath441 is a standard gaussian random variable . we have no theoretical prediction for the value of @xmath442 . of course it has to vanish as @xmath438 tends to @xmath439 . as explained in appendix [ appendix : g ] , the universal scale factor @xmath436 can be computed from thermal averages . there we also establish that @xmath443 thus genuine nonlinearity is equivalent to @xmath444 having a definite sign for all admissible @xmath223 , @xmath183 . to shorten notation , we define the projected current components in , with their asymptotic value subtracted , as @xmath445 these are referred to as normal modes of the current . [ [ leroux - model . ] ] leroux model . + + + + + + + + + + + + + we record the integrated current @xmath424 for the simulation parameters as in fig . [ fig : leroux_profiles ] above , with @xmath446 and @xmath433 marked as purple vertical line in fig . [ fig : leroux_profiles ] . the integrated current is then transformed to normal modes via eq . , using the theoretical values for @xmath439 , @xmath447 and @xmath448 . the resulting @xmath449 is shown in fig . [ fig : leroux_scaling_distribution ] and compared with the theoretical predictions and . the top row shows the standard deviation of @xmath449 as a function of time , in comparison with @xmath450 for @xmath87 and @xmath451 for @xmath89 , where @xmath452 denotes the standard deviation of the tracy - widom distribution . the corresponding probability density functions of @xmath449 in the bottom row of fig . [ fig : leroux_scaling_distribution ] are reproduced from @xcite , and accurately match the predicted tracy - widom and gaussian distributions , respectively . note that the rescaling uses the theoretical value @xmath453 , see also appendix [ appendix : leroux ] . however there is still a global shift by @xmath454 . such a shift is familiar from one - component models . the higher cumulants have all relaxed , while the mean is still drifting . + [ [ hard - point - particles - with - alternating - masses . ] ] hard - point particles with alternating masses . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + analogous to the leroux model , for the hard - point particle chain with alternating masses we integrate stretch , velocity , and energy currents in md simulations along the purple ray in figs . [ fig : hardpoint_finite_vol_summary ] and [ fig : hp_profiles ] , to obtain @xmath424 . for anharmonic chains there are three normal modes @xmath449 , @xmath455 , which we compute from @xmath424 via , again using the theoretical values for @xmath439 , @xmath447 and @xmath448 . only the @xmath87 mode is expected to follow a tracy - widom distribution . [ fig : hardpoint_scaling_phi_distribution ] shows the simulation results for @xmath449 in comparison with @xmath401 for @xmath87 and normal distributions for @xmath456 . as before , a small correction to the mean values is indicated at the @xmath328-axis labels . the numerical fit uses @xmath457 , while the theoretical prediction is @xmath458 . + + [ [ square - well - potential . ] ] square - well potential . + + + + + + + + + + + + + + + + + + + + + + to also have an example where genuine nonlinearity is violated , we repeat the analogous analysis for md simulations with square - well interaction potential , using the parameters as in figs . [ fig : squarewell_shock_rarefaction ] and [ fig : squarewell_profiles ] . the current is integrated along the ray @xmath386 with @xmath459 . to avoid collisions due to the periodic boundary conditions , the largest simulation time is relatively short , @xmath460 . the results for @xmath449 are shown in fig . [ fig : squarewell_scaling_phi_distribution ] in comparison with the theoretical predictions . different from the previous two examples , we show the simulation results without mean value correction . the agreement is slightly less precise as for the alternating masses . one reason could be due to the mentioned short simulation time . nevertheless , one clearly observes that the mean value of @xmath461 is close to that of @xmath393 , whereas @xmath449 for @xmath456 are approximately centered around zero , in accordance with the theoretical prediction . the numerically fitted coefficient @xmath462 , whereas the theoretical prediction is @xmath463 . the other two numerical coefficients are @xmath464 and @xmath465 . in fig . [ fig : squarewell_scaling_phi_distribution ] , the sign of @xmath466 for the transformation to normal modes is flipped , in accordance with the changing sign in eq . . we studied anharmonic chains with nonequilibrium initial conditions , specifically domain - wall . two rather distinct theoretical predictions have been tested against molecular dynamics simulations . + ( i ) the validity of the macroscopic euler equations , which predict flat profiles interrupted by shocks and rarefaction waves . our system size is moderate , roughly 4000 particles , which we try to compensate by averaging over order @xmath467 configurations sampled according to the initial domain - wall state . for times up to the crossing time for a sound wave , the agreement between the solution of the euler equations and the simulation is fairly accurate . exceptions are shock fronts , where additional oscillatory structures may appear . also the borders of rarefaction waves are somewhat rounded . + ( ii ) we measured currents integrated along a ray in space - time . if this ray lies in a domain with flat profile , one expects @xmath13 size gaussian fluctuations . we confirmed such a behavior in a related set - up @xcite and did not repeat it here . however , for a ray inside a rarefaction wave we do obtain @xmath0 size fluctuations with tracy - widom gue distributed random amplitude , provided the current is projected onto the respective left eigenvector . such a behavior is observed even in case the rarefaction wave does not perfectly approximate the euler solution . turning to stochastic particle dynamics with several components , as to be expected , the precision of the approximation is improved . our working example is the leroux lattice gas . in this model , even strong shocks are realized by the dynamics with a width of a few lattice sites . the tracy - widom distribution is confirmed at a precision known for the one - component asep . [ [ acknowledgements . ] ] acknowledgements . + + + + + + + + + + + + + + + + + the work of hs has been supported as a simons distinguished visiting scholar , when visiting the kitp early 2016 . cm acknowledges support from the alexander von humboldt foundation and computing resources of the leibniz - rechenzentrum . we compute the @xmath468 coupling matrices for the leroux model and anharmonic chains , following the derivation in @xcite . the linearization matrix @xmath35 and its left and right eigenvectors are stated in eqs . and the transformation to normal modes is accomplished through the matrix @xmath469 defined by @xmath470 by construction one has @xmath471 as usual , the static susceptibility matrix , @xmath472 , is given by @xmath473 in addition we then require @xmath474 thereby fixing the normalizations @xmath41 , @xmath40 in to @xmath475 @xmath476 to obtain the nonlinear couplings @xmath468 , in particular @xmath477 , we first compute the hessians of the current as second derivatives of @xmath478 , @xmath479 in normal coordinates @xmath480 and @xmath481 the coupling matrices are thus obtained as @xmath482 in particular , comparison with shows that @xmath483 to set the scale for the tracy - widom distribution , one has to compute @xmath484 . for a general anharmonic chain , in the special case @xmath485 , the coupling matrices are derived in @xcite , a result which should be extended to @xmath486 . in fact , it turns out that the coupling matrices do not depend on @xmath74 . following the notation of @xcite , the static susceptibility matrix is given by @xmath487 the linearization matrix @xmath35 in and its right and left eigenvectors in and , respectively , define the transformation to normal modes via @xmath488 such that @xmath489 to have @xmath490 , the normalization constants of the eigenvectors must satisfy @xmath491 an explicit computation of the diagonal entries of @xmath492 shows that the velocity terms cancel . hence the relations @xmath493 from @xcite remain valid in general , where @xmath494 . as in @xcite , we denote the hessian matrices of the average current by @xmath495 with the conserved fields @xmath496 and the current vector defined in . the coupling matrices are then given by @xmath497 for @xmath455 . while the hessian matrices @xmath498 depend on @xmath74 , the coupling matrices are actually independent of @xmath74 . thus using the formulas in @xcite one arrives at @xmath499 and for @xmath50 @xmath500 \begin{pmatrix } 0 & 0 & 0 \\ 0 & 1 & 0 \\ 0 & 0 & 0 \end{pmatrix } \\ & \quad + \frac{\sqrt{\upsilon}}{2 \sqrt{m}\,c } \left[(\partial_r p ) ( \partial_e c ) - ( \partial_e p ) ( \partial_r c ) \right ] \begin{pmatrix } 0 & 1 & 0 \\ 1 & 0 & -1 \\ 0 & -1 & 0 \end{pmatrix}. \end{split}\ ] ] the relation follows by using on both sides the expressions provided above . the coupling constants for these models have been discussed already in appendix a of @xcite . for completeness , here we adapt to the current sign convention for the eigenvectors and using the velocity ( instead of momentum ) as field variable . the linearization matrix @xmath35 and its right eigenvectors are stated in eqs . and . the corresponding left eigenvectors of @xmath35 are @xmath502 since the interaction potential is either zero or infinite , @xmath503 , and the normalization constants in become @xmath504 specializing and to the square - well interaction potential leads to the coupling matrices @xmath505 and for @xmath50 @xmath506\ ] ] with @xmath507 as above , the signs of some entries in @xmath508 are flipped compared to @xcite , due to different sign conventions for the eigenvectors of @xmath35 . a. bressan . http://dx.doi.org/10.1007/978-3-642-32160-3_2[hyperbolic conservation laws : an illustrated tutorial ] , pages 157245 . in _ modelling and optimisation of flows on networks , cetraro , italy 2009 _ , lecture notes in mathematics 2062 , springer , 2013 . f. arndt , t. heinzel , and v. rittenberg . breaking of translational invariance in one - dimensional stationary states on a ring ] . , 31:l45l51 , 1998 . h. a. bethe . the theory of shock waves for an arbitrary equation of state ] , pages 421495 . in _ classic papers in shock compression science _ , springer , 1998 . m. prhofer and h. spohn . http://dx.doi.org/10.1007/978-1-4612-0063-5_7[current fluctuations for the totally asymmetric simple exclusion process ] , pages 185204 . in _ in and out of equilibrium : probability with a physics flavor _ , progress in probability 51 , birkhuser boston , 2002 .

we determine the distribution of nearest neighbour spacings between the tangencies to a fixed circle in a class of circle packings generated by reflections . we use a combination of geometric tools and the theory of automorphic forms . 320755 ]

1509.02989

in recent years , there has been increasing interest in the quantitative study of aspects of circle packings , in particular for apollonian packings . for instance , the asymptotic count of the number of circles with bounded curvature has been determined @xcite , as well as several questions concerning the arithmetic of apollonian packings , see @xcite . in this paper , we investigate local statistics for the distribution of tangencies to a fixed circle in a circle packing , in a class of packings generated by reflections . the packings that we study are formed by starting with an initial finite configuration @xmath0 of circles in the plane with disjoint interiors , and such that the gaps between circles are curvilinear triangles , as in figure [ fig configurations ] . we then form the group @xmath1 generated by reflections in the dual circles defined by the triangular gaps of the configuration , see [ sec : construct ] , and by applying the elements of @xmath1 to the initial configuration @xmath0 , we obtain an infinite circle packing @xmath2 , see figure [ fig configurations ] for an example . we create such initial configurations by starting with a finite triangulation @xmath3 of the riemann sphere @xmath4 . by the circle packing theorem of koebe , andreev and thurston @xcite , there is a circle packing @xmath0 of @xmath5 having @xmath3 as its tangency graph , that is a collection of disks with disjoint interiors on @xmath5 corresponding to the vertices of @xmath3 where two disks are tangent if and only if the corresponding vertices are connected in @xmath3 . the gaps between the disks are the connected components of the complement @xmath6 of the disks in @xmath5 . they are triangular because we assume that @xmath3 is a triangulation , that is each face of @xmath3 is a triangle . after stereographic projection from the point @xmath7 , we obtain a circle packing @xmath8 in the finite plane @xmath9 . we fix a base circle @xmath10 and consider the subset @xmath11 of circles tangent to @xmath12 . let @xmath13 be an arc ( if @xmath12 is a line , take @xmath14 to be a bounded interval ) . let @xmath15 be the set of tangencies in @xmath14 , and @xmath16 be the subset of @xmath15 whose corresponding circles in @xmath17 have curvatures bounded by @xmath18 , see figure [ fig tangencies ] . [ b ] associated to the apollonian-9 packing of figure [ fig configurations ] , with @xmath12 being the bounding circle.,width=188 ] we will show that the cardinality of @xmath16 grows linearly with @xmath18 : @xmath19 where @xmath20 is the standard arclength of @xmath14 , and @xmath21 is a constant depending only on @xmath2 and @xmath12 . in particular , the tangencies are uniformly distributed in @xmath12 . to compare , the total number of circles @xmath22 with @xmath23 is @xmath24 for some @xmath25 @xcite . the goal of our paper is to study the distribution of nearest neighbour spacings ( gaps ) in @xmath26 . let @xmath27 be the sequence of tangencies in @xmath16 ordered by some orientation ( say , the counter - clockwise orientation ) . the nearest - neighbour gaps , or spacings , between the tangencies are @xmath28 where @xmath29 denotes the arc - length distance . the mean spacing is @xmath30 we define the gap distribution function to be @xmath31 clearly the definition of @xmath32 does not depend on the orientation that we choose . we will show there is a limiting gap distribution , which is conformally invariant and independent of @xmath14 : [ gapdistributionintro ] given @xmath33 as above , there exists a continuous piecewise smooth function @xmath34 , such that @xmath35 moreover , the limit distribution @xmath36 is conformally invariant : if @xmath37 , @xmath38 , @xmath39 then @xmath40 our method does not just give the existence , but actually explicitly calculates the limiting spacing distribution in terms of certain areas , showing that it is a continuous , piecewise smooth function . we also show that @xmath41 is supported away from the origin , which is a very strong form of `` level repulsion '' . in [ sec : examples ] we explicitly compute the limiting distribution for some examples . as a warm - up , we start with classical apollonian packings , where we start with three mutually tangent circles and then fill in each curvilinear triangle with the unique circle which is tangent to all three sides of that triangle , and then repeat this process with each newly created curvilinear triangle . in that case we recover a theorem of hall @xcite on the gap distribution of farey points . for a history and further results in this direction see @xcite , and see @xcite for treatments using homogeneous dynamics . we then compute the gap distribution for two other classes of packings : the guettler and mallows @xcite packing in figure [ fig configurations two ] , called also apollonian-3 , where the curvilinear triangle is filled by three new circles , each tangent to exactly two sides , and for which the associated tangency graph is the octahedron ; and for the configuration ( called apollonian-9 ) found by butler , graham , guettler and mallows ( * ? ? ? * figure 11 ) , for which the tangency graph is the icosahedron , as in figure [ fig configurations ] . the density functions of these three cases are displayed in figure [ fig : threeappdensity ] . of the gap distribution for classical apollonian packings ( dashed curve ) , which is the same as that discovered by hall for farey sequence , for the apollonian-3 packings of guettler and mallows ( dotted curve ) , and for the apollonian-9 packing of butler , graham , guettler and mallows ( solid curve).,width=453 ] we work on models where @xmath8 consists of two horizontal lines , with @xmath42 and @xmath43 , and the remaining circles lies in the strip between them . we call such models generalized ford configurations . let @xmath44 be the subgroup of @xmath1 consisting of orientation preserving elements of @xmath1 that fix the base circle @xmath12 . it is a non - uniform lattice in @xmath45 ( this group was used crucially in sarnak s work @xcite on apollonian packings ) . using geometric considerations , we reduce the spacing problem to counting elements @xmath46 of the fuchsian group @xmath44 such that the lower row @xmath47 lies in a dilated region defined by the intersection of certain quadratic inequalities , and such that @xmath48 lies in a given interval . after that , we use the spectral theory of automorphic forms , specifically the work of anton good @xcite , to show that the number of lattice points is asymptotically a multiple of the area of this dilated region . in [ sec : construct ] we give details how to construct @xmath2 from the initial configuration @xmath49 . we will also construct related groups and prove their geometric properties . in [ sec : uniform dist of tgts ] and [ fordgapdistribution ] we will prove the uniform distribution of tangencies and theorem [ gapdistributionintro ] in the special case that @xmath2 is a generalized ford packing , and @xmath12 is one of the two lines from @xmath2 . in [ transfer ] we show how to deduce the general case of and theorem [ gapdistributionintro ] from the results of [ fordequidistribution ] and [ fordgapdistribution ] . in [ sec : examples ] we compute some examples . we reduce the computational effort involved by using conformal invariance , observing that the packings in the examples admit extra symmetries . for instance , in the case of apollonian-9 packings , the associated symmetry group is the non - arithmetic hecke triangle group @xmath50 . we start with a ( finite ) triangulation @xmath3 of the sphere . by the circle packing theorem of koebe , andreev and thurston @xcite , there is a circle packing @xmath0 of @xmath51 having @xmath3 as its tangency graph , that is a collection of disks with disjoint interiors on @xmath51 corresponding to the vertices of @xmath3 ( there must be at least @xmath52 vertices ) , where two disks are tangent if and only if the corresponding vertices are connected in @xmath3 . the interstices , or gaps , between the disks , are the connected components of the complement @xmath53 of the disks in @xmath51 . because we assume that @xmath3 is a triangulation , that is each face of @xmath3 is a triangle , this means that the gaps are triangular . we may identify @xmath51 with the riemann sphere @xmath4 , and then by picking a point which will correspond to @xmath7 and performing a stereographic projection onto the plane we obtain a circle packing @xmath8 in the finite plane @xmath9 . if @xmath7 is contained in the interior of one of the disks , then @xmath49 is realized as a configuration of finitely many circles , with one circle containing the other ones , see figure [ fig configurations two ] . if @xmath7 is a point of tangency of two disks , then @xmath49 is a configuration consisting of two lines and several other circles in between ; we call this type of configuration a generalized ford configuration , see figure [ fig : apollonian-3fordbis ] . we start with a circle packing @xmath8 as above , of @xmath54 tangent circles @xmath55 in the plane , so that the gaps between circles are curvilinear triangles . we build up a circle packing @xmath56 from @xmath8 by refections : for each triangular gap , we draw a dual circle which passes through the vertices of the triangle , see figure [ fig : apo config ] . is the dashed circle.,width=188,height=188 ] we can generate new circles in the gap by reflecting @xmath8 through the dual circles . each of the circles forming the gap is fixed under this reflection , and the remaining circles are reflected into the gap . denote these reflections by @xmath57 and the corresponding triangles by @xmath58 , where @xmath59 is the number of gaps . note that all these reflections are anti - holomorphic maps in @xmath9 ( throughout this paper , by triangle we mean solid triangle with vertices removed ) . let @xmath60 be the group generated by the reflections . the reflection group @xmath61 has only the relations @xmath62 . this can be seen from the ping pong lemma : @xmath63 are disjoint sets , and @xmath64 maps every @xmath63 into @xmath65 when @xmath66 . if we apply @xmath61 to the configuration @xmath8 , we obtain a circle packing @xmath2 of infinitely many circles . pick a circle @xmath12 from @xmath2 , and let @xmath17 be the circles tangent to @xmath12 and let @xmath67 . we are interested in studying the counting and spacing problems on @xmath17 . the first step is to place @xmath2 in an ambient hyperbolic space with @xmath12 being the boundary , and show that there is a finite covolume fuchsian group @xmath44 acting on @xmath17 and having finitely many orbits . a general element in @xmath61 is given by @xmath68 with @xmath69 . therefore each circle @xmath70 can be represented by @xmath71 , where @xmath72 . we denote the triangular gap corresponding to @xmath73 by @xmath74 . we define the length of the above word to be @xmath75 . we say the word above is _ minimal _ for @xmath76 if we can not find any shorter expression , and we say @xmath76 is in @xmath75-th generation . we first show that there is a canonical way to express a circle with minimal length : [ g1]the minimal word for a given circle @xmath70 is unique . moreover , @xmath77 is minimal if and only if we have @xmath78 we first prove the only if " direction by induction on the generation of @xmath76 . for the base case first generation , if @xmath79 is minimal , then @xmath80 , otherwise @xmath81 will fix @xmath82 and then the word is not minimal . as a result , @xmath83 . suppose the only if " direction holds for generations up to @xmath84 and @xmath85 is a minimal word for @xmath76 , then @xmath86 has to be a minimal word for some circle . by our induction hypothesis , @xmath87 and @xmath88 . since @xmath89 ( otherwise @xmath90 and the word is not minimal ) , we must have @xmath91 . from we see that @xmath81 corresponds the unique triangle in @xmath92 where @xmath93 sits in , @xmath94 corresponds the unique triangle in @xmath92 where @xmath94 sits in , @xmath95 , @xmath96 corresponds he unique triangle in @xmath92 where @xmath97 sits in . this canonical description shows the uniqueness of minimal expression for a circle . we prove the if " direction by induction on the length of expression . if @xmath80 , then @xmath79 is a circle lying @xmath98 . the only circles with shorter words are the ones in @xmath99 , all of which do not lie in @xmath98 , so @xmath79 is minimal . now suppose @xmath100 is the minimal word for @xmath76 . then by our discussion above @xmath101 is the unique reflection that corresponds to the triangle in @xmath92 where @xmath76 sits in , so @xmath102 . then @xmath103 satisfies the condition and the minimal word for @xmath93 is @xmath104 . by our induction hypothesis , @xmath105 , @xmath106 for each @xmath107 . if @xmath108 , we write @xmath109 , and define @xmath110 to be the _ minimal _ triangle to @xmath76 . clearly @xmath76 is contained in @xmath111 . now let @xmath112 be the subgroup of @xmath1 consisting of elements that fix a given circle @xmath12 . by relabeling let @xmath113 be the set of circles that are tangent to @xmath12 , and their tangencies are @xmath114 . let @xmath115 be the set of generators of @xmath1 that fix @xmath12 , where @xmath116 is the reflection corresponding to the triangle formed by @xmath117 . for our convenience , at this point we make the convention that @xmath118 and @xmath119 for @xmath120 , because @xmath121 s from @xmath122 are forming a loop around @xmath12 . @xmath123 is generated by the reflections in @xmath124 . if @xmath125 where @xmath126 and @xmath127 , we first absorb the trivial identities @xmath128 if @xmath129 , then we cut the first and last few consecutive letters of the word @xmath130 that are from @xmath124 ( which may not exist ) . we continue this procedure several times until it stabilizes , then we get a word for @xmath12 , which has to satisfy condition so is minimal . but the minimal word for @xmath12 is just @xmath12 . since each cutting corresponds to multiplying an element in @xmath131 , we have @xmath132 . @xmath112 acts on @xmath17 , and has finitely many orbits on @xmath17 . @xmath112 acts on @xmath17 because @xmath1 send pairs of tangent circles to pairs of tangent circles . since @xmath112 fixes @xmath12 , it will send any circle tangent to @xmath12 to some circle tangent to @xmath12 . now suppose @xmath133 and the minimal word for @xmath76 is @xmath134 . from the argument in lemma 2.1 , @xmath135 corresponds to a triangle @xmath136 , which has to be a triangle that contains part of @xmath12 where @xmath76 lies in , so @xmath81 has to be from @xmath124 . by the same argument , one can show that @xmath137 , and @xmath138 . therefore , the number of orbits is @xmath139 , the cardinality of @xmath122 . now we put a hyperbolic structure associated to @xmath44 . without loss of generality we assume that @xmath12 is the bounding circle of radius @xmath140 , whose interior is the unit disk @xmath141 , and let @xmath142 be the geodesic connecting @xmath143 and @xmath144 ( again we extend the definition for all @xmath120 by setting @xmath142=@xmath145 ) , so that the region bounded by @xmath142 and @xmath12 contains the triangle @xmath63 . each @xmath146 preserves the metric in @xmath141 , so @xmath147 , the isometry group of @xmath141 . let @xmath148 be the open region bounded by the loops formed @xmath142 s ( see figure [ fig : fundamentaldomain ] ) . for @xmath112 . , width=151 ] the following theorem , due to poincar ( see ( * ? ? ? * theorem 7.1.3 ) ) , tells us that @xmath112 acts discontinuously on @xmath141 with fundamental domain @xmath148 . let @xmath149 be a finite - sided , convex polyhedron in @xmath141 of finite volume all of whose dihedral angles are submultiples of @xmath150 . then the group generated by the reflections in the sides of @xmath149 is a discrete reflection group with respect to the polyhedron @xmath149 . as corollaries , @xmath112 has @xmath151 as cusps , and the fundamental domain for @xmath112 has volume @xmath152 by gauss - bonnet . for our purpose , we want to take the orientation preserving subgroup of @xmath112 , which we denote by @xmath44 . the group @xmath44 is a free , index-2 subgroup of @xmath112 generated by @xmath153 . to see @xmath44 is free , we apply the pingpong lemma to @xmath154 , which maps @xmath65 to @xmath63 when @xmath66 . from the properties of @xmath112 , we immediately see : [ propertyofgamma ] @xmath44 has cusps at @xmath143 , with stabilizer @xmath155 . the area of a fundamental domain is @xmath156 . we now study the distribution of tangencies on a fixed circle from a circle packing . recall that @xmath2 is generated from a finite circle packing @xmath49 , which is associated to a triangulated spherical graph . we pick a circle @xmath157 and let @xmath17 be the collections of circles in @xmath56 which are tangent to @xmath12 . given a bounded arc @xmath13 , let @xmath16 be the collection of tangencies of circles from @xmath17 , whose curvatures are bounded by @xmath18 . let @xmath158 be the standard arclength measure . we will show that [ equidistribution ] as @xmath159 , @xmath160 with @xmath21 independent of @xmath14 given by @xmath161 where @xmath162 is the hyperbolic area of the region bounded by @xmath121 and the two geodesics linking @xmath143 to @xmath144 and @xmath143 to @xmath163 ( see figure [ fig : dregions ] ) . moreover , @xmath21 is conformal invariant : if @xmath37 , @xmath164 , and @xmath165 , then @xmath166 . we deduce theorem [ equidistribution ] from some old results of a. good @xcite , but it is in principle known ; for instance it is a special case of @xcite . we will first prove theorem [ equidistribution ] for a generalized ford packing , when the initial configuration contains two parallel lines , one of them being the base circle @xmath12 . the general case will follow by conformal invariance ( theorem [ transferprinciple ] ) whose proof is deferred to [ transfer ] . for an explicit formula for the areas @xmath162 in the case of a ford configuration , see . we first show that any of our initial configurations may be brought into a standard shape , that of a ( generalized ) ford configuration , see figure [ fig : apollonian-3fordbis ] . .,width=264 ] we can map the initial configuration @xmath49 by using a mbius transformation and possibly a reflection , to one which includes two parallels @xmath42 and @xmath167 , two circles tangent to both of @xmath168 and @xmath169 with their tangencies on @xmath168 is 0 and @xmath170 respectively and some circles in between . we may use a mbius transformation of the riemann sphere to map the configuration to the interior of the strip @xmath171 as follows : pick a triangle in the tangency graph @xmath3 one of whose vertices corresponds to @xmath12 , and whose other vertices corresponds to two other circles @xmath172 and @xmath173 , so that @xmath12 , @xmath172 , @xmath173 form a triple of mutually tangent circles . setting @xmath174 , @xmath175 , @xmath176 to be the tangency points , corresponding to the edges of the triangle , there is a ( unique ) mbius transformation @xmath37 which maps @xmath177 then @xmath178 will map @xmath12 and @xmath172 to a pair of tangent lines passing through @xmath179 and @xmath180 respectively , and @xmath173 will be mapped to a circle tangent to both at these points . hence @xmath178 will map @xmath12 to the real axis @xmath181 , @xmath172 to the line @xmath182 and @xmath173 to the circle @xmath183 centered at @xmath184 and having radius @xmath185 , and the remaining circles all lie in the strip @xmath186 . the triangular gap @xmath187 between @xmath12 , @xmath172 , @xmath173 is mapped into one of the two triangular gaps forming the complement in the strip @xmath186 of the circle @xmath183 . after applying if necessary the reflection @xmath188 , we may assume that @xmath187 is mapped to the gap in the half plane @xmath189 , so that all the other circles in @xmath8 are mapped into the gap with @xmath190 . since any triangulation of the sphere has at least @xmath52 vertices , there must be at least one such circle distinct from @xmath12 , @xmath172 , @xmath173 . since in the resulting configuration the gaps are still triangular , that means that there is one other circle tangent to both @xmath168 and @xmath169 , say @xmath191 ( @xmath192 ) and that the images of all the other circles are mapped to circles lying between @xmath193 and @xmath194 , as in figure [ fig : apollonian-3fordbis ] . we carry over all previous notations . the initial configuration @xmath49 includes two parallel lines @xmath42 and @xmath167 , two circles @xmath195 and @xmath196 tangent to both of @xmath168 and @xmath169 , and some circles in between . our base circle @xmath12 is @xmath168 , so @xmath197 and @xmath198 is the collection of all circles tangent to @xmath168 . the reflection @xmath199 corresponds to the triangle formed by @xmath168,@xmath169 and @xmath200 . applying @xmath201 to @xmath49 , the fundamental domain for @xmath44 is formed by geodesics connecting tangencies of @xmath49 and @xmath202 on @xmath168 . each tangency is a parabolic point for @xmath44 , and since @xmath203 , we have @xmath204 , from proposition [ propertyofgamma ] . we record a computation which will be used at several places : [ curvatureformula ] let @xmath205 . i ) if @xmath206 , then under the mbius transform @xmath178 , a circle @xmath207 will be mapped to the circle @xmath208 if @xmath209 , and to the line @xmath210 if @xmath211 . when @xmath212 , the image circle is @xmath213 ii ) if @xmath206 then the lines ( degenerate circles ) @xmath214 are mapped to @xmath215 , and to the line @xmath216 if @xmath212 . for both i ) and ii ) , if @xmath206 , we use the bruhat decomposition to write @xmath178 as f @xmath217 for each factor above the transformation formula is simple , then the curvature of the image circle is obtained by composing these simple formulae together . the case when @xmath212 is a simple check . to deduce theorem [ equidistribution ] , we need to calculate the contribution from each @xmath218 . for @xmath219 , write @xmath220 . from lemma [ curvatureformula ] it follows that @xmath221 sends @xmath121 , @xmath222 , to a circle of curvature @xmath223 and the line @xmath167 to a circle of curvature @xmath224 . therefore , for @xmath219 , we need to calculate @xmath225 and for @xmath226 , we need to calculate @xmath227 this counting problem is a special case of a theorem of a. good , ( see the corollary on page 119 of @xcite ) , which we quote here : [ anton ] let @xmath44 be a lattice in @xmath228 . suppose @xmath7 is a cusp of @xmath44 with stabilizer @xmath229 and @xmath230 is any cusp of @xmath44 with stabilizer @xmath231 . choose @xmath232 so that @xmath233 . any @xmath234 can be written uniquely in the form @xmath235 with @xmath236 , so this determines functions @xmath237 on @xmath228 . let @xmath238 be two bounded intervals in @xmath168 . then as @xmath159 , @xmath239 now apply theorem [ anton ] to deduce theorem [ equidistribution ] . for @xmath219 , we set @xmath240 , and @xmath241 so that @xmath242 . therefore , @xmath243 because @xmath244 contributes half of the fundamental period of the stabilizer at @xmath7 of @xmath245 ; the other half comes from @xmath246 . the map @xmath247 is a bijection of the cosets @xmath248 and @xmath249 . write @xmath250 , then @xmath251 ( when either @xmath163 or @xmath144 is @xmath7 , this expression is interpreted as a limit ) . then the condition @xmath252 from is the same as @xmath253 , or @xmath254 . the condition @xmath255 is the same as @xmath256 , or @xmath257 . therefore , counting is the same as counting @xmath258 recall the area of @xmath44 is @xmath259 . then by setting @xmath260 , from theorem [ anton ] we obtain @xmath261 we interpret this in conformally invariant terms : let @xmath162 be the area of the region bounded by @xmath121 and the two geodesics @xmath262 , @xmath142 ( see figure [ fig : dregions ] ) . ( shaded ) bounded by the circles @xmath121 in the initial configurations , and by the two geodesics linking @xmath143 to @xmath163 and to @xmath144 . displayed is an initial configuration , generating the classical ford circles , consisting of two parallel lines @xmath42 and @xmath167 , and two mutually tangent circles @xmath263 and @xmath264 . here @xmath265 , @xmath266 , @xmath267 . , width=377 ] then @xmath268 will translate this region to a triangle with vertices @xmath7 , @xmath269 , and @xmath270 . note that @xmath271 is a point on the segment connecting @xmath270 and @xmath269 , and that @xmath272 . we also have @xmath242 . since @xmath268 is area preserving , we have @xmath273 hence we find @xmath274 the case of @xmath226 is simpler . counting is the same as counting @xmath275 where @xmath276 is a fundamental period of @xmath229 , the length of which is twice the distance of @xmath277 and @xmath278 , which is @xmath279 . we obtain @xmath280 noting that the area of the region bounded by @xmath172 and the two geodesics @xmath281 , @xmath282 is @xmath283 , we find @xmath284 adding up the contributions from all @xmath218 , we obtain theorem [ equidistribution ] , with the constant @xmath285 where @xmath162 is the hyperbolic area of the region bounded by @xmath121 and the two geodesics @xmath262 and @xmath142 . explicitly , @xmath286 ( when one of @xmath287 is @xmath7 , we take the appropriate limiting value of @xmath162 ) . we now study the gap distribution on @xmath12 . as in [ sec : uniform dist of tgts ] , let @xmath288 be an arc ( or a bounded interval ) , and @xmath289 the tangencies in @xmath14 whose circles have curvature at most @xmath18 . let @xmath27 be the sequence of tangencies in @xmath16 ordered by counter - clockwise direction . the nearest - neighbour gaps , or spacings , between the tangencies are @xmath290 and the mean spacing is @xmath291 we define the gap distribution function on @xmath14 to be @xmath292 we will find the limiting gap distribution of @xmath293 : [ gapdistribution ] there exists a continuous piecewise smooth function @xmath41 , which is independent of @xmath294 such that @xmath295 the limiting distribution @xmath41 is conformal invariant : let @xmath296 be as in theorem [ equidistribution ] , and @xmath297 be the gap distribution function of @xmath298 from @xmath299 , then @xmath300 . the explicit formula for @xmath301 is given in theorem [ thm : formula for f ] . this section is devoted to proving theorem [ gapdistribution ] in the case of generalized ford packings . recall that we assume the initial configuration @xmath49 includes two parallel lines @xmath168 and @xmath169 , two circles tangent to both of @xmath168 and @xmath169 with their tangencies on @xmath168 is 0 and @xmath170 respectively and possibly some circles in between . our base circle @xmath12 is @xmath168 , so @xmath197 and @xmath198 is the collection of all circles tangent to @xmath168 . [ comparecurv ] i ) let @xmath302 lie in a triangular gap @xmath187 bounded by the real axis @xmath168 and two mutually tangent circles @xmath303 . then both @xmath304 and @xmath305 have radius greater than that of @xmath306 . \ii ) if @xmath307 are such that @xmath308 lies in a triangular gap @xmath187 bounded by the real axis , which is disjoint from @xmath309 , then there is another circle @xmath310 which separates @xmath308 and @xmath309 and has smaller curvature ( i.e. bigger radius ) than that of @xmath308 . to see ( i ) , we may move the circle @xmath306 so that it is tangent to both the real line and the larger of the two initial circles @xmath304 and @xmath305 , say it is @xmath304 . we get a configuration as in figure [ fig geom arg ] . if we denote by @xmath311 the radius of the larger circle @xmath304 , and by @xmath312 the angle between the segment joining the centers of the two circles @xmath304 , @xmath305 , and the segment joining the point of tangency of the circle @xmath304 with real line and with the center of @xmath304 , then a computation shows that the radius of the third mutually tangent circle @xmath305 is @xmath313 . this is an increasing function of @xmath314 . since the angle @xmath315 is smaller than @xmath316 for any circle @xmath306 contained in the triangular gap and tangent to both @xmath304 and the real line , it follows that the circle @xmath306 has smaller radius than that of @xmath305 as claimed . ( shaded black ) are such that @xmath308 lies in a triangular gap @xmath187 bounded by the real axis , with @xmath187 disjoint from @xmath309 , then there is another circle @xmath310 ( shaded gray ) , which is one of the bounding circles of @xmath187 , which separates @xmath308 and @xmath309 and has smaller curvature than that of @xmath308.,width=302 ] for ( ii ) , note if @xmath309 is disjoint from the triangular gap @xmath187 containing @xmath308 , then one of the bounding circles @xmath76 of @xmath187 separates @xmath308 and @xmath309 . by part ( i ) , @xmath76 has bigger radius than @xmath308 , see figure [ geom arg separate ] . the following proposition is a crucial ingredient in understanding the gap distribution : [ 12 ] for any adjacent pair @xmath317 with corresponding circles @xmath318 and @xmath319 , there is an element @xmath320 and @xmath321 such that @xmath322 and @xmath323 . the element @xmath324 is unique if @xmath318 , @xmath319 are disjoint , and if they are tangent then the only other element with this property is @xmath325 , where @xmath326 is the reflection in the dual circle corresponding to the triangle formed by @xmath327 and @xmath168 . let @xmath328 be the minimal word for @xmath329 , so that the triangle associated to @xmath318 is @xmath330 . let @xmath331 be the minimal word for @xmath332 and @xmath333 . let s suppose @xmath334 . suppose there exists an integer @xmath335 $ ] such that @xmath336 . we take @xmath59 to be minimal , then @xmath337 and @xmath338 , and @xmath339 and @xmath340 are two disjoint sets . since @xmath341 when @xmath342 , we see that @xmath343 maps @xmath339 and @xmath340 to two disjoint triangles lying in @xmath98 . these two triangles contain @xmath329 and @xmath332 respectively . we have a contradiction here because by lemma [ comparecurv](ii ) , there is another circle in @xmath17 separating @xmath329 and @xmath332 which has smaller curvature , so @xmath344 and @xmath345 are not neighbors . therefore @xmath341 when @xmath346 $ ] , and @xmath347 now we claim that @xmath348 fixes @xmath121 when @xmath349 . this will show that @xmath350 so that we may take @xmath351 , which satisfies @xmath352 and @xmath353 as required . otherwise , let @xmath354 $ ] again be the smallest integer such that @xmath355 , then @xmath340 contains @xmath356 , and @xmath340 and @xmath121 are disjoint . as a result @xmath357 maps @xmath121 and @xmath340 to two disjoint sets : @xmath344 and the triangle @xmath358 , the latter of which contains @xmath332 . therefore , by lemma [ comparecurv](ii ) , between @xmath329 and @xmath332 there must be a circle whose curvature is smaller than that of @xmath332 , so that @xmath318 and @xmath319 can not be adjacent in @xmath359 , contradiction . hence @xmath360 for @xmath361 $ ] and we have constructed @xmath351 . to address uniqueness of @xmath362 , assume @xmath363 also sends @xmath121 to @xmath329 , @xmath364 to @xmath332 , then @xmath365 fixes @xmath121 and @xmath364 simultaneously . therefore , if @xmath121 and @xmath364 are disjoint , @xmath366 ; if @xmath121 and @xmath364 are tangent , @xmath367 , where @xmath368 is the reflection corresponding to the triangle formed by @xmath121 , @xmath364 and @xmath168 . we now use proposition [ 12 ] to relate the gap distribution problem to a question of counting certain cosets in @xmath44 . for each pair of circles @xmath370 in the initial configuration which are tangent to @xmath12 , we will define regions @xmath371 in the right @xmath372-plane @xmath373 as follows . \a ) if @xmath374 , we require @xmath375 and in addition : \i ) if @xmath121 , @xmath364 are tangent , then @xmath376 satisfies @xmath377 and @xmath378 where @xmath379 , with @xmath380 is the reflection corresponding to the triangle formed by @xmath327 and @xmath168 . \ii ) if @xmath121 and @xmath364 are disjoint ( non - tangent ) , then @xmath376 satisfies @xmath381 \b ) if one of @xmath382 is @xmath7 , say @xmath383 , we require @xmath384 and in addition : \i ) if @xmath121 , @xmath364 are tangent , then @xmath376 satisfies @xmath385 and @xmath386 \ii ) if @xmath121 and @xmath364 are disjoint ( non - tangent ) , then @xmath376 satisfies @xmath387 note that @xmath388 is a finite union of convex sets and @xmath389 the result is : [ omegaij ] for any @xmath390 and @xmath391 , the circles @xmath392 are neighbors in @xmath393 if and only if @xmath394 lies in some @xmath388 in the right half plane , defined by the conditions \i ) if @xmath374 then we require , with or in the case that @xmath121 , @xmath364 are tangent , and if they are disjoint then we require . \ii ) if @xmath383 , then we require , with or in the case that @xmath121 , @xmath364 are tangent , and if they are disjoint then we require . let @xmath395 be the reflection in the line @xmath396 , corresponding to the triangular gap bounded by the lines @xmath168 , @xmath169 and the circle with tangency @xmath170 . from proposition [ 12 ] , passing to the orientation preserving subgroup @xmath44 , every ( unordered ) pair of neighbor circles can be expressed as @xmath397 or @xmath398 for some @xmath399 and @xmath390 with @xmath400 . the expression is unique when @xmath121 and @xmath364 are disjoint because the simultaneous stabilizer for @xmath121 and @xmath364 is trivial . however , we only need to calculate the contribution to @xmath301 from @xmath401 under @xmath44 : the contribution from @xmath402 is the same as that from @xmath403 because they are mirror symmetric under the reflection of the line @xmath404 . or more formally , @xmath405 and @xmath201 is metric preserving on @xmath168 . when @xmath121 and @xmath364 are tangent , then @xmath402 is also in the orbit of @xmath403 under @xmath44 . in this case , every pair of circles from @xmath406 can be uniquely expressed as @xmath407 for some @xmath399 . if two neighbouring circles in @xmath408 can be expressed as @xmath409 , then we need their curvatures to satisfy @xmath410 by lemma [ curvatureformula ] we have @xmath411 and so we get . all the circles in between @xmath412 and @xmath413 must have curvatures greater than @xmath18 . we see from lemma [ comparecurv ] that it suffices to check finitely many circles . a useful fact here is the following : [ inbetween ] if neither @xmath121 nor @xmath364 is the horizontal line from @xmath122 , which means @xmath374 , then for another circle @xmath138 , @xmath414 is in between @xmath412 and @xmath413 if and only if @xmath415 if @xmath364 is the horizontal line , so that @xmath383 , then @xmath414 is in between @xmath412 and @xmath413 if and only if @xmath416 the observation is that given real numbers @xmath417 and @xmath418 , a third real @xmath419 lies between them if and only if @xmath420 now take the numbers to be @xmath421 , @xmath422 and @xmath423 and compute @xmath424 from which the claim follows . in the case @xmath383 , we have @xmath425 which gives the claim . there are two cases : + case 1 : @xmath121 and @xmath364 are tangent . then @xmath426 does not lie between @xmath143 and @xmath427 , for all @xmath59 . if @xmath428 , then the circles @xmath429 for @xmath430 are mapped by @xmath362 to some circles in between @xmath412 and @xmath413 , and in this case we need @xmath431 and using lemma [ curvatureformula ] which gives @xmath432 we obtain condition . if @xmath433 , then from observation [ inbetween ] none of the circles @xmath434 lie between @xmath435 and @xmath436 , but the circles @xmath437 for @xmath430 are mapped by @xmath362 to some circles in between @xmath412 and @xmath413 ( recall that @xmath438 is the reflection corresponding to the triangle formed by @xmath327 and @xmath168 ) . in this situation we need @xmath439 by lemma [ curvatureformula ] , @xmath440 if @xmath379 , which gives condition . case 2 : if @xmath121 and @xmath364 are not tangent . we need to make sure that @xmath441 has curvature @xmath442 whenever @xmath441 is in between @xmath412 and @xmath413 . again from observation [ inbetween ] , if @xmath433 , then the circles whose tangencies lie between @xmath143 and @xmath427 will be mapped by @xmath362 to some circles between @xmath412 and @xmath413 ; if @xmath428 , then the complement of will be mapped in between @xmath412 and @xmath413 . thus we need @xmath443 if either @xmath444 or @xmath445 this gives condition once we use @xmath432 ( lemma [ curvatureformula ] ) . from lemma [ comparecurv ] , once or hold when @xmath121 , @xmath364 are tangent , or is satisfied if @xmath121 and @xmath364 are disjoint , then all circles between @xmath412 and @xmath413 have curvatures greater than @xmath18 . putting the above together , we have established proposition [ omegaij ] . we need to calculate the contribution from each pair of circles @xmath403 to the proportion @xmath446 of gaps of size at most @xmath447 , and we denote this quantity by @xmath448 : , @xmath449 we will later show that these have a limit as @xmath159 : @xmath450 a direct computation shows that the distance between @xmath412 and @xmath413 is given by @xmath451 . the average gap is asymptotically @xmath452 . therefore , the relative gap condition @xmath453 in the definition of @xmath454 can also be written as @xmath455 we will work with a simpler condition @xmath456 which will finally lead to a limiting gap distribution function @xmath41 which is continuous , then in retrospect working with will lead to the same limiting function @xmath41 , by the continuity of @xmath301 . if one of @xmath143 and @xmath427 is @xmath7 ( say @xmath383 ) , the above is changed to @xmath457 we define a region @xmath458 to be the elements @xmath459 satisfying ( or ) . then we have found that @xmath460 the region @xmath458 is compact because the condition already gives a compact region bounded by two sets of parallel lines with different slopes . note also that @xmath458 is a finite union of convex sets . from the defining equation , it is clear that the region @xmath458 grows homogeneously with respect to @xmath18 : @xmath461 as a consequence of the analysis above , we find that the normalized gaps in our circle packings all bounded away from zero : there is some @xmath462 so all gaps satisfy @xmath463 thus the limiting distribution @xmath41 is supported away from the origin . this is a very strong form of level repulsion , familiar from the theory of the farey sequence . to prove the assertion , note that we expressed the distribution function @xmath464 as a sum over all ( unordered ) pairs of distinct circles from the initial configuration @xmath370 : @xmath465 where @xmath466 here @xmath467 is the region in the plane of points lying in the compact sets @xmath468 of [ sec : determining ] , satisfying @xmath469 . @xmath470 with @xmath471 the sets @xmath468 are compact , and therefore the functions @xmath472 are bounded on them . thus if @xmath473 then @xmath474 and @xmath475 if @xmath476 . in the examples of section [ sec : examples ] , we have @xmath477 for the classical apollonian packing [ sec : app1 ] ( which reduces to the farey sequence by conformal invariance ) , @xmath478 for the apollonian-3 packing [ sec : app3 ] , and @xmath479 for the apollonian-9 packing [ sec : app9 ] . we now pass to the limit @xmath159 , by relating the counting problem encoded in our formula for @xmath480 to the area of the region @xmath481 . this is done via an equidistribution theorem of a. good . to formulate it , recall the iwasawa decomposition @xmath482 where @xmath483 and @xmath484 we can uniquely write any @xmath485 as @xmath486 we have the following joint equidistribution according to this decomposition , which is another special case of good s theorem ( see corollary on page 119 of @xcite ) : [ nak ] let @xmath14 be an bounded interval in @xmath168 , @xmath276 be an interval in @xmath487 , then as @xmath159 , @xmath488 theorem [ nak ] allows us to prove the following [ expanding ] for a nice " subset @xmath489 , @xmath490 as @xmath159 , where @xmath75 is the standard lebesgue measure in @xmath491 . here `` nice '' means bounded , convex with piecewise smooth boundary . we note that if @xmath492 then the iwasawa decomposition gives @xmath493 , @xmath494 so that @xmath495 are polar coordinates in the @xmath496 plane . we prove proposition [ expanding ] in the special case that @xmath497 is bounded by two continuous and piecewise smooth curves @xmath498 with @xmath499 $ ] , and @xmath500 . the sign @xmath501 is obtained only if when @xmath502 . this special case suffices for what we need for @xmath503 . this is essentially a riemann sum argument . first we can express @xmath504 as @xmath505 divide the interval @xmath506 $ ] into @xmath507 equal subintervals @xmath508 . for each subinterval @xmath509 , pick @xmath510 at which @xmath311 and @xmath511 achieve their maximum and minimum respectively . let @xmath512 be the union of truncated sectors @xmath513 $ ] , and @xmath514 be the union of @xmath515 $ ] . we have @xmath516 and @xmath517 we notice that the statement in proposition [ expanding ] satisfies finite additivity . from theorem [ nak ] , we know proposition [ expanding ] holds for sectors , thus it holds for finite union of truncated sectors , which is the case for @xmath518 and @xmath514 . letting @xmath507 go to infinity , we prove proposition [ expanding ] for @xmath497 . we can now state our main result , the formula for the limiting gap distribution @xmath41 , along the way proving theorem [ gapdistribution ] . we keep our previous notation . [ thm : formula for f ] for any interval @xmath288 , @xmath519 , where @xmath520 with @xmath521 given in . we saw that @xmath522 and we need to estimate @xmath454 . from proposition [ 12 ] and proposition [ omegaij ] we can rewrite @xmath454 as @xmath523 in terms of the iwasawa coordinates , the condition @xmath524 is essentially equivalent to @xmath525 , when @xmath526 tends to @xmath527 , as we shall explain now . writing @xmath528 , @xmath529 and @xmath530 , we have @xmath531 and since @xmath532 , this will essentially be @xmath533 provided @xmath534 is bounded away from zero . we first excise small sectors @xmath535 of angle @xmath536 containing as bisectors those @xmath537 s such that @xmath538 , where @xmath539 or @xmath540 . then for any @xmath541 , @xmath542 @xmath543 such that if @xmath544 and @xmath545 , then @xmath546 set @xmath547 so that the area of @xmath548 differs from that of @xmath549 by at most @xmath550 . then there are two intervals @xmath551 such that @xmath552 , and @xmath553 and so that @xmath554 applying proposition [ expanding ] , which approximates the above cardinalities by @xmath555 ( recall @xmath556 ) , and dividing by @xmath557 we obtain @xmath558 letting @xmath559 , we get @xmath560 with @xmath521 is as in , and @xmath561 a piecewise smooth , continuous function of @xmath447 . summing over all pairs of circles in @xmath122 , we get @xmath562 the reason there is an extra factor of @xmath563 for @xmath327 disjoint is because @xmath564 only parametrizes half of the gaps formed by the @xmath112-orbits of @xmath327 , and the contribution from @xmath565 , @xmath566 is identical to that from @xmath121 , @xmath364 . let @xmath567 be a mbius transformation . in this section we show that if a circle packing @xmath2 and a circle @xmath12 from @xmath2 satisfy theorem [ equidistribution ] with some constant @xmath21 , and theorem [ gapdistribution ] with some piecewise smooth continuous function @xmath301 , then the packing @xmath568 and @xmath569 also satisfy theorem [ equidistribution ] and theorem [ gapdistribution ] with the same constant and distribution function . in fact we will give a more refined statement : for any pair of distinct circles @xmath370 in the initial configuration , the densities @xmath570 defined in , are conformally invariant . this is particularly useful in reducing the computations of the limiting densities in the examples of [ sec : examples ] to manageable length . the argument itself is routine : the claims are obvious if @xmath178 is a dilation , and are proved for general @xmath37 by localizing . [ transferprinciple ] let @xmath288 be an arc ( if @xmath12 is a line then take a bounded interval ) , and @xmath37 . assume @xmath571 such that @xmath572 for any @xmath573 . i ) for the packing @xmath568 and the base circle @xmath569 , @xmath574 \ii ) for any pair of distinct circles @xmath370 in the initial configuration , the densities @xmath570 are conformally invariant : @xmath575 where on the rhs , @xmath576 refers to the gaps associated with the pair of circles @xmath577 , @xmath578 and the packing @xmath568 with base circle @xmath569 . \iii ) as a consequence , for all @xmath178 and subarcs @xmath288 , @xmath579 we use the notation @xmath580 and @xmath581 throughout the proof related to the asymptotic growth as @xmath582 . without further explanation , all the implied constants depend at most on @xmath178 and @xmath14 . \i ) we first show that for a circle @xmath76 with tangency @xmath583 on @xmath14 , the curvature of @xmath584 satisfies @xmath585 choose a parametrization @xmath586 for @xmath76 in a neighborhood of @xmath583 , with @xmath587 . then the curvature is given by @xmath588 therefore , a direct computation shows that @xmath589 which gives . + we divide the arc @xmath590 on @xmath569 into @xmath591 equal pieces @xmath592 with their preimages @xmath593 on @xmath12 . we pick a point @xmath143 from the interval @xmath594 for each @xmath595 , and let @xmath596 . we have @xmath597 for any circle @xmath76 on @xmath598 , we have @xmath599 therefore , @xmath600 such that @xmath601 dividing the above expression by @xmath18 , and letting @xmath159 , we obtain @xmath602 replace @xmath603 by , sum over all @xmath595 , and let @xmath604 , we obtain @xmath605 thus we prove the conformal invariance of the density of tangencies . + ii ) now we show that for each pair of circles @xmath606 , @xmath607 let @xmath608 be an ordered sequence of @xmath609 , and @xmath610 be the sequence of @xmath611 ordered in a way such that @xmath612 has the same orientation as @xmath608 . let @xmath613 be the subset of @xmath614 consisting of those @xmath615 such that @xmath616 . let @xmath617 be the subset of @xmath609 consisting of those @xmath618 such that @xmath619 . then @xmath620 from , the symmetric difference @xmath621 therefore , if we let @xmath622 and @xmath623 then @xmath624 and @xmath625 let @xmath626 and @xmath627 be the ordered sequence from @xmath628 and @xmath629 respectively . then @xmath178 is a bijection on these two sequences . let s say @xmath630 . therefore , @xmath631 and @xmath632 now @xmath633 @xmath634 @xmath635 combining , , , we see that @xmath636 which only depend on @xmath178 and @xmath14 , and for any arbitrary small number @xmath637 , @xmath638 , such that when @xmath639 , @xmath640 we also have @xmath641 therefore , @xmath642 , @xmath643 as a result , as @xmath159 , we obtain @xmath644 since @xmath645 is a convex combination of @xmath646 , if we let @xmath647 , @xmath648 and use the continuity of @xmath649 ( as follows from ) , we obtain @xmath650 any bounded circle packing can be written as @xmath568 for some generalized ford packing @xmath2 . the only issue we need to deal with is @xmath7 , if an arc @xmath276 in @xmath569 contain @xmath651 , the argument in the proof of theorem [ transferprinciple ] does not directly apply . however , this is easily solved by precomposing @xmath178 with some @xmath652 , so that @xmath653 is a bounded arc , and @xmath654 for some @xmath655 and any @xmath656 . then we can apply theorem [ transferprinciple ] . we compute the gap distribution function @xmath41 for three examples : the classical apollonian packing , where we start with a configuration of four mutually tangent circles , whose tangency graph is the tetrahedron ; the apollonian-3 packing in which one starts with three mutually tangent circles , and in either of the curvilinear triangles formed by three mutually tangent circles we pack three more circles , forming a sextuple whose tangency graph is the octahedron ; and the apollonian-9 packing introduced in @xcite where the initial configuration consists of 12 circles , with the icosahedron as its tangency graph . we explain the procedure used in the computations : according to theorem [ thm : formula for f ] , the gap distribution @xmath41 is given in terms of a sum of functions @xmath657 over ( unordered ) pairs of distinct circles @xmath658 of areas of regions @xmath659 in the @xmath376 plane , explicitly described in [ sec : determining ] . so for instance in the case of the apollonian-9 packing , there are @xmath660 such pairs . we are able to cut down on the computational effort involved by using conformal invariance , once we note that the packings we study below enjoy more symmetries than a generic packing . according to theorem [ transferprinciple ] , if there is some @xmath37 which preserves @xmath17 and @xmath12 ( note that @xmath661 only depends on @xmath17 , not on all of @xmath2 ) , and takes the pair of circles @xmath662 in @xmath122 to another pair @xmath663 in @xmath122 , then @xmath664 . in each of these packings , there exists a group @xmath665 which is larger than @xmath44 that also acts on this given packing . in each of the three examples presented , the bigger group @xmath666 is one of the hecke triangle groups @xmath667 , which is the group of mbius transformations generated by the inversion @xmath668 and the translation @xmath669 , where @xmath670 : @xmath671 for @xmath672 we recover the modular group : @xmath673 . for @xmath674 the groups @xmath667 are non - arithmetic . in all three cases , we show that @xmath675 is a normal subgroup of the appropriate @xmath667 . in the first example of the classical ford packing , @xmath676 which acts transitively on the ford circles @xmath17 . in the second example , of the apollonian-3 packing , @xmath677 . in the third example , the apollonian-9 packing , @xmath678 . it is a simple check that in each of these cases , the generators @xmath679 send circles in @xmath99 to some circles in @xmath17 . so if we show @xmath679 normalizes @xmath44 , then it follows that @xmath679 permutes the circles in @xmath680 . we further show that @xmath667 acts transitively on pairs of tangent circles in @xmath99 for @xmath681 , and on pairs of disjoint circles for @xmath682 . hence by conformal invariance of the components @xmath657 , this gives an expression for @xmath41 as a sum of areas of two regions for @xmath682 ( only one for @xmath672 ) . these regions are in turn expressed as a finite union of certain subregions @xmath683 . for each subregion @xmath684 , the first two equations give the conditions @xmath685 , and the other equations except the last one gives the condition @xmath686 , where @xmath76 exhausts circles in @xmath122 such that @xmath687 is in between @xmath412 and @xmath413 . the last quadratic equation corresponds to conditions or , capturing relative gap information . the areas of these regions can be explicitly computed in elementary terms , but the resulting formulae are too long to record . instead we display plots of the density of the normalized gaps ( the derivative of @xmath301 ) . + the resulting packing @xmath2 is a classical apollonian packing , generated by inscribing a unique circle into every curvilinear triangle produced by the four mutually tangent circles and repeating . the circles @xmath17 tangent to the real line are precisely the classical ford circles @xmath689 , having as points of tangency the farey sequence @xmath690 . the gap distribution of the farey sequence was found by hall @xcite , and we carry out the computation here as a warm - up to illustrate our method . the group @xmath112 is generated by the three reflections @xmath693 so @xmath44 generated by @xmath694 and @xmath695 , which are the generators of the principal congruence subgroup @xmath696 which is a normal subgroup of @xmath697 . any pair of circles from @xmath698 are _ conformally equivalent _ , in the sense that there exists some @xmath699 which preserves the packing and maps @xmath172 , @xmath173 to this given pair of circles : indeed , @xmath700 therefore we have @xmath701 . and so by theorem [ thm : formula for f ] , @xmath702 and it remains to compute the area of @xmath703 . we compute the region @xmath704 using , , . a direct computation gives @xmath705 the distance between @xmath706 and @xmath707 is @xmath708 . when @xmath709 , @xmath710 will be mapped by @xmath362 to some circle in between @xmath711 and @xmath712 by observation [ inbetween ] , and the corresponding region is @xmath713 note that @xmath717 and @xmath718 are symmetric about the @xmath419-axis , so we only need to calculate one area . therefore , by theorem [ thm : formula for f ] , @xmath719 the density function @xmath720 of the normalized gaps ( the derivative of @xmath41 ) is given in figure [ fig : classicapp3density ] . our next example was discovered by guettler and mallows @xcite , in which one starts with three mutually tangent circles , and in either of the curvilinear triangles formed by three mutually tangent circles we pack three more circles , forming a sextuple @xmath0 , see figure [ ap3ford ] . the base circle is @xmath42 and the circles of @xmath0 tangent to it are @xmath721 to see that @xmath44 is normal in @xmath726 , we check that the conjugates of the generators of @xmath44 by the generators @xmath201 , @xmath730 of @xmath726 still lie in @xmath44 : @xmath731 @xmath732 @xmath733 @xmath734 @xmath735 there are @xmath736 ( unordered ) pairs of circles from @xmath122 : four pairs of tangent circles @xmath737 , @xmath738 , @xmath739 , @xmath740 , and two pairs of disjoint circles @xmath741 . each of these cases gives one equivalence class under the action of @xmath726 : @xmath742 and @xmath743 hence from formula @xmath744 first we consider @xmath745 : the distance between @xmath749 and @xmath750 is @xmath708 . therefore , we need to consider the cases @xmath709 and @xmath714 separately . if @xmath709 , then the circles @xmath710 and @xmath751 will be mapped by @xmath362 to some circles in between @xmath711 and @xmath712 . the corresponding region @xmath752 is @xmath753 if @xmath714 , then the circles @xmath754 will be mapped in between @xmath711 and @xmath712 . the corresponding region @xmath755 is @xmath756 the region @xmath704 is the union of @xmath717 and @xmath718 which are symmetric , hence the area is @xmath757 next we consider @xmath758 . here @xmath759 and we need to split it into two cases . if @xmath760 , then @xmath761 will be in between @xmath762 and @xmath763 , and the corresponding region is @xmath764 if @xmath765 , then @xmath766 will be in between @xmath762 and @xmath763 , and , @xmath767 then @xmath758 is the union of @xmath768 and @xmath769 ( which have equal areas ) , and @xmath770 thus @xmath771 the density function @xmath720 of the normalized gaps ( the derivative of @xmath41 ) is given in figure [ fig : app3density ] . this packing was introduced in @xcite . the initial configuration @xmath0 , whose tangency graph is the icosahedron , is shown in figure [ ap9ford ] . the base circle is @xmath42 and the circles @xmath772 tangent to it are @xmath773 the constant @xmath21 in this case is @xmath774 . we claim that @xmath779 , the hecke-5 group . which is generated by @xmath780 we check that each generator of @xmath44 can be written as word of @xmath201 and @xmath781 and hence lies in @xmath50 : clearly , @xmath782 moreover @xmath783 and @xmath784 as claimed . since @xmath44 is generated by elements of @xmath50 , it is a subgroup of that group , and since both groups have finite volume , @xmath44 is a subgroup of finite index in @xmath50 , and in particular is non - arithmetic . to see that @xmath44 is normal in @xmath50 , we check that the conjugates of the generators of @xmath44 by the generators @xmath201 , @xmath781 of @xmath50 still lie in @xmath44 : @xmath785 @xmath786 @xmath787 @xmath788 @xmath789 and hence @xmath790 @xmath791 the pairs @xmath737 , @xmath792 are representatives of two conformally equivalent classes of pairs of circles from @xmath122 , characterized by the two circles being tangent or disjoint and @xmath50 acts transitively on each class : indeed , @xmath793 and @xmath794 each equivalence class contains 5 pairs of circles . therefore , @xmath795 we first compute @xmath704 . if @xmath709 , then by observation [ inbetween ] , the circles @xmath800 will mapped by @xmath362 in between @xmath711 and @xmath712 . the corresponding region is , using , @xmath801 the condition @xmath802 is redundant here . if @xmath714 , then @xmath803 will be mapped in between @xmath711 and @xmath712 . the corresponding region is , by , @xmath804 the region @xmath704 is the union of @xmath752 and @xmath755 , which are symmetric . similarly , if @xmath808 , then @xmath766 will be in between @xmath762 and @xmath763 , and the region is @xmath809 the region @xmath758 is the union of @xmath810 and @xmath811 . therefore , @xmath812 the density function of the normalized gaps is given in figure [ fig : app9density ] . j. marklof . _ fine - scale statistics for the multidimensional farey sequence_. limit theorems in probability , statistics and number theory , 4957 , springer proc . 42 , springer , heidelberg , 2013 .